Human resource management training of supervisors for improving health and well‐being of employees

Abstract

Background

Many workers suffer from work‐related stress and are at increased risk of work‐related cardiovascular, musculoskeletal, or mental disorders. In the European Union the prevalence of work‐related stress was estimated at about 22%. There is consensus that stress, absenteeism, and well‐being of employees can be influenced by leadership behaviour. Existing reviews predominantly included cross‐sectional and non‐experimental studies, which have limited informative value in deducing causal relationships between leadership interventions and health outcomes.

Objectives

To assess the effect of four types of human resource management (HRM) training for supervisors on employees' psychomental stress, absenteeism, and well‐being. We included training aimed at improving supervisor‐employee interaction, either off‐the‐job or on‐the‐job training, and training aimed at improving supervisors' capability of designing the work environment, either off‐the‐job or on‐the‐job training.

Search methods

In May 2019 we searched CENTRAL, MEDLINE, four other databases, and most relevant trials registers (ICTRP, TroPHI, ClinicalTrials.gov). We did not impose any language restrictions on the searches.

Selection criteria

We included randomised controlled trials (RCT), cluster‐randomised controlled trials (cRCT), and controlled before‐after studies (CBA) with at least two intervention and control sites, which examined the effects of supervisor training on psychomental stress, absenteeism, and well‐being of employees within natural settings of organisations by means of validated measures.

Data collection and analysis

At least two authors independently screened abstracts and full texts, extracted data and assessed the risk of bias of included studies. We analysed study data from intervention and control groups with respect to different comparisons, outcomes, follow‐up time, study designs, and intervention types. We pooled study results by use of standardised mean differences (SMD) with 95% confidence intervals when possible. We assessed the quality of evidence for each outcome using the GRADE approach.

Main results

We included 25 studies of which 4 are awaiting assessment. The 21 studies that could be analysed were 1 RCT, 14 cRCTs and 6 CBAs with a total of at least 3479 employees in intervention and control groups. We judged 12 studies to have an unclear risk of bias and the remaining nine studies to have a high risk of bias. Sixteen studies focused on improving supervisor‐employee interaction, whereas five studies aimed at improving the design of working environments by means of supervisor training.

Training versus no intervention

We found very low‐quality evidence that supervisor training does not reduce employees' stress levels (6 studies) or absenteeism (1 study) when compared to no intervention, regardless of intervention type or follow‐up. We found inconsistent, very low‐quality evidence that supervisor training aimed at employee interaction may (2 studies) or may not (7 studies) improve employees' well‐being when compared to no intervention. Effects from two studies were not estimable due to missing data.

Training versus placebo

We found moderate‐quality evidence (2 studies) that supervisor training off the job aimed at employee interaction does not reduce employees' stress levels more than a placebo training at mid‐term follow‐up. We found low‐quality evidence in one study that supervisor training on the job aimed at employee interaction does not reduce employees' absenteeism more than placebo training at long‐term follow‐up. Effects from one study were not estimable due to insufficient data.

Training versus other training

One study compared the effects of supervisor training off the job aimed at employee interaction on employees' stress levels to training off the job aimed at working conditions at long‐term follow‐up but due to insufficient data, effects were not estimable.

Authors' conclusions

Based on a small and heterogeneous sample of controlled intervention studies and in contrast to prevailing consensus that supervisor behaviour influences employees' health and well‐being, we found inconsistent evidence that supervisor training may or may not improve employees' well‐being when compared to no intervention. For all other types of interventions and outcomes, there was no evidence of a considerable effect. However, due to the very low‐ to moderate‐quality of the evidence base, clear conclusions are currently unwarranted. Well‐designed studies are needed to clarify effects of supervisor training on employees' stress, absenteeism, and well‐being.

Plain language summary

Effects of training supervisors on employees' stress, absenteeism, and well‐being

Background

Supervisors are assumed to play a crucial role in creating the working conditions of their employees and thus promoting their health and well‐being. Training programmes are widely used to improve supervisors' skills in improving health and well‐being of their employees.

Review question

We examined the effects of four types of training for supervisors on employees' stress, absenteeism, and mental well‐being. We assessed the effects of training aiming to improve interaction between supervisors and employees, either off‐the‐job or on‐the‐job.We also assessed training aimed at improving supervisors' capabilities to design the work environment, either off‐the‐job or on‐the‐job.

Study characteristics

We included 25 studies of which 4 studies are awaiting assessment. The 21 studies that could be analysed included a total of 3479 employees. Sixteen studies trained supervisor‐employee interaction, either off‐the‐job (9 studies) or on‐the job (7 studies). Five studies trained the design of working environments, off‐the‐job in 2 studies and on‐the job in 3 studies. The 21 studies compared 23 interventions with no training, sham training or other training at various times of follow‐up.

Key results

There is no considerable effect of supervisor training on employees' stress (6 studies) or absenteeism (1 study) when compared to no training. There is inconsistent evidence that supervisor training may (2 studies) or may not (7 studies) improve employees' well‐being when compared to no training. Data were missing from two studies, so we could not calculate the effects of training on employee well‐being.

There is no effect of supervisor training on employees' stress (2 studies) or absenteeism (1 study) when compared to a placebo training. Data were missing from one study, so we could not calculate the effects of training on employee well‐being.

One study that evaluated supervisor training compared to another type of training to reduce employees' stress did not provide enough data to calculate its effects.

Quality of the evidence

The quality of the evidence was very low for most outcomes due to risk of bias in the studies, inconsistent results, and imprecise effects. Researchers should consider the shortcomings of studies included in this review in order to conduct well‐designed studies in the future and report them appropriately.

Conclusions

Overall, the data suggest that training of supervisors may not lead to reduced levels of stress and absenteeism, or improved levels of well‐being in their employees. The discrepancy between the apparent scientific consensus and the empirical evidence might be attributed to weak study designs.High quality studies are needed to clarify if supervisor training affects employees' stress, absenteeism, and well‐being.

Background

Description of the condition

The workplace and especially the psychosocial work environment are important determinants of the health and well‐being of employees (Brunner 2006; Joyce 2010; Marmot 2006; Marmot 2012). Trends such as increased work pace, more highly skilled jobs, and the increased use of information and telecommunication technology have been placing increasingly higher demands on the mental functions of employees (European Agency for Safety and Health at Work 2007; Nieuwenhuijsen 2010). Psychosocial risk factors in the working environment are associated with higher levels of work stress, which is reported to be experienced by about 22% of European workers (Houtman 2005). In addition, an online‐survey conducted in 2001 in more than 30,000 workers worldwide revealed that globally 26% (range 17% to 35%) of workers felt to be under unreasonable work stress (D'Mello 2012).

To our knowledge, there are no generalisable data on rates of sick leave that could be solely and specifically attributed to work‐related psychomental stress or consecutive stress‐related symptoms. Likewise, it should be taken into consideration that the total rate of absence from work is not always due to illness alone, that is, justifiable sick leave, but also due to unjustifiable absenteeism, which can be defined as the practice of regularly staying away from work without good reason (Darr 2008; Oxford Dictionaries 2019). Therefore, the overall rate of absence from work should be seen as a combined measure influenced by a mixture of medical, psychological and social factors, as well as individual work attitudes like withdrawal, commitment, work‐engagement or job satisfaction (Darr 2008; Johns 2015). A survey conducted by the European Union (EU) in 2007 revealed that 23 million people, representing 8.1% of current or former workers reported having had a work‐related health problem within the preceding year (European Commission 2010; European Commission 2011). It was estimated that these problems resulted in at least 367 million calendar days of sick leave (European Commission 2010; European Commission 2011). The overall rate of sickness absence ranged from about 3.5% to 8.0% within the EU during the 1990s (Eurofound 1997), whereas in 2011 it was 4.7% in Germany (WIdO 2012), 1.8% in the UK (Office for National Statistics 2012), and 3.0% in the USA (Bureau of Labor Statistics 2012). The economic burden caused by absence from work (due to all causes) in the EU countries was estimated between 1.5% and 4% of the gross domestic product (GDP) in the 1990s (Eurofound 1997; Livanos 2010).

After musculoskeletal problems, mental (stress‐related) disorders such as burnout, anxiety and depression (14%) were the second most frequent work‐related health problems in this survey (European Commission 2010; European Commission 2011). In Germany, all mental diseases taken together accounted for about 10% of all sick‐leave days (in the year 2011) but may also include non stress‐related cases, such as personality disorders and substance abuse (WIdO 2012). This number has increased by nearly 60% since the year 2000 in Germany (WIdO 2012). In particular, it should be mentioned that the rate of sick leave due to burnout syndrome has increased more than 10‐fold in Germany since 2004 (WIdO 2012). In 2011, burnout syndrome caused more than 94 days of sick leave in 1000 people (WIdO 2012).

One common path to prevent or at least alleviate detrimental consequences of work stress is to improve individual coping with stress. While successful coping with work stress may have positive effects on performance and quality of life, unsuccessful coping may in the long run lead to coronary heart disease, musculoskeletal problems, adverse health‐related behaviour (e.g. smoking, substance abuse), impaired mental health (e.g. anxiety, burnout, depression) or other stress‐related symptoms (Fransson 2012; Heikkilä 2012a; Kivimäki 2012; Kuper 2002; Lang 2012; Siegrist 2006; Stansfeld 2006). Importantly, feeling stressed should not only be seen as a risk factor or intermediate variable alone, but also as a condition of reduced quality of life in itself.

Leadership by supervisors is traditionally regarded as another common path to prevent work stress. Moreover, leadership is considered a crucial factor for company success and performance. Many empirical studies have established positive relationships between supervisor behaviour and the motivation and performance of subordinated employees. A meta‐analysis has shown positive associations between transformational leadership and outcomes such as job satisfaction and work motivation (Judge 2004). A more recent review provides evidence for positive associations between leader behaviour and performance indicators at the level of organisation, team and individual (Gang Wang 2011).

Against the background of rising rates of mental illness and diseases such as burnout and depression within the workforces in post‐industrialised nations, leaders are increasingly aware of their extended responsibility to care for the safety and health of their subordinated employees. Several reports identify supervisor behaviour as an important risk factor for employees' health problems, including job strain, burnout, or depression, and substantiate their responsibility (Kuoppala 2008; Skakon 2010). Training interventions for supervisors to prevent health impairment of employees have grown in popularity in modern business organisations. Several pathways for leadership to influence employees’ health have been described, for example, by designing healthy workplaces and work tasks, or improving interactions between supervisors and subordinates (Wegge 2014).

In summary, a large evidence base shows clear associations between characteristics of the psychosocial work environment and employees' stress, absenteeism and psychomental well‐being. Against this background, supervisors are assumed to play a crucial role in both designing subordinates' working conditions and promoting supportive communication with employees. Human resource management (HRM) training programmes are widely used to improve supervisors' skills to take responsibility for the health and well‐being of their subordinates. However, scientific evidence on the effects of such leadership interventions on employees' psychomental health is lacking.

Description of the intervention

This review included studies on human resource management (HRM) training programmes that aim to enhance the knowledge, the attitude, the skills and the behaviour of supervisors. HRM training programmes are widely used to improve leadership in organisations (Yukl 2012). Training can take many forms, from self‐help activities to developmental activities (e.g. 360° feedback, coaching), from short workshops to programmes that last for a year or more, from programmes focused on skills needed in the current position to training preparing managers for promotion to higher positions, or from programmes tailored to a company's needs to workshops imparting generic skills (Yukl 2012). We classified the interventions relevant for this review based on the following two dimensions (Table 4).

1. Classification of interventions.

		Training
		Off‐the‐joba	On‐the‐joba
		Self‐help activities (textbook‐based, video‐based, web‐based) Formal training (face‐to‐face lecture, classroom lecture), case analysis and discussion Simulations (role playing, behavioural role modelling) Developmental assessment centres	Real time/things/situations Executive coaching Mentoring (360°) Feedback Job rotations Action learning
Supervisor‐Employee Interaction	Communication (e.g. active listening, clear instructions, role clarity, clear and ethical organisational and personal goals, etc.) Justice (interactional) Recognition and immaterial reward (appreciation, respect, etc.) Supervisory support (emotional) Leadership style	A1 e.g. web‐based teaching of how to improve active listening, role‐playing simulation for improving active listening skills	A2 e.g. personal coaching concerning active listening while working
Design of working environmentb	Participation Justice (procedural and distributive, including material reward) Autonomy (e.g. working time/schedule) Decision latitude and control Supervisory support (instrumental)	B1 e.g. classroom lecture for improving knowledge on positive health effect of participative leadership, case‐analysis simulation for improving participative leadership behaviour	B2 e.g. 360°‐feedback session for assessing and improving the leader's participative leadership behaviour

^aFor example, combined interventions containing both, an introduction to active listening skills using a video‐based lecture (A1) followed by personal coaching sessions (A2) are classified as A2. For another example, a short video‐based introduction on active listening (A1) followed by an considerably longer role‐playing session on participative leadership (B1) is classified as B1.
 ^bOnly included if these conditions could (and should) be influenced directly by the targeted supervisor.

1. First, we distinguished between HRM training programmes

   1. aiming to improve the dyadic interaction (e.g. relationship, communication) between supervisors and employees, and

   2. aiming to improve the capabilities of supervisors to change the psycho‐social working environment in a health‐promoting way.

2. Second, we classified interventions according to their proximity to real life by distinguishing between

   1. training off‐the‐job such as in simulated conditions and

   2. training on‐the‐job while working with real people on real problems

(Adapted from Australian Public Service Commission 2011 and Nohria 2010.)

The combination of both dimensions results in four categories of interventions (Table 4). The table presents practical examples of what HRM training programmes may look like.

How the intervention might work

The linkage between HRM training for supervisors and health outcomes of employees is rather complex, often indirect and influenced by a plethora of contextual factors (Figure 1). According to Brunner 2006, employee health is related to the social structure and work environment via several psychological factors and health‐related behaviours. Thus, HRM training programmes might work by influencing the following two pathways.

1.

Logic model of how HRM training of supervisors may lead to positive changes in employees

1. Reduction of work‐related stressors. Important work‐related psychosocial risk factors and possible sources of work‐related stress (work stressors) are listed in Table 5. Supervisors may influence some of these stressors, but these also depend on the organisational, structural, cultural and other environmental conditions. For example, 360°‐feedback sessions could identify such work‐related stressors as communication deficits between supervisors and employees that manifest as unclear or contradictory instructions. These deficits may be worked through during personal coaching sessions for supervisors.

2. Promotion of work‐related resources. As a dimension of mental health, well‐being has been defined as a state in which every individual realises his or her own potential, can cope with the normal stresses of life, can work productively and fruitfully, and is able to make a contribution to her or his community (WHO 2005). Therefore, it is crucial to encourage positive psychological or salutogenic approaches (Diener 1999; Ryff 1996; WHO 2005), for example, by promoting work‐related resources as listed in Table 5. In the absence of any work stressors, work‐related resources themselves are unlikely to significantly influence health status, but when employees are exposed to stressful situations resources may improve the capability of employees to successfully cope with stress. For example, a role‐playing simulation that aims to improve the ability of active listening and giving social support to the employee may lead to improved considerative and supportive behaviour of the supervisor. In turn, the employee's perception of altered supervisor behaviour (being recognised, receiving consideration and support) will act as a resource as described above.

2. Work‐related stressors and resources.

Work‐related stressorsa

Work‐related resourcesa

Bullying, harassment and violence Emotionally distressing human services work Ethical conflicts (e.g. illegitimate task assignments, carrying out task conflicting with personal values) High demand and low control Imbalance between effort and reward Injustice and unfairness Lack of autonomy and poor decision‐making latitude Lack of participation Lack of respect and recognition Monotony Organisational change, job insecurity Poor communication and information Poor prospects for career or personal development Poor social relationships (low social support, lack of role clarity, poor leadership, adverse social behaviour) Time pressure Unclear or ambiguous instructions and role, unclear organisational and personal goals

Autonomy and decision latitude of employee Good supervisor‐employee relationship Holistic assignments/tasks Information/transparency Organisational justice Participation Possibility of personal growth and development, career perspective Respect and recognition Social support (emotional, instrumental)

^aThe list is neither exhaustive nor does it make any claims about completeness. Some factors can be both, a work‐related stressor and a resource, depending on whether they are present/pronounced or absent/less pronounced (Cropanzano 2005; Eurofound 2012; Houtman 2007; Kals 2012; Lohmann‐Haislah 2012; Semmer 2010; Stadler 2003; WHO 2019; WHO Regional Office for Europe 2010).

In addition, supervisors may indirectly influence employees' behaviours as a consequence of coping with work stress (Knoll 2011; Siegrist 2006). Behaviours that can be affected include for example smoking, alcohol use, drug abuse, presenteeism (meaning attending work while ill (Johns 2010)), physical activity, diet, and adherence to regular check‐ups or health‐promotion programmes (Bamberger 2006; Heikkilä 2012a; Heikkilä 2012b; Kelloway 2010).

Furthermore, supervisors' support of work‐place health programmes is a critical determinant of the success of employee‐focused health promotion interventions in organisations (Kelloway 2010). Concurrently, HRM training programmes may also influence several context variables or modifying factors, which in turn influence coping with stress. However, we did not explicitly analyse these complex interactions between HRM training, the behaviour of supervisors and the context in this review. In order to identify the specific effect of HRM training as compared to everything else going on in the complex work environment, we carefully documented intervention components and any relevant contextual information provided.

Why it is important to do this review

As mentioned above, about one quarter of workers who had participated in an EU‐wide and global survey suffered from work‐related stress and are therefore at considerably increased risk for work‐related cardiovascular, musculoskeletal, and mental disorders (Ariens 2001; Belkic 2004; D'Mello 2012; Hoogendoorn 2000; Houtman 2005; Kivimäki 2002; Kivimäki 2012; Lang 2012; Marmot 2006; Stansfeld 2012).

There is consensus that onset of stress, stress consequences and degree of well‐being could be influenced by leadership behaviour (Gregersen 2011; Kuoppala 2008; Nieuwenhuijsen 2010; Nyberg 2005; Skakon 2010). A number of studies have been conducted to analyse leadership behaviours in detail but their predominant purpose was to prove the eligibility of these models to predict work performance or job satisfaction. A part of these studies investigated the impact of leadership training on employee health and well‐being. Recent reviews have been performed in a more or less systematic way but they have included predominantly non‐experimental studies (Gregersen 2011; Kuoppala 2008; Nieuwenhuijsen 2010; Nyberg 2005; Skakon 2010). Two of these reviews contain randomised controlled trials (RCTs) but only one each (Kuoppala 2008; Nyberg 2005). The most recent review of the literature by Tsutsumi 2011 focused on three RCTs and four quasi‐experimental studies. However, the authors did not assess included studies' risk of bias nor did they perform a meta‐analysis of results data. Consequently, a systematic review of the available evidence from experimental studies is warranted and should include a critical appraisal of their risk of bias.

Objectives

To assess the effect of four types of human resource management (HRM) training for supervisors on employees' stress, absenteeism, and psychomental well‐being. We included training aimed at improving supervisor‐employee interaction, either off‐the‐job (type A1) or on‐the‐job (type A2), and training aimed at improving supervisors' capability of designing the work environment, either off‐the‐job (type B1) or on‐the‐job (type B2).

Methods

Criteria for considering studies for this review

Types of studies

We included all eligible RCTs and cluster‐randomised controlled trials (cRCTs). We also included controlled before‐after studies (CBAs). We included CBAs only if they had at least two intervention sites and two control sites and only if the outcome had been measured both before and after the intervention (EPOC 2013b). We excluded all other study designs. We believed that including CBAs substantially increased the applicability of our review and its value for potential users.

We considered studies reported as full text, those published as abstract only, and unpublished data.

Types of participants

We included studies that enrolled any type of supervisors, of any gender and their dependently employed subordinates of any gender. For the purpose of this review a supervisor was defined as a person who has the authority to give instructions to at least one subordinate and is held responsible for their work and actions. We included studies that had been conducted in profit, non‐profit or governmental organisations, that is, in a real working environment. We excluded studies that had been conducted in special settings like unpaid work, work without contract, freelance work, internships, volunteering or any comparable conditions. We included studies conducted in mixed settings such as both paid and unpaid work. We did not apply any further restrictions concerning the study population. We also included all studies conducted only in subgroups or in companies organised other than by line‐management (e.g. by matrix‐management).

Types of interventions

We included studies comparing HRM training of supervisors (as listed in Table 4) with a passive control group, such as a waiting list or no intervention, or with an active control group receiving an alternative intervention. We considered the following two categories of HRM training:

1. off‐the‐job supervisor training (e.g. formal face‐to‐face lectures, simulations, role playing), and

2. on‐the‐job supervisor training (e.g. personal coaching, feedback sessions).

We considered HRM training that aimed to change supervisor‐employee interaction (e.g. communication skills, providing support, transformational leadership behaviour) or the design of the working environment (e.g. justice, participation). We included interventions regardless of the duration of HRM training (Table 4). We excluded studies that had not directly targeted supervisor behaviour but focused on general improvements of work organisation which were not amenable to direct change by the supervisor (e.g. overall working time or general reward system). Further, we excluded studies examining interventions that targeted the health behaviours of employees or supervisors (e.g. stress self‐management programmes).

Types of outcome measures

We included all studies reporting at least one primary outcome of our review as described below. All outcome measurements must have been performed in employees (not in supervisors).

Primary outcomes

We included studies that measured the effectiveness of HRM training on:

1. validated measures of psychomental stress, such as the Maslach Burnout Inventory (Maslach 1996), or the Perceived Stress Scale (Cohen 1983);

2. any estimate of absenteeism;

3. measures of well‐being such as the World Health Organization (WHO) five‐item Well‐Being Index (WHO‐5; Topp 2015), or work‐engagement scales (Psychiatric Research Unit 1998; Schaufeli 2002).

Secondary outcomes

If we identify studies in future updates of this review that measured one or more of the primary outcomes listed above, we will also report results measured as changes in employees' health‐related behaviour such as smoking, diet or substance use.

We excluded outcome data based on non‐validated tools for outcome measurement. To be classified as validated tools in this review, measurement tools (e.g. questionnaires, scales) must have been published in a peer‐reviewed journal, book, or as a test manual together with standard measures of instrument quality.

Search methods for identification of studies

Electronic searches

In April 2018 we conducted systematic literature searches using a broad and sensitivity‐maximising approach. In May 2019, we conducted an update of the systematic literature searches using the identical search strategies in order to identify recent studies. We did not use study design filters, as recommended for reviews in public health and health promotion (Lefebvre 2011). This was in order to obtain a comprehensive summary of published and unpublished evidence and to assess generalisability of findings across different implementations of the intervention. We conducted the literature searches in English‐language databases but we included all eligible studies regardless of the language in which the studies had been published. As far as possible, we involved native speakers for assessing inclusion criteria and data extraction of studies published in languages in which the review authors are not proficient. The literature search included the following databases from their inception to the respective search dates:

1. Cochrane Central Register of Controlled trials (CENTRAL; 2019, Issue 5, Appendix 1) in the Cochrane Library, 12 May 2019;

2. MEDLINE (PubMed, Appendix 2), 1946 to 21 May 2019;

3. PsycINFO (Ovid Appendix 3 EBSCOhost Appendix 4), 1806 to May 2019;

4. Scopus (Elsevier, Appendix 5), 21 May 2019;

5. the bibliographic databases of the National Institute for Occupational Safety and Health (NIOSH, NIOSHTIC and NIOSHTIC‐2, Appendix 6), 21 May 2019;

6. the Health and Safety Executive (HSE, Appendix 6)'s HSELINE Database, 21 May 2019;

7. the International Occupational Safety and Health Information Centre (CIS, Appendix 6) bibliographic database, CISDOC, 21 May 2019.

Searching other resources

In order to identify ongoing and unpublished trials, we searched the following trials registers:

1. the WHO International Clinical Trials Registry Platform (ICTRP) search portal (apps.who.int/trialsearch/) 13 April 2018;

2. the Trials Register of Promoting Health Interventions (TRoPHI; eppi.ioe.ac.uk/webdatabases/Search.aspx) 13 April 2018; and

3. ClinicalTrials.gov (clinicaltrials.gov/) 13 April 2018.

We checked reference lists of all included studies and of relevant review articles for additional references. We did not contact experts in the research field to identify unpublished data, ongoing studies or studies that might have been missed by the literature searches.

Data collection and analysis

Selection of studies

Three review authors (AK, JG, CS) screened titles, keywords and abstracts of retrieved references for inclusion of potentially eligible studies. Two review authors independently screened each reference. We coded references as 'retrieve' (eligible or potentially eligible/unclear) or 'do not retrieve'. We obtained the full‐text publications of all references coded as 'retrieve'. Subsequently, we independently screened the full text to decide on inclusion or exclusion of studies. For excluded studies we recorded the reasons in the table 'Characteristics of excluded studies'. We resolved disagreements by consensus or third party adjudication (DN, ER or EVE). We recorded the selection process in sufficient detail to be able to complete a PRISMA flow diagram (Moher 2009; Figure 2).

2.

PRISMA study flow diagram

Data extraction and management

Three authors (AK, JG, CS) independently extracted data from included studies into a pre‐defined data extraction form. We resolved any disagreements by discussion or third‐party adjudication where needed (DN, ER, EVE or Mrs. Nadine Pfeifer). One review author (AK) transferred data into Cochrane's statistical software, Review Manager 2019. A second review author (CS) spot‐checked that data had been entered correctly by comparing the data entered into Review Manager 2019 with the data from extraction forms.

Assessment of risk of bias in included studies

The review authors independently assessed the risk of bias of the included studies. Three authors (AK, JG, CS) split up all included studies for the first assessment and the remaining authors (DN, ER, EVE) for the second assessment.

We assessed the risk of bias of the different study designs using the following methods.

1. RCTs, by using the Cochrane tool for assessing risk of bias as described in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2017).

