Changes in exposure to positive leadership behaviours and subsequent changes in workers’ depressive symptoms and risk of onset of antidepressant treatment

Abstract

Objectives

Using consecutive waves of a Danish nationwide survey, we investigated in a quasi-experimental design whether changes in positive leadership behaviours were associated with subsequent changes in depressive symptoms and risk of antidepressant treatment.

Methods

Sample 1 consisted of 6062 workers participating in 2012, 2014 and 2016. Sample 2 consisted of 15 619 workers participating in either 2012–2014 or 2016–2018. We measured leadership behaviours by an eight-item scale and depressive symptoms by the Major Depression Inventory (MDI) and linked register data on purchase of antidepressants. Using linear and logistic regression, we adjusted for covariates and estimated the association between changing levels of leadership behaviours and subsequent depressive symptoms and onset of antidepressant treatment.

Results

In sample 1, changing from high to low levels of exposure to positive leadership behaviours from 2012 to 2014 was associated with a higher MDI score in 2016 (sample 1, regression coefficient: 1.12, 95% CI 0.68 to 1.56), whereas changing from low to high levels was associated with a lower MDI score (−1.84, 95% CI −2.51 to −1.17). In sample 2, ORs for antidepressant treatment ranged from 1.06 to 1.21 with wide CIs that included unity.

Conclusions

This study suggests that a decrease in positive leadership behaviours has a hazardous effect on workers’ level of depressive symptoms, whereas an increase has a beneficial effect. There were no clear associations with antidepressant treatment.

WHAT IS ALREADY KNOWN ON THIS TOPIC.

WHAT THIS STUDY ADDS

• Using consecutive waves of a Danish nationwide survey, we showed in a quasi-experimental design that a decrease in exposure to positive leadership behaviours was associated with a subsequent higher depressive symptom score, whereas an increase in positive leadership behaviours was associated with a subsequent lower depressive symptom score.

• There were no clear associations between changes in leadership behaviours and subsequent risk of treatment with antidepressants.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY

• The results of this study suggest that improving leadership behaviours at work might contribute to an improved population mental health in terms of reducing depressive symptoms.

Introduction

Mental health conditions are on the rise worldwide, and recent findings from the Global Burden of Disease Study showed that depressive disorder accounts for the largest disease burden in terms of disability-adjusted life years.¹ One suggested mechanism in the aetiology of depressive disorder is stressful conditions,² and large-scale systematic reviews have documented associations between specific working conditions and risk of depressive disorder, although the causality of the associations is still debated.³,⁵

Stressful working conditions are described in the job demands-resources (JD-R) model.⁶ The JD-R model posits that work demands put strain on the worker that over time can lead to illness, but also that resources may reduce (buffer) the strain caused by the demands by increasing motivation and positive affect towards work, and enabling fulfilment of work goals. Positive leadership behaviours can be considered such a resource.⁷ Furthermore, lack of positive leadership behaviours, such as lack of appreciation, may be a stressor in itself as it may violate basic psychological needs of employees.⁸

Indeed, exposure to low levels of positive leadership behaviour at the workplace has been associated with increased symptoms of mental health problems of workers; however, only a few studies examined the association with a longitudinal design.⁹,¹⁴ Some studies have further examined the association of exposure to different types of positive leadership behaviours and risk of depressive disorders, but results have been inconsistent.¹⁵,²³ To the best of our knowledge, no study so far has examined whether changes in leadership behaviours are related to subsequent changes in mental health symptoms or onset of treatment for mental disorders.

In this paper, we investigate whether an increase or decrease in positive leadership behaviours is associated with subsequent changes in depressive symptoms and risk of treatment with antidepressants. The positive leadership behaviours in our study entail behaviours such as clear information about the relation between the workers’ tasks and the goals of the organisation, appropriate decision latitude, the leaders’ involvement in the workers’ professional development, involvement in the planning of the work, feedback as needed, recognition, support and trust.

The aim of this study is twofold: first, to study whether a change in positive leadership behaviours is associated with change in the level of depressive symptoms and second, to investigate the association between changes in leadership behaviours and risk of onset of treatment with antidepressants.

