Reducing medication errors for adults in hospital settings

Agustín Ciapponi; Simon E Fernandez Nievas; Mariana Seijo; María Belén Rodríguez; Valeria Vietto; Herney A García-Perdomo; Sacha Virgilio; Ana V Fajreldines; Josep Tost; Christopher J Rose; Ezequiel Garcia-Elorrio

doi:10.1002/14651858.CD009985.pub2

. 2021 Nov 25;2021(11):CD009985. doi: 10.1002/14651858.CD009985.pub2

Reducing medication errors for adults in hospital settings

Agustín Ciapponi ^1,^✉, Simon E Fernandez Nievas ², Mariana Seijo ³, María Belén Rodríguez ⁴, Valeria Vietto ⁵, Herney A García-Perdomo ⁶, Sacha Virgilio ⁷, Ana V Fajreldines ⁸, Josep Tost ⁹, Christopher J Rose ¹⁰, Ezequiel Garcia-Elorrio ¹¹

Editor: Cochrane Effective Practice and Organisation of Care Group

PMCID: PMC8614640 PMID: 34822165

Abstract

Background

Medication errors are preventable events that may cause or lead to inappropriate medication use or patient harm while the medication is in the control of the healthcare professional or patient. Medication errors in hospitalised adults may cause harm, additional costs, and even death.

Objectives

To determine the effectiveness of interventions to reduce medication errors in adults in hospital settings.

Search methods

We searched CENTRAL, MEDLINE, Embase, five other databases and two trials registers on 16 January 2020.

Selection criteria

We included randomised controlled trials (RCTs) and interrupted time series (ITS) studies investigating interventions aimed at reducing medication errors in hospitalised adults, compared with usual care or other interventions. Outcome measures included adverse drug events (ADEs), potential ADEs, preventable ADEs, medication errors, mortality, morbidity, length of stay, quality of life and identified/solved discrepancies. We included any hospital setting, such as inpatient care units, outpatient care settings, and accident and emergency departments.

Data collection and analysis

We followed the standard methodological procedures expected by Cochrane and the Effective Practice and Organisation of Care (EPOC) Group. Where necessary, we extracted and reanalysed ITS study data using piecewise linear regression, corrected for autocorrelation and seasonality, where possible.

Main results

We included 65 studies: 51 RCTs and 14 ITS studies, involving 110,875 participants. About half of trials gave rise to 'some concerns' for risk of bias during the randomisation process and one‐third lacked blinding of outcome assessment. Most ITS studies presented low risk of bias. Most studies came from high‐income countries or high‐resource settings. Medication reconciliation –the process of comparing a patient's medication orders to the medications that the patient has been taking– was the most common type of intervention studied. Electronic prescribing systems, barcoding for correct administering of medications, organisational changes, feedback on medication errors, education of professionals and improved medication dispensing systems were other interventions studied.

Medication reconciliation

Low‐certainty evidence suggests that medication reconciliation (MR) versus no‐MR may reduce medication errors (odds ratio [OR] 0.55, 95% confidence interval (CI) 0.17 to 1.74; 3 studies; n=379). Compared to no‐MR, MR probably reduces ADEs (OR 0.38, 95%CI 0.18 to 0.80; 3 studies, n=1336 ; moderate‐certainty evidence), but has little to no effect on length of stay (mean difference (MD) ‐0.30 days, 95%CI ‐1.93 to 1.33 days; 3 studies, n=527) and quality of life (MD ‐1.51, 95%CI ‐10.04 to 7.02; 1 study, n=131).

Low‐certainty evidence suggests that, compared to MR by other professionals, MR by pharmacists may reduce medication errors (OR 0.21, 95%CI 0.09 to 0.48; 8 studies, n=2648) and may increase ADEs (OR 1.34, 95%CI 0.73 to 2.44; 3 studies, n=2873). Compared to MR by other professionals, MR by pharmacists may have little to no effect on length of stay (MD ‐0.25, 95%CI ‐1.05 to 0.56; 6 studies, 3983). Moderate‐certainty evidence shows that this intervention probably has little to no effect on mortality during hospitalisation (risk ratio (RR) 0.99, 95%CI 0.57 to 1.7; 2 studies, n=1000), and on readmissions at one month (RR 0.93, 95%CI 0.76 to 1.14; 2 studies, n=997); and low‐certainty evidence suggests that the intervention may have little to no effect on quality of life (MD 0.00, 95%CI ‐14.09 to 14.09; 1 study, n=724).

Low‐certainty evidence suggests that database‐assisted MR conducted by pharmacists, versus unassisted MR conducted by pharmacists, may reduce potential ADEs (OR 0.26, 95%CI 0.10 to 0.64; 2 studies, n=3326), and may have no effect on length of stay (MD 1.00, 95%CI ‐0.17 to 2.17; 1 study, n=311).

Low‐certainty evidence suggests that MR performed by trained pharmacist technicians, versus pharmacists, may have little to no difference on length of stay (MD ‐0.30, 95%CI ‐2.12 to 1.52; 1 study, n=183). However, the CI is compatible with important beneficial and detrimental effects.

Low‐certainty evidence suggests that MR before admission may increase the identification of discrepancies compared with MR after admission (MD 1.27, 95%CI 0.46 to 2.08; 1 study, n=307). However, the CI is compatible with important beneficial and detrimental effects.

Moderate‐certainty evidence shows that multimodal interventions probably increase discrepancy resolutions compared to usual care (RR 2.14, 95%CI 1.81 to 2.53; 1 study, n=487).

Computerised physician order entry (CPOE)/clinical decision support systems (CDSS)

Moderate‐certainty evidence shows that CPOE/CDSS probably reduce medication errors compared to paper‐based systems (OR 0.74, 95%CI 0.31 to 1.79; 2 studies, n=88).

Moderate‐certainty evidence shows that, compared with standard CPOE/CDSS, improved CPOE/CDSS probably reduce medication errors (OR 0.85, 95%CI 0.74 to 0.97; 2 studies, n=630).

Low‐certainty evidence suggests that prioritised alerts provided by CPOE/CDSS may prevent ADEs compared to non‐prioritised (inconsequential) alerts (MD 1.98, 95%CI 1.65 to 2.31; 1 study; participant numbers unavailable).

Barcode identification of participants/medications

Low‐certainty evidence suggests that barcoding may reduce medication errors (OR 0.69, 95%CI 0.59 to 0.79; 2 studies, n=50,545).

Reduced working hours

Low‐certainty evidence suggests that reduced working hours may reduce serious medication errors (RR 0.83, 95%CI 0.63 to 1.09; 1 study, n=634). However, the CI is compatible with important beneficial and detrimental effects.

Feedback on prescribing errors

Low‐certainty evidence suggests that feedback on prescribing errors may reduce medication errors (OR 0.47, 95%CI 0.33 to 0.67; 4 studies, n=384).

Dispensing system

Low‐certainty evidence suggests that dispensing systems in surgical wards may reduce medication errors (OR 0.61, 95%CI 0.47 to 0.79; 2 studies, n=1775).

Authors' conclusions

Low‐ to moderate‐certainty evidence suggests that, compared to usual care, medication reconciliation, CPOE/CDSS, barcoding, feedback and dispensing systems in surgical wards may reduce medication errors and ADEs. However, the results are imprecise for some outcomes related to medication reconciliation and CPOE/CDSS. The evidence for other interventions is very uncertain. Powered and methodologically sound studies are needed to address the identified evidence gaps. Innovative, synergistic strategies –including those that involve patients– should also be evaluated.

Plain language summary

Interventions for reducing medication errors in adults in hospital settings

Background to the question

An adverse drug event (ADE) is an injury resulting from a medical intervention related to a drug. ADEs are sometimes associated with medication errors. ADEs and medication errors may cause important harm, costs and even death.

Interventions for reducing medication errors include medication reconciliation, which is the process of comparing a patient's medication orders to the medications that the patient has been taking. Medication reconciliation can be performed jointly with other interventions, such as electronic prescribing systems, barcoding for a correct administering of medications, organisational changes, feedback on medication errors, education of professionals and improved medication dispensing systems.

Review question

What is the effectiveness of interventions to reduce medication errors for adults in hospital settings?

We included inpatient care settings (secondary or tertiary units, intensive care units, operating theatres), outpatient care settings, and accident and emergency departments.

Study characteristics

We searched databases of scientific studies. We included 65 studies, 51 of which were randomised trials, involving 23,182 adults in hospital settings. The remaining 14 studies were large interrupted time‐series that concern long‐term period before and after a point of intervention to assess the intervention's effects, involving more than 87,000 participants.

Certainty of the evidence

We assessed the included evidence to establish how certain we are that the effects are true and would not be altered with the addition of more evidence. In general, we judged the certainty of the evidence to be low to moderate, but it was very low for some outcomes.

Key results

Medication reconciliation compared with no medication reconciliation probably reduce ADEs and may reduce medication errors. It may have little to no effect on length of stay or quality of life. However, the effect of medication reconciliation on these latter outcomes is imprecise; it is not clear if the effects are beneficial or detrimental (low‐ to moderate‐certainty evidence).