2. cRCTs, by using the same tool but amended by adding the following sources of bias particular to cRCTs, as described in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011a):

   1. recruitment bias;

   2. baseline imbalance;

   3. loss of clusters;

   4. incorrect statistical analysis;

   5. comparability with individual randomised trials.

3. CBAs, by using the 'Risk of bias' criteria developed by Cochrane Effective Practice and Organisation of Care (EPOC; EPOC 2013a).

We graded each included study in each domain of bias as having a high, low or unclear risk of bias using the GRADEpro GDT software (GRADEpro GDT 2015). We also provided a quote from the study report to justify our judgment. When study authors did not report information that could be quoted but we could infer it, we reported our own justification for the judgment.

At the study level, we considered studies to be at high risk of bias when we judged one or more key domains to be at high risk of bias. In RCTs, we considered random sequence generation, allocation concealment, incomplete outcome data, and selective outcome reporting to be key domains. In CBAs, we considered similarity of baseline outcome measurements, similarity of baseline participant characteristics, adequate addressing of incomplete outcome data, adequate prevention of knowledge of the allocated interventions during the study, adequate protection against contamination, and freedom from selective outcome reporting to be key domains.

We summarised our assessment in both a 'Risk of bias' summary figure (Figure 3) and a 'Risk of bias' graph (Figure 4), as described in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2017). Where information on risk of bias related to unpublished data or correspondence with a study author, we noted this in the 'Risk of bias' tables.

3.

'Risk of bias' summary: review authors' judgements about each 'Risk of bias' item for each included study

4.

'Risk of bias' graph: review authors' judgements about each 'Risk of bias' item presented as percentages across all included studies

Measures of treatment effect

We used mean differences (MDs) or standardised mean differences (SMDs) for continuous outcomes, or other types of data as reported by the authors of the studies. In future updates, we will use risk ratios (RRs) for dichotomous outcomes, if applicable. If the studies reported only effect estimates and their 95% confidence intervals (CIs) or standard errors, we entered these data into Review Manager 2019 using the generic inverse variance method. When the results could not be entered in either way, we described them in the 'Characteristics of included studies' table, or entered the data into additional tables. As recommended by Verbeek 2009, in future updates, we will record time‐to‐event data, such as time to return to work (e.g. measurement of absenteeism) as continuous outcome measures using the means given in the study groups including standard deviations.

Unit of analysis issues

The unit of analysis of studies included in this review was individual employees' levels of stress, absenteeism, or well‐being. Where studies performed cluster‐randomisation and did not consider the design effect in the analyses, we attempted to correct for such clustering, using a fairly large assumed intra‐cluster correlation of 0.10. We considered this assumption to be a realistic estimate based on current best practice with studies about implementation research (Campbell 2001). We followed the methods stated in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011a).

Dealing with missing data

To complete data on interventions or outcomes, or to clarify methodological details, we contacted first or last study authors by email, where contact data were available. In the case that we could not obtain essential information within a reasonable time after contacting the study authors, we recorded the respective 'Risk of bias' domain as 'unclear'. If applicable in future updates of this review, we will assess the effect of missing data on findings in sensitivity analyses. If numerical outcome data, such as standard deviations or correlation coefficients were missing and we could not obtain them from the study authors, we calculated them from other available statistics such as P values, according to the methods described in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011a).

Assessment of heterogeneity

We considered interventions to be similar when we could group them within the same category of HRM training interventions as defined in Table 4. For example, we considered a web‐based lecture aiming to improve active listening to be similar to a face‐to‐face lecture on how to improve supervisory emotional support to employees. In contrast, we considered an individual in‐house coaching session aiming to improve interactional justice to be dissimilar to a role‐playing simulation on how to provide immaterial reward and respect to the employee.

We considered outcome measurements to be similar when they fell into the same outcome category (either measures of psychomental stress, measures of absenteeism or measures of well‐being). Drawing on Naghieh 2013, we distinguished follow‐up times of less than three months (short‐term), three months to one year (medium‐term) and more than one year (long‐term). We used the I² statistic (Higgins 2003), and the Chi² test to assess statistical heterogeneity in results (Deeks 2017).

Assessment of reporting biases

We did not assess reporting bias according to the recommendations given in the Cochrane Handbook for Systematic Reviews of Interventions (Sterne 2017), due to the limited number of included studies. In future updates of this review, we will perform appraisals of publication bias, language bias, and outcome reporting bias. Because there were fewer than five studies in each of the meta‐analyses, we did not assess publication bias by drawing a funnel plot or undertaking Begg's and Egger's tests. To avoid language bias we included studies irrespective of language of publication. To avoid multiple publication bias, we identified multiple reports of the same study, and we linked each publication to a single reference as recommended in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011b). We analysed outcome reporting bias by comparing outcome measures stated in the Methods sections (or other sources such as trials registries, or study protocols) with those in study reports.

Data synthesis

We conducted meta‐analyses of studies that were judged to be sufficiently similar. We judged the following categories to be too different to be combined:

1. comparison group (no intervention, placebo, other training);

2. outcome (stress, absenteeism, well‐being);

3. follow‐up timing (short‐, mid‐, long‐term); and

4. study design (RCT, cRCT, CBA).

We based our conclusions on data from RCTs and cRCTs but also considered results from CBAs as evidence. Given the expected heterogeneity among included studies, we calculated pooled point estimates and 95% CIs using a random‐effects model. We used Review Manager 2019 to generate summary statistics, conduct meta‐analyses, and produce forest plots. Where we considered clinical or statistical heterogeneity to be too large for conducting meta‐analysis (e.g.I² > 75%; Deeks 2017), we performed a narrative synthesis (Petticrew 2013).

We created three 'Summary of findings' tables using the prespecified primary outcome categories, psychomental stress, absenteeism and well‐being.

We used the GRADE approach to assess the quality of the body of evidence for a given primary outcome category, classifying the quality of evidence as high, moderate, low, or very low (Schünemann 2013). We have presented the quality of evidence in three 'Summary of findings' tables, using methods and recommendations described in the Cochrane Handbook for Systematic Reviews of Interventions (Schünemann 2017). We justified all decisions to downgrade the quality of the evidence using the five criteria of study limitations, consistency of effect, imprecision, indirectness and publication bias, or to upgrade the quality of evidence using the three criteria of strength of effect, dose‐response relationship and residual confounding. We added comments to aid readers' understanding of the review where necessary.

Subgroup analysis and investigation of heterogeneity

We conducted subgroup analyses for intervention types (Table 4), that is, training aiming to improve supervisor‐employee interaction, either off‐the job (type A1) or on‐the‐job (type A2) training, or training aimed at supervisors' designing of the work environment, either off‐the job (type B1) or on‐the‐job (type B2) training.

If applicable in future updates of this review, we will conduct the following subgroup analyses.

1. Company size: small and medium‐sized versus larger (> 250 annual work units) enterprises (European Commission 2003). The hierarchical structures and shared values of large enterprises may differ from those in small companies. We assume that the effect of HRM training aiming to change the attitude and behaviour of supervisors may also be influenced by higher‐level managers and corporate culture.

2. Economic sector: primary, secondary, tertiary (person‐orientated services or non‐personal services). Employees' needs (e.g. security of income and employment) and work‐stressor profiles (e.g. physical stress in agriculture or construction versus psychomental stress in human health services) may differ according to the sector of economic activity.

3. Income level of the country in which the study was conducted (high‐income versus low‐ or middle‐income; World Bank 2019). We expect that differences in the ethical, cultural and economic background of the country in which the study was conducted may influence our outcomes of interest. We assume that especially the cultural and economic background would have a large impact on employees' attitudes and needs.

4. Concept of training: theory‐based versus not theory‐based. Theory‐based training interventions provide established psychological explanations for relationships between supervisor behaviour and employee reactions or outcomes (e.g. hypothesised processes of full‐range leadership model or dyadic leadership), whereas interventions not based on theory are merely exploratory.

5. Type of training (Table 4)

Sensitivity analysis

Because all included studies had high or unclear risk of bias, were published, and had first‐degree or mixed supervisor‐employee relationships, we did not perform sensitivity analyses.

If applicable in future updates, we will examine the robustness of pooled estimates and variance by excluding:

1. studies with a high risk of bias;

2. unpublished studies;

3. studies in which outcomes were measured in indirectly subordinated employees (second level or higher)

Results

Description of studies

See: Characteristics of included studies; Characteristics of excluded studies

Results of the search

Flow of studies

In total, electronic database searches found 6591 records (2081 from Scopus, 2138 from PsycINFO, 1979 from MEDLINE, 265 from CISDOC/HSELINE/NIOSHTIC/NIOSHTIC2, and 128 from CENTRAL). Handsearching found an additional 19 references. See Figure 2 for a PRISMA flowchart. After removal of 677 duplicates, three authors (AK, JG, CS) assessed the titles and abstracts of 5914 references for eligibility and identified 68 potentially eligible reports. After assessing the full texts, we excluded 43 reports. Reasons for excluding studies were, for example, absence of a control group (Atwater 2006; Hetherington 1987), absence of relevant outcomes (Clark 1985; Donohoe 1998), and non‐fulfilment of the EPOC criteria for CBAs (Beaton 2001; Mauno 2006). A complete list of reasons for exclusion can be found in the table Characteristics of excluded studies; for a categorisation, see Figure 2. Finally, we included 22 reports from 21 studies (Barling 1996; Barrech 2018; Biggs 2014; Dahinten 2014; Deci 1989; Eastburg 1994; Elo 2005; Gumuseli 2002; Hammer 2011; Hardré 2009; Jeon 2015; Kawakami 2005; Kawakami 2006; Ketelaar 2017; Norman 2003; Odle‐Dusseau 2016; Pomerantz 1992; Romanowska 2011; Scandura 1984; Takao 2006; Weir 1997, see also Characteristics of included studies).

Studies awaiting classification

In our updated search between April 2018 and May 2019 we found another four studies (three cRCTs, one CBA) that met the inclusion criteria. See Characteristics of studies awaiting classification.

Ongoing studies

There are no ongoing studies.

Included studies

We included 21 studies from 22 reports in this review (see Characteristics of included studies). One study was reported in two articles; one in Finnish (Elo 2005), and the other in English (see Elo 2005 secondary reference). We counted both reports as one study and subsequently only refer to Elo 2005 in the following text.

Study design

Of the 21 studies, one was a RCT (Gumuseli 2002), 14 were cRCTs, and six were CBAs (Biggs 2014; Dahinten 2014; Deci 1989; Elo 2005; Odle‐Dusseau 2016; Scandura 1984).

Time period, country, and setting

Eight studies stated the period over which the study took place (Barrech 2018; Gumuseli 2002; Jeon 2015; Kawakami 2005; Kawakami 2006; Ketelaar 2017; Pomerantz 1992; Takao 2006). There was a latency time of up to five years between data collection and publication of the manuscript in these studies. The remaining studies either did not report the study period or it was unclear. Six studies were published before 2000 (Barling 1996; Deci 1989; Eastburg 1994; Pomerantz 1992; Scandura 1984; Weir 1997), seven studies between 2000 and 2010 (Elo 2005; Gumuseli 2002; Hardré 2009; Kawakami 2005; Kawakami 2006; Norman 2003; Takao 2006), and eight studies after 2010 (Barrech 2018; Biggs 2014; Dahinten 2014; Hammer 2011; Jeon 2015; Ketelaar 2017; Odle‐Dusseau 2016; Romanowska 2011).

Ten studies were conducted in the USA or Canada (Barling 1996; Dahinten 2014; Deci 1989; Eastburg 1994; Hammer 2011; Norman 2003; Odle‐Dusseau 2016; Pomerantz 1992; Scandura 1984; Weir 1997), five in Europe (Barrech 2018; Elo 2005; Gumuseli 2002; Ketelaar 2017; Romanowska 2011), three in Japan (Kawakami 2005; Kawakami 2006; Takao 2006), two in Australia (Biggs 2014; Jeon 2015), and one case study did not report its location (Hardré 2009).

According to the three‐sector model (Fourastié 1949), we divided the studies according to their economic area: primary sector (extraction of raw material); secondary sector (manufacturing); and tertiary sector (services). Three studies were carried out in the manufacturing industry (secondary economic sector; Barrech 2018; Deci 1989; Takao 2006), 14 studies in the tertiary economic sector, and one study in a mixed setting (secondary and tertiary; Hardré 2009). The most frequent type of industry/company, with six studies, was healthcare (Barrech 2018; Dahinten 2014; Eastburg 1994; Jeon 2015; Odle‐Dusseau 2016; Weir 1997), followed by the food industry, with three studies (Gumuseli 2002; Hammer 2011; Takao 2006). The other studies were run in various organisations, ranging from a police service (Biggs 2014) to a computer software engineering company (Kawakami 2005). See Characteristics of included studies for further details.

Sample sizes

Nineteen of 21 studies reported the number of participants in data analysis for each study arm. Thus, the studies included at least 1628 employees in the intervention groups and 1717 in the control groups. One study did not state any data on sample size (Barling 1996). Based on the reported number of groups (2) and degrees of freedom rin the statistical analysis (75), we assumed that the total number of employees in both groups was 77 (the denominator degrees of freedom of a univariate analysis of variance are obtained by subtracting the number of groups from the sample size; hence, the sample size can be calculated by adding the denominator degrees of freedom and the number of groups). Neither did another study (Scandura 1984) report group sizes but the number of employees included in the statistical analyses was between 57 and 83. In summary, the included studies analysed a total of at least 3479 employees. The smallest study included 20 employees: eight in the intervention group and 12 in the control group (Gumuseli 2002), whereas in the largest study, the sample size of the intervention and control group was 224 and 303 employees, respectively (Jeon 2015).

Study population/participants

The study population consisted of first‐degree subordinates in 17 studies, seven of which did not state this explicitly but reported sufficient information either in the paper or on request to deduce first‐degree relationships (Barrech 2018; Gumuseli 2002; Hardré 2009; Jeon 2015; Ketelaar 2017; Odle‐Dusseau 2016; Weir 1997). In three studies, supervisor‐subordinate relationship was presumably mixed (first‐ and second‐degree; Hammer 2011; Kawakami 2005; Kawakami 2006). In one study the relationship was most likely first‐degree, but linkage between supervisors and employees was not permanently fixed (Elo 2005). Fourteen studies stated the age of employees and seven did not (Barling 1996; Deci 1989; Eastburg 1994; Gumuseli 2002; Pomerantz 1992; Romanowska 2011; Scandura 1984). Due to differing point estimates (mean, mode) we could not give an overall measure. Eleven studies stated participants’ sex. Percentage of men ranged from about 60% to 92% in seven studies (Barrech 2018; Biggs 2014; Elo 2005; Kawakami 2005; Kawakami 2006; Norman 2003; Takao 2006), and was notably lower in the remaining four studies (Dahinten 2014; Hardré 2009; Ketelaar 2017; Odle‐Dusseau 2016; ranging from 2% to 41%).

Interventions

Sixteen of 21 studies examined interventions aimed at ‘supervisor‐employee interaction’ of which nine were performed off‐the‐job (type A1; Barling 1996; Eastburg 1994; Hammer 2011; Kawakami 2005; Kawakami 2006; Pomerantz 1992; Romanowska 2011; Scandura 1984; Takao 2006), and seven were run on‐the‐job (type A2; Biggs 2014; Deci 1989; Elo 2005; Gumuseli 2002; Norman 2003; Jeon 2015; Ketelaar 2017). Only five studies assessed interventions that aimed at the ‘design of the working environment’, of which two were off‐the‐job (type B1; Dahinten 2014; Hardré 2009), and three were on‐the‐job (type B2; Barrech 2018; Odle‐Dusseau 2016; Weir 1997). The concept of training was theory‐based in eight studies (Barling 1996; Barrech 2018; Biggs 2014; Hammer 2011; Hardré 2009; Odle‐Dusseau 2016; Romanowska 2011; Scandura 1984), unclear in three studies (Jeon 2015; Ketelaar 2017; Weir 1997), and not reported in the remaining studies.

Content, duration and timing of intervention, as well as type of delivery, was comparable only between two pairs of studies. Two studies (Kawakami 2005; Kawakami 2006), applied a self‐administered, web‐based training programme to supervisors for three to five hours over four weeks, according to the Japanese guidelines for promoting mental health care in enterprises by the Japanese Ministry of Labour. Two further studies (Hammer 2011; Odle‐Dusseau 2016) delivered a training intervention to supervisors to enhance their family‐supportive behaviour in two different organisational settings. In the other studies, interventions were heterogeneous ranging from one single face‐to‐face meeting for discussion (Eastburg 1994) to comprehensive training programmes lasting several days, including actions‐learning workshops, 360° feedback processes, and coaching sessions (Biggs 2014). There was one, three‐armed study that compared two intervention groups (situational leadership model and micro‐counselling, both type A1) with a control group that did not receive any intervention (Pomerantz 1992). See Characteristics of included studies for further details on interventions.

Comparisons/type of control group

Sixteen studies did not perform any intervention in the control groups. Employees allocated to the control group received a placebo intervention in four studies (Kawakami 2005; Kawakami 2006; Ketelaar 2017; Scandura 1984). Placebo interventions were training sessions for relaxation in two studies (Kawakami 2005; Kawakami 2006), distribution of written information about the participatory approach intervention in one study (Ketelaar 2017), and three 2‐hour sessions of general input, probably lectures, containing general information about job enrichment, performance evaluation, decision making, and communication (Scandura 1984).

In one study, the control group received type B1 training and the intervention group type A1 training (Romanowska 2011).

Outcomes and follow‐up

In total, the included studies reported 57 outcome measurements. Seven studies reported one outcome each (Barling 1996; Dahinten 2014; Deci 1989; Gumuseli 2002; Kawakami 2005; Kawakami 2006; Takao 2006), five studies reported two outcomes (Elo 2005; Hammer 2011; Ketelaar 2017; Pomerantz 1992; Scandura 1984), three studies reported three outcomes (Barrech 2018; Eastburg 1994; Weir 1997), three studies reported four outcomes (Biggs 2014; Odle‐Dusseau 2016; Romanowska 2011), two studies reported six outcomes (Hardré 2009; Jeon 2015), and one study reported seven outcomes (Norman 2003).

One study reported outcomes for both the whole cohort as well as two subgroups defined by degree of compliance with the training programme (compliers versus non‐compliers; Weir 1997). In order to avoid double‐counting, we only considered the outcomes of the total cohort. A second study did not report outcomes for the whole study group but separately for two subgroups that were defined by the baseline values of the outcome measure (high versus low values on the Leader‐member Exchange Scale; Scandura 1984). A third study reported mediation models but omitted total effects on outcomes and corresponding test statistics, so that we could not quantify the results (Odle‐Dusseau 2016). A fourth study reported mid‐ and long‐term follow‐ups for the same intervention (Jeon 2015). By separately analysing mid‐ and long‐term follow‐ups in the comparisons, we avoided double‐counting.

Of the remaining 53 outcome measures, 18 measures related to stress, 32 to well‐being, and three to absenteeism.

Ten studies measured 18 stress outcomes using 10 different questionnaires.

1. Four studies (Eastburg 1994; Elo 2005; Pomerantz 1992; Romanowska 2011), used sub‐scales of the Maslach Burnout Inventory (Maslach 1981; Maslach 1996), in seven instances: emotional exhaustion in five instances; depersonalisation in one instance; personal accomplishment in one instance;

2. Three studies (Kawakami 2005; Kawakami 2006; Takao 2006), used the Brief Job Stress Questionnaire (Shimomitsu 2000), in three instances;

3. Biggs 2014 used the General Health Questionnaire (GHQ‐12; Goldberg 1972), in one instance;

4. Romanowska 2011 used the Hopkins Symptom Checklist (SCL‐90; Lipman 1986), in one instance; the Karolinska Sleep Questionnaire (Akerstedt 2002), in one instance; as well as a composite index of three scales (Maslach 1996; Lipman 1986; Akerstedt 2002).

5. Barrech 2018 used the Giessen Subjective Complaints List (Brähler 1995), in one instance: sub‐scale exhaustion tendency; and the Hospital Anxiety and Depression Scale (Herrmann‐Lingen 2011), in two instances: anxiety and depression;

6. One study (Elo 2005) used an own scale (Elo 2003)

See Characteristics of included studies.

Eleven studies measured 32 well‐being outcomes using 21 different questionnaires.

1. Job satisfaction in nine instances:

   1. Hammer 2011: one instance according to Hackman 1975;

   2. Norman 2003: one instance according to Hackman 1974;

   3. Biggs 2014: one instance according to Warr 1979;

   4. Scandura 1984: two instances according to Hoppock 1935;

   5. Gumuseli 2002: one instance using the Minnesota Job Satisfaction Survey Turkish version;

   6. Deci 1989: one instance using the company‐specific Global Satisfaction Index;

   7. Odle‐Dusseau 2016: one instance using the Global Job Satisfaction Composite (Cammann 1979; Friedman 2000);

   8. Jeon 2015: two instances using the Workforce Dynamics Questionnaire (Nancarrow 2006).

2. Affective Organizational Commitment Scale in three instances:

   1. Odle‐Dusseau 2016: one instance, according to Allen 1990;

   2. Dahinten 2014 one instance according to Meyer 1991;

   3. Norman 2003: one instance according to Meyer 1993.

3. Barling 1996: one instance using the Organizational Commitment Questionnaire (Mowday 1982);

4. Norman 2003: five instances using the Affective Well‐being Scale (Daniels 1997);

5. Hardré 2009: one instance using the Student Engagement Questionnaire (Miserandino 1996);

6. Biggs 2014: one instance using the Utrecht Work Engagement Scale (Schaufeli 2006);

7. Odle‐Dusseau 2016: one instance using employee engagement (Britt 2001);

8. Hardré 2009: five instances using the Workplace Self‐Regulation Questionnaire (adapted after Academic Self‐Regulation Questionnaire; Ryan 1989);

9. Three studies reported turnover intention in three instances:

   1. Hammer 2011: one instance according to Boroff 1997;

   2. Biggs 2014: one instance according to Brough 2004;

   3. Odle‐Dusseau 2016: one instance according to Chapman 1991.

10. Jeon 2015 reported staff turnover rate in two instances (delivered by the organisation).

Two studies measured three absenteeism outcomes using the following methods:

1. Ketelaar 2017 measured duration and frequency of sickness‐related absence from work during the previous six months in two instances;

2. Weir 1997 measured total hours of absence from work during the last 12 months of study in one instance.

Several studies addressed multiple categories of primary outcomes (e.g. well‐being and stress) or a bundle of different indicators for one outcome (e.g. emotional exhaustion and depersonalisation as indicators of burnout). Where applicable, we selected only one indicator from such cases in order to avoid multiple outcomes of the same study being included in the same comparison. In doing so, we chose the most proximal indicator for the outcome category (e.g. emotional exhaustion but not depersonalisation as indicating stress). Hence, from the 21 studies and 53 outcomes included, we selected 23 outcomes for inclusion in our analyses (one outcome from the RCT, 15 outcomes from cRCTs, seven outcomes from CBAs).

The latency period between baseline outcome measurement and intervention was not stated in one study (Pomerantz 1992) and remained unclear in nine studies (Dahinten 2014; Eastburg 1994; Elo 2005; Gumuseli 2002; Kawakami 2005; Kawakami 2006; Romanowska 2011; Takao 2006; Jeon 2015). It was one to two weeks in four studies (Barling 1996; Hardré 2009; Norman 2003; Scandura 1984), one to four months in three studies (Barrech 2018; Biggs 2014; Deci 1989), and seven to nine months in two studies (Hammer 2011; Odle‐Dusseau 2016). Two studies stated that baseline measurement was done immediately before the intervention but did not specify the delays (Ketelaar 2017; Weir 1997).

Follow‐up measurement time point was less than three months (short term according to the protocol) after intervention in five studies (Eastburg 1994; Hammer 2011; Hardré 2009; Odle‐Dusseau 2016; Pomerantz 1992), three months to one year (mid‐term) in thirteen studies (Dahinten 2014; Weir 1997; Barrech 2018; Kawakami 2005; Kawakami 2006; Norman 2003; Gumuseli 2002; Takao 2006; Barling 1996; Deci 1989; Scandura 1984; Biggs 2014; Ketelaar 2017), and more than one year (long‐term) in two studies (Elo 2005; Romanowska 2011). One study reported both mid‐ and long‐term follow‐up measurements (Jeon 2015).

Funding sources

Fourteen studies stated that they had been partially or fully funded. Ten studies were funded by public sources, and two by commercial sources (Deci 1989; Norman 2003). Two studies stated mixed funding by public and commercial sources (Kawakami 2005; Kawakami 2006). In one study, commercial support was likely but not explicitly stated (Gumuseli 2002). Six studies did not state funding (Eastburg 1994; Elo 2005; Hardré 2009; Odle‐Dusseau 2016Pomerantz 1992; Weir 1997).

Excluded studies

See Flow of studies (Results of the search).

Risk of bias in included studies

At study level, we judged all six CBAs to be at high risk of bias (Biggs 2014; Dahinten 2014; Deci 1989; Elo 2005; Odle‐Dusseau 2016; Scandura 1984). Regarding cRCTs, we judged three studies to be at high risk of bias (Barling 1996; Romanowska 2011; Weir 1997), and eleven at unclear risk of bias (Barrech 2018; Eastburg 1994; Hammer 2011; Hardré 2009; Jeon 2015; Kawakami 2005; Kawakami 2006; Ketelaar 2017; Norman 2003; Pomerantz 1992; Takao 2006). We judged the one RCT to be at unclear risk of bias (Gumuseli 2002). For further details see the 'Risk of bias' table (Figure 3) and the 'Risk of bias' graph (Figure 4). We give detailed justifications for 'Risk of bias' judgements in the individual 'Risk of bias' tables for each included study (Characteristics of included studies).

Allocation

Only one cRCT reported an acceptable method of randomisation (random assignment of workplaces using a random‐number table) and we therefore judged this study as low risk of bias (Kawakami 2006). For the RCT and all other cRCT studies, we judged the risk of bias due to sequence generation as unclear. We judged allocation concealment as unclear in the included RCT and all but one included cRCT (Jeon 2015). CBAs were not assessed with these items. For detailed information on 'Risk of bias' justifications, please see the 'Risk of bias' tables for each study (Characteristics of included studies).