Methods

Participants and procedure

The study participants were derived from the Work Environment and Health in Denmark (WEHD) study.^{24 25} WEHD consisted of a nationwide sample of Danish workers and was conducted in four waves in 2012, 2014, 2016 and 2018. In short, the waves included both randomly selected new participants in each wave and a cohort of participants that took part in multiple waves. The participants filled in a comprehensive questionnaire on employment and working conditions and their self-rated health. Sampling and details pertaining to WEHD are described in further details elsewhere.^{24 25} We linked the survey data of all participants via civil registration numbers with Danish national health register data.

We constructed two analytic samples; sample 1 for the outcome of depressive symptoms and sample 2 for the outcome of onset of antidepressant treatment. Online supplemental figures 1 and 2 depict the selection process of the two samples. Briefly, sample 1 consisted of 8762 individuals who responded to all three WEHD waves in 2012, 2014 and 2016. We excluded respondents who either had no leader or a missing value on the leadership behaviour scale and individuals with missing values on key variables, yielding a sample of 6062 workers (online supplemental figure 1). Sample 2 consisted of 29 906 individuals who responded in two consecutive waves, either in 2012 and 2014 (n=11 781) or in 2016 and 2018 (n=18 125). If individuals were represented in both pairs, we kept only the first pair. After excluding individuals with no leader or missing values on the leadership behaviour scale, individuals with missing values on key variables, and individuals who either had a history of treatment for depressive, manic or bipolar disorders or had a Major Depression Inventory (MDI) score ≥21 points at baseline, the sample consisted of 15 619 workers.

Measures

The Leadership Index (WEHD-LI)

The leadership index consisted of eight items from the WEHD questionnaire. All items addressed positive leadership behaviours and asked ‘how often’ the behaviour occurred. A full list of the items is provided in the online supplemental textbox 1. Each item was rated on a 5-point Likert scale, with the possible answers ‘1=Always’, ‘2=Often’, ‘3=Sometimes’, ‘4=Rarely’, ‘5=Never’. We created a sum score, ranging from high to low levels of the leadership behaviours, yielding a scale ranging from 8 to 40 points. The measure was validated as a formative measure in a previous paper on a larger sample of WEHD (the validation is reported in the supplementary material of the paper²⁶). Because of the formative nature of the leadership index, we did not impute missing values on the index, but excluded respondents with missing values.

Depressive symptoms

We measured depressive symptoms with the MDI, a clinically validated 10-item self-administered rating scale to assess both a continuous depressive symptom score and to identify individuals with a probable depressive disorder during the last 2 weeks.²⁷ The MDI score ranges from 0 to 50, with a higher score indicating more depressive symptoms. Previous studies have shown that the continuous MDI score is a strong predictor of sickness absence, including MDI scores that were below any cut-off points for clinical depressive disorder.^{28 29}

Onset of antidepressant treatment

To assess onset of antidepressant treatment, we used redemption of a prescription of antidepressant medication at a Danish pharmacy, ascertained in the Danish National Prescription Registry³⁰ by the Anatomical Therapeutic Chemical code (ATC-code) N06A.

Covariates

As covariates, we included variables which in earlier studies were associated with risk of depressive symptoms.^{31 32} All covariates were retrieved from national registers and were linked to the survey data using the participants’ civil registration number.³³ Sociodemographic covariates included sex, age, educational level, income level, cohabitation with a partner or spouse, and living together with children under age 12. Educational level was categorised into ‘low’ (<10 years of education), ‘medium-low’ (10–12 years), ‘medium-high’ (13–15 years) and ‘high’ (>15 years). Household income level was weighted according to the composition and size of the household and categorised into quintiles. We included job type, defined by the 10-level categorisation of DISCO08 (the Danish version of the ISCO08 classification), industry, defined by the 10-level categorisation of NACE DB07,³⁴ and industrial sector (public vs private sector). We included a covariate of working hours below 30 hours a week, as an indicator of part-time employment.