Compared to medication reconciliation by other professionals, medication reconciliation performed by pharmacists may increase ADEs (but this result is imprecise); may reduce medication errors; and may have little to no effect on length of stay, mortality during hospitalisation, and readmissions. However, these effects are imprecise (low‐certainty evidence).

Compared to no assistance, database‐assisted medication reconciliation conducted by pharmacists may reduce potential ADEs and may have no effect on length of stay, but the last effect is imprecise (low‐certainty evidence).

Medication reconciliation performed by trained pharmacist technicians instead of pharmacists, may have no effect on length of stay, but this effect is imprecise (low‐certainty evidence).

Medication reconciliation before admission, versus after admission, may increase identified discrepancies; however, the effect is imprecise (low‐certainty evidence).

Compared to usual care, some interventions have different effects:

Multimodal interventions probably increase discrepancy resolutions (moderate‐certainty evidence).

Electronic prescribing systems probably reduce medication errors and ADEs. Prioritised alerts may additionally prevent ADEs (low‐ to moderate‐certainty evidence).

Barcode identification of participants or medications may reduce medication errors (low‐certainty evidence).

Reduced working hours and feedback on medication errors may reduce serious medication errors; however, the effect is imprecise (low‐certainty evidence).

Authors' conclusions

Compared to usual care, medication reconciliation, electronic prescribing systems, barcoding and feedback to professionals may reduce ADEs or medication errors, or both. Nonetheless, the best modalities to deliver these interventions, and the effect of other interventions, are less clear.

How up to date is this review?

The review authors searched for studies that had been published up to January 2020.

Summary of findings

Background

Description of the condition

The National Coordinating Council for Medication Error Reporting and Prevention (NCC MERP) defines a medication error as “any preventable event that may cause or lead to inappropriate medication use or patient harm while the medication is in the control of the healthcare professional, patient or consumer” (NCC‐MERP 2021; see also Lisby 2012). Medication errors can be associated with adverse drug events (ADEs), defined as unwanted occurrences after exposure to a drug that are not necessarily caused by the drug. ADEs include adverse drug reactions as well as 'preventable ADEs' and 'ameliorable ADEs', which are ADEs due to medication error (Figure 1). More specifically, an ameliorable ADE is an injury whose severity or duration could have been substantially reduced if different actions had been taken. A preventable ADE is an injury that is the result of an error at any stage in the medication use (Morimoto 2004). An adverse drug reaction is defined as any response to a drug which is noxious and unintended that occurs at doses normally used for prophylaxis, diagnosis or therapy of the disease (European Council 2005; Falconer 2019). Potential ADEs are defined as medication errors with high likelihood to cause harm (Bates 1995).

Medication error framework (from Morimoto 2004 (Licence: 4295121359710) that modified Bates 1995, with permission)

The severity of ADEs has been classified as follows (ISMP 2011).

Category 1: circumstances or processes that have the potential to cause an adverse drug event.
Category 2: an event occurred, but the patient was not harmed.
Category 3: an event occurred that resulted in the need for increased patient assessments but no change in vital signs and no patient harm.
Category 4: an event occurred that resulted in the need for treatment or intervention, or both, and caused temporary patient harm.
Category 5: an event occurred that resulted in initial or prolonged hospitalisation, affected patient participation in an investigational drug study, and/or caused temporary patient harm.
Category 6: an event occurred that resulted in permanent patient harm or a near‐death event, such as anaphylaxis.
Category 7: an event occurred that resulted in patient death.

Such events may be related to professional practice; healthcare products, procedures and systems, including prescribing; order communications; product labelling, packaging and nomenclature; compounding; dispensing; distribution; administration; and education and monitoring (Nebeker 2004). However, most of the literature on medication errors suggests that prescribing errors are the most prevalent cause (Kohn 2000; World Alliance for Patient Safety 2008).

The burden of medication errors and adverse drug events in hospitals is especially important. Medication errors and adverse drug events are associated with substantial death and injury (Kohn 2000). More than 500,000 people are injured or die each year in hospitals from adverse drug events (ADEs), which may cost up to USD 5.6 million annually per hospital (Classen 2005).

One systematic review found that the prevalence of prescribing errors ranged widely, from 2% to 94% (Assiri 2018).

The Canadian Adverse Events Study showed an adverse event rate of 7.5 per 100 hospital admissions, of which 37% were judged to be preventable (Baker 2004). Another multicenter study in the USA found that medication errors occurred in 5.3 of each 100 medication orders written, half of which were caused by missing medication dosages, 15% involved dose errors, and 13% involved route or frequency errors (Bates 1995). Five of the 25 adverse drug events (20%) identified during the study period were directly associated with medication errors, all of them judged as preventable. A systematic review of studies on adverse events in hospitalised people showed that 1 in 10 is affected by an adverse event, with a median percentage of 43% being preventable and a rate of lethal events of 7.4 per 100 adverse events (de Vries 2008). Other studies suggest that medication errors and adverse drug events are associated with 140,000 deaths annually, and occur in 1 in 16 hospitalised people (Classen 1997; World Alliance for Patient Safety 2008). Classen and colleagues estimated that the additional costs of hospitalisation for each person with an adverse drug event were USD 2000 (Classen 1997). Two recent systematic reviews found considerable variability between studies in terms of financial cost, patients, settings and errors included (Vilela 2018; Walsh 2017).

Description of the intervention

Attention has been paid to this patient safety issue, and the literature identifying the causes, frequency and consequences of ADEs and medication errors, as well as the effects of interventions to prevent them, has grown (World Alliance for Patient Safety 2008).

In this review, we adopt the Cochrane Effective Practice and Organisation of Care (EPOC) group's taxonomy of health systems interventions to conceptualise and organise interventions used to try to reduce medication errors in hospitals (EPOC 2015). The taxonomy identifies four main domains of interventions: delivery arrangements, financial arrangements, governance arrangements, and implementation strategies, defined as follows.

Delivery arrangement interventions involve changes in how, when and where health care is organised and delivered, and who delivers health care.
Financial arrangement interventions involve changes in: how funds are collected; how services are purchased; insurance schemes; and the use of targeted financial incentives or disincentives.
Governance arrangement interventions involve rules or processes that affect the way in which powers are exercised, particularly with regard to authority, accountability, openness, participation, and coherence.
Implementation strategy interventions are those designed to bring about changes in healthcare organisations, the behaviour of healthcare professionals or the use of health services by healthcare recipients (EPOC 2015).

Reviews of medication safety intervention evidence have identified more than 20 distinct practices, healthcare professionals and technologies that have the potential to improve medication safety (de Vries 2008; Hodgkinson 2006; Ioannidis 2001; Shojania 2001). The following is a non‐exhaustive list of examples.

Medication reconciliation: the process of comparing a patient's medication orders in hospital to the medications that the patient has been taking. Medication reconciliation can be performed by individual healthcare professionals (such as pharmacists or pharmacist technicians) or teams, or both, trained to prevent or manage medication errors.
Database‐assisted medication reconciliation by using prescription databases to assist professionals in obtaining medication histories upon hospital admission.
Electronic prescribing systems, including computerised physician ordering entry (CPOE) and clinical decision support systems (CDSS). In general, these refer to the process of a medical professional entering and sending medication orders and treatment instructions electronically via a computer application instead of on paper charts. They are computer‐based programs that analyse data within electronic health records to provide prompts and reminders to assist healthcare providers in implementing treatments at the point of care.
Electronic prescription: the computer‐based electronic generation, transmission and filling of a medical prescription.
Automated dispensing systems, including devices that dispense medications and fill prescriptions. These systems also communicate with the pharmacy and its information management system to track medications removed and support inventory replenishment.
Bedside terminal systems: bedside computers to provide access to hospital resources.
Computer‐generated medication administration records (MARs): these synchronise data throughout an organisation; for example, they can interface with the pharmacy system, the computerised prescriber order entry system, and any admission‐discharge‐transfer system.
Computer alert systems.
Barcodes for identification of patients or medications.
Education and training.
Pharmacists.
Dedicated nurses.
Double‐checking.
Medication administration review and safety.
Utilisation of standardised checklists (protocols) by triage nurses.
Syringes marked with doses to reduce mistakes in identifying the right medication or doses.
Self‐medication programmes to reduce errors by healthcare workers.
Illumination in the workplace to reduce mistakes in identifying the right medication or doses.
Reduced working hours by eliminating extended work shifts and reducing the number of hours worked per week.
Education interventions to improve medication prescription or administration.
Multidisciplinary approaches.

How the intervention might work

The interventions applied at different hospital care levels, including delivery, financial and governance arrangements as well as implementation strategies, are expected to improve patient safety in terms of medication errors. The interventions are mainly directed to human resources, use different technologies and structural or organisational changes, or a combination of some or all of these. Interventions directed to improve human resources performance may include medication reconciliation, training, education, and feedback on medication errors, or having dedicated health professionals. Technological interventions may reduce human medication errors through electronic prescribing systems, such as CPOE and CDSS, electronic medication administration records (e‐MARs), automated dispensing or barcodes for identification of patients or medications. Structural or organisational changes may include reduced working hours or decentralised pharmacy systems.