Blinding

Regarding blinding of participants and personnel (performance bias) in cRCTs and RCTs, we judged four at high risk of bias (Norman 2003; Pomerantz 1992; Romanowska 2011; Takao 2006), 10 at unclear risk, and one at low risk (Jeon 2015). With respect to blinding of outcome assessment (detection bias) in cRCTs and RCTs, we assessed one study to be at high risk of bias (Takao 2006), 11 at unclear risk and three at low risk of bias (Jeon 2015; Kawakami 2005; Weir 1997). CBAs were not assessed with these items. For detailed information on 'Risk of bias' justifications, please see the 'Risk of bias' tables for each study (Characteristics of included studies).

Incomplete outcome data

We found three cRCTs to be at high risk of bias (Barling 1996; Romanowska 2011; Weir 1997), four at unclear risk (Barrech 2018; Eastburg 1994; Hardré 2009; Pomerantz 1992), and the remaining eight studies (including the RCT) at low risk of attrition bias. CBA studies were not assessed with these items. For detailed information on 'Risk of bias' justifications, please see the 'Risk of bias' tables for each study (Characteristics of included studies).

Selective reporting

Because prespecified analysis plans or published protocols were not available from the studies, bias due to selective reporting remained unclear for all but three cRCTs (Barrech 2018; Jeon 2015; Ketelaar 2017), for which we judged low risk of bias. CBAs were not assessed with these items. For detailed information on risk of bias justifications, please see the risk of bias tables for each study (Characteristics of included studies).

Other potential sources of bias

Six studies published conflict of interest statements (Barrech 2018; Dahinten 2014; Hammer 2011; Jeon 2015; Ketelaar 2017; Romanowska 2011), but did not explicitly state in detail whether any of the authors had potential conflicts of interest. The remaining 15 studies did not include any conflict of interest statement. Additionally, we found high risk of bias due to self‐selection (Biggs 2014; Romanowska 2011), major changes of external factors during the intervention (Deci 1989), unknown baseline measurements and characteristics (Pomerantz 1992), and incomplete reporting of sample sizes and unusually high participation rate (Scandura 1984). For detailed information on 'Risk of bias' justifications, please see the 'Risk of bias' tables for each study (Characteristics of included studies).

Bias pertaining to cluster‐randomised studies (cRCTs)

In addition to the sources of bias discussed above, we assessed risk of bias in the 14 cRCT studies with five additional items. For detailed information on 'Risk of bias' justifications, please see the 'Risk of bias' tables for each study (Characteristics of included studies).

Recruitment bias

We found high risk of recruitment bias in four studies (Barrech 2018; Hammer 2011; Ketelaar 2017; Romanowska 2011), unclear risk in three studies (Norman 2003; Pomerantz 1992; Weir 1997), and low risk in the remaining seven studies.

Baseline imbalance

We found baseline imbalances to be a source of high risk of bias in one study (Pomerantz 1992), of unclear risk in seven studies, and of low risk in six studies (Barrech 2018; Eastburg 1994; Jeon 2015; Ketelaar 2017; Norman 2003; Weir 1997).

Loss of clusters

We judged the risk of bias as a result of loss of clusters as high in two studies (Barrech 2018; Hardré 2009), unclear in three studies (Eastburg 1994; Ketelaar 2017; Pomerantz 1992), and low in the remaining nine studies.

Incorrect statistical analysis

We found a high risk of bias due to incorrect statistical analyses in 11 studies and a low risk in the remaining three studies (Barrech 2018; Jeon 2015; Ketelaar 2017).

Comparability with individually randomised trials

We judged the comparability to RCTs as a source of high risk of bias in one study (Barrech 2018), of unclear risk in eight studies, and of low risk in five studies (Barling 1996; Hardré 2009; Jeon 2015; Ketelaar 2017; Pomerantz 1992).

For detailed information on 'Risk of bias' justifications, please see the 'Risk of bias' tables for each study (Characteristics of included studies).

Bias pertaining to controlled before‐after studies (CBAs)

We assessed risk of bias in the six CBAs with eight items in the following six domains. For detailed information on 'Risk of bias' justifications, please see the 'Risk of bias' tables for each study (Characteristics of included studies).

Allocation bias

For all included CBAs, we judged the generation of the allocation sequence as inadequate and, therefore, judged them as high risk of bias. Concerning the adequate concealment of the allocation, we judged all but one study to be at high risk of bias, while we rated the remaining study as low risk of bias (Odle‐Dusseau 2016).

Baseline imbalance

We found high risk of bias due to dissimilar baseline outcome measurements in one study (Dahinten 2014), unclear risk in two studies (Deci 1989; Odle‐Dusseau 2016), and low risk in three studies. We detected a high risk of bias due to dissimilarities in baseline characteristics in three studies (Dahinten 2014; Elo 2005; Scandura 1984), and unclear risk in the remaining three studies.

Incomplete outcome data

Regarding adequate handling of incomplete outcome data, we rated one study at low risk of bias (Odle‐Dusseau 2016), and the other five studies at high risk.

Blinding

We found the adequate protection of knowledge of the allocated interventions during the study to be a source of high risk of bias in three studies (Biggs 2014; Dahinten 2014; Deci 1989), and of unclear risk in the remaining three studies.

Contamination

Concerning adequate protection against contamination, we judged two studies as high risk of bias (Deci 1989; Odle‐Dusseau 2016), and the remaining four studies as unclear risk.

Selective outcome reporting

We rated five studies at unclear risk of bias due to selective outcome reporting and one study at low risk (Odle‐Dusseau 2016).

Regarding the incidences of high, unclear, and low risk of bias across all items and studies (Figure 4), incorrect statistical analyses showed the highest risk of bias, and loss of clusters showed the lowest risk of bias (both were cRCT items). Concerning RCT items only, participant and personnel blinding showed the highest risk of bias, and incomplete outcome data showed the lowest risk of bias. Among CBA items, adequate generation of the allocation sequence was associated with the highest risk of bias, whereas similar baseline outcome measurements were associated with the lowest risk of bias.

In summary, the 'Risk of bias' assessment illustrates the different vulnerabilities of RCT, cRCT, and CBA designs (Figure 3; Figure 4). For instance, whereas we did not find a high risk of bias regarding any applicable item for the single RCT in this review (Gumuseli 2002), we found 19 occurrences of high risk of bias in items applicable exclusively to the 14 cRCTs and eight additional occurrences of high risk of bias in items applicable to the RCT and cRCTs, all of which where attributable to cRCT studies. Correspondingly, we found 25 occurrences of high risk of bias in items applicable to the six CBAs. For detailed information on 'Risk of bias' justifications, please see the 'Risk of bias' tables for each study (Characteristics of included studies).

Effects of interventions

See: Table 1; Table 2; Table 3

Summary of findings for the main comparison. Training compared to no intervention in supervisors.

Training compared to no intervention in supervisors
Participants: Supervisors and their subordinates (employees) Setting: Predominately healthcare providers, food industry, and other service companies throughout the world. Intervention: Training of supervisors Comparison: No intervention among supervisors
Outcomes	Anticipated absolute effects (95 % CI)		№ of participants (studies)	Quality of the evidence (GRADE)	Comments
	No intervention	Training
Stress Lower score better Short‐term cRCT	‐	SMD 0.05 higher (0.50 lower to 0.60 higher)	157 (2 cRCTs)	⊕⊝⊝⊝ Very lowa,b,c	SMD interpretation: small effect, SMD = 0.2; medium effect, SMD = 0.5; and large effect, SMD = 0.8.
Stress Lower score better Mid‐term cRCT	‐	SMD 0.02 higher (0.22 lower to 0.26 higher)	304 (2 cRCTs)	⊕⊝⊝⊝ Very lowa,b,c
Stress Lower score better Mid‐term CBA	‐	SMD 0.09 lower (0.12 lower to 0.30 higher)	368 (1 CBA)	⊕⊝⊝⊝ Very lowd,e
Stress Lower score better Long‐term CBA	‐	SMD 0.23 higher (0.12 lower to 0.57 higher)	145 (1 CBA)	⊕⊝⊝⊝ Very lowf,g,h
Absenteeism Lower score better Mid‐term cRCT	‐	SMD 0.20 higher (0.11 lower to 0.51 higher)	164 (1 cRCT)	⊕⊝⊝⊝ Very lowa,c
Well‐being Higher score better Short‐term cRCT	‐	SMD 0.14 higher (0.28 lower to 0.57 higher)	337 (2 cRCTs)	⊕⊝⊝⊝ Very lowa,b,c,e
Well‐being Higher score better Short‐term CBA	‐	Not estimable	143 (1 CBA)	⊕⊝⊝⊝ Very lowf,m,n	No usuable outcome data available
Well‐being Higher score better Mid‐term cRCT + RCT On the job/interaction training	‐	Not estimable	77 (1 RCT)	⊕⊝⊝⊝ Very lowa,c	Metaanalysis was not performed due to high heterogeneity (I²=91%) but RCT did not provided usuable outcome data.
Well‐being Higher score better Mid‐term cRCT + RCT Off the job/interaction training	‐	SMD 0.73 higher (0.34 lower to 1.8 higher)	601 (3 RCTs)	⊕⊝⊝⊝ Very lowa,c	Metaanalysis was not performed due to high heterogeneity (I²=91%). Therefore subgroup results only.
Well‐being Higher score better Mid‐term CBA Off the job/interaction training	‐	SMD 0.45 higher (0.14 lower to 1.04 higher)	744 (2 CBAs)	⊕⊝⊝⊝ Very lowb,d	Metaanalysis was not performed due to high heterogeneity (I²=88%). Therfore, subgroup results only.
Well‐being Higher score better Mid‐term CBA On the job/environment training	‐	SMD 0.28 higher (0.13 lower to 0.69 higher)	129 (1 CBA)	⊕⊝⊝⊝ Very lowc,d	Metaanalysis was not performed due to high heterogeneity (I²=88%). Therfore, subgroup results only.
Well‐being Higher score better Long‐term cRCT	‐	SMD 0.12 higher (0.05 lower to 0.29 higher)	547 (1 cRCT)	⊕⊕⊝⊝ Lowc
A1: supervisor on‐the‐job training for supervisor‐employee interaction; A2: supervisor off‐the‐job training for supervisor‐employee interaction; ASRQ: Academic Self‐Regulation Questionnaire; B1: supervisor on‐the‐job training for work environment design; B2: supervisor off‐the‐job training for work environment design; CBA: controlled before‐after study; CI: confidence interval; cRCT: cluster‐randomised controlled trial; MD: mean difference; RCT: randomised controlled trial; SMD: standardised mean difference; WSRQ: Workplace Self‐Regulation Questionnaire
GRADE Working Group grades of evidence High quality: further research is very unlikely to change our confidence in the estimate of effect. Moderate quality: further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate. Low quality: further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate. Very low quality: we are very uncertain about the estimate.

^a Rated down one level because of risk of bias in studies
 ^b Rated down one level because of inconsistency.
 ^c Rated down two levels for imprecision.
 ^d Rated down two levels because of risk of bias in studies
 ^e Rated down one level because of imprecision.

Summary of findings 2. Training compared to placebo in supervisors.

Training compared to placebo in supervisors
Participants: Supervisors and their subordinates Setting: Computer software engineering, office machine sales and service, service industry, and university in Japan, USA, and the Netherlands. Intervention: Training Comparison: Placebo
Outcomes	Anticipated absolute effects (95% CI)		No. of participants (studies)	Quality of the evidence (GRADE)	Comments
	No intervention	Training
Stress Lower score better; Mid‐term; cRCT	‐	SMD 0.07 lower (0.27 lower to 0.14 higher)	357 (2 cRCTs)	⊕⊕⊝⊝ Lowa	SMD interpretation: small effect, SMD = 0.2; medium effect, SMD = 0.5; and large effect, SMD = 0.8.
Absenteeism Lower score better; Mid‐term cRCT	‐	SMD 0.11 lower (0.41 lower to 0.19 higher)	174 (1 cRCT)	⊕⊕⊝⊝ Lowa
Well‐being Higher score better; Mid‐term CBA	‐	Not estimable	57 (1 observational study)	⊕⊝⊝⊝ Very lowa,b	No usuable data were available for the whole cohort.
A1: supervisor on‐the‐job training for supervisor‐employee interaction; A2: supervisor off‐the‐job training for supervisor‐employee interaction; CBA: controlled before‐after study; CI: confidence interval; cRCT: cluster‐randomised controlled trial; MD: mean difference; SD: standard deviation; SMD: standardised mean difference
GRADE Working Group grades of evidence High quality: further research is very unlikely to change our confidence in the estimate of effect. Moderate quality: further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate. Low quality: further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate. Very low quality: we are very uncertain about the estimate.

^a Rated down two levels for imprecision
 ^b Rated down two levels because of risk of bias

Summary of findings 3. Training compared to other training in supervisors.

Training compared to other training in supervisors
Participants: Supervisors and their subordinates Setting: Different professional areas (education, medical care, police, culture, religious service, business, IT, other) in Sweden Intervention: Training Comparison: Other training
Outcomes	Anticipated absolute effects (95% CI)		No. of participants (studies)	Quality of the evidence (GRADE)	Comments
	No intervention	Training
Stress Lower score better; Long‐term cRCT	‐	Not estimable	98 (1 cRCT)	⊕⊝⊝⊝ Very lowa,b	Study did not state quantitative data by control or intervention group.
A1: supervisor on‐the‐job training for supervisor‐employee interaction; CI: confidence interval; cRCT: cluster‐randomised controlled trial; MD: mean difference; SD: standard deviation; SMD: standardised mean difference
GRADE Working Group grades of evidence High quality: further research is very unlikely to change our confidence in the estimate of effect. Moderate quality: further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate. Low quality: further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate. Very low quality: we are very uncertain about the estimate.

^a Rated down two levels because of risk of bias
 ^b Rated down two levels because of imprecision

We grouped studies according to our categorisation (Data synthesis), which resulted in 14 out of 216 potential analyses. For the number of participants in the analyses please see the Summary of findings tables (Table 1; Table 2; Table 3;

1. Training versus no intervention

We found 16 studies that evaluated the effectiveness of training compared to no intervention (Barling 1996; Barrech 2018; Biggs 2014; Dahinten 2014; Deci 1989; Eastburg 1994; Elo 2005; Gumuseli 2002; Hammer 2011; Hardré 2009; Jeon 2015; Norman 2003; Odle‐Dusseau 2016; Pomerantz 1992; Takao 2006; Weir 1997). Two studies contributed two outcomes each (Biggs 2014; Jeon 2015), resulting in 18 comparisons. See Table 1 and corresponding forest plots (10 analyses: Analysis 1.1; Analysis 1.2; Analysis 1.3; Analysis 1.4; Analysis 1.5; Analysis 1.6; Analysis 1.7; Analysis 1.8; Analysis 1.9; Analysis 1.10).

1.1. Analysis.

Comparison 1 Training vs no intervention, Outcome 1 Stress/short‐term/cRCT.

1.2. Analysis.

Comparison 1 Training vs no intervention, Outcome 2 Stress/mid‐term/cRCT.

1.3. Analysis.

Comparison 1 Training vs no intervention, Outcome 3 Stress/mid‐term/CBA.

1.4. Analysis.

Comparison 1 Training vs no intervention, Outcome 4 Stress/long‐term/CBA.

1.5. Analysis.

Comparison 1 Training vs no intervention, Outcome 5 Absenteeism/mid‐term/cRCT.

1.6. Analysis.

Comparison 1 Training vs no intervention, Outcome 6 Well‐being/short‐term/cRCT.

1.7. Analysis.

Comparison 1 Training vs no intervention, Outcome 7 Well‐being/short‐term/CBA.

1.8. Analysis.

Comparison 1 Training vs no intervention, Outcome 8 Well‐being/mid‐term/cRCT + RCT.

1.9. Analysis.

Comparison 1 Training vs no intervention, Outcome 9 Well‐being/mid‐term/CBA.

1.10. Analysis.

Comparison 1 Training vs no intervention, Outcome 10 Well‐being/long‐term/cRCT.

1.1 Stress

1.1.1 Short‐term follow‐up

1.1.1.1 RCT or cRCT

We found very low‐quality evidence, based on two cRCTs (Eastburg 1994; Pomerantz 1992), of no effect of supervisor training to improve supervisor‐employee interaction by means of off‐the‐job training (intervention type A1) on employees' emotional exhaustion when compared to no intervention using a short‐term follow‐up (standardised mean difference (SMD) 0.05, 95% confidence interval (CI) ‐0.50 to 0.60; Analysis 1.1). Heterogeneity (Chi² = 2.87, df = 1 (P = 0.09), I² = 65%) was below the predefined threshold of I² = 75%. We did not find studies on intervention types A2, B1, or B2.

1.1.1.2 CBA

We did not find studies on any intervention type A1, A2, B1, or B2.

1.1.2 Mid‐term follow‐up

1.1.2.1 RCT or cRCT

Regarding intervention type A1, one cRCT (Takao 2006) found very low‐quality evidence of no effect of supervisor training to improve supervisor‐employee interaction by means of off‐the‐job training on employees' psychological distress when compared to no intervention using a mid‐term follow‐up (SMD −0.05, 95% CI −0.31 to 0.22; Analysis 1.2).

Regarding intervention type B2, one cRCT (Barrech 2018) found very low‐quality evidence of no effect of supervisor training to improve supervisors' capability to design the work environment by means of on‐the‐job training on employees' exhaustion tendency when compared to no intervention using a mid‐term follow‐up (SMD 0.23, 95% CI −0.23 to 0.70; Analysis 1.2).

Combining intervention types A1 and B2, we found very low‐quality evidence, based on two cRCTs (Barrech 2018; Takao 2006), of no effect of supervisor training on employees' stress when compared to no intervention using a mid‐term follow‐up (SMD 0.02, 95% CI −0.22 to 0.26; Analysis 1.2). Heterogeneity (Chi² = 1.05, df = 1 (P = 0.31), I² = 5%) was below the predefined threshold of I² = 75%.

We did not find studies on intervention type A2 or B1.

1.1.2.2 CBA

We found very low‐quality evidence, based on one CBA (Biggs 2014), of no effect of supervisor training to improve supervisor‐employee interaction by means of on‐the‐job training (intervention type A2) on employees' psychological strain when compared to no intervention using a mid‐term follow‐up (SMD −0.09, 95% CI −0.12 to 0.30; Analysis 1.3). We did not find studies on intervention types A1, B1, or B2.

1.1.3 Long‐term follow‐up

1.1.3.1 RCT or cRCT

We did not find studies on intervention type A1, A2, B1, or B2.

1.1.3.2 CBA

We found very low‐quality evidence, based on one CBA (Elo 2005), of no effect of supervisor training to improve supervisor‐employee interaction by means of on‐the‐job training (intervention type A2) on employees' emotional exhaustion when compared to no intervention using a long‐term follow‐up (SMD 0.23, 95% CI −0.12 to 0.57; Analysis 1.4). We did not find studies on intervention types A1, B1, or B2.

1.2 Absenteeism

1.2.1 Short‐term follow‐up

1.2.1.1 RCT or cRCT

We did not find studies on any intervention type A1, A2, B1, or B2.

1.2.1.2 CBA

We did not find studies on any intervention type A1, A2, B1, or B2.

1.2.2 Mid‐term follow‐up

1.2.2.1 RCT or cRCT

We found very low‐quality evidence, based on one cRCT (Weir 1997), of no effect of supervisor training to improve supervisors' capability to design the work environment by means of on‐the‐job training (intervention type B2) on employees' total hours of absence during the last 12 months when compared to no intervention using a mid‐term follow‐up (SMD 0.20, 95% CI −0.11 to 0.51; Analysis 1.5). We did not find studies on intervention types A1, A2, or B1.

1.2.2.2 CBA

We did not find studies on any intervention type A1, A2, B1, or B2.

1.2.3 Long‐term follow‐up

1.2.3.1 RCT or cRCT

We did not find studies on any intervention type A1, A2, B1, or B2.

1.2.3.2 CBA

We did not find studies on any intervention type A1, A2, B1, or B2.

1.3 Well‐being

1.3.1 Short‐term follow‐up

1.3.1.1 RCT or cRCT

Regarding intervention type A1, one cRCT (Hammer 2011), found very low‐quality evidence of no effect of supervisor training to improve supervisor‐employee interaction by means of off‐the‐job training on employees' job satisfaction when compared to no intervention using a short‐term follow‐up (SMD −0.05, 95% CI −0.30 to 0.21; Analysis 1.6).

Regarding intervention type B1, one cRCT (Hardré 2009), found low‐quality evidence of no effect of supervisor training to improve supervisors' capability to design the work environment by means of off‐the‐job training on employees' overall autonomous motivation when compared to no intervention using a short‐term follow‐up (SMD 0.39, 95% CI −0.01 to 0.79; Analysis 1.6).

Combining intervention types A1 and B1, we found very low‐quality evidence, based on two cRCTs (Hammer 2011; Hardré 2009), of no effect of supervisor training on employees' well‐being when compared to no intervention using a short‐term follow‐up (SMD 0.14, 95% CI −0.28 to 0.57; Analysis 1.6). Heterogeneity (Chi² = 3.30, df = 1 (P = 0.07), I² = 70%) was below the predefined threshold of I² = 75%.

We did not find studies on intervention types A2 or B2.

1.3.1.2 CBA

We found very low‐quality evidence, based on one CBA (Odle‐Dusseau 2016), which examined the effect of supervisor training to improve supervisors' capability to design the work environment by means of on‐the‐job training (intervention type B2) on employees' job satisfaction and compared it to no intervention using a short‐term follow‐up (Analysis 1.7). Results suggested a statistically significant mediating effect of supervisor training via family‐supportive supervisory behaviours on improved levels of job satisfaction. However, due to insufficient data, effects were not estimable. We did not find studies on intervention types A1, A2, or B1.

1.3.2 Mid‐term follow‐up

1.3.2.1 RCT or cRCT

Regarding intervention type A1, one cRCT (Barling 1996) found very low‐quality evidence on the effect of supervisor training to improve supervisor‐employee interaction by means of off‐the‐job training on employees' organisational commitment and compared it to no intervention using a mid‐term follow‐up (Analysis 1.8). Results suggested statistically significantly higher levels of organisational commitment in employees of trained supervisors compared to employees of untrained supervisors. However, due to insufficient data, effects were not estimable.

Regarding intervention type A2, we found very low‐quality evidence based on three studies. Because the combined results suggested considerable heterogeneity (Chi² = 22.58, df = 2 (P < 0.0001), I² = 91%) above the predefined threshold of I² = 75%, we refrained from a meta‐analysis and report single study results instead. One RCT (Gumuseli 2002), found no effect of supervisor training to improve supervisor‐employee interaction by means of on‐the‐job training on employees' job satisfaction when compared to no intervention using a mid‐term follow‐up (SMD 0.59, 95% CI −0.32 to 1.51; Analysis 1.8). One cRCT (Jeon 2015), found no effect of supervisor training to improve supervisor‐employee interaction by means of on‐the‐job training on employees' job satisfaction when compared to no intervention using a mid‐term follow‐up (SMD 0.05, 95% CI −0.13 to 0.22; Analysis 1.8). A second cRCT (Norman 2003) found improved levels of job satisfaction in employees as a result of supervisor training to improve supervisor‐employee interaction by means of on‐the‐job training when compared to no intervention using a mid‐term follow‐up (SMD 1.61, 95% CI 0.98 to 2.25; Analysis 1.8).

We did not find studies on intervention types B1 or B2.

1.3.2.2 CBA

Regarding intervention type A2, we found two very low‐quality evidence CBAs. Because the combined results suggested considerable heterogeneity (Chi² = 15.92, df = 1 (P < 0.0001), I² = 94%) above the predefined threshold of I² = 75%, we refrained from a meta‐analysis and report single study results instead. One CBA (Biggs 2014) found no effect of supervisor training to improve supervisor‐employee interaction by means of on‐the‐job training on employees' job satisfaction when compared to no intervention using a mid‐term follow‐up (SMD 0.15, 95% CI −0.06 to 0.36; Analysis 1.9). A second CBA (Deci 1989) found improved levels of job satisfaction in employees as a result of supervisor training to improve supervisor‐employee interaction by means of on‐the‐job training when compared to no intervention using a mid‐term follow‐up (SMD 0.75, 95% CI 0.54 to 0.96; Analysis 1.9).

Regarding intervention type B1, one CBA (Dahinten 2014) found very low‐quality evidence of no effect of supervisor training to improve supervisors' capability to design the work environment by means of off‐the‐job training on employees' overall organisational commitment when compared to no intervention using a mid‐term follow‐up (SMD 0.28, 95% CI −0.13 to 0.69; Analysis 1.9).

Because the combined results of all three studies also suggested considerable heterogeneity (Chi² = 16.46, df = 2 (P = 0.0003), I² = 88%) above the predefined threshold of I² = 75%, we refrained from a meta‐analysis.

We did not find studies on intervention types A1 or B2.

1.3.3 Long‐term follow‐up

1.3.3.1 RCT or cRCT

We found low‐quality evidence, based on one cRCT (Jeon 2015), of no effect of supervisor training to improve supervisor‐employee interaction by means of on‐the‐job training (intervention type A2) on employees' job satisfaction when compared to no intervention using a long‐term follow‐up (SMD 0.12, 95% CI −0.05 to 0.29; Analysis 1.10).

We did not find studies on intervention types A1, B1, or B2.

1.3.3.2 CBA

We did not find studies on any intervention type A1, A2, B1, or B2.

2. Training versus placebo

We found four studies that evaluated the effectiveness of training compared to placebo with one outcome each (Kawakami 2005; Kawakami 2006; Ketelaar 2017; Scandura 1984). See Table 2 and corresponding forest plots (Analysis 2.1; Analysis 2.2; Analysis 2.3).