As indicators for possible predisposition or vulnerability for developing a depressive disorder, we included: (1) history of an affective disorder before baseline assessment, defined as either a diagnosis of depressive disorder, manic episode or bipolar affective disorder registered in the Danish National Patient Register³⁵ (International Classification of Diseases, 10th Revision (ICD-10) codes F30–F33; ICD-8 codes 296, 790.2 or 300.4 (ICD-9 was never used in Denmark)) or purchase of an antidepressant registered in the Danish National Prescription Registry (ATC-code N06A); (2) history of a mental disorder other than affective disorder, ascertained in the Danish National Patient Register (any F diagnosis in ICD-10 or diagnoses 290 to 315 in ICD-8) and (3) Any mental disorder of a participant’s parent, ascertained in the Danish National Patient Register (any F diagnosis in ICD-10 or diagnoses 290–315 in ICD-8).

Statistical analysis

Depressive symptoms

To investigate whether change in leadership behaviours was associated with subsequent change in the level of depressive symptoms, we stratified sample 1 into two strata. Stratum 1 included participants with a high leadership behaviour score in 2012, whereas stratum 2 included participants with a low leadership behaviour score. We defined a high level of leadership behaviour by scoring ≥24 points on the WEHD-LI index. The WEHD-LI index ranged from 8 to 40 points and a value of ≥24 corresponds to an average response of ‘sometimes’ or more to the eight items of WEHD-LI. See online supplemental textbox 1 and an earlier validation study of WEHD-LI²⁶ for details.

Within stratum 1, we compared changes in depressive symptom score among participants who experienced a change from high leadership score in 2012 to low leadership score in 2014 to participants who maintained a high leadership score (reference group). In stratum 2, we compared changes in depressive symptom score among participants who experienced a change from low leadership score in 2012 to high score in 2014 to participants who maintained a low leadership score (reference group). We ascertained change in depressive symptom score in both 2014 (contemporaneous change) and 2016 (long-term change) in a linear model. We adjusted for sex, age, educational level, cohabitation, children under age 12, income level, job type, job sector, public or private sector employment, working <30 hours, registration with diagnosed depressive disorder, manic episode or bipolar disorder at a psychiatric hospital before WEHD 2012, registration with purchase of antidepressants before WEHD 2012, registration with any other mental disorder before 2012, parent registered with any mental disorder before 2012, leadership behaviour score in 2012, and MDI score in 2012. Figure 1 depicts the analytic approach.

Figure 1. Visualisation of the creation of the comparison groups in the two strata.

We conducted three sensitivity analyses where we (1) excluded participants who were registered with purchase of antidepressants, or diagnosed with depressive disorder, manic episode or bipolar disorder before 2012 or with an MDI score above a clinical level in 2012 (MDI score≥21), (2) further excluded participants with sub-clinical levels of depressive disorder in 2012 (MDI score≥13) and (3) adjusted for whether or not participants had changed their workplace during follow-up. We measured workplace change during follow-up with information retrieved from the WEHD questionnaires. See online supplemental textbox 2 for details.

Onset of antidepressant treatment

In sample 2, we investigated whether change in leadership behaviour was associated with subsequent risk of onset of antidepressant treatment. We compared this risk among (a) those whose exposure changed from high levels to low levels of the leadership behaviours from t₀ (first wave, either 2012 or 2016) to t₁ (the waves 2 years later) to those with persistent high leadership behaviour (reference group) and (b) those whose exposure changed from low to high levels of the leadership behaviours from t₀ and t₁ to those with persistent low leadership behaviour (reference group). We analysed the purchase of antidepressants both for the time period between t₀ and t₁ (contemporaneous onset) and for the time period after t₁ (long-term effect onset). We estimated ORs with logistic regression analysis, adjusted for sex, age, educational level, cohabitation, having children under age 12, public or private sector employment, working <30 hours, registration with other mental illness before t₀, parents registered with any mental disorder before t₀, the continuous leadership score at t₀ and the continuous MDI score at t₀. For the 2012–2014 waves, we had 2 years follow-up on purchase of antidepressants after 2014. For the 2016–2018 waves, it was only possible to follow the participants up until 1 January 2019. To accommodate for the difference in follow-up time, we further adjusted for wave-pair in this analysis.