Why it is important to do this review

Medication errors are a leading, avoidable, source of harm to hospital patients. Some authors have called for the implementation of evidence‐based practices to find solutions to this patient safety problem (Brennan 2005). Several systematic reviews, published prior to our protocol, partially addressed this topic (de Vries 2008; Hodgkinson 2006; Ioannidis 2001; Shojania 2001; Wong 2010). But none of these reviews, nor more recent ones (Ahtiainen 2020; Eng 2018; Khalil 2020; Korb‐Savoldelli 2018; Redmond 2018; Roumeliotis 2019; Shitu 2019), have comprehensively covered the wide range of interventions used to reduce medication errors at different points of care, precluding comparisons of the clinical utility of separate interventions and strong recommendations.

Our systematic review provides an exhaustive and up‐to‐date analysis of the available evidence for interventions devoted to preventing medication errors in hospital settings.

Objectives

To determine the effectiveness of interventions to reduce medication errors in adults in hospital settings.

Methods

Criteria for considering studies for this review

Types of studies

We included study designs that met the explicit criteria used by the Cochrane Effective Practice and Organisation of Care (EPOC) group: randomised controlled trials (RCTs), quasi‐randomised controlled trials (quasi‐RCTs), interrupted time series (ITS) studies with at least three data points before and three after the intervention, and controlled before‐and‐after (CBA) studies, with more than one intervention or control site, that could be analysed as ITS studies.

A quasi‐randomised trial is one in which participants are allocated to different arms of the trial using a method of allocation that is not truly random (e.g. sequence generated by odd or even date of birth; sequence generated by some rule based on date (or day) of admission; sequence generated by some rule based on hospital or clinic record number).

Types of participants

Setting

We included studies conducted in a hospital. We further classified studies by these setting categories: (i) inpatient care (secondary or tertiary units, intensive care units, operating theatres); (ii) outpatient care; and (iii) accident and emergency departments.

Healthcare professionals

We considered a study for inclusion if it involved healthcare professionals responsible for prescribing, dispensing or administering medications, in charge of care of adult (> 18 years old) hospitalised patients. When studies also included participants under 18 years old, we extracted data only for the adult population.

We excluded studies based in geriatric, institutional settings caring for the elderly, psychiatric institutions and in settings that provide care to children. The last of these is the focus of another Cochrane Review (Maaskant 2015).

Types of interventions

We included studies of interventions applied in hospital care to improve patient safety in terms of medication errors, compared to no intervention, other intervention, or usual care. Studies might have described one intervention, or a package of interventions, which we refer to as 'multifaceted'. The types of interventions we anticipated finding are listed in the Description of the intervention. We categorised these interventions ‐ applied at the hospital care level ‐ according to the EPOC taxonomy of four domains of interventions aimed at achieving practice change (EPOC 2015).

Delivery arrangements

Health service delivery arrangements include changes in who receives care and when, who provides care, the working conditions of those who provide care, co‐ordination of care amongst different providers, where care is provided, the use of health information and communication technology to deliver care, and quality and safety systems. Listed below are some of these delivery arrangement subcategories, together with examples of interventions to reduce medication errors in hospital settings.

Who receives care and when? Example: medication reconciliation before versus at admission.
Who provides care? Example: medication reconciliation performed by pharmacist versus by other professionals.
Who provides care, or co‐ordination of care? Example: medication reconciliation by a multidisciplinary team or trained pharmacist or pharmacist technicians versus standard pharmacist.
Health information and communication technology. Examples: electronic prescribing systems such as CPOE and CDSS; barcoding; dispensing systems, database‐assisted medication reconciliation; one or two versus four charts open simultaneously for medication reconciliation.
Working conditions of healthcare workers. Example: reduced versus unreduced working hours.
Co‐ordination of care / integration. Example: integrated multimodal intervention.

Financial arrangements

Health financing arrangements comprise the collection of funds, insurance schemes, the purchasing of services, and the use of targeted financial incentives or disincentives.

Governance arrangements

The term 'governance' can be defined in several, sometimes overlapping, ways. Its use differs within the social sciences, especially between economics and political science. We have defined governance here as rules or processes that affect the way in which powers are exercised, particularly with regard to authority, accountability, openness, participation, effectiveness and coherence. Governance arrangements subcategories could include:

Interagency collaboration. Example: collaboration and partnerships, for example, using big data;
Policies that regulate programme monitoring and evaluation;
Processes for accrediting healthcare providers in patient safety;
Policies that regulate the sale and dispensing of drugs or other healthcare products;
Policies that regulate training and licensing requirements for health professionals or what they can do.

Implementation strategies

Implementation strategies include interventions designed to bring about changes in healthcare organisations or the behaviour of healthcare professionals or recipients. Implementation strategy subcategories could include:

Interventions targeted at healthcare worker practice. Examples: feedback on prescribing errors; education.
Types of problems targeted at healthcare worker practice. Example: medication reconciliation.

Nevertheless, medication reconciliation is an intervention that crosscuts the EPOC taxonomy categories, including also delivery arrangements.

The comparison groups in the studies could have been another intervention, no intervention or usual care.

Types of outcome measures

Our initial approach was to extract each outcome with the exact name given by the authors of the included studies. However, these studies assessed the impacts of interventions to reduce medication errors in a wide range of ways (70 different outcomes). In order to organise and prioritise these outcomes for analysis in each comparison, we sought the input of a group of expert pharmacists, and we arrived at a consensus (Appendix 1 describes the outcomes as reported by authors of the included studies and grouped outcomes for this systematic review). We also analysed separately the ungrouped outcomes in natural units, or in the way that the authors of primary studies originally reported the outcomes, and we have reported them as additional figures. We did not pre‐specify time points for the outcomes; instead, we reported every available result.

Primary outcomes

Medication errors (grouped outcomes)

Proportion of patients with a medication error (i.e. administration, discrepancy, dispensing or duplication errors)
Incidence of medication errors

Adverse drug events (grouped outcomes)

Proportion of patients with serious adverse drug events, defined as categories 6 and 7 (see Description of the condition): that is, adverse drug events that are permanently disabling, require or prolong hospitalisation or are lethal or potentially life‐threatening (ISMP 2011)
Proportion of ADEs and preventable ADEs, defined as undesired reaction to medication that may have been prevented by appropriate drug selection or management (Hodgkinson 2006)

Secondary outcomes

Adverse drug events (grouped outcomes)

Total number of adverse drug events
Incidence of serious ADEs

Non‐grouped outcomes

Mortality
Morbidity
Hospitalisations
Length of stay
Resource use
Quality of life
Identified discrepancies and discrepancy resolutions

We used in this review a taxonomy for medication error proposed by Bates 1995 and modified by Morimoto 2004 (see Figure 1).

Medication errors (MEs) include any errors that occur during any of the processes involved in medicines management (e.g. prescribing, transcribing, dispensing, administration, documentation and monitoring).
Potential adverse drug events (ADEs) are defined as medication errors with a high likelihood to cause harm. Medication errors cause around 30% of ADEs.
Adverse drug events (ADEs) are defined as injuries resulting from medical interventions related to a drug.

A discrepancy is defined as an inconsistency between two medication lists of a patient, regarding the presence, absence, dosage, route, frequency or formulation of a medication during a transition of care between home and hospital or between different hospital settings. Unintended medication discrepancy is a type of medication error not detected by medication reconciliation. Thus, discrepancy resolution and identified discrepancies ‐ as a proxy of the former outcome ‐ are beneficial outcomes oriented to resolve medication errors.

Search methods for identification of studies

Electronic searches

We searched the Cochrane Database of Systematic Reviews (CDSR) and the Database of Abstracts of Reviews of Effects (DARE) for primary studies included in related systematic reviews.

We searched the following databases on 16 January 2020.

Cochrane Central Register of Controlled Trials (CENTRAL; 2020, Issue 1), in the Cochrane Library.
MEDLINE, Ovid (including Epub ahead of print, in‐process and other non‐indexed citations, 1946 onwards).
Embase, Ovid (1974 onwards).
CINAHL (Cumulative Index to Nursing and Allied Health Literature), EBSCOHost (1980 onwards).
Conference Proceedings Citation Index ‐ Science, Web of Science, Clarivate Analytics (1990 onwards).
COS Conference Papers Index, ProQuest (1995 onwards).
Dissertations and Theses, Global, ProQuest (1861 onwards).

The EPOC Cochrane Information Specialist (CIS), in consultation with the authors, developed the search strategies. Broad initial searches were subsequently revised in an iterative process, following peer review by a second information specialist, to produce a more specific set of search terms. Search strategies are comprised of keywords and controlled vocabulary terms. We applied no language or time limits. For translations of publications, we contacted native‐speaker collaborators. We searched all databases from database start date to the date of search (16 January 2020). All strategies used are provided in Appendix 2.

Searching other resources

We also:

Reviewed reference lists of relevant systematic reviews or other publications;
Contacted authors of relevant studies or reviews to clarify reported published information or to seek unpublished results/data;
Contacted researchers with expertise relevant to the review topic or EPOC interventions; and
Conducted cited reference searches in Science Citation Index, Web of Science.