2.1. Analysis.

Comparison 2 Training vs placebo, Outcome 1 Stress/mid‐term/cRCT.

2.2. Analysis.

Comparison 2 Training vs placebo, Outcome 2 Absenteeism/mid‐term/cRCT.

2.3. Analysis.

Comparison 2 Training vs placebo, Outcome 3 Well‐being/mid‐term/CBA.

2.1 Stress

2.1.1 Short‐term follow‐up

2.1.1.1 RCT or cRCT

We did not find studies on any intervention type A1, A2, B1, or B2.

2.1.1.2 CBA

We did not find studies on any intervention type A1, A2, B1, or B2.

2.1.2 Mid‐term follow‐up

2.1.2.1 RCT or cRCT

We found moderate‐quality evidence, based on two cRCTs (Kawakami 2005; Kawakami 2006), of no effect of supervisor training to improve supervisor‐employee interaction by means of off‐the‐job training (intervention type A1) on employees' psychological complaints when compared to a placebo at mid‐term follow up (SMD −0.07, 95% CI −0.27 to 0.14; Analysis 2.1). Heterogeneity (Chi² = 0.01, df = 1 (P = 0.92), I² = 0%) was below the predefined threshold of I² = 75%. We did not find studies on intervention types A2, B1, or B2.

2.1.2.2 CBA

We did not find studies on any intervention type A1, A2, B1, or B2.

2.1.3 Long‐term follow‐up

2.1.3.1 RCT or cRCT

We did not find studies on any intervention type A1, A2, B1, or B2.

2.1.3.2 CBA

We did not find studies on any intervention type A1, A2, B1, or B2.

2.2 Absenteeism

2.2.1 Short‐term follow‐up

2.2.1.1 RCT or cRCT

We did not find studies on any intervention type A1, A2, B1, or B2.

2.2.1.2 CBA

We did not find studies on any intervention type A1, A2, B1, or B2.

2.2.2 Mid‐term follow‐up

2.2.2.1 RCT or cRCT

We found low‐quality evidence, based on one cRCT (Ketelaar 2017), of no effect of supervisor training to improve supervisor‐employee interaction by means of on‐the‐job training (intervention type A2) on employees' total number of work days on sick leave during the previous six months when compared to a placebo using a mid‐term follow‐up (SMD −0.11, 95% CI −0.41 to 0.19; Analysis 2.2).

We did not find studies on intervention types A1, B1, or B2.

2.2.2.2 CBA

We did not find studies on any intervention type A1, A2, B1, or B2.

2.2.3 Long‐term follow‐up

2.2.3.1 RCT or cRCT

We did not find studies on any intervention type A1, A2, B1, or B2.

2.2.3.2 CBA

We did not find studies on any intervention type A1, A2, B1, or B2.

2.3 Well‐being

2.3.1 Short‐term follow‐up

2.3.1.1 RCT or cRCT

We did not find studies on any intervention type A1, A2, B1, or B2.

2.3.1.2 CBA

We did not find studies on any intervention type A1, A2, B1, or B2.

2.3.2 Mid‐term follow‐up

2.3.2.1 RCT or cRCT

We found very low‐quality evidence in one CBA study (Scandura 1984), which examined the effect of supervisor training to improve supervisor‐employee interaction by means of off‐the‐job training (intervention type A1) on employees' job satisfaction and compared it to a placebo using a mid‐term follow‐up (Analysis 2.3). The reported interaction effect of treatment by initial leader‐member exchange (LMX) levels by time suggested more positively improved levels of job satisfaction in the initially low‐LMX group when compared to the initially high‐LMX group. However, due to insufficient data, effects were not estimable. We did not find studies on intervention types A2, B1, or B2.

2.3.2.2 CBA

We did not find studies on any intervention type A1, A2, B1, or B2.

2.3.3 Long‐term follow‐up

2.3.3.1 RCT or cRCT

We did not find studies on any intervention type A1, A2, B1, or B2.

2.3.3.2 CBA

We did not find studies on any intervention type A1, A2, B1, or B2.

3. Training versus other training

We found one study that evaluated the effectiveness of training compared to other training with one outcome (Romanowska 2011). See Table 3 and corresponding forest plot (Analysis 3.1).

3.1. Analysis.

Comparison 3 Training vs other training, Outcome 1 Stress/long‐term/cRCT.

3.1 Stress

3.1.1 Short‐term follow‐up

3.1.1.1 RCT or cRCT

We did not find studies on any intervention type A1, A2, B1, or B2.

3.1.1.2 CBA

We did not find studies on any intervention type A1, A2, B1, or B2.

3.1.2 Mid‐term follow‐up

3.1.2.1 RCT or cRCT

We did not find studies on any intervention type A1, A2, B1, or B2.

3.1.2.2 CBA

We did not find studies on any intervention type A1, A2, B1, or B2.

3.1.3 Long‐term follow‐up

3.1.3.1 RCT or cRCT

We found one very low‐quality evidence cRCT (Romanowska 2011), which examined the effect of supervisor training to improve supervisor‐employee interaction by means of off‐the‐job training (intervention type A1) on employees' emotional exhaustion and compared it to type B1 training using a long‐term follow‐up (Analysis 3.1). Results suggested no effect of training on employees' levels of emotional exhaustion. However, due to insufficient data, effects were not estimable. We did not find studies on intervention types A2, B1, or B2.

3.1.3.2 CBA

We did not find studies on any intervention type A1, A2, B1, or B2.

3.2 Absenteeism

We did not find studies on any intervention type A1, A2, B1, or B2.

3.3 Well‐being

We did not find studies on any intervention type A1, A2, B1, or B2.

Discussion

Summary of main results

We identified 21 studies that evaluated the effectiveness of human resource management training of supervisors in improving employees' psychomental stress, absenteeism, and well‐being. Another four studies are still awaiting classification. From the included studies, we extracted data for 23 outcomes to be included in our comparisons. Due to insufficient data, we could not calculate effect estimates for four comparisons.

Sixteen studies evaluated training versus no intervention (Table 1). There was very low‐certainty evidence that there was no considerable effect of training compared to no intervention on stress levels at short‐term follow‐up (two studies: SMD 0.05, 95% CI −0.50 to 0.60), mid‐term follow‐up (two studies: SMD 0.02, 95% CI −0.22 to 0.26; one study: SMD −0.09, 95% CI −0.12 to 0.30) or at long‐term follow‐up (one study: SMD 0.23, 95% CI −0.12 to 0.57). There was also no considerable effect on absenteeism at mid‐term follow‐up (one study: SMD 0.20, 95% CI −0.11 to 0.51, very low‐certainty evidence) or on well‐being at short‐term follow‐up (two studies: , 95% CI −0.28 to 0.57, very low‐quality evidence; one study with insufficient data; all very‐low quality evidence) mid‐term follow‐up (six studies: SMD 0.59, 95% CI −0.32 to 1.51; SMD 0.05, 95% CI −0.12 to 0.22; SMD 1.61, 95% CI 0.98 to 2.25; SMD 0.15, 95% CI −0.06 to 0.36; SMD 0.75, 95% CI −0.54 to 0.96; SMD 0.28, 95% CI −0.13 to 0.69; one study with insufficient data; all very low‐certainty evidence) or long‐term follow‐up (one study: SMD 0.12, 95%CI −0.05 to 0.29, low‐certainty evidence). There were no considerable differences between types of training on‐ or off‐the‐job.

Four studies evaluated training versus placebo (Table 2). There was moderate‐certainty evidence that there was no considerable effect of training compared to placebo on stress levels at mid‐term follow‐up (two studies: SMD −0.07, 95% CI −0.27 to 0.14). There was also no considerable effect on absenteeism at mid‐term follow‐up (one study: SMD −0.11, 95% CI −0.41 to 0.19, low‐certainty evidence) or on well‐being at mid‐term follow‐up (one study with insufficient data, very low‐certainty evidence). There were no considerable differences between types of training on‐ or off‐the‐job.

One study evaluated training versus other training (Table 3). There was very low‐certainty evidence that there was no considerable effect of training compared to other training on stress levels at long‐term follow‐up (one study with insufficient data). We did not find studies that used absenteeism or well‐being outcomes.

Overall completeness and applicability of evidence

Of 216 possible analyses due to our hierarchical categorisation of studies (Data synthesis), the included studies allowed for 14 analyses. This means that we did not find studies for most possible combinations of comparison group, outcome, follow‐up timing, study design, and intervention type. Furthermore, the number of studies within each analysis was small, with a maximum of two studies per meta‐analysis. This may have also resulted in low statistical power of the analyses.

Only one study used a RCT design, which allows for the strongest inference on causal effects of leadership interventions on employee health and well‐being. This seriously limits the explanatory power of the results. Fourteen studies employed a cRCT design, which is appropriate for the subject under investigation, albeit risk of bias is potentially higher compared to RCT designs. We also included CBAs, mainly because they are most common in natural organisational settings and might give valuable hints for future studies employing more rigorous designs. It is important to consider that the inference on causal effects from CBAs is limited.

Furthermore, the small number of studies may indicate publication bias, such that negative or equivocal findings may not have been considered for publication. While the largely negative findings found in this review seem to speak against this bias, it should be noted that many of the included studies reported on multiple outcomes and in each report, at least one positive finding was reported.

Although the 21 studies were conducted in nine different countries, about half of them originated in North America. The studies were mainly conducted in the tertiary economic sector, although industries varied broadly. Albeit the sample sizes were rather small (14 studies had recruited fewer than 200 employees overall), the hosting companies were of medium to large size in terms of total number of employees (150 to more than 4000). Consequently, the generalisability to the wider working population or to companies of small size is limited.

Quality of the evidence

The many sources of risk of bias found in the studies resulted in the overall low rating of the quality of evidence according to our GRADE approach. Other reasons for downgrading the quality of evidence were inconsistencies (positive and negative effects across comparable studies) and imprecision of effects (due to missing effect measures, wide confidence intervals, or small sample sizes). Among the quality of evidence ratings for the 14 analyses, only one had a moderate rating (based on cRCTs Kawakami 2005; Kawakami 2006), three had a low rating (based on cRCTs Hardré 2009; Jeon 2015; Ketelaar 2017), and the remaining 10 were rated as very low. Since the two positive findings draw upon analyses rated very low regarding quality of evidence, a clear conclusion is currently unwarranted.

The higher prevalence of cRCT and CBA designs compared to the RCT design follows naturally from the object of investigation, namely, field interventions taking place in naturalistic and complex settings within organisations. While a greater number of RCTs is certainly desirable, the difficulties that arise from applying a strict and rigid experimental design to a naturalistic setting with complex interdependencies can be inferred from characteristic methodological problems observed in the included studies. Among the 14 cRCTs, we observed incomplete reporting or incorrect statistical analyses in 11 studies, recruitment bias was likely in four studies, and we judged lack of blinding of participants likely in four studies. Other observed sources of bias concerned attrition bias (three studies), loss of clusters (two studies), detection bias (one study), baseline imbalance (one study) and incomparability with RCTs (one study). In summary, given the number of sources of bias and their likelihood, the question arises whether some studies, although designed as cRCTs (prospective view), can be de facto considered as cRCTs from a retrospective point of view. Furthermore, given the low rate of significant findings among the RCT and cRCTs, the question arises whether this result can be attributed to the nonexistence of effects in reality, or whether a trade‐off between rigidity of design, occurrence of bias, and likelihood of effects in naturalistic settings may be in effect. For example, unbiased randomisation in an organisational setting may be impeded by management restrictions on team eligibility, which may reduce supervisors’ willingness to participate and implement training content in daily work, which in turn may contribute to negative findings.

Regarding CBAs, it is clear that they impose less strict standards on participants and the organisation, which may introduce additional sources of bias and thus increase the likelihood of bias. Further risks of bias included inadequate handling of incomplete outcome data (five studies), dissimilar baseline characteristics (three studies) or baseline outcome measurements (one study), and inadequate prevention against knowledge of the allocated interventions (three studies) or contamination (one study). On the other hand, CBAs may be more readily implemented in organisations and customised to their specific requirements, which may raise the likelihood of being endorsed by involved individuals. This notion is all the more important in today’s fast‐paced work systems, where intervention studies are very easily perceived as a mere additional burden by management, supervisors, and employees alike. In sum, the question remains whether an intervention in a complex organisational setting benefits more from a stricter design with higher requirements imposed on the organisation and the employees, or whether a more simplified design that allows for concessions to the hosting organisation will likely be more successful. Given the trade‐offs between both extremes in terms of internal and external validity, the decision will inevitably vary with the conditions of a specific study. However, the results presented here may give some guidance on which difficulties and sources of bias one is likely to expect if deciding on a specific study design. It is important that researchers be aware of these caveats and consider them carefully.

In this review we mainly relied on self‐reported outcomes, especially concerning stress and well‐being. Furthermore, the included studies employed a large variety of instruments. Self‐report data may increase the likelihood of bias, especially in those studies where blinding of participants was not effective or possible (e.g. because the control group received no intervention).

Another problem generally encountered in the included studies concerned incomplete reporting. While this shortcoming may be attributed to page restrictions for journal articles in general, it nevertheless impedes the retrieval of information to adequately assess, for example, recruiting procedures, dropouts, missing data, or the description of interventions. As a result of the latter, the allocation of studies to our typology of interventions may be inaccurate. Incomplete reporting of complex interventions in the workplace has been noted by other review authors in the past (e.g. Egan 2009; Joyce 2010), and we renew this criticism here. In light of the page restrictions of scientific journals, many journals offer online appendices for additional material. We encourage study authors to make use of this possibility to provide complete, transparent information on their studies.

Potential biases in the review process

Although we aimed at extensively searching the literature through a number of databases, it is possible that we did not identify all relevant research on the topic. In a similar vein, our search terms may have been insufficient to detect potentially relevant studies. Furthermore, our categorisation approach largely determined the comparisons between studies. It is possible that a different categorisation approach may have led to different analyses. Therefore, our results should be interpreted with caution and considering the outlined search strategy and categorisation of studies. Despite our efforts, we could not obtain some essential data from the authors of four studies. Because of the rather small number of studies included in this review, the studies with missing data could have a major influence on our overall findings. We hope that new studies added in the future will allow clearer and more reliable answers to the questions raised here.

Agreements and disagreements with other studies or reviews

Against the background of a plethora of studies of various designs on leadership and employee health and well‐being that have been included in other reviews and even meta‐analyses, the empirical results of our systematic review are disappointing. In spite of prevailing consensus that leadership behaviour influences stress and well‐being of employees (e.g. Kuoppala 2008; Nyberg 2005; Skakon 2010), we could hardly find any positive effects of leadership training in natural settings of organisations on outcomes such as employee well‐being, stress, and absenteeism. This discrepancy between the apparent scientific consensus and the empirical evidence in this review may predominantly be attributed to weak study designs. The only RCT (Gumuseli 2002), that we could identify yielded no significant effects, nor did 12 of the 14 included cRCTs. Whereas associations between leadership behaviour and employee well‐being and stress are well‐documented in research using uncontrolled observational designs, we could not find strong evidence for causal effects of leadership interventions on employees’ well‐being, stress, and absenteeism. A more recent literature review (Tsutsumi 2011), included three RCTs and four quasi‐experimental studies and was able to identify positive effects. However, this review performed no meta‐analysis or risk of bias assessment, and most of the included studies did not match our predefined primary outcomes. Other reviews (e.g. Kuoppala 2008; Skakon 2010) were predominantly based on non‐experimental studies.

Authors' conclusions

Implications for practice.

The results of our systematic review do not support the utility of leadership interventions to have a beneficial impact on employees' stress, absenteeism, and well‐being. However, HR managers might use the evidence base of studies in this review to generate ideas on suitable leadership interventions in terms of training content, designs, and processes, as well as the possibility of evaluating potential effects of planned interventions with established measures. They should be aware of the methodological shortcomings in evaluative research in organisations and should strive to collaborate with researchers in order to improve the evidence base in this field. By applying evidence‐based management practices (Rousseau 2012) to supervisor training interventions instead of merely copying existing approaches currently practised in organisations, professionals may contribute to both uncovering the effects of particular supervisor interventions and gaining knowledge about how to design effective interventions to strengthen employee health and well‐being.

Implications for research.

In this systematic review, we focused on the general population of employees working in organisations and changes in their perceptions of stress, absenteeism and well‐being as a result of their supervisors receiving training in interaction with employees or in work‐design. While the small number of studies currently does not allow for analyses that compare interventions across different industries, differences across industries (e.g., in the degree of supervisors' influence or in the relevance of certain outcomes) are nevertheless plausible. It is therefore important that future studies report industries and/or occupations targeted by the interventions as accurate as possible.

So far, the overall very low‐quality evidence from the included studies does not offer clear guidance for the design of neither the content of interventions (improving dyadic interaction between supervisors and employees versus improving capabilities of the supervisor to adapt the work environment in a beneficial way), nor the setting of interventions (off‐ versus on‐the‐job). For this reason, we invite researchers conducting intervention studies in the future to define leadership behaviour more clearly and to describe leadership interventions as well as organisational settings more precisely. Reporting on leadership interventions should be improved, such that content as well as procedures are documented more comprehensively. It is of utmost importance that the rationale of how interventions are thought to translate into outcomes is clarified early on. Currently, the complex system of a specific organisational setting in which an intervention takes place seems to be reduced to its in‐ and outputs without considering internal processes (i.e., a 'black box') in many studies. There is a complex path from a training programme being perceived as helpful by a supervisor, to the ability and opportunity to transfer new skills into changes in daily work settings, to employees noticing and responding to these changes, and also, eventually, to measurable changes in psychomental health and organisational performance. Therefore, we encourage researchers to analyse the process from the delivery of a supervisor intervention to observable effects in employees' health and well‐being at multiple levels (Sitzmann 2019), both as a form of manipulation check and to obtain information about the transmission process. In the light of scarce evidence for effective supervisor training to affect employees, it might prove fruitful to first investigate in detail the determinants of a successful transmission process and their interrelations before expending resources on actual training. Furthermore, interventions that were reported to be effective may be taken as preliminary model interventions for future research. However, the multitude of parameters that may potentially affect typical field intervention studies currently impedes clear evidence‐based recommendations grounded in the scarce and diverse information contained in available study reports.

Regarding comparisons, most studies contrasted supervisor training with no intervention. More rigorous designs that compared supervisor training to a placebo or even other training were a minority among the included studies. There is an obvious need for more research, with higher methodological standards, on the effectiveness of leadership interventions on employee health and well‐being. Future studies should aim to establish stronger, experimental (RCT, cRCT) or at least quasi‐experimental designs with comparison groups receiving a placebo or other training in order to improve the very limited and scarce evidence base. In conducting controlled intervention studies, researchers should be aware of the various risks of bias, for example, with respect to allocation of participants to intervention groups, appropriate statistical analysis, selective reporting of findings, or insufficient protection of studies against contamination. Moreover, important statistical information was lacking in a number of studies, which made it difficult or impossible to compare results between studies directly. Therefore, we encourage researchers to report complete and accurate information on statistical procedures and results by adhering to current reporting standards (e.g. Appelbaum 2018).

While outcomes of interest were classified in three broad domains (psychomental stress, absenteeism and well‐being), there was great diversity within these domains. The outcomes investigated most frequently were job satisfaction as an indicator of well‐being and burnout as an indicator of stress. Both outcomes for absenteeism were also diverse in terms of unit (duration vs. frequency) and reference timeframe (6 vs. 12 months). The use of standardised and validated outcome measures for well‐being, stress and absenteeism could facilitate future synthesis of evidence. More differentiated implications for research regarding extended PICO aspects like time (training duration, follow‐up intervals) or settings (e.g., training on‐ versus off‐the‐job) cannot be inferred due to the restricted number and quality of evidence of previous studies.

What's new

Date	Event	Description
11 October 2019	Amended	Italics format of title changed

Appendices

Appendix 1. Search strategy: CENTRAL

Cochrane Central Register of Controlled Trials (CENTRAL)

Issue 5 of 12, May 2019

Search date: May 21, 2019

1. [mh "Interprofessional Relations"] or (supervisor* NEAR/8 employee* NEAR/8 interaction*) or ((labor or labour) NEAR/8 management NEAR/8 relation*) or (leadership NEAR/4 style*) or (transform* NEAR/2 leader*) or (relation* NEAR/2 leader*) or (transact* NEAR/2 leader*) or (considera* NEAR/2 leader*) or (leader* NEAR/2 behavio*) or (leader* NEAR/2 style*) or (leader* NEAR/2 development) or (supervisor* NEAR/2 behavio*) or (leader* NEAR/2 intervention*) or (supportive NEAR/2 leader*) or (supportive NEAR/2 manag*) or (supportive NEAR/2 superv*) (1066)

2. ((management or manager* or supervisor* or leader* or foreman or foremen or director* or executive* or administrator* or coach*) NEAR/8 (training* or lecture* or workshop* or program* or teaching or intervention* or guidance* or education or schooling or coach* or tutor* or mentor* or counsel*)) (26016)

3. 1 and 2 (407)

4. "human resource manag*" or [mh "Staff Development"] (85)

5. (1 or 2) and 4 (26)

6. 3 or 5 (429)

7. (employee* NEAR/2 health) or [mh ^"Work Performance"] or (employee* NEAR/2 performance) or "work related stress*" or [mh "Occupational Stress"] or "occupational stress*" or [mh "Occupational Health"] or "occupational health" or [mh "Occupational Diseases"] or "occupational disease*" or "work‐related disease*" or "work‐related disorder*" or "work stress*" or "job stress" or [mh ^"Job Satisfaction"] or "job satisfaction" or [mh ^Absenteeism] or "Employee Absenteeism" or "sick* leave*" or "sick* absence*" or "employee leave benefit*" (5899)

8. ((stress or strain or wellbeing or "well‐being" or health or anxiet* or depress* or burnout or "burn‐out" or psychosocial) NEAR/4 (employee* or worker* or subordinate* or staff)) (5481)

9. 7 or 8 (10445)

10. 6 and 9 (95, final results)

Appendix 2. Search strategy: MEDLINE

MEDLINE

(Ovid MEDLINE(R) In‐Process & Other Non‐Indexed Citations, Ovid MEDLINE(R) Daily and Ovid MEDLINE(R) 1946 to Present)

Search date: May 21, 2019

1. exp Interprofessional Relations/ or (supervisor* adj8 employee* adj8 interaction*).mp. or ((labor or labour) adj8 management adj8 relation*).mp. or (leadership adj4 style*).mp. or (transform* adj2 leader*).mp. or (relation* adj2 leader*).mp. or (transact* adj2 leader*).mp. or (considera* adj2 leader*).mp. or (leader* adj2 behavio*).mp. or (leader* adj2 style*).mp. or (leader* adj2 development).mp. or (supervisor* adj2 behavio*).mp. or (leader* adj2 intervention*).mp. or (supportive adj2 leader*).mp. or (supportive adj2 manag*).mp. or (supportive adj2 superv*).mp. (72589)

2. ((management or manager* or supervisor* or leader* or foreman or foremen or director* or executive* or administrator* or coach*) adj8 (training* or lecture* or workshop* or program* or teaching or intervention* or guidance* or education or schooling or coach* or tutor* or mentor* or counsel*)).mp. (116663)

3. 1 and 2 (4374)

4. "human resource manag*".mp. or exp Staff Development/ (9451)

5. (1 or 2) and 4 (2354)

6. 3 or 5 (6367, final results)

7. (employee* adj2 health).mp. or Work Performance/ or (employee* adj2 performance).mp. or "work related stress*".mp. or exp Occupational Stress/ or "occupational stress*".mp. or exp Occupational Health/ or "occupational health".mp. or exp Occupational Diseases/ or "occupational disease*".mp. or "work‐related disease*".mp. or "work‐related disorder*".mp. or "work stress*".mp. or "job stress".mp. or Job Satisfaction/ or "job satisfaction".mp. or Absenteeism/ or "Employee Absenteeism".mp. or "sick* leave*".mp. or "sick* absence*".mp. or "employee leave benefit*".mp. (211602)

8. ((stress or strain or wellbeing or "well‐being" or health or anxiet* or depress* or burnout or "burn‐out" or psychosocial) adj4 (employee* or worker* or subordinate* or staff)).mp. (64597)

9. 7 or 8 (261450)

10. 6 and 9 (813)

Appendix 3. Search strategy: PsycINFO (via Ovid)

PsycINFO (via Ovid)

(Ovid, 1806 to April Week 2 2018)

Search date: 13 April 2018

1. Supervisor Employee Interaction/ or Labor Management Relations/ or Leadership Style/ or (supervisor* adj3 employee* adj3 interaction*).mp. or ((labor or labour) adj3 management adj3 relation*).mp. or "leader* style*".mp. or "transform* leader*".mp. or "relation* leader*".mp. or "transact* leader*".mp. or "considera* leader*".mp. or "leader* behavio*".mp. or "leader* style*".mp. or "leader* development".mp. or "supervisor* behavio*".mp. or "leader* intervention*".mp. or "supportive leader*".mp. or "supportive manag*".mp. or "supportive superv*".mp. (23442)

2. Management Training/ or ((management or manager* or supervisor* or leader* or foreman or foremen or director* or executive* or administrator* or coach*) adj4 (training* or lecture* or workshop* or program* or teaching or intervention* or guidance* or education or schooling or coach* or tutor* or mentor* or counsel*)).mp. (63990)

3. 1 and 2 (4517)

4. exp Human Resource Management/ or "human resource manag*".mp. (41313)

5. (1 or 2) and 4 (5242)

6. 3 or 5 (9399)

7. (employee* adj2 health).mp. or Job Performance/ or ((employee* or job or work) adj2 performance).mp. or "work related stress*".mp. or exp Occupational Stress/ or "occupational stress*".mp. or Occupational Health/ or Work Related Illnesses/ or "occupational health".mp. or "occupational disease*".mp. or "work‐related disease*".mp. or "work‐related disorder*".mp. or "work stress*".mp. or "job stress".mp. or Job Satisfaction/ or "job satisfaction".mp. or absenteeism.mp. or Employee Absenteeism/ or "sick* leave*".mp. or "sick* absence*".mp. or "employee leave benefit*".mp. (69104)