Results

Samples description

Sample 1 consisted of 6062 participants. Of the 4332 participants with high levels of the leadership behaviours in 2012, 772 (18%) reported change to low levels in 2014. Of the 1730 participants who had low levels of the leadership behaviours in 2012, 676 (39%) reported change to high levels in 2014.

Sample 2 included 15 619 participants. Of the 11 626 participants with high levels of leadership behaviours at t₀, 1905 (16%) reported change to low levels at t₁. Of the 3993 participants with low levels of leadership behaviours at t₀, 1686 (42%) reported change to high levels at t₁. We identified 147 new cases of purchase of antidepressants between t₀ and t₁, and a further 167 new cases in the follow-up period after t₁.

Table 1 gives an overview of the key sample characteristics. Women were slightly over-represented in both samples (55% and 53% in sample 1 and sample 2, respectively). Mean age was 46.9 years (SD: 9.2) in sample 1 and 46.5 years (SD: 10.4) in sample 2. The educational group with the largest number of participants was ‘medium-low education’ in both groups (42.4% and 42.2%, respectively). In 2014 the mean MDI score was 7.7 (SD: 7.3) in sample 1 and 6.8 (SD: 6.2) in sample 2 at t₁. For a detailed overview of the distribution of the different covariates in the samples, see online supplemental table 1.

Table 1. Distribution of selected covariates in the two study samples.

	Sample 1				Sample 2
	n	%	Mean	SD	n	%	Mean	SD
Total	6062	100.00			15 619	100.00
Sex
Men	2704	44.61			7363	47.14
Women	3358	55.39			8256	52.86
Age			46.9	9.2			46.5	10.4
Educational level
Low	669	11.04			1722	11.03
Medium low	2567	42.35			6586	42.17
Medium high	2015	33.24			5037	32.25
High	811	13.38			2274	14.56

In both study samples, there were small but statistically significant differences between the compared groups for the initial conditions at baseline, both for the leadership scale and the MDI score, also after adjusting for the covariates. Details are displayed in online supplemental table 2.

Depressive symptoms in 2014 and 2016

Table 2 shows the results for changes in the continuous depressive symptom score. Participants who changed from exposure to high leadership behaviour levels in 2012 to low levels in 2014 had a 3.10 point higher MDI score (95% CI 2.69 to 3.51) in 2014 and a 1.12 point higher MDI score in 2016 (95% CI 0.68 to 1.56) compared with participants with persistent exposure to high leadership behaviour levels in the most adjusted model.

Table 2. Persisting and changing leadership behaviours from 2012 to 2014 and differences in depressive symptom score (MDI score) in 2014 and 2016, stratified by high and low leadership scores at baseline (sample 1, n=6062).

	Groups		Difference in MDI score 2014 Adjusted model*		Difference in MDI score 2016 Adjusted model*
	n	%	Est.	95% CI	Est.	95% CI
(A) High level of positive leadership behaviours in 2012
Persistent high level in 2012 and 2014	3560	82.2	0	Ref.	0	Ref.
Change from high level in 2012 to low level in 2014	772	17.8	3.10	(2.69 to 3.51)	1.12	(0.68 to 1.56)
(B) Low level of positive leadership behaviours in 2012
Persistent low level in 2012 and 2014	1054	60.9	0	Ref.	0	Ref.
Change from low level in 2012 to high level in 2014	676	39.1	−3.10	(−3.74 to –2.46)	−1.84	(−2.51 to to −1.17)

^*Model is adjusted for sex, age, educational level, cohabitation, children under age 12, income level, job type, industry sector, public or private sector, working <30 hours, registered diagnosis of depressive disorder, manic episode or bipolar affective disorder or prescribed antidepressants or purchase of prescribed antidepressants before WEHD 2012, registered diagnosis with other mental disorder before 2012, having a parent with a registered diagnosis of any mental disorder before 2012 and leadership index and MDI score in 2012.

MDI, Major Depression Inventory; WEHD, Work Environment and Health in Denmark.

Participants who changed from exposure to low leadership behaviour levels in 2012 to high levels in 2014 had a 3.10 point lower MDI score (95% CI −3.74 to –2.46) in 2014 and a 1.84 point lower MDI score in 2016 (95% CI −2.51 to –1.17) compared with participants with persistent exposure to low leadership behaviour levels in the most adjusted model.