Trials Registries

We searched these trials registers on 16 January 2020:

ClinicalTrials.gov (www.clinicaltrials.gov/); and
WHO International Clinical Trials Registry Platform (ICTRP) (www.who.int/ictrp/en/).

Data collection and analysis

Selection of studies

Working in pairs, the review authors independently screened all titles and abstracts retrieved from the search strategy, using software for systematic reviews (Covidence), to assess which studies met the inclusion criteria. We obtained copies of all references considered potentially relevant. We resolved any disagreement between the pairs of review authors through discussion. If consensus could not be reached, we involved an EPOC Group editor to resolve the disagreement.

Data extraction and management

Pairs of review authors independently undertook data extraction using a modified and piloted version of the EPOC Group data collection checklist. We resolved any disagreement between the review author pairs through discussion.

A statistician extracted data from the included interrupted time series (ITS) studies using WebPlotDigitizer (accessed in March and April 2020). He estimated pre‐interruption level and slope, post‐interruption change in level, post‐interruption slope using piecewise linear regression, adjusted for autocorrelated disturbances and seasonality, using the ITSA add‐on command (Linden 2016), for Stata (StataCorp 2015). We adjusted for autocorrelated disturbances by setting the maximum lag option to a value determined by visual inspection of autocorrelation and partial correlation plots, and by using Cumby‐Huizinga general tests for autocorrelation with a significance threshold of 0.05 (Cumby 1992). We adjusted for seasonality by modelling the effect of each quarter as a fixed effect if at least three observations were available for each quarter. We modelled ITS data on the natural logarithmic scale to constrain the error distribution to positive values, to stabilise variance, and to facilitate meta‐analysis (see Measures of treatment effect). None of the included ITS studies included controls in which no intervention (or a substantively different intervention) was used in the post‐interruption period, so we could not adjust for other possible explanations for the observed changes.

Assessment of risk of bias in included studies

Pairs of review authors independently assessed the risk of bias of the included studies.

For RCTs, we used the Cochrane risk of bias tool (Higgins 2011), paying special attention to the following domains: sequence generation, allocation concealment, blinding and incomplete outcome data. For the other eligible designs, we assessed their quality using pre‐established criteria used by the EPOC group (EPOC 2017). We resolved any discrepancies in quality ratings through discussion and the involvement of an arbitrator where necessary. For all study designs, we added a conflict of interest domain ('unclear risk' of bias for studies sponsored by industry and 'high risk' of bias only when there was evidence of causal bias).

Measures of treatment effect

Reporting

We tabulated data in natural units for each study. We reported pre‐intervention and post‐intervention means or proportions where baseline results were available for both study and control groups from RCTs, quasi‐RCTs and CBAs. We calculated the absolute change from baseline with 95% confidence limits. For ITS studies, we reported the main outcomes in natural units with two indicators of the effects of the intervention being documented: the change in the level of outcome immediately after the intervention and the change in the slope of the regression lines. 

Analytical approach

Primary analyses

We based the primary analyses on consideration of dichotomous process measures (for example, proportion of participants experiencing an adverse reaction). When studies reported more than one measure for each endpoint, we extracted the primary measure (as defined by the authors of the study) or the median measure identified. We presented the results for all comparisons using a standard method of presentation where possible. For comparisons of RCTs or quasi‐RCTs and ITS studies, we reported (separately for each study design):

median effect size across included studies;
interquartile ranges of effect sizes across included studies;
range of effect sizes across included studies.

Secondary analyses

Secondary analyses explored the consistency of primary analyses with other types of endpoints. We calculated standardised mean differences (SMD) for continuous measures by dividing the difference in mean scores between the intervention and comparison group in each study by an estimate of the (pooled) standard deviation. In order to gain comparability between combined SMDs, we also transformed MD of single studies to SMDs.

Confounding variables considered for ITS analysis included patient level variables (sex, age, and ethnicity), provider role (attending physician, resident, medical student, nurse, pharmacist or other), type of setting (inpatient care settings such as secondary or tertiary units, intensive care units, operating theatres, outpatient care settings and accident and emergency departments) or practice context (i.e. order placed during a day or a night shift).

Methods for reanalysis

We reanalysed RCTs and quasi‐RCTs with potential unit of analysis errors, where possible, by recalculating results using the appropriate unit of analysis; otherwise, we contacted the authors of such studies for clarification. For the ITS studies, we exponentiated change in level and slope (which were estimated on the logarithmic scale, see Data extraction and management) to obtain estimates of ratios of post‐ to pre‐interruption levels and slopes. These estimates describe the nature of any change in reporting. In principle, however, genuine changes in level and slope can lead to no overall change (i.e. a change in slope can effectively cancel a change in level). We therefore measure change as the ratio of expected events by extrapolating the pre‐interruption curve into the post‐interruption period and treating it as a counterfactual. Because this ratio is a function of time, we estimated it at one and two years post‐intervention. We excluded a study if it would be necessary to extrapolate beyond the end of follow‐up for that study.

Where appropriate, we used Cochrane's standard statistical methods for pooling of data from randomised and quasi‐randomised controlled trials. For categorical and continuous data, we calculated the risk ratio (RR) or odds ratio (OR) and mean difference (MD), respectively, with 95% confidence intervals (CIs). We used a random‐effects model to take into account the heterogeneity of the various studies.

We reported data in individual tables comparing effect sizes of interventions for grouped outcomes according to the EPOC group taxonomy (delivery, financial and governance arrangements, and implementation standards) (EPOC 2015). We examined data from ITS studies and cluster‐randomised trials with unit of analysis errors according to the EPOC Group guidelines and used absolute risk differences.

We created summary of findings tables for the main comparisons in the review to interpret the results and draw conclusions about the effects (benefits, potential harms and costs) of different interventions, including the size of effects and quality of the evidence for outcomes for which there is evidence.

Unit of analysis issues

We reanalysed the study if data were available (i.e. using intracluster correlation). If not, we reported the unit of analysis error for each study and classified it as high risk of bias in the 'other bias' domain. For cluster‐randomised trials, we appraised whether an appropriate analysis had been done that adjusted for clustering in results. If the analysis did not appear to have adjusted for clustering appropriately, we considered whether the effect estimate was likely to be affected by such issues and, as appropriate, noted this as a potential source of bias relating to the outcome in question.

We described the unit of analysis of each study and only combined them using the generic inverse‐variance method with specific standard errors.

Dealing with missing data

If information was missing or unclear, we contacted the study investigators for additional information or clarification. To reduce the risk of overly positive answers, we used open‐ended questions.

Assessment of heterogeneity

We obtained an initial visual overview of heterogeneity through scrutinising the forest plots and looking at the overlap between CIs around the estimate for each included study. To quantify the inconsistency across studies, and thus the impact of heterogeneity on the meta‐analysis, we used the I² statistic to detect heterogeneity (Higgins 2003). In the latter case, we defined an I² of higher than 60% as revealing substantial heterogeneity. We also interpreted the significance of the I² test in light of: (i) the magnitude and direction of effects; and (ii) the strength of evidence for heterogeneity (for example, a CI for the I², or the P value for the Chi² test).

We assessed observable heterogeneity amongst the study questions and methods, to determine whether a meta‐analysis was appropriate. We also looked at the study participants, settings, interventions, and reported outcomes. We paid particular attention to the homogeneity of the methodology (such as variances in blinding and concealment of allocation) within and across included studies.

If we found evidence of statistical heterogeneity, we examined it in a subgroup analysis and a sensitivity analysis, as outlined in the respective sections below (Subgroup analysis and investigation of heterogeneity; Sensitivity analysis).

Assessment of reporting biases

To reduce possible publication bias, we employed strategies to search for and include relevant unpublished studies. These strategies included searching the grey literature and prospective trial registration databases to overcome time‐lag bias.

To investigate the likelihood of overt publication bias, we planned to draw a funnel plot, plotting trial effects against inverse standard errors of the effects for any outcome with more than eight studies. In the event, this step was unnecessary.

Data synthesis

For each comparison, we reported summary statistics for each of the included studies (RCTs or quasi‐RCTs and ITS studies). We used forest plots to display the data graphically.

For dichotomous data, we used the Mantel‐Haenszel method, and for continuous data, we used the inverse‐variance method.

We pooled the results from individual studies in a meta‐analysis using the random‐effects model by DerSimonian and Laird (DerSimonian 1986). We chose this method because we could not assume a single, underlying (fixed) treatment effect. When the impact of the intervention was assessed in individual studies on more than one outcome measure, we selected the outcome that best reflected the targeted intervention for pooling data.

We used generic inverse‐variance when we only had results expressed as adjusted relative effects or to combine different types of outcomes, following the expert advice of the pharmacists we consulted about outcome groupings. We gave priority to risk ratios (RRs; for easier interpretation), but if data did not allow this approach, we reported odds ratios (ORs).

When a study compared more than one arm, we only analysed the intervention arm that fitted most closely with the comparison and with the studies included under it. For example, we excluded from the analysis any multimodal interventions besides the intervention under study.