8. ((stress or strain or wellbeing or "well‐being" or health or anxiet* or depress* or burnout or "burn‐out" or psychosocial) adj2 (employee* or worker* or subordinate* or staff)).mp. (15197)

9. 7 or 8 (79218)

10. 6 and 9 (1158, final result)

Appendix 4. Search strategy: PsycINFO (via EBSCOhost)

(EBSCOhost, 1806 to May 2019)

Search date: May 21, 2019

1. DE "Supervisor Employee Interaction" or DE "Labor Management Relations" or DE "Leadership Style" or (supervisor* N3 employee* N3 interaction*) or ((labor or labour) N3 management N3 relation*) or "leader* style*" or "transform* leader*" or "relation* leader*" or "transact* leader*" or "considera* leader*" or "leader* behavio*" or "leader* style*" or "leader* development" or "supervisor* behavio*" or "leader* intervention*" or "supportive leader*" or "supportive manag*" or "supportive superv*" (24946)

2. DE "Management Training" or ((management or manager* or supervisor* or leader* or foreman or foremen or director* or executive* or administrator* or coach*) N4 (training* or lecture* or workshop* or program* or teaching or intervention* or guidance* or education or schooling or coach* or tutor* or mentor* or counsel*)) (110309)

3. 1 and 2 (9391)

4. DE "Human Resource Management" or DE "Career Development" or DE "Employee Benefits" or DE "Job Analysis" or DE "Labor Management Relations" or DE "Outsourcing" or DE "Personnel Evaluation" or DE "Personnel Recruitment" or DE "Personnel Selection" or DE "Personnel Termination" or "human resource manag*" (43025)

5. (1 or 2) and 4 (16328)

6. 3 or 5 (24234)

7. (employee* N2 health) or DE "Job Performance" or ((employee* or job or work) N2 performance) or "work related stress*" or DE "Occupational Stress" or DE "Compassion Fatigue" or "occupational stress*" or DE "Occupational Health" or DE "Work Related Illnesses" or "occupational health" or "occupational disease*" or "work‐related disease*" or "work‐related disorder*" or "work stress*" or "job stress" or DE "Job Satisfaction" or "job satisfaction" or absenteeism or DE "Employee Absenteeism" or "sick* leave*" or "sick* absence*" or "employee leave benefit*" (98877)

8. ((stress or strain or wellbeing or "well‐being" or health or anxiet* or depress* or burnout or "burn‐out" or psychosocial) N2 (employee* or worker* or subordinate* or staff)) ()

9. 7 or 8 (111265)

10. 6 and 9 (3852)

11. Limiters: Publication Year: 2018‐2019 (197746)

12. 10 and 11 (199, final result)

Appendix 5. Search strategy: Scopus

Scopus

Search date: May 21, 2019

1. TITLE‐ABS‐KEY("interprofessional relation*" or (supervisor* W/2 employee* W/2 interaction*) or ((labor or labour) W/2 management W/2 relation*) or (leadership W/2 style*) or (transform* W/2 leader*) or (relation* W/2 leader*) or (transact* W/2 leader*) or (considera* W/2 leader*) or (leader* W/2 behavio*) or (leader* W/2 style*) or (leader* W/2 development) or (supervisor* W/2 behavio*) or (leader* W/2 intervention*) or (supportive W/2 leader*) or (supportive W/2 manag*) or (supportive W/2 superv*)) (79890)

2. TITLE‐ABS‐KEY((management or manager* or supervisor* or leader* or foreman or foremen or director* or executive* or administrator* or coach*) W/2 (training* or lecture* or workshop* or program* or teaching or intervention* or guidance* or education or schooling or coach* or tutor* or mentor* or counsel*)) (169699)

3. 1 and 2 (6664)

4. TITLE‐ABS‐KEY("human resource manag*" or "staff develop*") (41583)

5. (1 or 2) and 4 (4335)

6. 3 or 5 (10632)

7. TITLE‐ABS‐KEY((employee* W/2 health) or "work performance" or (employee* W/2 performance) or "work related stress*" or "occupational stress*" or "occupational health" or "occupational disease*" or "work‐related disease*" or "work‐related disorder*" or "work‐related illness*" or "work stress*" or "job stress" or "job satisfaction" or absenteeism or "sick* leave*" or "sick* absence*" or "employee leave benefit*") (260983)

8. TITLE‐ABS‐KEY(((stress or strain or wellbeing or "well‐being" or health or anxiet* or depress* or burnout or "burn‐out" or psychosocial) W/2 (employee* or worker* or subordinate* or staff))) (79673)

9. 7 or 8 (313086)

10. 6 and 9 (944, final results)

Appendix 6. Search strategy: OSH Update

OSH Update (databases NIOSHTIC, NIOSHTIC‐2, HSELINE, CISDOC)

Search date: May 21, 2019

1. GW{(interprofessional.‐2. relation*) or (supervisor* and employee* and interaction*) or ((labor or labour) and management and relation*) or (leader* style*) or (transform* leader*) or (relation* leader*) or (transact* leader*) or (considera* leader*) or (leader* behavio*) or (leader* development) or (supervisor* behavio*) or (leader* intervention*) or (supportive leader*) or (supportive manag*) or (supportive superv*)} (4169)

2. GW{(management or manager* or supervisor* or leader* or foreman or foremen or director* or executive* or administrator* or coach*) and (training* or lecture* or workshop* or program* or teaching or intervention* or guidance* or education or schooling or coach* or tutor* or mentor* or counsel*)} (48636)

3. 1 and 2 (2712)

4. GW{(human resource manag*) or (staff develop*)} (2411)

5. (1 or 2) and 4 (921)

6. 3 or 5 (3575)

7. GW{(employee.‐2. health) or (work performance) or (employee.‐2. performance) or (work related stress*) or (occupational stress*) or (occupational health) or (occupational disease*) or (work‐related disease*) or (work‐related disorder*) or (work‐related illness*) or (work stress*) or (job stress) or (job satisfaction) or (absenteeism) or (sick* leave*) or (sick* absence*) or (employee leave benefit*)} (105488)

8. GW{((stress or strain or wellbeing or well‐being or health or anxiet* or depress* or burnout or burn‐out or psychosocial) and (employee* or worker* or subordinate* or staff))} (129906)

9. 7 or 8 (188325)

10. 6 and 9 (2641)

11. DC{OUNIOC or OUNIOS OR OUHSEL OR OUCISD} (584625)

12. 10 and 11 (227)

Data and analyses

Comparison 1. Training vs no intervention.

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Stress/short‐term/cRCT	2	157	Std. Mean Difference (IV, Random, 95% CI)	0.05 [‐0.50, 0.60]
1.1 A1	2	157	Std. Mean Difference (IV, Random, 95% CI)	0.05 [‐0.50, 0.60]
2 Stress/mid‐term/cRCT	2	304	Std. Mean Difference (IV, Random, 95% CI)	0.02 [‐0.22, 0.26]
2.1 A1	1	226	Std. Mean Difference (IV, Random, 95% CI)	‐0.05 [‐0.31, 0.22]
2.2 B2	1	78	Std. Mean Difference (IV, Random, 95% CI)	0.23 [‐0.23, 0.70]
3 Stress/mid‐term/CBA	1	368	Std. Mean Difference (IV, Random, 95% CI)	0.09 [‐0.12, 0.30]
3.1 A2	1	368	Std. Mean Difference (IV, Random, 95% CI)	0.09 [‐0.12, 0.30]
4 Stress/long‐term/CBA	1	145	Std. Mean Difference (IV, Random, 95% CI)	0.23 [‐0.12, 0.57]
4.1 A2	1	145	Std. Mean Difference (IV, Random, 95% CI)	0.23 [‐0.12, 0.57]
5 Absenteeism/mid‐term/cRCT	1	164	Std. Mean Difference (IV, Random, 95% CI)	0.20 [‐0.11, 0.51]
5.1 B2	1	164	Std. Mean Difference (IV, Random, 95% CI)	0.20 [‐0.11, 0.51]
6 Well‐being/short‐term/cRCT	2	337	Std. Mean Difference (IV, Random, 95% CI)	0.14 [‐0.28, 0.57]
6.1 A1	1	239	Std. Mean Difference (IV, Random, 95% CI)	‐0.05 [‐0.30, 0.21]
6.2 B1	1	98	Std. Mean Difference (IV, Random, 95% CI)	0.39 [‐0.01, 0.79]
7 Well‐being/short‐term/CBA	1	143	Std. Mean Difference (IV, Random, 95% CI)	0.0 [0.0, 0.0]
7.1 B2	1	143	Std. Mean Difference (IV, Random, 95% CI)	0.0 [0.0, 0.0]
8 Well‐being/mid‐term/cRCT + RCT	4		Std. Mean Difference (IV, Random, 95% CI)	Totals not selected
8.1 A1	1		Std. Mean Difference (IV, Random, 95% CI)	0.0 [0.0, 0.0]
8.2 A2	3		Std. Mean Difference (IV, Random, 95% CI)	0.0 [0.0, 0.0]
9 Well‐being/mid‐term/CBA	3		Std. Mean Difference (IV, Random, 95% CI)	Totals not selected
9.1 A2	2		Std. Mean Difference (IV, Random, 95% CI)	0.0 [0.0, 0.0]
9.2 B1	1		Std. Mean Difference (IV, Random, 95% CI)	0.0 [0.0, 0.0]
10 Well‐being/long‐term/cRCT	1	547	Std. Mean Difference (IV, Random, 95% CI)	0.12 [‐0.05, 0.29]
10.1 A2	1	547	Std. Mean Difference (IV, Random, 95% CI)	0.12 [‐0.05, 0.29]

Comparison 2. Training vs placebo.

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Stress/mid‐term/cRCT	2	357	Std. Mean Difference (IV, Random, 95% CI)	‐0.07 [‐0.27, 0.14]
1.1 A1	2	357	Std. Mean Difference (IV, Random, 95% CI)	‐0.07 [‐0.27, 0.14]
2 Absenteeism/mid‐term/cRCT	1	174	Std. Mean Difference (IV, Random, 95% CI)	‐0.11 [‐0.41, 0.19]
2.1 A2	1	174	Std. Mean Difference (IV, Random, 95% CI)	‐0.11 [‐0.41, 0.19]
3 Well‐being/mid‐term/CBA	1	57	Std. Mean Difference (IV, Random, 95% CI)	0.0 [0.0, 0.0]
3.1 A1	1	57	Std. Mean Difference (IV, Random, 95% CI)	0.0 [0.0, 0.0]

Comparison 3. Training vs other training.

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Stress/long‐term/cRCT	1	98	Std. Mean Difference (IV, Random, 95% CI)	0.0 [0.0, 0.0]
1.1 A1	1	98	Std. Mean Difference (IV, Random, 95% CI)	0.0 [0.0, 0.0]

Characteristics of studies

Characteristics of included studies [ordered by study ID]

Barling 1996.

Methods	Study design: cRCT Country: Canada Type of industry/economic sector: banking, tertiary sector Size of company: not explicitly stated Supervisor‐employee relationship: first degree
Participants	Number of participants Supervisors (intervention): 9; supervisors (control): 11 Employees (intervention): NR; employees (control): NR The number of employees in the final sample is assumed to be approximately 77 (estimated from the degrees of freedom used in the F‐test Age of participants Supervisors (intervention): NR; supervisors (control): NR Employees (intervention): NR; employees (control): NR Sex of participants Male supervisors (intervention): 56%; male supervisors (control): 55% Male employees (intervention): NR; male employees (control): NR Inclusion criteria: not explicitly stated Exclusion criteria: not explicitly stated
Interventions	Type of intervention: A1 Content of intervention Group training: (first segment) ‐ the foundations of transformational, transactional, and laissez‐faire leadership styles, and relevant research findings; (second segment) ‐ exercises on specific topics to help internalise aspects learned in first segment e.g. goal‐setting (formulating goals for themselves concerning transformational leadership), role playing of specific leadership behaviours Individual booster sessions: feedback on managers' leadership style based on self‐reported data and subordinate questionnaire, personalised action plans for following month, personal (attainable) goals set, sustainability of behaviour emphasised ‐ following meetings with goal to discuss accomplishments based on session 1. Primary purpose of training programme: becoming intellectually stimulating Timing of intervention 1‐day, group‐based training session for all branch managers in experimental group 1 day after group training start of 1/4 individual booster sessions, timing of following sessions NR, duration NR, scheduled on a monthly basis Type of delivery Group training: lecturing (familiarisation with transformational leadership) incl. role play, thought games (worst and best leaders ever encountered by managers?), interaction (association of answers with leadership styles), lecturing (on transformational, transactional, and laissez‐faire leadership styles and research findings) exercise sessions including group work on different topics with thought games, role playing (different leadership styles), group discussions in plenary sessions Concept of training: theory‐based
Outcomes	Psychomental stress: none Absenteeism: none Well‐being: organisational commitment, scaling and polarity not reported. Measured through Organizational Commitment Questionnaire (Mowday 1982) Timing of outcome assessment Baseline measurement: 2 weeks before intervention Follow‐up measurement: 5 months after intervention
Training in control group	None
Training compliance in intervention group	Not assessed
Notes	None
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	Generation of random sequence not specified
Allocation concealment (selection bias)	Unclear risk	Unclear from the text
Blinding of participants and personnel (performance bias) All outcomes	Unclear risk	Study personnel: not stated, but maybe possible Supervisors: managers in the training group were encouraged to discuss ideas and experiences with other members of the group but to refrain from having similar discussions with managers in the no‐training control group or subordinates in either of the two experimental groups. Thus, they knew about allocation. Employees: not stated, unclear if possible at all
Blinding of outcome assessment (detection bias) All outcomes	Unclear risk	Outcome was measured by self‐assessment, we assume that blinding was not broken (see above), but it remains unclear
Incomplete outcome data (attrition bias) All outcomes	High risk	Dropouts and response rate not stated, total number of participating employees not directly stated
Selective reporting (reporting bias)	Unclear risk	All predefined primary outcome measures reported, but no protocol available
cRCT: recruitment bias	Low risk	Recruitment of employees was "at the same time that managers were initially approached"
cRCT: baseline imbalance	Unclear risk	Organisational commitment not different at baseline. Differences in other characteristics not stated
cRCT: loss of clusters	Low risk	No evidence for cluster loss
cRCT: incorrect statistical analysis	High risk	Multilevel analysis not performed
cRCT: comparability with RCTs	Low risk	Recruitment of employees was "at the same time that managers were initially approached"

Barrech 2018.

Methods	Study design: cRCT Country: Switzerland Type of industry/economic sector: healthcare Size of company: not stated Supervisor‐employee relationship: first degree
Participants	Number of participants Supervisors (intervention): 52; supervisors (control): 47 Employees (intervention): 391; employees (control): 424 Age of participants Supervisors (intervention): mean 43.10 (SD 8.05); supervisors (control): mean 50.50 (SD 6.03) Employees (intervention): 40.69 (SD 9.64); employees (control): 41.64 (SD 10.57) Sex of participants Male supervisors (intervention): 80%; male supervisors (control): 75% Male employees (intervention): 63%; male employees (control): 39% Inclusion criteria: all supervisors and their respective team members working at the plant Exclusion criteria: not stated
Interventions	Type of intervention: B2 Content of intervention Seminars: active listening to theoretical input, group discussions, case studies Peer counselling: discussion of real‐life examples, receive and appreciate social support by colleagues Seminar 1: biological fundamentals of chronic stress and its relationships with health; important role of social support in this context Seminar 2: determinants of healthy psychosocial working conditions with reference to specific working environment of participants Seminar 3: role of leadership in the context of employee health, aspects of change‐oriented leadership behaviour, principles and methods of successful communication during change Timing of intervention 6 sessions (3 seminars, 3 peer counselling sessions) during 3 months Each seminar session followed by a peer counselling session Duration of each session: 2‐4 h Up to 10 supervisors in one session Sessions were held during working hours Type of delivery: 3 seminars, 3 peer counselling sessions Concept of training: unclear
Outcomes	Psychomental stress: none Exhaustion tendency. 3 items from the Giessen Subjective Complaints List: (Brähler 1995) Anxiety. 7 items from HADS: Herrmann‐Lingen 2011 Depression. 7 items from HADS: Herrmann‐Lingen 2011 Absenteeism: none Well‐being: none Timing of outcome assessment Baseline measurement: 1 month prior to training of intervention group Follow‐up measurement: 3 months after end of intervention group training, 3 months after end of control group training
Training in control group	None
Training compliance in intervention group	Not assessed
Notes	None
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	No information given
Allocation concealment (selection bias)	Unclear risk	No information given
Blinding of participants and personnel (performance bias) All outcomes	Unclear risk	Investigators were aware of supervisors in the intervention group but had minimal contact to supervisors in the control group and to employees. Blinding of supervisors not possible/not meaningful. No information given on blinding of employees.
Blinding of outcome assessment (detection bias) All outcomes	Unclear risk	Blinding not possible: employees had to provide informed consent
Incomplete outcome data (attrition bias) All outcomes	Unclear risk	No information given
Selective reporting (reporting bias)	Low risk	Comprehensive reporting of study results, utilising an online appendix
cRCT: recruitment bias	High risk	Several weaknesses: randomisation carried out before supervisors enrolled for the study; effectively, supervisors chose to participate in training voluntarily and thereby determined allocation to intervention group or control group (trained supervisors from intervention group and from waiting‐list control group were included in intervention group; small control group remaining = supervisors in control group who chose not to participate); self‐selection and/or selection bias probable
cRCT: baseline imbalance	Low risk	Lower baseline job insecurity in intervention group vs control group: self‐selection and/or selection bias probable
cRCT: loss of clusters	High risk	Low return rates in general; clusters switched from control group to intervention group by self‐selection of supervisors
cRCT: incorrect statistical analysis	Low risk	Sound rationale for analyses and correctly applied
cRCT: comparability with RCTs	High risk	Randomisation ineffective

Biggs 2014.

Methods	Study design: CBA Country: Australia Type of industry/economic sector: police service, tertiary sector Size of company: not explicitly stated Supervisor‐employee relationship: first degree
Participants	Number of participants Supervisors (intervention): NR; supervisors (control): NR Employees (intervention): 146; employees (control): 222 Age of participants Supervisors (intervention): NR; supervisors (control): NR Employees (intervention): 40.00 ± 7.72 (mean, SD); employees (control): 39.32 ± 8.38 (M, SD) Sex of participants Male supervisors (intervention): NR; male supervisors (control): NR Male employees (intervention): 77%; male employees ((control): 82% Inclusion criteria In general, all police officers invited within the 2 organisational regions Leaders Rank above sergeant Job role entailed people‐management responsibilities Willingness to take part in the study Employees: Immediate subordinates of the abovementioned police officers Exclusion criteria Civilian staff Police officers who changed regions Participants in concurrent interventions or trials Trained supervisors themselves Participants (n = 4) with > 50% missing data Participants (n = 5) with multivariate outliers at P < 0.001
Interventions	Type of intervention: A2 Content of intervention Training on theoretical leadership styles and behaviours, as well as practical resources to enhance their leadership capabilities. Education to develop effective leadership styles (e.g. transformational leadership), effective communication, and strategic leadership. Participants were asked to conduct their own action‐learning project (e.g. implementing a change strategy) during workshop sessions. Timing of intervention 5 days action learning, plus 360° feedback, plus coaching sessions 5 x action learning, 1 x feedback, number of coaching sessions not stated Action learning approximately 1 day each, duration of feedback and coaching sessions not stated Type of delivery One‐time, 360° review and feedback process 5‐day action‐learning workshops Individual coaching provided to programme participants (leaders) throughout the duration of the programme Concept of training: theory‐based
Outcomes	Psychomental stress: Psychological strain. Scaling 0‐3, higher scores indicate higher strain. GHQ‐12 (Goldberg 1972) Absenteeism: none Well‐being: Job satisfaction. Scaling 1‐7, higher scores were indicative of high job satisfaction. 15‐item measure developed by Warr 1979 Work engagement. Scaling 0‐6, higher scores indicate a higher degree of engagement. 9‐item UWES (Schaufeli 2006) Turnover intention (defined as intention to leave the organisation over the preceding 12 months). Scaling 1‐5, higher scores indicate stronger desire to change the job. 3‐item measure developed by Brough 2004 Timing of outcome assessment Baseline measurement: 4 months before intervention Follow‐up measurement: 7 months after intervention
Training in control group	None
Training compliance in intervention group	Not assessed
Notes	None
Risk of bias
Bias	Authors' judgement	Support for judgement
CBA: was the allocation sequence adequately generated	High risk	Judgement according to EPOC recommendations for CBAs.
CBA: was the allocation adequately concealed	High risk	Judgement according to EPOC recommendations for CBAs.
CBA: were baseline outcome measurements similar All outcomes	Low risk	Not statistically different (α = 0.05) regarding job satisfaction, work engagement, turnover intentions, and psychological well‐being. Analyses were adjusted for baseline values.
CBA: were baseline characteristics similar	Unclear risk	Significant differences regarding origin (organisational region, df = 1, Chi² = 24.6) and rank (df = 5, Chi² = 25.2). Own Chi‐² tests based on data from Table 1, p.10. Tenure was also significantly different (15.64 ± 9.99 years in the intervention group vs 12.45 ± 10.05 years in the control group, MD 3.19, df = 366, T = 2.99, T‐Test assuming equal variance and normal distribution). Rank and tenure of intervention group significantly but only slightly higher.
CBA: were incomplete outcome data adequately addressed All outcomes	High risk	2637 employees were invited to participate. Final sample size included 368 employees (overall response rate = 14%). Participants' data skipped if > 50 % data missing. Imputation of missings otherwise
CBA: was knowledge of the allocated interventions adequately prevented during the study	High risk	Self‐reported, non‐objective measurements, not assessed blindly
CBA: was the study adequately protected against contamination	Unclear risk	Unclear from the text
CBA: was the study free from selective outcome reporting	Unclear risk	All prespecified outcomes were reported, but no protocol available
Other bias	High risk	Selection bias: as participation in intervention was voluntary, a relevant selection bias may have occurred.

Dahinten 2014.

Methods	Study design: CBA Country: Canada Type of industry/economic sector: healthcare/nursing, tertiary sector Size of company: NR Supervisor‐employee relationship: first degree
Participants	Number of participants Supervisors (intervention): 110; supervisors (control): 18 Employees (intervention): 99; employees (control): 30 Age of participants Supervisors (intervention): M = 44.5; SD = 7.5; supervisors (control): M = 48.3; SD = 9.9 Employees (intervention): M = 44.2; SD = 11.2; employees (control): M = 50.6; SD = 8.4 Sex of participants Male supervisors (intervention): 10%; male supervisors (control): 12% Male employees (intervention): 3%; male employees (control): 0% Inclusion criteria Novice nurse leaders (< 3 years of first‐line leadership experience) Assessment of leadership potential (by Chief Nursing Officer) Voluntary participation in training and study Exclusion criteria: NR
Interventions	Type of intervention: B1 Content of intervention Training of leaders in (application of) staff empowering behaviours Timing of intervention 4‐day residential workshop plus additional issues, see above, most probably during working time Type of delivery Workshop with lectures and interactive learning sessions + innovation project (duration: 1 year) + mentorship from senior nursing leaders + online knowledge network. 4‐day residential workshop with didactic leadership content and interactive learning sessions Year‐long innovation project of relevance to the leaders’ respective organisations Mentorship from senior nursing leaders Organisational supports, such as release time for project work Online knowledge network to facilitate connections among leaders Concept of training: not theory‐based
Outcomes	Psychomental stress: none Absenteeism: none Well‐being Organisational commitment. Scaling 1‐7, higher scores indicate higher levels of effective organisational commitment. 8 items, Affective Organizational Commitment Scale (Meyer 1991) Timing of outcome assessment Baseline measurement: during intervention (intervention group), start of intervention (control group) Follow‐up measurement: 12 months after intervention
Training in control group	None
Training compliance in intervention group	Not assessed
Notes	None
Risk of bias
Bias	Authors' judgement	Support for judgement
CBA: was the allocation sequence adequately generated	High risk	Judgement according to EPOC recommendations for CBAs
CBA: was the allocation adequately concealed	High risk	Judgement according to EPOC recommendations for CBAs
CBA: were baseline outcome measurements similar All outcomes	High risk	Significant baseline difference in outcome (organisational commitment) between intervention and control group. Intervention group scored significantly lower than the comparison group on Structural Empowerment (P < 0.05) and Organisational Commitment (P < 0.01) at Time 1
CBA: were baseline characteristics similar	High risk	Significant baseline differences in education and leader experience (of leaders) as well as age and nursing experience (of subordinates) Intervention vs control group 44 vs 51 years of age Intervention vs control group 16 vs 19 years in nursing
CBA: were incomplete outcome data adequately addressed All outcomes	High risk	Analysis of complete data only, substantial dropout rates of subordinates (T1: 1067; T1 + T2: 129)
CBA: was knowledge of the allocated interventions adequately prevented during the study	High risk	No blinding. There was no sham intervention so knowledge was obvious
CBA: was the study adequately protected against contamination	Unclear risk	No harming influences mentioned
CBA: was the study free from selective outcome reporting	Unclear risk	All predefined outcomes were reported, but no protocol available
Other bias	Unclear risk	NA

Deci 1989.