When we excluded in sensitivity analyses participants with (1) history of depressive disorder at or before baseline and (2) indication of subclinical depression at baseline, the estimates attenuated but remained statistically significant (online supplemental tables 3 and 4). When we further adjusted for workplace change during follow-up, the estimates changed only marginally (online supplemental table 5).

Onset of antidepressant treatment

When we used registered purchase of antidepressants as the outcome, CIs were wide and none of the estimates were statistically significant (table 3). When we further adjusted for workplace change during follow-up, the estimates changed only marginally (online supplemental table 6)

Table 3. Persisting and changing leadership behaviours and subsequent risk of onset of antidepressant treatment between t₀ and t₁ (sample 2, n=15 619) and 2 years after t₁ (subset of sample 2, n=15 472).

Onset of purchase of antidepressants between t0 and t1	n	% of strata	New cases		Adjusted model*
			Antidepressants (t0 to t1)	% of group	OR	95% CI
(A) High level of positive leadership behaviours at t0
Persistent high level at t0 and t1	9721	83.6	76	0.78	1	Ref.
Change from high level at t0 to low level at t1	1905	16.4	18	0.94	1.12	(0.66 to 1.92)
(B) Low level of positive leadership behaviours at t0
Persistent low level at t0 and t1	2307	57.8	30	1.30	1	Ref.
Change from low level at t0 to high level at t1	1686	42.2	23	1.36	1.06	(0.60 to 1.86)
Onset of purchase of antidepressants after t 1 †
	n	% of strata	Antidepressants after t 1	% of group	OR	95% CI
(A) High level of positive leadership behaviours at t0
Persistent high level at t0 and t1	9645	83.6	94	0.97	1	Ref.
Change from high level at t0 to low level at t1	1887	16.4	21	1.11	1.12	(0.69 to 1.85)
(B) Low level of positive leadership behaviours at t0
Persistent low level at t0 and t1	2277	57.8	29	1.27	1	Ref.
Change from low level at t0 to high level at t1	1663	42.2	23	1.38	1.21	(0.68 to 2.13)

^*Model is adjusted for sex, age, educational level, cohabitation, children under age 12, public or private sector, working <30 hours, registered diagnosis with other mental disorder before t₀, having a parent with a registered diagnosis of any mental disorder before t₀ and leadership index and MDI score in t₀.

^†Analysis of use of antidepressants after t₁: Participants with use of antidepressants between t₀ and t₁ were excluded.

MDI, Major Depression Inventory; t₀, Year 2012 for the 2012–2014 pair, year 2016 for the 2016–2018 pair; t₁, Year 2014 for the 2012–2014 pair, year 2018 for the 2016–2018 pair.

Discussion

Summary of results

In this nationwide sample of workers, changing in exposure from high levels of positive leadership behaviours in 2012 to low levels in 2014 was associated with higher levels of depressive symptoms both in 2014 and 2 years later in 2016, when compared with participants with persistent high levels of positive leadership behaviours. Changing in exposure from low to high levels of positive leadership behaviours compared with persistent low levels was associated with a decrease in depressive symptoms both in 2014 and 2016. Thus, ‘losing’ positive leadership behaviours was associated with a negative change in workers’ mental health, whereas ‘gaining’ positive leadership behaviours was associated with a positive change in mental health. The changes in depressive symptoms were stronger in 2014 than in 2016, suggesting a stronger contemporaneous or short-term effect and a weaker long-term effect of changes in leadership behaviour. However, both contemporaneous/short-term and long-term changes were statistically significant.

We did not find a clear association between exposure to changes in leadership behaviour and onset of antidepressant treatment. CIs for all estimates were wide and included unity, and analyses were challenged by a low number of new cases.