We analysed ITS studies separately to RCTs.

We analysed ITS study data using the guidelines of the EPOC group (EPOC 1998), and reported outcomes in natural units. We reported pre‐intervention and post‐intervention means or proportions for both study and control groups, and calculated the unadjusted and adjusted (for any baseline imbalance) absolute change from baseline with 95% CIs. We used either a regression analysis with time trends before and after the intervention, which adjust for autocorrelation and any periodic changes, or an autoregressive, integrated, moving average (ARIMA) model to isolate the effect of the intervention from existing time trends.

Subgroup analysis and investigation of heterogeneity

We performed the following subgroup analyses, where possible, to check if the intervention effect varied with different populations, interventions, or settings.

Type of setting (general wards, emergency department, intensive care units).
Type of provider (less trained, more trained, etc.).
Type of outcome (all errors, prescribing errors, etc.).
Type of outcome measure (per patients, per admissions, per prescriptions, etc.).
Time points of outcomes (during hospitalisation, post‐hospitalisation).

When we were not able to perform a meta‐analysis, we summarised the results for these subgroups within the text of the review.

Sensitivity analysis

We performed sensitivity analysis based on the method of meta‐analysis. That is, we compared the results from the random‐effects and fixed‐effect models if there was unexplained heterogeneity between studies, to assess the robustness of the results.

We also planned to analyse only studies at low risk of bias for both random sequence generation and allocation concealment. However, we were unable to conduct this analysis due to an insufficient number of such studies for each comparison.

Summary of findings and assessment of the certainty of the evidence

We imported data from Review Manager 5.4 (Review Manager 2020) to GRADE profiler (GRADEpro GDT), and created summary of findings tables for the following 16 comparisons.

Comparison 1: medication reconciliation (MR) versus no MR.
Comparison 2: MR performed by pharmacist versus other professionals.
Comparison 3: MR by pharmacist: database‐assisted versus unassisted.
Comparison 4: MR by pharmacist: trained pharmacist technician versus pharmacist.
Comparison 5: MR: before versus at admission.
Comparison 6: MR: one to two versus four charts open simultaneously.
Comparison 7: MR: multimodal intervention versus usual care.
Comparison 8: computerised physician order entry (CPOE)/clinical decision support systems (CDSS) versus control/paper‐based system.
Comparison 9: CPOE/CDSS: improved versus standard CPOE/CDSS.
Comparison 10: CPOE/CDSS: prioritised versus non‐prioritised alerts.
Comparison 11: barcoding versus no barcoding.
Comparison 12: organisational changes: reduced versus unreduced working hours.
Comparison 13: feedback on prescribing errors versus no feedback.
Comparison 14: feedback on prescribing errors versus education.
Comparison 15: education versus no education on prescribing or administration.
Comparison 16: dispensing system versus control.

In the summary of findings tables, for each comparison, we have presented seven of eight primary and secondary outcomes, listed below. We prioritised these in consultation with a group of expert pharmacists.

Medication errors (MEs; primary outcome)
Adverse drug events (ADEs) / preventable ADEs (primary outcome)
Mortality (secondary outcome)
Readmission (secondary outcome)
Length of stay (LoS; secondary outcome)
Quality of life (QoL; secondary outcome)
Discrepancy resolution (secondary outcome)
Identified discrepancies per patient (secondary outcome). This outcome was presented only if the previous seven outcomes were not reported.

We reported separately the evidence from RCTs or ITS studies which evaluated the same outcome.

Pairs of review authors independently graded the certainty of the evidence for each outcome using the GRADE approach (Guyatt 2011; Hultcrantz 2017; Schünemann 2013); we resolved discrepancies by reaching consensus. For assessments of the overall certainty of the evidence for outcomes that included pooled data from RCTs, we initially graded the evidence as high certainty, downgrading the rating (by one level from high to moderate certainty, by two levels to low certainty, or three levels to very low‐certainty evidence) depending on the extent of accomplishment across the following criteria: study limitations (risk of bias); indirectness of evidence; inconsistency; imprecision of effect estimates; or publication bias. For certainty ratings for outcomes that included pooled data from ITS studies, we initially graded the evidence as low certainty, upgrading the rating to moderate or high certainty if the pooled estimates revealed a large magnitude of effect, negligible concerns about confounders, or a strong dose‐response gradient. We used these assessments, along with the evidence (or lack thereof) for absolute benefit or harm of the interventions, and the sum of available data on all primary and secondary outcomes from each study included for each comparison, to draw conclusions about the effectiveness of the interventions.

Results

Description of studies

See Characteristics of included studies for more information.

Results of the search

We screened a total of 21,545 titles and abstracts, and from these, identified 1066 full‐text publications for further screening. Of these full‐text publications, we excluded 985 reports. The majority of these involved an ineligible study design (n = 619), an ineligible intervention (n = 125) or other disqualifier (n = 99) (e.g. non‐separated adult and paediatric population or insufficient information to decide). Ultimately, we included 65 studies, four secondary references and 12 ongoing studies (see Figure 2).

Included studies

We included 65 studies and 4 secondary references (see Characteristics of included studies).

Of the included studies, 51 were RCTs: Aag 2014; Adelman 2013; Adelman 2019; Al‐Hashar 2018; Barker 1984; Becerra‐Camargo 2015; Beckett 2012; Bell 2016; Bolas 2004; Boockvar 2017; Cadman 2017; Chiu 2018; Colpaert 2006; De Winter 2011; Ding 2012; Farris 2014; Fernandes 2011; George 2011; Gordon 2017; Graabaek 2019; Greengold 2003; Gursanscky 2018; Hale 2013; Heselmans 2015; Hickman 2018; Juanes 2018; Khalil 2016; Kwan 2007; Landrigan 2004; Leung 2017; Lind 2017; Marotti 2011; McCoy 2012; Merry 2011; Nielsen 2017; O'Sullivan 2016; Pevnick 2018; Piqueras Romero 2015; Quach 2015; Redwood 2013; Schmader 2004; Schneider 2006; Schnipper 2009; Scullin 2007; SUREPILL 2015; Tamblyn 2018; Tompson 2012; Tong 2016; Vega 2016; Wang 2017; Willoch 2012.

The 14 remaining included studies included seven ITS studies (Agrawal 2009; Bhakta 2019; Burkoski 2019; Kannampallil 2018; Ongering 2019; Schnipper 2018; Van Doormaal 2009), one controlled ITS study (Thompson 2018), and six CBA studies, reanalysed as ITS studies (Bowdle 2018; Furuya 2013; Green 2015; Higgins 2010; Narang 2013; Seibert 2014).

Population

The total number of included participants was 110,875. The RCTs included a total of 23,182 participants (some studies used prescriptions or providers as the unit of analysis, and are not included in this total: Adelman 2013; Adelman 2019; Ding 2012; Gordon 2017; Greengold 2003; Gursanscky 2018; Hickman 2018; Leung 2017). ITS studies included a total of 87,692 participants (some studies used medication alerts or prescriptions as the unit of analysis, and are not included in this total: Bhakta 2019; Burkoski 2019; Green 2015; Higgins 2010; Narang 2013; Seibert 2014; Thompson 2018).

Fifty‐five studies included inpatient adults (from18 years old up to no age limit), five studies included only elderly inpatients (age more than 65 years old) (Beckett 2012; Chiu 2018; Piqueras Romero 2015; Quach 2015; Schmader 2004), and five studies included both inpatient and outpatients adults (Adelman 2013; Adelman 2019; Agrawal 2009; Farris 2014; Vega 2016).

Setting

The most frequent setting was medical and surgical wards (17 studies). The remaining studies' settings were: medical wards (13 studies), surgical wards (9), emergency departments (7), intensive care units (ICUs; 3), operating rooms (2) and other settings (14).

Most of the RCTs (47/51) were conducted in high‐income countries: 16 in the USA (Adelman 2013; Adelman 2019; Barker 1984; Beckett 2012; Bell 2016; Boockvar 2017; Farris 2014; Greengold 2003; Landrigan 2004; McCoy 2012; Pevnick 2018; Quach 2015; Schmader 2004; Schneider 2006; Schnipper 2009; Tompson 2012); eight in Australia (George 2011; Gursanscky 2018; Hale 2013; Hickman 2018; Khalil 2016; Leung 2017; Marotti 2011; Tong 2016); five in the United Kingdom (Bolas 2004; Cadman 2017; Gordon 2017; Redwood 2013; Scullin 2007); three in Belgium (Colpaert 2006; De Winter 2011; Heselmans 2015); three in Canada (Fernandes 2011; Kwan 2007; Tamblyn 2018); three in Denmark (Graabaek 2019; Lind 2017; Nielsen 2017); three in Spain (Juanes 2018; Piqueras Romero 2015; Vega 2016); two in Norway (Aag 2014; Willoch 2012); and one each in Ireland (O'Sullivan 2016), the Netherlands (SUREPILL 2015), New Zealand (Merry 2011) and Oman (Al‐Hashar 2018). The four remaining RCTs were conducted in middle‐income countries: three in China (Chiu 2018; Ding 2012; Wang 2017), and one in Colombia (Becerra‐Camargo 2015).