Methods	Study design: CBA Country: USA Type of industry/economic sector: office machine production, secondary sector Size of company: not explicitly stated, "nearly 1000 employees" Supervisor‐employee relationship: first degree
Participants	Number of participants Supervisors (intervention): NR; supervisors (control): NR Employees (intervention): 412; employees (control): 177 Age of participants: Supervisors (intervention): NR; supervisors (control): NR Employees (intervention): NR; employees (control): NR Sex of participants Male supervisors (intervention): NR; male supervisors (control): NR Male employees (intervention): NR; male employees (control): NR Inclusion criteria: NR Exclusion criteria: NR
Interventions	Type of intervention: A2 Content of intervention Training to promote self‐determination of subordinates by providing autonomy support, non‐controlling informative feedback, and recognising and accepting their perspectives (needs and feelings) Timing of intervention 1 day with branch manager 5 days with management team: 2‐day offsite team building management development session plus 3 single days in the following weeks 7 days with various field managers (incl. meeting with subordinate technicians) Type of delivery Discussions and activities around 3 basic themes (see below); examination of own behaviours; experimentation (practices within own teams); feedback about leader behaviour and group reactions Concept of training: not theory‐based
Outcomes	Psychomental stress: none Absenteeism: none Well‐being: Job satisfaction. Scaling 1‐5, higher scores indicate higher levels of job satisfaction. 20 items, (short form) Global Satisfaction Index derived from employee attitude survey of company, no reference Timing of outcome assessment Baseline measurement: 6 (5‐8) months before intervention phase Follow‐up measurement: 6 (4‐7) months after intervention phase
Training in control group	None
Training compliance in intervention group	Not assessed
Notes	Possible bias as intervention was not only on managers, but also on employees. Quote: “The intervention consisted of an external change agent’s spending 13 days working with the employees of a particular branch. The bulk of the time was spent with the managers (a branch manager and approximately 8 field managers who report to the branch manager and each of whom supervises about 18 technicians), although the technicians also had contact with the change agent on three occasions”
Risk of bias
Bias	Authors' judgement	Support for judgement
CBA: was the allocation sequence adequately generated	High risk	Judgement according to EPOC recommendations for CBAs
CBA: was the allocation adequately concealed	High risk	Judgement according to EPOC recommendations for CBAs
CBA: were baseline outcome measurements similar All outcomes	Unclear risk	Unclear from the text
CBA: were baseline characteristics similar	Unclear risk	Unclear from the text
CBA: were incomplete outcome data adequately addressed All outcomes	High risk	No further information about dropouts from training or from study participation, respectively
CBA: was knowledge of the allocated interventions adequately prevented during the study	High risk	See above (change agents in contact with all)
CBA: was the study adequately protected against contamination	High risk	No comparison of location characteristics
CBA: was the study free from selective outcome reporting	Unclear risk	Protocol not available
Other bias	High risk	Intervention schedule and standardisation of intervention unclear, probably different group processes during organisational development Major changes of external factors No idea of selection processes

Eastburg 1994.

Methods	Study design: cRCT Country: USA Type of industry/economic sector: healthcare, tertiary sector Size of company: not explicitly stated, 150‐bed hospital Supervisor‐employee relationship: first degree
Participants	Number of participants Supervisors (intervention): NR; supervisors (control): NR Employees (intervention): 34; employees (control): 28 Age of participants Supervisors (intervention): NR; supervisors (control): NR Employees (intervention): NR; employees (control): NR Sex of participants Male supervisors (intervention): NR; male supervisors (control): NR Male employees (intervention): NR; male employees (control): NR Inclusion criteria: NR Exclusion criteria: NR
Interventions	Type of intervention: A1 Content of intervention Individual meetings with nurse supervisor: discussion on possible reason for nurse burnout, briefing on research findings concerning burnout (particularly the lack of positive feedback), examples of positive feedback and thought game for supervisors (how to give pos. feedback in work setting), written summary about discussion points as a reminder for future Timing of intervention 1 single individual meeting, duration not stated, timing not stated Type of delivery Individual meeting (face‐to‐face teaching and discussion) with each of the supervisors (intervention group) Concept of training: not theory‐based
Outcomes	Psychomental stress: Emotional exhaustion. Scaling and polarity not reported. Measured via MBI (Maslach 1981) Depersonalisation. Scaling and polarity not reported. Measured via MBI (Maslach 1981) Personal accomplishment. Scaling and polarity not reported. Measured via MBI (Maslach 1981) Absenteeism: none Well‐being: none Timing of outcome assessment Baseline measurement: not explicitly stated Follow‐up measurement: 30 days after the intervention
Training in control group	None
Training compliance in intervention group	Not assessed
Notes	T‐tests applied were one‐sided
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	Quote: "was done randomly" Stratified by size of unit, but no details on sequence generation stated
Allocation concealment (selection bias)	Unclear risk	Unclear from the text
Blinding of participants and personnel (performance bias) All outcomes	Unclear risk	Study personnel: not blinded Supervisors: not blinded Employees: it remains unclear if subordinate nurses were informed about group allocation of their supervisor
Blinding of outcome assessment (detection bias) All outcomes	Unclear risk	Self‐assessment, unclear if employees had known the allocation of their supervisors
Incomplete outcome data (attrition bias) All outcomes	Unclear risk	There were nurses who dropped out but no info on reasons for dropout. One dropout in the first phase, 14 dropouts (19%) between baseline measurement and follow‐up
Selective reporting (reporting bias)	Unclear risk	At baseline, stress was also measured using the Nurses' Stress Scale. It remains unclear, why it was not measured after the intervention
cRCT: recruitment bias	Low risk	Randomisation following recruitment
cRCT: baseline imbalance	Low risk	Stratified randomisation regarding size of units Quote: "no significant pre‐treatment differences in burnout were found between treatment and control groups"
cRCT: loss of clusters	Unclear risk	Not stated in the text. Additionally, it remains unclear how many 'units' remained in the treatment and control groups
cRCT: incorrect statistical analysis	High risk	Multilevel analysis not applied
cRCT: comparability with RCTs	Unclear risk	Methods of statistical analysis remain unclear

Elo 2005.

Methods	Study design: CBA Country: Finland Type of industry/economic sector: maintaining and constructing streets, green areas and public buildings, tertiary sector Size of company: not explicitly stated Supervisor‐employee relationship: not permanently fixed, assuming first degree
Participants	Number of participants Supervisors (intervention): 8; supervisors (control): unclear (n = 32 in Elo 2014 (page 185), n = 43 in Elo 2005 (61 – 8 – 10 = 43)) Employees (intervention): 49; employees (control): 96 Age of participants Supervisors (intervention): NR; supervisors (control): NR Employees (intervention): 44.7 (9.0); employees (control): 43.9 (10.7) Sex of participants Male supervisors (intervention): NR; male supervisors (control): NR Male employees (intervention): 67%; male employees (control): 84% Inclusion criteria Willingness to take part Completion of the questionnaire Only those units were included in which all or none of the supervisors participated in the intervention. Exclusion criteria All employees whose supervisors were present > 0 days but < 6.5 days were excluded.
Interventions	Type of intervention: A2 Content of intervention Self‐awareness, "better communication", means with which to promote well‐being and readiness to take action in one's work unit Timing of intervention "Several" sessions 7.5 days in total 1‐3 days each "in residential" over a time period of 6 months Type of delivery Group‐based action learning, feedback, group discussions Role‐playing and short lectures Concept of training: not theory‐based
Outcomes	Psychomental stress: Emotional exhaustion. Scaling 0‐6, higher scores indicate more emotional exhaustion. 5‐items, MBI (Maslach 1996) Perceived stress symptoms. Scaling 1‐5, higher scores indicate more perceived symptoms. 1 item, (Elo 2003). Absenteeism: none Well‐being: none Timing of outcome assessment Baseline measurement: not explicitly stated Follow‐up measurement: not explicitly stated, assuming 2 years after baseline
Training in control group	None
Training compliance in intervention group	Not assessed
Notes	Concomitant intervention: organisation's stress management programme Hours of participation (mean, SD): intervention group 3.93 days (SD 4.10) vs control group 1.29 days (SD 2.02), P < 0.001
Risk of bias
Bias	Authors' judgement	Support for judgement
CBA: was the allocation sequence adequately generated	High risk	No, as groups were built according to the extent the supervisors participated in the training programme. Additionally, judgement according to EPOC recommendations for CBAs
CBA: was the allocation adequately concealed	High risk	No, as groups were built according to the extent the supervisors participate in the training programme. Additionally, judgement according to EPOC recommendations for CBAs.
CBA: were baseline outcome measurements similar All outcomes	Low risk	No Emotional exhaustion different (2.29 ± 1.49, n = 49 vs 1.63 ± 1.16, n = 96) MD = 0.66 P (2‐sided) 0.0084 (own calculation) Stress symptoms different (2.69 ± 1.07, n = 49 vs 2.14 ± 0.92, n = 96) MD = 0.55 P (2‐sided) 0.0028 (own calculation) but the statistical analysis accounted for baseline differences
CBA: were baseline characteristics similar	High risk	Significant differences concerning sex and the number of days the subordinates themselves had participated in the organisation's stress management programme
CBA: were incomplete outcome data adequately addressed All outcomes	High risk	No, there was no information about attrition, or dropouts before the training programme. In addition, there was no information about the employees whose supervisors took part ≥ 1 but < 6.5 days
CBA: was knowledge of the allocated interventions adequately prevented during the study	Unclear risk	Unclear from the text
CBA: was the study adequately protected against contamination	Unclear risk	Unclear from the text
CBA: was the study free from selective outcome reporting	Unclear risk	Unclear from the text, no protocol available
Other bias	Unclear risk	Significant difference between intervention and control groups concerning the organisation's stress management programme.

Gumuseli 2002.

Methods	Study design: RCT Country: Turkey Type of industry/economic sector: food industry, tertiary sector Size of company: large (assumption based on company) Supervisor‐employee relationship: most likely first degree
Participants	Number of participants Supervisors (intervention): 8; supervisors (control): 11 Employees (intervention): 8; employees (control): 12 Age of participants Supervisors (intervention): NR; supervisors (control): NR Employees (intervention): NR; employees (control): NR Sex of participants Male supervisors (intervention): NR; male supervisors (control): NR Male employees (intervention): NR; male employees (control): NR Inclusion criteria: willingness to take part and to help in the transfer of training (only supervisors) Exclusion criteria: not explicitly stated
Interventions	Type of intervention: A2 Content of intervention Meetings were held with the supervisors of the experimental group 1 month before the training programme (content not stated, most likely to improve "management support and guidance"). Training (not explicitly stated): "Supervisors were introduced, with notes, to both the roles and responsibilities of the managers in participants’ transferal of knowledge, skills and attitudes to the workplace, and also to some methods and techniques." "After the training program, supervisors were visited individually with the purpose of information exchange about these follow‐ups and guidance activities. During the follow‐up period, which continued for three months, supervisors were communicated with via telephone and e‐mail." Timing of intervention First meeting 1 month before training (duration not stated) Second meeting = "training" (assumption as not clearly stated in the text) For 3 months after training supervisors were "communicated" with via telephone and email. Third meeting after training (time point and duration not stated) Type of delivery Meetings (multiple supporting and guidance activities provided by 3 trainers of the company's training department over a period of 3 months) Concept of training: not theory‐based
Outcomes	Psychomental stress: none Absenteeism: none Well‐being: Job satisfaction (shortened Turkish version of the Minnesota Job Satisfaction Survey, "proven to be reliable and valid" but no overall data on validation of the Turkish version were stated). Scaling not reported, polarity not specified (assuming that higher scores indicate higher job satisfaction). Timing of outcome assessment Baseline measurement: assuming immediately before training Follow‐up measurement: 3 months after
Training in control group	None
Training compliance in intervention group	Not assessed
Notes	None
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	Technique of randomisation not stated
Allocation concealment (selection bias)	Unclear risk	Technique of randomisation not stated, allocation unclear from the text
Blinding of participants and personnel (performance bias) All outcomes	Unclear risk	It remains unclear whether subordinates were informed about group allocation of their supervisor
Blinding of outcome assessment (detection bias) All outcomes	Unclear risk	Self‐assessment. Unclear if employees had known the allocation of their supervisor.
Incomplete outcome data (attrition bias) All outcomes	Low risk	No dropouts reported
Selective reporting (reporting bias)	Unclear risk	Unclear from the text, no protocol available

Hammer 2011.

Methods	Study design: cRCT Country: USA Type of industry/economic sector: food industry/grocery stores, tertiary sector Size of company: NR, 360 employees Supervisor‐employee relationship: NR but presumably mixed, as supervisors included store directors, assistant directors, customer service managers, assistant customer service managers and department managers (predominant group)
Participants	Number of participants Supervisors (intervention): 39; supervisors (control): NR Employees (intervention): 117 (own calculation, 67% of 239); employees (control): 122 Age of participants Supervisors (intervention): 42.5 years; supervisors (control): NR Employees (intervention): mean age 38 years (for baseline participants); employees (control): not explicitly stated, 2 years younger than the intervention group Sex of participants Male supervisors (intervention): 36%; male supervisors (control): NR Male employees (intervention): NR; male employees (control): NR Inclusion criteria: NR Exclusion criteria: NR
Interventions	Type of intervention: A1 Content of intervention Computer‐based (1 h, self‐paced) supervisor training Background information on benefits of reducing work‐family conflict for employees’ and their families’ health and well‐being Organisation’s motivation for reducing work‐family conflict Information on company’s current work‐family policies and programmes Four Family‐Supportive Supervisor Behaviours (FSSB) dimensions (emotional support, instrumental support, role modelling behaviours, creative work‐family management strategies) Data on consistent perceptual gap between employees and supervisors regarding work‐family support Description of self‐monitoring programme Request to change 6 behaviours over following 3‐5 weeks Face‐to‐face training, (1‐2 days later) 60‐90 min of face‐to‐face group training (during slow time of the work day) Mostly informational 2 role plays by presenters Request to change 6 behaviours over following 3‐5 weeks: current frequency of behaviours, goal‐setting to increase behaviours Behavioural self‐monitoring through “Supervisor Daily Data Card”, instructions for behavioural self‐monitoring (not mandatory, directly after face‐to‐face training) Speak with store employees Ask something about an employee’s family Say something about the employee’s supervisor’s family Give positive feedback about employee’s work performance Suggest constructive improvement in employee’s work performance Initiate a question about or offer a way to improve an employee’s schedule Timing of intervention Sessions: 1 computer‐based, 1 face‐to‐face Duration Computer‐based training: approximately 60 min Face‐to‐face training: 60‐90 min Behavioural self‐monitoring: 2 weeks post‐training Timing: during work Type of delivery Training intervention comprising computer‐based training and face‐to‐face training Behavioural self‐monitoring (not mandatory, thus no 'on‐the‐job' intervention A2) Relevant implementation aspects Training intervention implemented as part of a company‐mandated supervisory training programme Self‐monitoring optional USD 25 gift card for supervisors/employees participating in pre‐ or post‐survey Concept of training: theory‐based
Outcomes	Psychomental stress: none Absenteeism: none Well‐being Job satisfaction. Scaling 1‐5, higher scores represent higher levels of job satisfaction. 5‐item scale by Hackman 1975 Turnover intention. Scaling 1‐5, higher scores represent higher levels of intention to quit. 2‐item scale by Boroff 1997 Timing of outcome assessment Baseline measurement: 9 months before intervention Follow‐up measurement: 1 month after intervention
Training in control group	None
Training compliance in intervention group	Supervisor training compulsory; computer‐based pre‐test and post‐test containing an identical set of 15 questions in order to assess learning and retention of material 39/39 completed computer‐based training Compliance for face‐to‐face training not stated (assuming also 100%) 32/39 volunteered for self‐monitoring Further details concerning reactions, learning, and behavioural criteria are stated in the paper (page 10)
Notes	None
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	No information provided on how randomisation was achieved
Allocation concealment (selection bias)	Unclear risk	Unclear from the text
Blinding of participants and personnel (performance bias) All outcomes	Unclear risk	Study personnel and supervisors could not be blinded, but it remains unclear from the text whether employees knew about group allocation of their supervisors
Blinding of outcome assessment (detection bias) All outcomes	Unclear risk	Self‐reported outcome but unclear whether employees were blinded to the intervention
Incomplete outcome data (attrition bias) All outcomes	Low risk	Risk of attrition bias due to only 67% response rate at post‐intervention survey; no statistically significant differences in the percentage of dropouts in stores with (35%) and without (36%) the supervisor training; dropouts had significantly lower mean job satisfaction and higher turn‐over intentions than completers. 61% response rate at baseline (n = 360 of 590, not stated); 67% (n = 239 of 360) at follow‐up assuming an overall response rate of 41%. Nevertheless probably low risk of attrition bias due to use of full‐information maximum likelihood routine, a modelling technique to estimate model parameters when data are missing‐at‐random to minimise bias
Selective reporting (reporting bias)	Unclear risk	No protocol available; physical health as composite score, probably complete All primary outcome measures (as specified in the methods section) reported, work‐family conflict measures not stated as dependent outcome variable
cRCT: recruitment bias	High risk	All employees eligible for participation, response rate of 61% across intervention and control group with no discussion of implications. No comparison between participants and non‐participants.
cRCT: baseline imbalance	Unclear risk	No significant differences on key demographic variables between control and intervention groups except for age (intervention group 2 years older). Nevertheless, the risk of bias remains unclear as baseline measurements were conducted 9 months before intervention.
cRCT: loss of clusters	Low risk	No clusters were lost
cRCT: incorrect statistical analysis	High risk	No adjustment for clustering, although reasons for not adjusting are provided
cRCT: comparability with RCTs	Unclear risk	Unclear from the text

Hardré 2009.

Methods	Study design: cRCT Country: NR Type of industry/economic sector: manufacturing and customer service, secondary/tertiary sector Size of company: not explicitly stated, (large multinational Fortune 500 company) Supervisor‐employee relationship: first degree (assumption)
Participants	Number of participants Supervisors (intervention): 12; supervisors (control): 13 Employees (intervention): 53; employees (control): 45 Age of participants Supervisors (intervention): NR; supervisors (control): NR; for both: mean 53, range 31‐60 years Employees (intervention): NR; employees (control): NR; for both: mean 33, range 21‐65 years Sex of participants Male supervisors (intervention): NR; male supervisors (control): NR; for both: 68% Male employees (intervention): NR; male employees (control): NR; for both: 41% Inclusion criteria: NR Exclusion criteria: NR
Interventions	Type of intervention: B1 Content of intervention Information and illustrations on how to support employees' workplace autonomy Information on how to support employees' autonomy Refinement of managers' efforts to support employee autonomy through Q&A (first session) information on theoretical foundations, different types of work motivation, introduction of autonomy‐supportive motivating style, empirical research findings on the topic, introduction of 4 autonomy‐supportive motivating strategies with illustrations, small group work with focus on strategies and workplace application ‐ with potential to ask questions on relevance, application, limitations etc. ‐ received booklet in the end with instructions on usage (second session) Q&A with feedback on progress of managers efforts Booklet: outlining strategies and providing examples, specifically developed for study, also with parts on theoretical foundation, advantages of the approach, and research findings Timing of intervention 1 x group‐delivered training during week 2, duration 1 h 1 x group‐delivered question‐and‐answer session during week 3, duration 1 h Self‐study using a study‐specific booklet over the 5 weeks of study Type of delivery On‐site training session, workshop experience and face‐to‐face consultations Group‐delivered informational session on how to support employees' autonomy Group‐delivered Q&A session (first session) informational session on theoretical foundations, group‐based interaction and discussion, individual study using booklet on topic ‐ this session started with overview of self‐determination theory ‐ the basis of the intervention, in the end potential to ask individual questions (second session) group‐based 1‐h Q&A follow‐up feedback focused with potential to ask individual questions Concept of training: theory‐based
Outcomes	Psychomental stress External regulation (only doing the job for money). Scaling 1‐7, higher scores reflect higher controlled motivation. 1 of 4 subscales measured via WSRQ adapted from the ASRQ (Ryan 1989), 16‐item instrument designed to assess why employees do their work. Motivation (unclear why one does the job at all). Scaling 1‐7, higher scores reflect stronger amotivation. 1 of 4 subscales measured via WSRQ adapted from the ASRQ (Ryan 1989), 16‐item instrument designed to assess why employees do their work Absenteeism: none Well‐being Intrinsic motivation (my work is so interesting). Scaling 1‐7, higher scores reflect higher intrinsic motivation. 1 of 4 subscales measured via WSRQ adapted from the ASRQ (Ryan 1989), 16‐item instrument designed to assess why employees do their work. Identified regulation (motivation to do the job based on its value and importance). Scaling 1‐7, higher scores reflect more identified motivation. 1 of 4 subscales measured via WSRQ adapted from the ASRQ (Ryan 1989), 16‐item instrument designed to assess why employees do their work. Overall autonomous motivation. Scaling 1‐4. Summarises outcomes above into 1 index by reversing the scoring for outcome 3 and 4 so that it could be averaged to reflect “high scores reflect relatively autonomous motivation while low scores reflect relatively controlled motivation”. Measured via WSRQ adapted from the ASRQ (Ryan 1989), 16‐item instrument designed to assess why employees do their work. Extent of engagement. Scaling 1‐7, higher scores means higher engagement. Measured through 4‐item scale based on Miserandino 1996’s widely used SEQ “featured the following stem used for all four items, ‘When I am on the job’ ‐ Ii try very hard, put in high effort, participate a lot, I just ask as if I am working (reverse scored) Timing of outcome assessment Baseline measurement: 1 week before intervention Follow‐up measurement: 4‐5 weeks after intervention
Training in control group	None
Training compliance in intervention group	Not explicitly reported for the whole training intervention, but "All 13 managers in the experimental group attended this follow‐up session"
Notes	T‐test were 1‐sided
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	NR "were randomly assigned", no further information on random sequence generation
Allocation concealment (selection bias)	Unclear risk	NR, unclear from the text
Blinding of participants and personnel (performance bias) All outcomes	Unclear risk	Supervisors and employees not blinded. Unclear from the text whether employees knew about group allocation of their supervisors
Blinding of outcome assessment (detection bias) All outcomes	Unclear risk	NR, self‐assessment. Unclear if employees were clear about allocation of their supervisors
Incomplete outcome data (attrition bias) All outcomes	Unclear risk	No reasons for missing data stated, dropout/non‐response rate was 33% for supervisors (10 of 30) and 60% for employees (143 of 241), also due to a "large number of participants who forgot their code number"
Selective reporting (reporting bias)	Unclear risk	No protocol available
cRCT: recruitment bias	Low risk	After randomisation of managers, employees were also selected randomly
cRCT: baseline imbalance	Unclear risk	Only overall baseline measurements stated
cRCT: loss of clusters	High risk	10 of 30 managers dropped out during follow‐up (33%)
cRCT: incorrect statistical analysis	High risk	Multilevel analysis not applied
cRCT: comparability with RCTs	Low risk	Random sample of employees of each manager

Jeon 2015.

Methods	Study design: cRCT Country: Australia Type of industry/economic sector: residential and community aged care services Size of company: approximately 4000 (whole company) Supervisor‐employee relationship: most likely first degree
Participants	Number of participants Supervisors (intervention): 50; supervisors (control): NR Employees (intervention): 750 (n=224 pat time 2); employees (control): 980 (n=303 at time 2) Age of participants Supervisors (intervention): NR; supervisors (control): NR Employees (intervention): 46.5; employees (control): 47.1 Sex of participants Male supervisors (intervention): NR; male supervisors (control): NR Male employees (intervention): NR; male employees (control): NR Inclusion criteria: care site located in New South Wales (NSW) or the Australian Capital Territory (ACT) Exclusion criteria Sites: sites undergoing or expecting major management or structural changes disagreement to random allocation Managers: disagreement of managers to take part in the intervention Employees: employed < 6 months involved in non‐direct care
Interventions	Type of intervention: A2 Content of intervention Structured manager education and support programme that starts with 4 modules covering core leadership topics delivered in 8 full‐day workshops Timing of intervention Time 1 (baseline), Time 2 is 9 months from Time 1, and Time 3 is 9 months after completion of Time 2, which is 6 months after the completion of the intervention. Type of delivery Action‐learning techniques, 360° feedback, case scenarios, one‐on‐one interactions with a programme facilitator, and individual practice improvement projects are included in the 12‐month CLiAC program, which is facilitated in the participant’s workplace. Concept of training: unclear
Outcomes	Psychomental stress: none Absenteeism: none Well‐being Job satisfaction. WDQ, 1/58 items (Nancarrow 2006) Turnover rate: information on staff turnover rates was supplied in aggregate form for each site (intervention and control) by the collaborating organisation. 6‐month rates were calculated by summing separations (voluntary and involuntary) at relevant facilities, and dividing by the mean monthly number of staff at each facility over the same time period; these rates were then converted to percentages by multiplying by 100. Timing of outcome assessment Baseline measurement: immediately before start of intervention Follow‐up measurement: first follow‐up 9 months after baseline, second follow‐up 6 months after intervention
Training in control group	None
Training compliance in intervention group	Forty‐six supervisors out of all trained supervisors (n = 50) completed the entire programme (3 resigned and 1 retired during the programme)
Notes
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	No information provided on how randomisation was exactly achieved. Although process was described as stratified and restricted by 4 variables it remains unclear.
Allocation concealment (selection bias)	Low risk	Quote: "Allocation was thus fully concealed."
Blinding of participants and personnel (performance bias) All outcomes	Low risk	Quote: "The members of the research team responsible for data entry and analysis will remain blind until completion of the main analysis." Blinding of supervisors not possible/not meaningful. Blinding of employees is very likely as supervisors signed forms agreeing not to discuss with their work team and staff any group‐specific activities
Blinding of outcome assessment (detection bias) All outcomes	Low risk	Self‐administered questionnaires, employees most likely blinded
Incomplete outcome data (attrition bias) All outcomes	Low risk	Low response rate but balanced between intervention and control group
Selective reporting (reporting bias)	Low risk	Protocol was published after the recruitment was completed. Relevant outcomes stated in the protocol were also given in the study report. Absenteeism was excluded due to difficulty in obtaining reliability
cRCT: recruitment bias	Low risk	Supervisors and employees were recruited at the same time. Employees most likely blinded
cRCT: baseline imbalance	Low risk	Stratified and restricted randomisation was used. Baseline characteristics considered not imbalanced (see Characteristics of included studies, Table Jeon 2015, table row "Participantcs"). Adjustment for baseline values not possible because too few care staff used the same coded identifier on multiple occasions: only 97 (7.5%) staff were identified as having completed the surveys at all 3 times.
cRCT: loss of clusters	Low risk	Two of the 24 sites were disqualified after Time 2 as they underwent major management changes, so no further data were collected for these sites, which were, therefore, excluded from the statistical analysis.
cRCT: incorrect statistical analysis	Low risk	Analyses account for clustering of participants
cRCT: comparability with RCTs	Low risk	Comparability is considered high

Kawakami 2005.