Interpretation and comparison with previous research

To the best of our knowledge, this is the first study that examined whether an increase or decrease in positive leadership behaviours over time can affect the level of depressive symptoms and risk of antidepressant treatment. Our results suggest that positive leadership behaviours, characterised by providing different resources to the workers, have a positive effect on workers’ mental health, whereas lowering these positive leadership behaviours is potentially harmful to workers’ mental health. This overall conclusion is in agreement with previous cross-sectional and longitudinal studies that examined exposure to specific leadership and management styles and behaviours at one point in time and workers’ mental health.^{9 10}

A lack of association between leadership behaviours and risk of antidepressant treatment has been reported in several previous studies.^{17 18 20 23} However, our research group recently reported that low levels of positive leadership behaviours in WEHD, measured at one point in time, were associated with an increased risk of pharmaceutical treatment for depressive and anxiety disorders.²² We could not replicate this finding in the present study that analysed changes in exposure.

There are several possible explanations to this. First, our previous results on leadership behaviours and antidepressants might have been an overestimation of the association, for example, if respondents had a depressive disorder that was not treated and identified at baseline when leadership behaviours were measured. This bias towards overestimation was less likely in the present study, where we analysed change in leadership behaviours. Second, there were only a limited number of new cases of antidepressant treatment, namely 18–23 cases in the different groups of changes in leadership behaviours, thus the analyses might have been statistically underpowered to detect associations.

Strengths and limitations

Strengths of the study include the study design with multiple measurements of the exposure that allowed us to analyse change in exposure, the possibility to adjust for numerous covariates, the size of the sample, and that the participants were derived from a large nationwide sample of Danish workers; thus, the results were not restricted to specific industries. The register data further allowed us not only to analyse change in depressive symptoms but also onset of antidepressant treatment. The study design is stronger than previous ones, both including previous cross-sectional studies and longitudinal studies that measured leadership behaviours only once. We conducted the analyses within strata based on leadership behaviour levels at baseline to heighten the interchangeability within the strata. If there were specific time-invariant confounding variables that were deterministically predictive of general high levels or low levels of leadership behaviours, for instance personality factors or ability to form good relationships with superiors, the state of these confounders would likely be more similar within the strata at baseline. Thus, these possible confounders were likely less influential on the results in our study, compared with a design where only one measurement point was used to identify the level of exposure and groups of comparison.

However, although the study design was stronger than in previous studies, it was not as strong as we had hoped. The level of interchangeability between the comparison groups was lower than we had expected. The investigation of the initial conditions showed that the different comparison groups displayed different mean levels in both the leadership behaviour score and the MDI score at baseline (online supplemental table 2). It therefore seems that the comparison groups were already different at baseline, before the change in leadership behaviour occurred, while we expected them to be similar. This indicates that the attempt to increase interchangeability between the comparison groups by our study design, and thereby lower the influence of bias, was only partly met. We accounted for this statistically by adjusting for leadership behaviour index and level of depressive symptoms in the analyses.

It is another limitation that we do not know why the leadership index changed. There could be many reasons: change of job for both the worker and/or the leader, change in the organisational strategy concerning leadership, new practices from the leadership because of different circumstances, or leadership training. Different reasons for change in the rating of the leadership behaviours could maybe even be effect modifiers. We took workplace change into account in a supplementary analysis and observed that estimates did not change substantially. The wording of the included items made it further difficult to analytically differentiate between the resource and the importance of the leader being the specific provider of the resource. It is therefore unclear whether the leader was the actor or if the general lack of a resource was the key factor.

It is a further limitation of this study that we only had access to the participants’ own rating of positive leadership behaviours. Group-level measures based on multiple workers’ rating of the same leaders were not available in this study. It would have been preferable to have both individual and group-level measures to achieve more accurate ratings of leadership behaviour.

Implications for policy and practice

The recently published WHO guidelines on ‘Mental health at work’ pointed to the key contribution of working conditions to mental health and called for using the workplace as an arena to promote population mental health.³⁶ The results of our study suggest that improving leadership behaviours at work might contribute to such an improved population mental health in terms of reducing depressive symptoms in the workforce.

Conclusion

This study suggests that increasing positive leadership behaviours at the workplace has a beneficial effect on workers’ level of depressive symptoms and risk of depressive disorder, whereas a decrease in these leadership behaviours has a hazardous effect. Our study is the first study to investigate this with a quasi-experimental study design. Intervention studies should examine whether improving leadership behaviours could contribute to an improved mental health of workers.

Data availability statement

Data may be obtained from a third party and are not publicly available.