All the 14 ITS studies were conducted in high‐income countries: 10 in the USA (Agrawal 2009; Bhakta 2019; Bowdle 2018; Green 2015; Higgins 2010; Kannampallil 2018; Narang 2013; Schnipper 2018; Seibert 2014; Thompson 2018); two in the Netherlands (Ongering 2019; Van Doormaal 2009); one in Canada (Burkoski 2019); and one in Japan (Furuya 2013).

Interventions and comparisons

We categorised all the interventions in the included studies into two of the four EPOC taxonomy categories (EPOC 2015), as described in the Description of the intervention and in the Types of interventions sections; namely, delivery arrangements and implementation strategies (see Appendix 3 for categorisations). Thus, we did not categorise any interventions in the included studies as falling under the two remaining categories of financial arrangements or governance arrangements.

In order to further categorise the interventions, we classified each study by EPOC group taxonomy and the comparison number (See Appendix 4).

In Appendix 5 and Appendix 6, we describe the study designs, populations, settings and countries, and the study level contribution by comparison.

Ongoing Studies

We identified 12 ongoing studies, which we describe in the Characteristics of ongoing studies tables.

Excluded studies

In the Characteristics of excluded studies tables, we present the details of the 12 excluded studies (Farley 2014; Franklin 2019; Gillespie 2009; Heng 2013; Kripalani 2012; Kucukarslan 2003; Makowsky 2009; Pellegrin 2017; Shah 2013; Singh 2012; Stowasser 2002; Whittington 2004).

Risk of bias in included studies

We assessed the risk of bias for RCTs and ITS studies separately. We provide a summary of the results of our assessment below and graphically in Figure 3 and Figure 4 (for RCTs) and Figure 5 and Figure 6 (for ITS studies). Further details can be found in the risk of bias tables for each study (see the Characteristics of included studies tables).

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

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

Risk of bias summary for CBA and ITS studies: review authors' judgements about each risk of bias item for each included study

Risk of bias graph for CBA and ITS studies: review authors' judgements about each risk of bias item presented as percentages across all included studies

The main causes for high risk of other bias were: the analysis method did not account for the cluster design (10 studies: Barker 1984; Colpaert 2006; Ding 2012; Gordon 2017; Greengold 2003; Gursanscky 2018; Landrigan 2004; Leung 2017; SUREPILL 2015; Tamblyn 2018); the specific effect of the intervention could not be isolated (Farris 2014); recruitment was performed on certain days of the week (George 2011); retrospective methods were used to identify adverse drug reactions (Schmader 2004); contamination bias (Adelman 2013); and temporal difference between arms in the identification of outcomes (Willoch 2012).

Interrupted time series studies

Reliable primary outcome measure(s)

We rated 13 studies as low risk of bias for reliable primary outcome measure (Agrawal 2009; Bhakta 2019; Burkoski 2019; Furuya 2013; Green 2015; Higgins 2010; Kannampallil 2018; Narang 2013; Ongering 2019; Schnipper 2018; Seibert 2014; Thompson 2018; Van Doormaal 2009), and one as unclear risk of bias (Bowdle 2018).

Blinded assessment of primary outcome(s)

We rated seven studies as low risk of bias for blinded assessment of primary outcomes (Bhakta 2019; Burkoski 2019; Green 2015; Kannampallil 2018; Narang 2013; Ongering 2019; Thompson 2018), four as unclear risk of bias (Agrawal 2009; Bowdle 2018; Higgins 2010; Seibert 2014), and three as high risk of bias because they used non‐blinded assessment or were open trials (Furuya 2013; Schnipper 2018; Van Doormaal 2009).

Data were analysed appropriately

We rated nine studies as low risk of bias for appropriate data analysis (Bhakta 2019; Burkoski 2019; Green 2015; Kannampallil 2018; Ongering 2019; Schnipper 2018; Seibert 2014; Thompson 2018; Van Doormaal 2009), and five as high risk of bias because they did not use ARIMA models or time series regression models to analyse the data (Agrawal 2009; Bowdle 2018; Furuya 2013; Higgins 2010; Narang 2013).

Protection against detection bias (same pre‐post data collection)

We rated 12 studies as low risk of bias for protection against detection bias (Agrawal 2009; Bhakta 2019; Bowdle 2018; Burkoski 2019; Furuya 2013; Higgins 2010; Kannampallil 2018; Ongering 2019; Schnipper 2018; Seibert 2014; Thompson 2018; Van Doormaal 2009), and two as unclear risk of bias (Green 2015; Narang 2013).

Completeness of data set

We rated 12 studies as low risk of bias for completeness of data set (Agrawal 2009; Bhakta 2019; Bowdle 2018; Burkoski 2019; Furuya 2013; Green 2015; Higgins 2010; Kannampallil 2018; Narang 2013; Ongering 2019; Seibert 2014; Thompson 2018), and two as unclear risk of bias (Schnipper 2018; Van Doormaal 2009).

Reason for the number of points pre‐ and post‐intervention given

We rated six studies as low risk of bias for giving reasons for the number of points pre‐ and post‐intervention (Burkoski 2019; Green 2015; Ongering 2019; Schnipper 2018; Seibert 2014; Van Doormaal 2009), and seven as unclear risk of bias (Agrawal 2009; Bhakta 2019; Bowdle 2018; Furuya 2013; Higgins 2010; Kannampallil 2018; Thompson 2018). We rated one study as high risk of bias in this domain because it did not present a rationale for the numbers of data points (Narang 2013).

Protection against secular changes

We rated six studies as low risk of bias for protection against secular changes (Bhakta 2019; Kannampallil 2018; Ongering 2019; Schnipper 2018; Thompson 2018; Van Doormaal 2009), and seven as unclear risk of bias Agrawal 2009; Bowdle 2018; Burkoski 2019; Furuya 2013; Higgins 2010; Narang 2013; Seibert 2014). We assessed one study as high risk of bias because it used a before‐and‐after design, and the results could potentially be confounded by an unknown simultaneous intervention that was not measured in the analyses (Green 2015).

Shape of the intervention effect was specified

We rated two studies as low risk of bias for specifying the shape of the intervention effect (Seibert 2014; Van Doormaal 2009), and 12 as unclear risk of bias (Agrawal 2009; Bhakta 2019; Bowdle 2018; Burkoski 2019; Furuya 2013; Green 2015; Higgins 2010; Kannampallil 2018; Narang 2013; Ongering 2019; Schnipper 2018; Thompson 2018).

Conflict of interest

We rated 10 studies as low risk of bias for conflict of interest (Bhakta 2019; Furuya 2013; Green 2015; Higgins 2010; Kannampallil 2018; Ongering 2019; Schnipper 2018; Seibert 2014; Thompson 2018; Van Doormaal 2009), and three as unclear risk of bias (Agrawal 2009; Burkoski 2019; Narang 2013). We assessed one study as high risk of bias for conflict of interest because one of the study authors was the director and a shareholder of the company that supported the study and another author was a consultant for the same company (Bowdle 2018).

Other bias

We rated nine studies as low risk of bias for other bias (Bhakta 2019; Green 2015; Higgins 2010; Narang 2013; Ongering 2019; Schnipper 2018; Seibert 2014; Thompson 2018; Van Doormaal 2009), and five as unclear risk of bias (Agrawal 2009; Bowdle 2018; Burkoski 2019; Furuya 2013; Kannampallil 2018).

Effects of interventions

See: Table 1; Table 2; Table 3; Table 4; Table 5; Table 6; Table 7; Table 8; Table 9; Table 10; Table 11; Table 12; Table 13; Table 14; Table 15; Table 16

Summary of findings 1. Medication reconciliation versus no medication reconciliation.

Medication reconciliationversus no medication reconciliationfor reducing medication errors in adults in hospital settings
Patient or population: adults Setting: hospitals Intervention: medication reconciliation (MR) Comparison: no medication reconciliation
Outcomes^#	Relative effect (95% CI)	Absolute effect (95% CI)	№ of participants (studies)	Certainty of the evidence (GRADE)	Comments
Medication errors (Follow‐up 17 to 27 months)	OR 0.55 (0.17 to 1.74)	Not estimable	379 (3 RCTs)	⊕⊕⊝⊝ Low^a,b	Grouped outcomes Analysis 1.1
ADEs (Follow‐up 7 to 16 months)	OR 0.38 (0.18 to 0.80)	Not estimable	1336 (3 RCTs)	⊕⊕⊕⊝ Moderate^c	Grouped outcomes Analysis 1.2
Mortality during hospitalisation (Follow‐up 9 months)	RR 3.85 (0.44 to 33.89)	27 more per 1000 (from 5 fewer to 316 more)	212 (1 RCT)	⊕⊝⊝⊝ Very low^d,e	Baseline risk 1.0% Analysis 1.3
Length of stay (days) (Follow‐up 9 to 13 months)	Not estimable	MD ‐0.30 (‐1.93 to 1.33)	527 (3 RCTs)	⊕⊕⊝⊝ Low^a,b	Analysis 1.4
Quality of life (VAS 0‐10; EQ‐5D‐3L) (Follow‐up 10 months)	Not estimable	MD ‐1.51 (‐10.04 to 7.02)	131 (1 RCT)	⊕⊕⊝⊝ Low^a,b	(high score better) Analysis 1.5
Discrepancy resolution (Follow‐up 10 months)	RR 7.48 (5.62 to 9.95)	860 more per 1000 (from 613 more to 1000 more)	564 (1 RCT)	⊕⊕⊕⊝ Moderate^b	Analysis 1.6
ADEs: adverse drug events; CI: confidence interval; EQ‐5D‐3L: EuroQol 5‐dimension survey; OR: odds ratio; RR: risk ratio; VAS: visual analogue scale
GRADE Working Group grades of evidence High certainty: we are very confident that the true effect lies close to that of the estimate of the effect. Moderate certainty: we are moderately confident in the effect estimate. The true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different. Low certainty: our confidence in the effect estimate is limited. The true effect may be substantially different from the estimate of the effect. Very low certainty: we have very little confidence in the effect estimate. The true effect is likely to be substantially different from the estimate of effect.