Methods	Study design: cRCT Country: Japan Type of industry/economic sector: computer software engineering, tertiary sector Size of company: not explicitly stated, ≥ 235 people Supervisor‐employee relationship Subordinate workers working for section chiefs; not stated but presumably mixed. Assuming first degree relationship as "managers ranked higher than section chief were excluded"
Participants	Number of participants Supervisors (intervention): 8 per protocol, 9 intention‐to‐treat; supervisors (control): 7 Employees (intervention): 75 PP, 82 ITT; employees (control): 84 PP, 84 ITT Age of participants Supervisors (intervention): NR; supervisors (control): NR Employees (intervention): mean = 32.7 (SD 7.0); employees (control): mean = 32.7 (SD 6.1) Sex of participants Male supervisors (intervention): NR; male supervisors (control): NR Male employees (intervention): 84%; male employees (control): 76% Inclusion criteria: all section chiefs and all employees Exclusion criteria: NR
Interventions	Type of intervention: A1 Content of intervention Training according to the guidelines for promoting mental healthcare in enterprises. by the Japanese Ministry of Labour Introduction: Scope and contents of the training Chapter 1: A story of worker A with depression Chapter 2: Why is mental health important in the workplace? Chapter 3: "The Guidelines for Promoting Mental Health Care in Enterprises" by the Japanese Ministry of Labour and roles of supervisors in occupational mental health Chapter 4: (A1) Listening and advising subordinate workers and using mental health services Chapter 5: (A1) Supporting workers who are returning to work after experiencing mental health problems Chapter 6: (B1) Improvement of the work environment for stress prevention Chapter 7: "Self‐care" awareness of and coping with stress Chapter 8: Essential knowledge about mental health Chapter 9: Summary Timing of intervention Total duration: 3‐5 h over 4 weeks Timing: self‐administered Supervisors were advised to study 3‐5 chapters a week and spend 2‐4 weeks in completing the entire programme Type of delivery: web‐based training Concept of training: not theory‐based
Outcomes	Psychomental stress Psychological distress. Scaling 1‐4, higher score indicates greater psychological distress. 18‐item scale from the BJSQ, gauging psychological complaints experienced by workers in the previous month. 5 subscale scores used to calculate vigour (3 items), anger‐irritability (3 items), fatigue (3 items), anxiety (3) items) and depression (6 items) (Shimomitsu 2000) Absenteeism: none Well‐being: none Timing of outcome assessment: Baseline measurement: not explicitly stated, "before" the beginning of the training Follow‐up measurement: 3 months after the training
Training in control group	Placebo intervention: training session on relaxing methods
Training compliance in intervention group	Assessed: “During the four‐week training period, a study coordinator oversaw their progress and encouraged them by e‐mail to complete the training”; supervisor’s knowledge and attitudes to work site mental health assessed
Notes	None
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	No information provided on how randomisation was achieved
Allocation concealment (selection bias)	Unclear risk	Unclear from the text
Blinding of participants and personnel (performance bias) All outcomes	Unclear risk	Blinding of study personnel not reported. Supervisors in intervention group were aware of being part of intervention group and thus, risk of bias is unclear. As supervisors in the intervention group were asked not to discuss the training with anyone else (although compliance with this recommendation is unclear), employees were most likely blinded and thus, we consider bias due to non‐blinding of employees to be low.
Blinding of outcome assessment (detection bias) All outcomes	Low risk	Self‐reported outcome and subordinates likely to be blinded (see above)
Incomplete outcome data (attrition bias) All outcomes	Low risk	Very low attrition rates are unlikely to cause bias. Dropouts: 18 employees in the treatment group (18%), 6 employees in the control group (7%)
Selective reporting (reporting bias)	Unclear risk	No protocol available; BJSQ used and presumably all BJSQ outcomes reported
cRCT: recruitment bias	Low risk	All section chiefs recruited into study
cRCT: baseline imbalance	Unclear risk	Some baseline imbalances in terms of composition of subordinates (% female, functions in company, supervisor support); significance of these imbalances in terms of causing bias unclear
cRCT: loss of clusters	Low risk	One section chief not trained; ITT and PP analysis show similar results and loss of cluster is therefore unlikely to have caused bias
cRCT: incorrect statistical analysis	High risk	No adjustment for clustering of subordinates by section chief
cRCT: comparability with RCTs	Unclear risk	Unclear

Kawakami 2006.

Methods	Study design: cRCT Country: Japan Type of industry/economic sector: office machine sales and service, tertiary sector Size of company: not explicitly stated, ≥ 297 people Supervisor‐employee relationship: not stated (assuming first and second degree, section chiefs and work group chiefs)
Participants	Number of participants Supervisors (intervention): 21 PP, 22 ITT; supervisors (control): 23 PP/ITT Employees (intervention): 81; employees (control): 108 Age of participants Supervisors (intervention): NR; supervisors (control): NR Employees (intervention): mean = 31.0 (SD 6.5); employees (control): mean = 32.0 (SD 6.0) Sex of participants Male supervisors (intervention): NR; male supervisors (control): NR Male employees (intervention): 65%; male employees (control): 76% Inclusion criteria: NR, all section chiefs and all employees included Exclusion criteria: NR
Interventions	Type of intervention: A1 Content of intervention Training according to the guidelines for promoting mental health care in enterprises. by the Japanese Ministry of Labour Introduction: Scope and contents of the training Chapter 1: A story of worker A with depression Chapter 2: Why is mental health important in the workplace? Chapter 3: "The Guidelines for Promoting Mental Health Care in Enterprises" by the Japanese Ministry of Labour and roles of supervisors in occupational mental health Chapter 4: (A1) Listening and advising subordinate workers and using mental health services Chapter 5: (A1) Supporting workers who are returning to work after experiencing mental health problems Chapter 6: (B1) Improvement of the work environment for stress prevention Chapter 7: "Self‐care" awareness of and coping with stress Chapter 8: Essential knowledge about mental health Chapter 9: Summary Timing of intervention Total duration: 3‐5 h over 4 weeks Timing: self‐administered Type of delivery: web‐based training Concept of training: not theory‐based
Outcomes	Psychomental stress Psychological distress. Scaling 1‐4, higher score indicates greater psychological distress. 18‐item scale from the BJSQ, gauging psychological complaints experienced by workers in the previous month. 5 subscale scores used to calculate vigour (3 items), anger‐irritability (3 items), fatigue (3 items), anxiety (3) items) and depression (6 items) (Shimomitsu 2000) Absenteeism: none Well‐being: none Timing of outcome assessment: Baseline measurement: not explicitly stated, "before" the beginning of the training Follow‐up measurement: 3 months after the training
Training in control group	Placebo intervention: training session on relaxing methods
Training compliance in intervention group	Assessed: "During the four‐week training period, a study coordinator oversaw their progress and encouraged them by e‐mail to complete the training"; supervisor’s knowledge and attitudes to work site mental health assessed
Notes	None
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Random assignment of workplaces using a random number table
Allocation concealment (selection bias)	Unclear risk	Unclear from the text
Blinding of participants and personnel (performance bias) All outcomes	Unclear risk	NR, unclear from the text, blinding not meaningful/possible
Blinding of outcome assessment (detection bias) All outcomes	Unclear risk	Self‐reported outcome, and not clear whether subordinates were blinded to supervisor training
Incomplete outcome data (attrition bias) All outcomes	Low risk	Very low attrition rates are unlikely to cause bias. Dropouts: 4/85 subordinates in treatment group (5%), 6/114 in control group (5%)
Selective reporting (reporting bias)	Unclear risk	Unclear from the text, no protocol available
cRCT: recruitment bias	Low risk	All section and work group chiefs recruited into study. Recruitment was made after randomisation, but nearly all (22/23) supervisors agreed to take part in the study.
cRCT: baseline imbalance	Unclear risk	Some baseline imbalances in terms of composition of subordinates (% female, age); significance of these imbalances in terms of causing bias unclear. "There was a difference in characteristics of workers and some of the job stressors at baseline", each outcome was controlled for corresponding baseline values of the same variable but not for other baseline values.
cRCT: loss of clusters	Low risk	Follow‐up of workplaces complete; one section chief died before follow‐up assessment; overall, loss of clusters is unlikely to have caused bias
cRCT: incorrect statistical analysis	High risk	No adjustment for clustering of subordinates by supervisor or workplace
cRCT: comparability with RCTs	Unclear risk	Unclear

Ketelaar 2017.

Methods	Study design: cRCT Country: The Netherlands Type of industry/economic sector: a university medical centre, and a university Size of company: not stated Supervisor‐employee relationship: most likely first degree
Participants	Number of participants Supervisors (intervention): 22; supervisors (control): 32 Employees (intervention): 123; employees (control): 150 Age of participants Supervisors (intervention): NR; supervisors (control): NR Employees (intervention): 42 ± 11; employees (control): 44 ± 11 Sex of participants Male supervisors (intervention): NR; male supervisors (control): NR Male employees (intervention): 14%; male employees (control): 16% Inclusion criteria Eligible supervisors have a minimum age of 18 years, work ≥ 24 h/week. Eligible employees have a minimum age of 18 years and work ≥ 24 h/week. Exclusion criteria Supervisors and employees whose contracts will end within 1 year from baseline or who are not able to fill out questionnaires in the Dutch language were excluded
Interventions	Type of intervention: A2 Content of intervention: Meeting 1 Step 1 supervisor addresses the employee’s work functioning problems due to health complaints or risk of sick leave and informs the employee about the 'participatory approach' protocol Preparation Step 2 employee makes an inventory of his or her work tasks and activities, prioritises work functioning problems regarding these activities, and thinks of possible solutions for the 2 most important work functioning problems Step 3 supervisor makes an inventory of the employee’s work tasks and activities, prioritises work functioning problems regarding these activities, and thinks of possible solutions for the 2 most important work functioning problems Meeting 2 Step 4 supervisor and employee discuss work functioning problems and possible solutions, and assess the applicability of these solutions Step 5 supervisor and employee agree on an action plan to realise solutions Realisation Step 6 solutions are prepared and realised Meeting 3 Step 7 supervisor and employee evaluate the action plan and the realised solutions Timing of intervention: see below Type of delivery: The multifaceted strategy to implement the participatory approach was applied in the intervention group and consisted of 3 components, following the baseline measurement (month 1): (A) 1 working group meeting per organisation with stakeholder representatives (month 2); (B) supervisor training (4 h + 2 h optional) in application of the participatory approach (month 3); and (C) optional supervisor coaching (months 4–12) Concept of training: unclear
Outcomes	Psychomental stress: none Absenteeism: sick leave Number of episodes on sick leave in last 6 months (frequency) Total number of work days on sick leave during last 6 months (total duration Well‐being: none Timing of outcome assessment: Baseline measurement: immediately before intervention Follow‐up measurement: 6 months after baseline
Training in control group	Other type of intervention: distribution of written information about the participatory approach intervention
Training compliance in intervention group	Not assessed
Notes	None
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	Randomisation at department level but clusters were matched as pairs based on the number of sick leave frequencies
Allocation concealment (selection bias)	Unclear risk	NR/unclear
Blinding of participants and personnel (performance bias) All outcomes	Unclear risk	Blinding of participants, supervisors, and employees not possible/not meaningful
Blinding of outcome assessment (detection bias) All outcomes	Unclear risk	Frequency and duration of sick leave were self‐reported. In contrast, in the protocol, sick leave should also have been analysed using administrative data. Although previous research has found that self‐reported data on sick leave closely corresponds to administrative data (31), this does not take into account that taking part in a study aiming to prevent or reduce sick leave might cause participants to unconsciously underestimate their sick leave at 6 months’ follow‐up.
Incomplete outcome data (attrition bias) All outcomes	Low risk	27%‐33% dropouts but balanced between intervention and control group.
Selective reporting (reporting bias)	Low risk	6‐month data were stated in the protocol, 12‐month data were not reported but were presumably not gathered
cRCT: recruitment bias	High risk	Supervisors and employees were recruited at the same time, but decision of departments was on voluntary basis; invitation of supervisors biased (mail to all supervisors vs selected supervisors by department head
cRCT: baseline imbalance	Low risk	Type of organisation was imbalanced at baseline but adjusted for in the multilevel multivariable analysis
cRCT: loss of clusters	Unclear risk	Number of clusters at baseline and follow‐up NR
cRCT: incorrect statistical analysis	Low risk	Analyses account for clustering of participants
cRCT: comparability with RCTs	Low risk	Comparability is considered high

Norman 2003.

Methods	Study design: cRCT Country: USA Type of industry/economic sector: technology company, assuming tertiary sector Size of company: NR Supervisor‐employee relationship: first degree
Participants	Number of participants Supervisors (intervention): 10; supervisors (control): 16 Employees (intervention): 20; employees (control): 34 Age of participants Supervisors (intervention): proportion of age groups (years) 25‐29 (20%) 30‐34 (66%) 35‐39 (13%) Supervisors (control): proportion of age groups (years) 25‐29 (21%) 30‐34 (52%) 35‐39 (10%) 40‐44 (5.3%) 45‐49 (10.5%) Employees (intervention): proportion of age groups (years) 20‐24 (15%) 25‐29 (40%) 30‐34 (25%) 35‐39 (10%) 40‐44 (10%) Employees (control): proportion of age groups (years) 20‐24 (15%) 25‐29 (55%) 30‐34 (17%) 35‐39 (7%) 40‐44 (2%) 45‐49 (2%) Sex of participants Male supervisors (intervention): 80%; male supervisors (control): 65% Male employees (intervention): 75%; male employees (control): 92% Inclusion criteria Leaders: all front‐line managers of company; randomly selected from each branch voluntary participation in complete training and study willingness to complete the entire training programme and all training homework Employees: all immediate subordinates of participating leaders (range 0‐9 subordinates) voluntary participation in study Exclusion criteria: leaders having vacation plans conflicting with training programme schedule
Interventions	Type of intervention: A2 Content of intervention: 10 leadership behaviours assumed to build trust with employees; application to day‐to‐day management with subordinates supervised by personal coaches. Give recognition to each subordinate, have scheduled coaching sessions with each subordinate, write a development plan with each subordinate, use communication skills such as listening and giving balanced feedback, apply the company values, apply values consistently, make the effort to understand employee values, set objectives with each subordinate, review objectives against accomplishments, delegate meaningful assignments Timing of intervention 5 days in‐class teaching (9‐5): days 1 + 2, 3 + 4, 5 in 3 blocks with a delay of 1 month; 3 weekly individual coaching sessions after each block of in‐class teaching; Duration of coaching sessions not stated Timing of in‐class teaching and coaching not precisely stated Total duration: 3 months Type of delivery In‐class seminar with discussion; post‐session assignments; individual coaching (with 3 other programme participants) Concept of training: not theory‐based
Outcomes	Psychomental stress: none Absenteeism: none Well‐being Job satisfaction. Scaling 1‐5, higher scores indicate higher level of job satisfaction. 7 Items, Job Satisfaction Scale of Job Diagnostic Survey (Hackman 1974), α = 0.86 Organisational commitment. Scaling 1‐5, higher scores indicate higher affective organizational commitment. 9 items (9 stated in the appendix, 8 stated in the text), Affective Organizational Commitment Scale (Meyer 1993), α = 0.83 Affective well‐being at work. Scaling 1‐5, higher scores indicate better affective well‐being at work. Revised version of Warr's Affective Well‐being at Work Scales (Daniels 1997), 5 factors/scales: comfort (vs anxiety): 6/7 items; pleasure (vs depression): 6/7 items; positive affect: 7 items; tiredness: 2 items; anger: 3 items Timing of outcome assessment Baseline measurement: 2 weeks before Follow‐up measurement: 3 months after beginning of the intervention
Training in control group	None
Training compliance in intervention group	Assessed: 33/35 leaders (intervention and later control group) completed the training and coaching
Notes	None
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	"Randomly selected", but unclear how randomisation was done
Allocation concealment (selection bias)	Unclear risk	Unclear from the text
Blinding of participants and personnel (performance bias) All outcomes	High risk	No blinding of primary investigator. Subordinates were informed about training days of their supervisors
Blinding of outcome assessment (detection bias) All outcomes	Unclear risk	The data were self‐reported and it remains unclear whether blinding of employees was broken
Incomplete outcome data (attrition bias) All outcomes	Low risk	Statistical analysis with complete data only (see also dropouts) low dropout rate (6/60, 10%)
Selective reporting (reporting bias)	Unclear risk	No protocol available
cRCT: recruitment bias	Unclear risk	Supervisors were recruited before randomisation. No information about participation rates and representativity of subordinates
cRCT: baseline imbalance	Low risk	No significant differences regarding the measured baseline characteristics: age, sex, ethnicity, education, years at company, years with manager (own calculation from Appendix A table 1)
cRCT: loss of clusters	Low risk	2/35 clusters were lost between randomisation and follow up (9%)
cRCT: incorrect statistical analysis	High risk	Statistical analysis did not account for clustering
cRCT: comparability with RCTs	Unclear risk	Unclear

Odle‐Dusseau 2016.

Methods	Study design: CBA Country: USA, Mid‐Atlantic region Type of industry/economic sector: elderly care, 8 separate locations (retirement communities that were part of 1 organisation) Size of company: not stated Supervisor‐employee relationship: not stated – probably first degree
Participants	Number of participants Supervisors (intervention): 86; supervisors (control): 36 Employees (intervention): 93; employees (control): 50 Age of participants Supervisors (intervention): NR; supervisors (control): NR Employees: 45 (no SD given; not differentiated between intervention and control group) Sex of participants Male supervisors (intervention): NR; male supervisors (control): NR Employees: 124 female, 19 males (87% female, 13% male) Not differentiated between intervention and control group Inclusion criteria: not explicitly stated Exclusion criteria: initial needs analysis for FSSB training (perceived levels of FSSB not too high) – no statements concerning the actual exclusion of employees
Interventions	Type of intervention: B2 Content of intervention Face‐to face training workshop: general information & reflection on FSSB strategies in own work active listing to FSSB‐related topics (Hammer 2011) interactive exercise (recall experiences with employees facing a work‐family conflict dilemma; analysis of reaction as a supervisor and mapping to FSSB dimension; generate list of alternative strategies) measure of participant reactions to the training Behavioural self‐monitoring cards: motivate supervisors to transfer learned strategies into daily work provide daily estimates (for 2 weeks following face‐to‐face training on behaviours exhibited that were reflective of each of the 4 dimensions of FSSB (Hammer 2011: emotional support, instrumental support, role‐modelling behaviours, creative work‐family management) return cards to first author Timing of intervention 1 session for each supervisor intended (first author offered 2 delivery times at each location to accommodate supervisor schedules); session duration approximately 3 h (= total duration); timing not stated Approximately 10‐14 sessions intended (= provide daily estimates for 2 weeks following face‐to‐face training); duration and timing not stated Type of delivery Face‐to face training workshop Behavioural self‐monitoring cards Concept of training: theory‐based
Outcomes	Psychomental stress: none Absenteeism Turnover intention, 3 items assessing thoughts and intentions to leave the organisation (Chapman 1991). 1 item to assess a behavioural component of intention to leave; no citation, only item wording given: “I am presently seeking to change jobs” Well‐being Job satisfaction: 2 items from a 3‐item scale by Friedman 2000 + 2 items from the Michigan Organizational Assessment Questionnaire (Cammann 1979) Organisational commitment: 4 items from the 6‐item Affective Commitment Scale (Allen 1990) Employee engagement: 4 items from a scale used in past research on job engagement, e.g. Britt 2001 Timing of outcome assessment Baseline measurement: 7 months before intervention Follow‐up measurement: 1 month after intervention
Training in control group	None
Training compliance in intervention group	Assessed: measure of participant reactions to training sessions “overwhelmingly positive”; number of cards returned as indirect measure of compliance (return rate of only 10%)
Notes	None
Risk of bias
Bias	Authors' judgement	Support for judgement
CBA: was the allocation sequence adequately generated	High risk	Allocation was left completely to supervisors; control group not explicitly planned for; selection effects very likely. Judgement according to EPOC recommendations for CBAs.
CBA: was the allocation adequately concealed	Low risk	Study authors state that employees were not aware whether their supervisors had received training; no other information available
CBA: were baseline outcome measurements similar All outcomes	Unclear risk	NR
CBA: were baseline characteristics similar	Unclear risk	NR
CBA: were incomplete outcome data adequately addressed All outcomes	Low risk	Full maximum likelihood estimation is among the best performing methods to account for missing data
CBA: was knowledge of the allocated interventions adequately prevented during the study	Unclear risk	NR
CBA: was the study adequately protected against contamination	High risk	No; ceteris paribus limitations on p. 306: possible bias due to history effects, maturation effects, selection effects
CBA: was the study free from selective outcome reporting	Low risk	The measures that were contained in the survey according to the authors equal those reported in the paper
Other bias	Unclear risk	n.a.

Pomerantz 1992.

Methods	Study design: cRCT Country: USA Type of industry/economic sector: State Government, tertiary sector Size of company: NR Supervisor‐employee relationship: first degree
Participants	Number of participants Supervisors (situational leadership): 10; supervisors (micro‐counselling): 11; supervisors (control): 12 Employees (situational leadership): 42; employees (micro‐counselling): 41; employees (control): 54 Age of participants Supervisors (situational leadership): mean = 40.6 (SD = 7.27) Supervisors (micro‐counselling): mean = 42.9 (SD = 7.44) Supervisors (control): mean = 45.4 (SD = 7.21) Employees (situational leadership): NR; employees (micro‐counselling): NR; employees (control): NR Sex of participants Male supervisors (situational leadership): 60%; supervisors (micro‐counselling): 55%; male supervisors (control): 58% Male employees (situational leadership): NR; employees (micro‐counselling): NR; male employees (control): NR Inclusion criteria Employed by Governmental department within the State of California Supervisors with 3‐11 subordinates Currently not engaged in a grievance First come, first served Exclusion criteria: NR
Interventions	Intervention 1 (situational leadership) Type of intervention: A1 (situational leadership) Content of intervention: To have students understand how to use the situational leadership model To have students assess their subordinates' level of readiness for a leadership style To have students be able to understand how to implement the situational leadership model at work Timing of intervention: 1 day of training Type of delivery: multimodal training: lecture, discussion, group review, film presentation, fish bowl Concept of training: not theory‐based Intervention 2 (micro‐counselling) Type of intervention: A1 (micro‐counselling) Content of intervention: training of micro‐counselling (active listening, feedback): responding effectively to people that present a problem, confusion, or frustration; communicating feedback, advice, and directions clearly Timing of intervention: 1 day (ca. 8.30–4.30), structured lesson plan: lecture, brainstorming, discussion, practice at California State training centre Type of delivery: micro‐counselling training adapted from Microtraining Associates (Ivey 1979) field‐tested training: lecture, video‐based training, discussion, writing, exercise Concept of training:
Outcomes	Psychomental stress Emotional exhaustion. Scaling 0‐6, higher scores indicate higher level of emotional exhaustion. 6 Items, Emotional Exhaustion Scale of the MBI (Maslach 1996) Absenteeism: none Well‐being: none Timing of outcome assessment Baseline measurement: no baseline measurement performed Follow‐up measurement: 1 month after training
Training in control group	None
Training compliance in intervention group	Situational leadership: 10/12 supervisors completed training. The effectiveness of the training was measured by pre‐ and post‐tests, participant satisfaction with the training and the instructor, and participant confidence to implement the behaviour learned in the training at work Micro‐counselling: 11/12 supervisors completed training. The effectiveness of the training was measured by pre‐ and post‐tests, participant satisfaction with the training and the instructor, and participant confidence to implement the behaviour learned in the training at work
Notes
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	36 managers were randomly assigned to treatment groups (situational leadership or micro‐counselling) and control group; all subordinates per manager were invited to participate. Random sequence generation not stated
Allocation concealment (selection bias)	Unclear risk	Unclear from the text
Blinding of participants and personnel (performance bias) All outcomes	High risk	No blinding of primary investigator or supervisors possible, information about supervisor participating in study, unclear information about group allocation of supervisor
Blinding of outcome assessment (detection bias) All outcomes	Unclear risk	Self‐assessment of subordinates and it remains unclear if blinding of employees was broken
Incomplete outcome data (attrition bias) All outcomes	Unclear risk	Unclear from the text, statistical analysis with complete data only
Selective reporting (reporting bias)	Unclear risk	No protocol available
cRCT: recruitment bias	Unclear risk	Unclear from the text, no information about participation rates and representativity of subordinates
cRCT: baseline imbalance	High risk	No information on subordinates at baseline
cRCT: loss of clusters	Unclear risk	1 supervisor of intervention group (micro‐counselling) and 2 supervisors of alternative intervention group (situational leadership)
cRCT: incorrect statistical analysis	High risk	No pairwise comparison tests between groups are provided (main effect of analysis of variance NR), no adjustment for clustering of participants
cRCT: comparability with RCTs	Low risk	Full randomisation of managers into treatment groups vs control group
Other bias	High risk	Only follow‐up measurement of emotional exhaustion of subordinates, baseline and further details of subordinate groups unknown

Romanowska 2011.

Methods	Study design: cRCT Country: Sweden Type of industry/economic sector: different professional areas (education, medical care, police, culture, religious service, business, IT, other) Size of company: NR Supervisor‐employee relationship: first degree
Participants	Number of participants Supervisors (intervention): 23; supervisors (control): 25 Employees (intervention): 37‐41; employees (control): 56‐58 Age of participants Supervisors (intervention): NR; supervisors (control): NR Employees (intervention): NR; employees (control): NR Age of supervisors and employees together 51 years in intervention group Age of supervisors and employees together 47 years in control group Sex of participants Male supervisors (intervention): NR; male supervisors (control): NR Male employees (intervention): NR; male employees (control): NR Inclusion criteria Leaders: formal management position with ≥ 4 subordinates voluntary participation in leadership intervention and study (3 assessments) Subordinates: no long work absences same leader during study period voluntary participation in study (3 assessments) Exclusion criteria: NR
Interventions	Type of intervention: A1 Content of intervention Schibbolet (art‐based) method focusing on reflection about universal themes (e.g. love, abuse of power) stimulated by a performance with music or literary text fragments without continuous or logical context in order to induce tension and surprise and to provide understanding, meaningfulness, changes of perspective Timing of intervention 3 days conventional leadership training 12 art‐based leadership sessions at 3 h 1 year duration Type of delivery: art‐based leadership intervention: writing, witnessing a performance, group reflections Concept of training: theory‐based
Outcomes	Psychomental stress Emotional exhaustion. Scaling 1‐6, higher scores indicate higher levels of emotional exhaustion. 5 items, Emotional Exhaustion Scale from MBI (Maslach 1996) Depressive symptoms. Scaling 1‐5, higher scores indicate more depressive symptoms. 6 items, derived from SCL‐90 (Lipman 1986) Sleep disturbance. Scaling 1‐6, higher scores indicate more disturbed sleep. 4 items, Sleep Disturbance Summary Measure from the Karolinska Sleep Questionnaire (Akerstedt 2002) Total poor mental health. Scaling 1‐(unclear if 5 or 6), higher scores indicate more disturbed sleep. Composite index of 15 items on emotional exhaustion, depressive symptoms, sleep disturbances Absenteeism: none Well‐being: none Timing of outcome assessment: Baseline measurement: start of intervention Follow‐up measurement: 18 months after
Training in control group	Type of Training: B1 Content of Training "Conventional" leadership training for 3 days together with the intervention group Democracy, employee participation, group functioning, communication, feedback, etc. The theoretical approach was transformational leadership, group psychology, organisation and leadership theories Timing of training: 12 sessions, 3 h each, over a 1‐year period Type of delivery Leadership training programme: lectures, group discussions, group process exercises, individual assignments Concept of training: theory‐based
Training compliance in intervention group	Not assessed
Notes	None
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	Pairwise randomisation of leaders matched for gender, age, occupation. Randomisation procedure NR
Allocation concealment (selection bias)	Unclear risk	Unclear from the text
Blinding of participants and personnel (performance bias) All outcomes	High risk	Researchers were blinded for randomisation (but only for the randomisation process), blinding was otherwise not possible
Blinding of outcome assessment (detection bias) All outcomes	Unclear risk	Unclear if subordinates were informed about allocation
Incomplete outcome data (attrition bias) All outcomes	High risk	Analysis of complete data only, dropouts reported in detail. Supervisor dropouts 14/48 (29%), dropout rate of employees not estimable
Selective reporting (reporting bias)	Unclear risk	No protocol available
cRCT: recruitment bias	High risk	Each leader selected 4 subordinates (selection bias). Recruitment was done before randomisation, but "Each participating leader was asked to select 4 of their subordinates"
cRCT: baseline imbalance	Unclear risk	Baseline characteristics of employees not reported. Employees were selected by the participating supervisors
cRCT: loss of clusters	Low risk	Detailed description of dropouts (see above); no relevant differences between responders and non‐responders
cRCT: incorrect statistical analysis	High risk	Missing report of values (M, SD, t‐tests) of subordinates. Multilevel modelling not applied
cRCT: comparability with RCTs	Unclear risk	Unclear from the text
Other bias	High risk	High risk of selection bias of leadership training participants as well as subordinates of leaders

Scandura 1984.

Methods	Study design: CBA Country: USA Type of industry/economic sector: service industry, tertiary sector Size of company: NR, but described as "large" Supervisor‐employee relationship: first degree
Participants	Number of participants Supervisors (intervention): NR; supervisors (control): NR Employees (intervention): 26; employees (control): 57 Age of participants Supervisors (intervention): NR; supervisors (control): NR Employees (intervention): NR; employees (control): NR "Most employees over 40 years of age" Sex of participants Male supervisors (intervention): NR; male supervisors (control): NR Male employees (intervention): NR; male employees (control): NR "Most employees female" Inclusion criteria: voluntary participating employees and their immediate supervisors Exclusion criteria: NR
Interventions	Type of intervention: A1 Content of intervention Enable and encourage supervisors to improve dyadic exchange relations with their members by active listening, exchanging mutual expectations, exchanging resources, and practicing a series of one‐on‐one‐sessions between unit manager and member (about 30 min) Normative LMX model and how to use it (2 h) Active listening skill training (2 h) Exchanging mutual expectations (2 h) Exchanging resources (2 h) Practicing one‐on‐one sessions (4 h) Timing of intervention Six 2‐h sessions (over a 6‐week period) Type of delivery LMX treatment: seminar setting, lecture, discussion, role modelling/role play Concept of training: theory‐based
Outcomes	Psychomental stress: none Absenteeism: none Well‐being Job satisfaction. Scaling not reported, polarity not reported (probably higher values indicating higher levels of overall job satisfaction). 4 items, short form of the Hoppock Job Satisfaction Scale (Hoppock 1935) Timing of outcome assessment Baseline measurement: shortly before intervention Follow‐up measurement: 26 weeks after baseline
Training in control group	Placebo intervention: sessions of general input, probably lectures, general information about job enrichment, performance evaluation, decision making, and communication, three 2‐h sessions
Training compliance in intervention group	Not assessed
Notes	Concomitant intervention: three 2‐h sessions (placebo) of general input (job enrichment, performance evaluation, decision making, and communication), balanced between groups
Risk of bias
Bias	Authors' judgement	Support for judgement
CBA: was the allocation sequence adequately generated	High risk	Judgement according to EPOC recommendations for CBAs. Allocation of employees to condition (intervention, control) by work units. Method of assignment to treatment or control not stated
CBA: was the allocation adequately concealed	High risk	Judgement according to EPOC recommendations for CBAs. Allocation by unit, but knowledge about intervention possible. No concealment
CBA: were baseline outcome measurements similar All outcomes	Low risk	Standardised self‐report. Baseline differences in job satisfaction were adjusted using multivariate analysis of variance
CBA: were baseline characteristics similar	High risk	No statistical reporting of systematic comparisons. Not reported and baseline differences in LMX were adjusted during MANOVA
CBA: were incomplete outcome data adequately addressed All outcomes	High risk	High dropout rate Intervention group: 26% (9/35) Control group: 12% (8/65)
CBA: was knowledge of the allocated interventions adequately prevented during the study	Unclear risk	Allocation by unit, but knowledge about intervention condition possible. Unclear from the text
CBA: was the study adequately protected against contamination	Unclear risk	Unclear from the text
CBA: was the study free from selective outcome reporting	Unclear risk	No protocol available
Other bias	High risk	Unknown sample size of supervisors (intervention, control), assumably 1:N ratio (supervisor: employees), all working in same department, suspicious rate of employee participation (98%) combined with monitoring of quantitative and qualitative performance

Takao 2006.

Methods	Study design: cRCT Country: Japan Type of industry/economic sector: brewery, secondary sector Size of company: not explicitly stated, at least 301 people Supervisor‐employee relationship: first degree
Participants	Number of participants Supervisors (intervention): 23 PP, 24 ITT; supervisors (control): 22 Employees (intervention): 134; employees (control): 92 Age of participants Supervisors (intervention): 50.0 (SD 5.2); supervisors (control): 48.9 (SD 4.5) Employees (intervention): > 50 years: 42 (31%) 35‐49 years: 28 (21%) < 34 years: 64 (48%) Employees (control): > 50 years: 22 (24%) 35‐49 years: 13 (14%) < 34 years: 57 (62%) Sex of participants Male supervisors (intervention): 96%; male supervisors (control): 100% Male employees (intervention): 66%; male employees (control): 70% Inclusion criteria: not explicitly stated, all supervisors and employees included Exclusion criteria: not explicitly stated. For analysis: missing information on age, occupation, or psychological distress measurement
Interventions	Type of intervention: A1 Content of intervention Supervisor education programme (60 min) through lecture Significance of positive mental health Roles of supervisors relating to positive mental health in the workplace (A1) Awareness of the mental problems of subordinates and how to respond to them (A1) Support for employees return to work (A1) Working environment improvements (B1) Self‐care recommendations Accurate knowledge regarding mental problems Active listening training (180 min) (A1) through lecture (60 min) and (A1) role‐playing exercise (120 min). Reference to detailed reporting of education programme reported in Tsutsumi 2005. Timing of intervention Number of sessions: single session Duration of session: 60 min plus 80 min Part 1: 60 min Part 2: 60‐min lecture + 120‐min practice session Type of delivery Lecture, role‐playing exercise Part 1: face‐to‐face education programme, once Part 2: lecture + practice session Concept of training: not theory‐based
Outcomes	Psychomental stress Psychological distress. Scaling 1‐4, higher score indicates greater psychological distress. 18‐item scale from the BJSQ, gauging psychological complaints experienced by workers in the previous month. 5 subscale scores used to calculate vigour (3 items), anger‐irritability (3 items), fatigue (3 items), anxiety (3) items) and depression (6 items) (Shimomitsu 2000) Absenteeism: none Well‐being: none Timing of outcome assessment Baseline measurement: not explicitly stated, "before" the beginning of training Follow‐up measurement: 3 months after training
Training in control group	None
Training compliance in intervention group	Not assessed
Notes	Concomitant intervention All employees: informational brochure on mental health All supervisors: guidelines for the promotion of employee mental health, encouragement to improve workplace environment according to guidelines
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	Random assignment of supervisors; no specification of how randomisation was conducted
Allocation concealment (selection bias)	Unclear risk	Unclear from the text
Blinding of participants and personnel (performance bias) All outcomes	High risk	Quote: "We could not conduct a blind intervention due to the nature of the study". No blinding as supervisors/subordinates in intervention and control group often worked together.
Blinding of outcome assessment (detection bias) All outcomes	High risk	Self‐reported outcome and no blinding of employees
Incomplete outcome data (attrition bias) All outcomes	Low risk	Very low attrition rates and reasons for dropping out of study described (e.g. retirement, sick leave); unclear why so many data points are lost for analysis, presumably due to incomplete questionnaires and/or due to non‐first degree subordinate relationship; some potential to cause some bias. Overall response rate for primary outcome of employees: 87% in intervention group, 91% in control group
Selective reporting (reporting bias)	Unclear risk	No protocol available
cRCT: recruitment bias	Low risk	All supervisors and subordinates recruited into study; not clear how possible second degree relationships (i.e. non‐immediate subordinates) were handled
cRCT: baseline imbalance	Unclear risk	Statistically significant differences in terms of age, type of occupation (i.e. more blue‐collar workers in intervention group) and in terms of job demand and control (lower among intervention group); significance of these imbalances in terms of causing bias unclear
cRCT: loss of clusters	Low risk	One supervisor in intervention group did not participate in study for unknown reasons; otherwise, no supervisors in either intervention or control groups were lost to follow‐up; overall, loss of clusters is unlikely to have caused bias
cRCT: incorrect statistical analysis	High risk	No adjustment for clustering of subordinates by supervisor
cRCT: comparability with RCTs	Unclear risk	Unclear

Weir 1997.

Methods	Study design: cRCT Country: Canada Type of industry/economic sector: healthcare service, tertiary sector Size of company: NR, 310‐bed hospital Supervisor‐employee relationship: NR, assuming first degree
Participants	Number of participants Supervisors (intervention): 7; supervisors (control): 6 Employees (intervention): 86; employees (control): 78 Age of participants Supervisors (intervention): NR; supervisors (control): NR Employees (intervention): NR, "approximately 44 years"; employees (control): NR, "approximately 44 years" Age of employees is only reported for all responders (n = 201): M = 44.74, SD = 8.72 Sex of participants Male supervisors (intervention): NR; male supervisors (control): NR Male employees (intervention): NR; male employees (control): NR Sex of employees is only reported for all responders (n = 201): 1/201 (0.5%) male Inclusion criteria: NR Exclusion criteria: NR
Interventions	Type of intervention: B2 Content of intervention Coaching sessions: problem orientation problem definition and formulation generation of alternatives decision‐making solution implementation and evaluation Workshops: learn how to use a consultant develop confidence and clear expectations concerning leadership and manager role Purpose: leadership development to improve morale, quality of case and reduce absenteeism Goal: to facilitate a decentralised and participatory style of problem‐solving management meetings (participative decision‐making) vs more traditional hierarchical approach to decision‐making. Assumption that staff would be committed, satisfied and motivated, which would translate into quality of care and work morale if they were included in decision‐making. Process consultations: to gain insight into relationship of staff members to achieve decentralised way of decision‐making. Aim was to increase communication, member roles and functions in groups, group problem‐solving and decision‐making, group norms and group growth, leadership and authority, and inter‐group co‐operation and competition to achieve effective organisational performance. Consultants were advisers and examined communication issues, decision‐making processes, authority relationships, unit atmosphere, and satisfactions and dissatisfactions with work etc. Expected stages of intervention: nurturing problem‐solving perspective realistic goal setting and formulation of expectations assistance of staff nurses in generating alternative plans to improve staff relationships help staff nurses evaluate consequences of altered behaviour implementation of solutions and evaluation of outcomes Group consultations: to unite everyone involved in endeavours ‐ discussion of engagement processes & goals and alternative strategies, consequences and strategies elaborated and discussed Timing of intervention Three 1‐day workshops at 6, 9, and 12 months group consultations of day workshops Unspecified number of coaching sessions. Type of delivery Individual coaching sessions (meetings) One‐day workshops Nurse manager‐consultant problem‐solving meetings, process consultations (set of activities that helps client to perceive, understand and act on process events that occur in client environment), observation of daily events and to learn from them Concept of training: theory‐based?
Outcomes	Psychomental stress: none Absenteeism Absenteeism: total hours of absence for all the resident staff members on the unit were summed and divided by the total paid hours of work for full‐time equivalent positions for the study year. Well‐being: none Timing of outcome assessment Baseline measurement: not explicitly stated, immediately before the intervention Follow‐up measurement: after completion of training (at 12 months)
Training in control group	None
Training compliance in intervention group	Assessed: "on the basis of the dose or pattern of contacts (high‐low) of the consultative process", "whether or not the nurse managers engaged in a sufficient number of meetings to have allowed the consultation to have any effect", "divided [arbitrarily?] into complier and non‐complier"
Notes	None
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	'Randomly assigned', but no information on sequence generation stated
Allocation concealment (selection bias)	Unclear risk	Unclear from the text
Blinding of participants and personnel (performance bias) All outcomes	Unclear risk	Unclear from the text, Quote: "Staff members of consenting units were advised of the objectives of the study and that their nurse manager and unit had a 50/50 chance of being selected to receive the consultation"
Blinding of outcome assessment (detection bias) All outcomes	Low risk	Absenteeism considered as objective outcome
Incomplete outcome data (attrition bias) All outcomes	High risk	Overall response rate of employees 23% (164/712) Quote: "with regard to the total sample, the representativeness of the respondents and the direction of bias on any conclusions and effectiveness of treatment cannot be ascertained" Response rate at unit level before randomisation: 13/15 units
Selective reporting (reporting bias)	Unclear risk	No protocol available, information on training meetings (number, attendance rate, number of phone consultations, etc.) were prespecified, but not stated; low compliance was stated, but not further specified
cRCT: recruitment bias	Unclear risk	Unclear from the text
cRCT: baseline imbalance	Low risk	Quote: "At baseline, the responder experimental and control groups of subjects were equivalent in their demographic, social, and work variables and in their mean hours of absence."
cRCT: loss of clusters	Low risk	No clusters dropped out
cRCT: incorrect statistical analysis	High risk	Quote: "Statistical methodology for this cluster design study was not possible to conduct because the clusters were not able to be matched"
cRCT: comparability with RCTs	Unclear risk	Unclear

A1: supervisor on‐the‐job training for supervisor‐employee interaction; A2: supervisor off‐the‐job training for supervisor‐employee interaction; ASRQ: Academic Self‐Regulation Questionnaire; B1: supervisor on‐the‐job training for work environment design; B2: supervisor off‐the‐job training for work environment design; BJSQ: Brief Job Stress Questionnaire; CBA: controlled before‐after study; cRCT: cluster‐randomised controlled trial; EPOC: Cochrane Effective Practice and Organisation of Care (Group); FSSB: Family‐Supportive Supervisor Behaviours; GHQ‐12: 12‐item General Health Questionnaire; HADS: Hospital Anxiety and Depression Scale; ITT: intention‐to‐treat; LMX: Leadership‐Member Exchange; MBI: Maslach Burnout Inventory; MD: mean difference; NA: not applicable; NR: not reported; PP: per protocol; SCL‐90: Hopkins Symptom Checklist; SD: standard deviation; SEQ: Student Engagement Questionnaire; UWES: Utrecht Work Engagement Scale; WDQ: Workforce Dynamics Questionnaire; WSRQ: Workplace Self‐Regulation Questionnaire

Characteristics of excluded studies [ordered by study ID]

Study	Reason for exclusion
Atwater 2006	Ineligible design (no control group)
Beaton 2001	CBA with only 1 treatment and one control site
Carron 1964	No relevant outcomes
Cascio 2007	Only an overview article
Clark 1985	No relevant outcomes
Cummings 2018	Ineligible design
Donohoe 1998	No relevant outcomes measured in subordinates, < 2 control sites
Eden 2000	Ineligible intervention: training aimed at supervisors alone, not at enhancing supervisor‐employee interaction; quality of relevant measures doubtful (no references given)
Greenberg 2006	Ineligible design
Haland‐Haldorsen 1997	Intervention was not exclusively for supervisors
Hammer 2015	Intervention was not exclusively for supervisors, and outcomes were not suitable
Hansen 2016	Ineligible design
Heaney 1995	Intervention was not exclusively for supervisors
Helphinstine 1993	Outcome of employees of trained and untrained supervisors were not investigated
Hetherington 1987	No control group
Hocking 1983a	No relevant leadership training intervention
Hocking 1983b	No relevant leadership training intervention
Kawakami 1997	Intervention was not exclusively for supervisors and included a stress management programme provided by supervisors and others to employees
Kaya 2010	Ineligible design, not interventional study
Kelly 2014	No suitable outcomes
Leister 1977	No primary outcome was measured
Lewis 2012	No primary outcomes reported
Logan 2005	Outcome was not measured in subordinates
Maddi 1998	Outcome was not measured in subordinates
Madede 2017	Intervention was not exclusively for supervisors
Majchrzak 1986	No relevant outcomes (as unauthorised absence from the Marine Corps is not comparable to absenteeism from work)
Mauno 2006	CBA with only one intervention and one control site
Maxwell 2005	Outcomes were not measured at employee‐level
Mullen 2009	No relevant outcomes
Mönninghoff 2008	2 CBAs but each included < 2 intervention or control sites
Napier 1985	Relevant outcome (satisfaction) was not assessed using a validated questionnaire
Nordstrom 1990	Ineligible design, no control group. No relevant outcomes
Olson 2013	Intervention was not exclusively for supervisors
Patterson 2010	Review
Rappe 2007	No primary outcome measured in employees
Reynolds 1997	CBA. Only one intervention site
Romanowska 2013	No primary outcomes
Romanowska 2014	No primary outcomes
Sirianni 2001	Ineligible design, no control group
Terres 1984	Intervention is not leadership training
Theorell 2001	CBA. Only one intervention site
Torp 2008	No relevant intervention
Tsutsumi 2005	Groups were defined only after the intervention according to the attendance rate of supervisors

CBA: controlled before‐after study

Characteristics of studies awaiting assessment [ordered by study ID]

Dimoff 2018.

Methods	Cluster‐randomised controlled trial (cRCT)
Participants	37 leaders from a small publishing company and a small property management company (24 intervention group, 13 control group), and their 82 employees (60 intervention group, 22 control group).
Interventions	Intervention group: leader‐focused, 3‐hour mental health training; Control group: wait list
Outcomes	At employee level: perceptions of leaders’ communication about mental health and resources, perceptions of leaders’ consideration for struggling employees, employees' willingness to use resources, and actual resource use 6 weeks and 12 weeks after training.
Notes	–

Gonzalez‐Morales 2016.

Methods	Quasi‐experimental field study using controlled before–after (CBA) design
Participants	Intervention group: 23 supervisors from 4 restaurants, 208 employees Control group: Unknown number of supervisors from 4 restaurants, 241 employees
Interventions	Training in 4 supportive supervision strategies (benevolence, sincerity, fairness, and experiential processing) during four 2‐hour sessions over a period of 2 months.
Outcomes	Perceived supervisor support (four modified items from the Survey of Perceived Organizational Support scale) and abusive supervision (six items from the Negative Acts Questionnaire–Revised) 9 months after training.
Notes	–

Hammer 2019a.

Methods	Cluster‐randomised controlled trial (cRCT)
Participants	Intervention group: 65 supervisors from 16 U.S. military service organisations, 203 employees (veterans) Control group: 46 supervisors from 19 U.S. military service organisations, 158 employees (veterans)
Interventions	Intervention group: Veteran‐Supportive Supervisor Training containing three parts: (a) computer‐based training, (b) behaviour tracking in the workplace to improve transfer of training knowledge, (c) supplementary activities titled “Above and Beyond” (additional mini‐modules on specific topics, such as military leave, as well as participation in a moderated online forum discussion); Control group: wait list
Outcomes	Perceived health (4 items), overall functional impairment (14 items), job performance (3 items), turnover intentions (2 items), social support (4 scales), 3 months and 9 months post baseline data collection.
Notes	–

Hammer 2019b.

Methods	Cluster‐randomised controlled trial (cRCT)
Participants	20 workgroups from a municipal public works department (11 assigned to intervention, 9 to control) with 195 construction workers (125 in intervention group, 70 in control group)
Interventions	Intervention group: Safety and Health Improvement Program (SHIP) containing 3 parts: (a) 1‐hour computer‐based training to teach supervisors ways to better support worker safety and work‐life challenges, (b) supervisor behavioural self‐monitoring to facilitate transfer of training to practice, and (c) team‐based discussions with supervisors and work crew members to identify challenges and opportunities for improvement; Control group: wait list
Outcomes	Family‐supportive supervisor behaviours (4 items), perceptions of effectiveness of team processes (7 items), perceptions of work‐life effectiveness (3 items), supervisor and co‐worker support (2 items) measured after 6 months.
Notes	–

Differences between protocol and review

Based on peer‐review comments, we slightly restructured and extended the 'Background' section by moving and adjusting arguments from the first two sections of the Discussion to the Background section in order to improve the focus of the review from the very beginning.

In contrast to the protocol, we introduced the term 'cRCTs' to denote cluster‐randomised controlled trials as a convenient abbreviation.

In contrast to the protocol, we no longer considered peer‐reviewed journals to be the only source of valid measurement tools. Instead, we also considered validated questionnaires, and scales published in books and test manuals to be of sufficient psychometric quality.

In contrast to the protocol, we also used CBAs for drawing conclusions.

In contrast to the protocol, we replaced the database Embase with Scopus. This was due to the fact that the Information Specialists at the University of Eastern Finland (Kaisa Hartikainen and Heikki Laitinen) no longer had access to Embase and the editors (Jani Ruotsalainen and Jos Verbeek) agreed to this change because Scopus includes all references in Embase. We planned to contact experts in the research field to identify other sources of data. In the light of our own knowledge and experience in the research field, we considered this to yield no substantially new findings.

Since publication of the protocol, Christian Seubert (CS) joined the team to offer further assistance in completing the review in agreement with the editors.

We used the GRADEpro GDT software to assess the quality of evidence (GRADEpro GDT 2015).

Due to the limited number of included studies, we did not assess reporting bias according to the recommendations given in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2017).

After discussion with the Cochrane Work Group, we changed the hierarchy of the categorisation variables. The hierarchy in the protocol was intervention types (training aiming to improve supervisor‐employee interaction, either off‐the‐job (type A1) or on‐the‐job (type A2), or training aimed at supervisors' designing of the work environment, either off‐the job (type B1) or on‐the‐job (type B2)), comparison groups (no intervention, placebo, other training), outcomes (stress, absenteeism, well‐being), follow‐up timings (short‐, mid‐, long‐term), and study designs (RCT, cRCT, CBA). In this review, intervention types were switched to the bottom of the hierarchy (using subgroup analyses). Hence, the hierarchy in this review is comparison groups, outcomes, follow‐up timing, and study designs. In contrast to the protocol, we did not use the intervention for categorisation and combined all intervention types in the meta‐analyses.

Contributions of authors

Conceiving the protocol: AK, JG, DN

Designing the protocol: AK, JG, ER, EVE

Co‐ordinating the protocol: AK

Designing search strategies: AK, JG, as well as Leena Isotalo, Heikki Laitinen, and Kaisa Hartikainen from Cochrane Occupational Safety and Health

Writing the protocol: AK, JG

Providing general advice on the protocol: ER, DN, EVE

Securing funding of the protocol: JG, DN

Abstract and full‐text screening: AK, CS, JG

Data extraction and analysis: AK, CS, ER, JG, DN, ER, EVE

Risk of bias analysis: AK, CS, ER, EVE, JG

Writing the manuscript: AK, CS, JG

Critical review of the manuscript: ER, EVE, DN

Sources of support

Internal sources

• Ludwig‐Maximilians University Munich, Germany.

  Salary for Dennis Nowak and Eva Rehfuess

• University of Innsbruck, Austria.

  Salaries for Jürgen Glaser and Christian Seubert

• Cochrane Switzerland, Institute for Social and Preventive Medicine, Switzerland.

  Salary for Erik von Elm

• Technical University of Munich, Germany.

  Salary for Andreas Kuehnl

External sources

• No sources of support supplied

Declarations of interest

AK: None known

CS: None known

ER: None known

EVE: None known

DN: None known

JG: None known

Edited (no change to conclusions)