Medication reconciliation: pharmacist versus other professionals for reducing medication errors in adults in hospital settings
Patient or population: adults Setting: hospitals Intervention: medication reconciliation by pharmacist Comparison: medication reconciliation by other professionals
Outcomes	Relative effect (95% CI)	Absolute effect (95% CI)	№ of participants (studies)	Certainty of the evidence (GRADE)	Comments
Medication errors (Follow‐up 1 to 5 months)	OR 0.21 (0.09 to 0.48)	Not estimable	2648 (8 RCTs)	⊕⊕⊝⊝ Low^a	Grouped outcomes Analysis 2.1
ADEs (Follow‐up 18 months to 5 years)	OR 1.34 (0.73 to 2.44)	Not estimable	2873 (3 RCTs)	⊕⊕⊝⊝ Low^b,c	Grouped outcomes Analysis 2.2
Mortality during hospitalisation (Follow‐up 13 to 21 months)	RR 0.99 (0.57 to 1.73)	0 fewer per 1000 (from 20 fewer to 34 more)	1000 (2 RCTs)	⊕⊕⊕⊝ Moderate^c	Baseline risk 4.6% Analysis 2.3 Mortality at six months RR 0.54 (95% CI 0.22 to 1.32)
Readmission at 1 month (Follow‐up 13 to 21 months)	RR 0.93 (0.76 to 1.14)	20 fewer per 1000 (from 67 fewer to 39 more)	997 (2 RCTs)	⊕⊕⊕⊝ Moderate^c	Baseline risk 28% Analysis 2.4
Length of stay (days) (Follow‐up 18 to 21 months)	Not estimable	MD ‐0.25 (‐1.05 to 0.56)	3983 (6 RCTs)	⊕⊕⊝⊝ Low^b,c	General wards inpatients (MD ‐0.25, 95% CI ‐1.09 to 0.59) Inpatients coming from ICU (MD ‐0.30, 95% CI ‐6.71 to 6.11) Test for subgroup differences: I² = 0% Analysis 2.5
Quality of life (VAS 0‐10; EQ‐5D‐3L) (Follow‐up 18 months)	Not estimable	MD 0.00 (‐14.09 to 14.09)	724 (1 RCT)	⊕⊕⊝⊝ Low^b,c	(High score better) Analysis 2.6
Discrepancy resolution (Follow‐up 6 to 13 months)	OR 4.80 (1.81 to 12.76)	Not estimable	1449 (3 RCTs)	⊕⊕⊝⊝ Low^a,b	Grouped outcomes Analysis 2.7
ADEs: adverse drug events; CI: Confidence interval; EQ‐5D‐3L: EuroQol 5‐dimension survey; ICU: intensive care unit; MD: mean difference; OR: odds ratio; RR: risk ratio; VAS: visual analogue scale
GRADE Working Group grades of evidence High certainty: we are very confident that the true effect lies close to that of the estimate of the effect. Moderate certainty: we are moderately confident in the effect estimate. The true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different. Low certainty: our confidence in the effect estimate is limited. The true effect may be substantially different from the estimate of the effect. Very low certainty: we have very little confidence in the effect estimate. The true effect is likely to be substantially different from the estimate of effect.

Medication reconciliationby pharmacist: database‐assisted versus unassisted MR for reducing medication errors in adults in hospital settings
Patient or population: adults Setting: hospitals Intervention: database‐assisted medication reconciliation performed by pharmacists Comparison: unassisted nedication reconciliation performed by pharmacists
Outcomes^#	Relative effect (95% CI)	Absolute effect (95% CI)	№ of participants (studies)	Certainty of the evidence (GRADE)	Comments
Potential ADEs ( ≥ 1 per patient) (Follow‐up 3 to 20 months)	OR 0.26 (0.10 to 0.64)	77 more per 1000 (from 7 fewer to 163 more)	3326 (2 RCTs)	⊕⊕⊝⊝ Low^a,b	Baseline risk 39.8% Analysis 3.1
Length of stay (days) (Follow‐up 31 months)	Not estimable	MD 1.00 (‐0.17 to 2.17)	311 (1 RCT)	⊕⊕⊝⊝ Low^b,c	Analysis 3.2
Discrepancy resolution (Follow‐up 3 to 31 months)	OR 1.37 (0.97 to 1.93)	1 fewer per 1000 (from 2 fewer to 1 fewer)	791 (2 RCTs)	⊕⊕⊝⊝ Low^a,c	Analysis 3.3
ADEs: adverse drug events; CI: confidence interval; OR: odds ratio; MD: Mean difference
GRADE Working Group grades of evidence High certainty: we are very confident that the true effect lies close to that of the estimate of the effect. Moderate certainty: we are moderately confident in the effect estimate. The true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different. Low certainty: our confidence in the effect estimate is limited. The true effect may be substantially different from the estimate of the effect. Very low certainty: we have very little confidence in the effect estimate. The true effect is likely to be substantially different from the estimate of effect.

Medication reconciliation by trained pharmacist technician versus pharmacist for reducing medication errors in adults in hospital settings
Patient or population: adults Setting: hospitals Intervention: medication reconciliation by trained pharmacist technician Comparison: medication reconciliation by pharmacist
Outcomes^#	Relative effect (95% CI)	Absolute effect (95% CI)	№ of participants (studies)	Certainty of the evidence (GRADE)	Comments
Medication errors (Follow‐up 7 months)	OR 0.65 (0.25 to 1.70)	Not estimable	306^ƚ (2 RCTs)	⊕⊝⊝⊝ Very low^a,b,c	Grouped outcomes. Analysis 4.1 ^ƚThe number of participants in 1 of the studies is unknown because the study analysed prescriptions.
Length of stay (days) (Follow‐up not available)	Not estimable	MD ‐0.30 (‐2.12 to 1.52)	183 (1 RCT)	⊕⊕⊝⊝ Low^a,c	Analysis 4.2
CI: confidence interval; MD: mean difference; OR: odds ratio
GRADE Working Group grades of evidence High certainty: we are very confident that the true effect lies close to that of the estimate of the effect. Moderate certainty: we are moderately confident in the effect estimate. The true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different. Low certainty: our confidence in the effect estimate is limited. The true effect may be substantially different from the estimate of the effect. Very low certainty: we have very little confidence in the effect estimate. The true effect is likely to be substantially different from the estimate of effect.

Medication reconciliation: before versus at admission for reducing medication errors in adults in hospital settings
Patient or population: adults Setting: hospitals (emergency department) Intervention: medication reconciliation before admission Comparison: medication reconciliation after admission
Outcomes^#	Relative effect (95% CI)	Absolute effect (95% CI)	№ of participants (studies)	Certainty of the evidence (GRADE)	Comments
Identified discrepancies per patient (Follow‐up 1 month)	Not estimable	MD 1.27 (0.46, 2.08)	307 (1 RCT)	⊕⊕⊝⊝ Low^a,b	Analysis 5.1
CI: confidence interval; MD: mean difference
GRADE Working Group grades of evidence High certainty: we are very confident that the true effect lies close to that of the estimate of the effect. Moderate certainty: we are moderately confident in the effect estimate. The true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different. Low certainty: our confidence in the effect estimate is limited. The true effect may be substantially different from the estimate of the effect. Very low certainty: we have very little confidence in the effect estimate. The true effect is likely to be substantially different from the estimate of effect.

Medication reconciliation: 1 or 2 versus 4 charts open simultaneously for reducing medication errors in adults in hospital settings
Patient or population: adults Setting: hospitals Intervention: 1 or 2 charts open simultaneously Comparison: 4 charts open simultaneously
Outcomes^#	Relative effect (95% CI)	Absolute effect (95% CI)	№ of participants (studies)	Certainty of the evidence (GRADE)	Comments
Medication errors (ITS study) (Follow‐up 70 months)	Not estimable	MD ‐0.19 (‐0.58, 0.20)	11,504 (1 ITS study)	⊕⊝⊝⊝ Very low^a	Analysis 6.1
CI: confidence interval; MD: mean difference
GRADE Working Group grades of evidence High certainty: we are very confident that the true effect lies close to that of the estimate of the effect. Moderate certainty: we are moderately confident in the effect estimate. The true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different. Low certainty: our confidence in the effect estimate is limited. The true effect may be substantially different from the estimate of the effect. Very low certainty: we have very little confidence in the effect estimate. The true effect is likely to be substantially different from the estimate of effect.

Medication reconciliation: multimodal intervention vs usual care for reducing medication errors in adults in hospital settings
Patient or population: adults Setting: hospitals Intervention: medication reconciliation: multimodal intervention Comparison: medication reconciliation: usual care
Outcomes^#	Relative effect (95% CI)	Absolute effect (95%CI)	№ of participants (studies)	Certainty of the evidence (GRADE)	Comments
Medication error (Follow‐up 24 months)	RR 0.92 (0.87, 0.97)	Not estimable	1648 (1 ITS study)	⊕⊝⊝⊝ Very low^a	Unintended discrepancies ( ≥ 1 per patient) Analysis 7.1
Potential ADEs (Follow‐up 24 months)	RR 0.97 (0.86, 1.09)	Not estimable	1648 (1 ITS study)	⊕⊝⊝⊝ Very low^a,b	Analysis 7.2
Discrepancy resolution (Follow‐up 6 months)	RR 2.14 (1.81 to 2.53)	417 more per 1000 (from 297 more to 560 more)	487 (1 RCT)	⊕⊕⊕⊝ Moderate^a	Analysis 7.3
ADEs: adverse drug events; CI: confidence interval; RR: risk ratio
GRADE Working Group grades of evidence High certainty: we are very confident that the true effect lies close to that of the estimate of the effect. Moderate certainty: we are moderately confident in the effect estimate. The true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different. Low certainty: our confidence in the effect estimate is limited. The true effect may be substantially different from the estimate of the effect. Very low certainty: we have very little confidence in the effect estimate. The true effect is likely to be substantially different from the estimate of effect.

CPOE/CDSS compared to control/paper‐based systems for reducing medication errors in adults in hospital settings
Patient or population: adults Setting: hospitals Intervention: computerised physician order entry (CPOE)/clinical decision support systems (CDSS) Comparison: control/paper‐based system
Outcomes^#	Relative effect (95% CI)	Absolute effect (95% CI)	№ of participants (studies)	Certainty of the evidence (GRADE)	Comments
Medication errors (Follow‐up 4 months)	OR 0.74 (0.31 to 1.79)	Not estimable	88 (2 RCTs)	⊕⊕⊕⊝ Moderate^a	Grouped outcomes. In fact, this is one RCT but with results separated for first‐year doctors and other doctors. Analysis 8.1
ADEs (Follow‐up 1 to 12 months)	OR 0.24 (0.04 to 1.50)	Not estimable	827 (2 RCTs)	⊕⊝⊝⊝ Very low^a,b,c	Grouped outcomes. The ITS study Ongering 2019 favours the paper‐based arm on serious Preventable ADEs per prescriptions (MD 0.12, 95% CI ‐0.03 to 0.27; n = 2711 patients). Analysis 8.2
Mortality during hospitalisation (Follow‐up 12 months)	RR 1.04 (0.54 to 2.01)	2 more per 1000 (from 21 fewer to 46 more)	737 (1 RCT)	⊕⊝⊝⊝ Very low^a,b	Analysis 8.3
Length of stay (days) (Follow‐up 12 months)	Not estimable	MD ‐1.00 (‐2.05 to 0.05)	737 (1 RCT)	⊕⊝⊝⊝ Very low^a,b	Analysis 8.4
ADEs: adverse drug events; CI: confidence interval; OR: odds ratio; RR: risk ratio
GRADE Working Group grades of evidence High certainty: we are very confident that the true effect lies close to that of the estimate of the effect. Moderate certainty: we are moderately confident in the effect estimate. The true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different. Low certainty: our confidence in the effect estimate is limited. The true effect may be substantially different from the estimate of the effect. Very low certainty: we have very little confidence in the effect estimate. The true effect is likely to be substantially different from the estimate of effect.

CPOE/CDSS: improved compared to standard CPOE/CDSS for reducing medication errors in adults in hospital settings
Patient or population: adults Setting: hospitals Intervention: improved CPOE/CDSS Comparison: standard CPOE/CDSS
Outcomes^#	Relative effect (95% CI)	Absolute effect (95% CI)	№ of participants (studies)	Certainty of the evidence (GRADE)	Comments
Medication errors (Follow‐up 3 to 47 months)	OR 0.85 (0.74 to 0.97)	Not estimable	630 (2 RCTs)	⊕⊕⊕⊝ Moderate^a	Analysis 9.1.1 2 ITS studies (OR 0.77, 95% CI 0.37 to 1.62; participants = 2382 + ^ƚGreen 2015 sample not reported)
Medication errors (Follow‐up 3 to 47 months)	OR 0.77 0.37 to 1.62	Not estimable	2382^ƚ (2 ITS studies)	⊕⊕⊕⊝ Moderate^a	Analysis 9.1.2 ^ƚGreen 2015 sample not reported
ADEs (Follow‐up 1 to 3 months)	OR 0.82 (0.71 to 0.94)	Not estimable	2382ƚ (2 ITS studies)	⊕⊕⊕⊝ Moderate^a	Analysis 9.2 ^ƚGreen 2015 sample was not reported
ADEs: adverse drug events; CI: confidence interval; OR: odds ratio
GRADE Working Group grades of evidence High certainty: we are very confident that the true effect lies close to that of the estimate of the effect. Moderate certainty: we are moderately confident in the effect estimate. The true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different. Low certainty: our confidence in the effect estimate is limited. The true effect may be substantially different from the estimate of the effect. Very low certainty: we have very little confidence in the effect estimate. The true effect is likely to be substantially different from the estimate of effect.

CPOE/CDSS: prioritised versus no prioritised alerts for reducing medication errors in adults in hospital settings
Patient or population: adults Setting: hospitals Intervention: CPOE/CDSS: prioritised alerts Comparison: CPOE/CDSS: non‐prioritised alerts
Outcomes^#	Relative effect (95% CI)	Absolute effect (95% CI)	№ of participants (studies)	Certainty of the evidence (GRADE)	Comments
Resolved potential ADEs (per prescriptions) (Follow‐up 21 months)	Not estimable	MD 1.98 (1.65 to 2.31)	Not available (1 ITS study)	⊕⊕⊝⊝ Low^a	The unit of analysis was prescriptions Analysis 10.1
ADEs: adverse drug events; CI: confidence interval; MD: mean difference
GRADE Working Group grades of evidence High certainty: we are very confident that the true effect lies close to that of the estimate of the effect. Moderate certainty: we are moderately confident in the effect estimate. The true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different. Low certainty: our confidence in the effect estimate is limited. The true effect may be substantially different from the estimate of the effect. Very low certainty: we have very little confidence in the effect estimate. The true effect is likely to be substantially different from the estimate of effect.

Barcoding versus no barcoding for reducing medication errors in adults in hospital settings
Patient or population: adults Setting: hospitals Intervention: barcoding Comparison: no barcoding
Outcomes^#	Relative effect (95% CI)	Absolute effect (95% CI)	№ of participants (studies)	Certainty of the evidence (GRADE)	Comments
Medication errors (Follow‐up 22 to 79 months)	OR 0.69 (0.59 to 0.79)	Not estimable	50,545^ƚ (2 ITS studies)	⊕⊕⊝⊝ Low^a	Grouped outcomes Analysis 11.1 ^ƚThe number of participants is unknown for 1 study because it used prescriptions as the unit of analysis.
CI: confidence interval; OR: odds ratio
GRADE Working Group grades of evidence High certainty: we are very confident that the true effect lies close to that of the estimate of the effect. Moderate certainty: we are moderately confident in the effect estimate. The true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different. Low certainty: our confidence in the effect estimate is limited. The true effect may be substantially different from the estimate of the effect. Very low certainty: we have very little confidence in the effect estimate. The true effect is likely to be substantially different from the estimate of effect.

No.	Search terms	Results
1	medication errors/	13014
2	inappropriate prescribing/	2991
3	medication reconciliation/	1066
4	((drug? or prescri* or medicat) adj3 alert).ti,ab,kf.	1119
5	((drug? or medication? or medicine? or dose or dosage? or dosing or prescri* or order?) adj2 wrong*).ti,ab,kf.	722
6	(medication adj1 (review? or reconcil* or counsel* or error? or safety)).ti,ab,kf.	9945
7	((inappropriate* or appropriate) adj1 prescri).ti,ab,kf.	3282
8	(prescri* adj4 (safe* or error*)).ti,ab,kf.	4055
9	decision support systems, clinical/ or medical order entry systems/	9292
10	prescri*.ti,ab,kf,hw.	222837
11	9 and 10	1433
12	or/1‐8,11	28110
13	(unit or units).ti,ab,kf,hw.	662106
14	hospital*.ti,ab,kf,hw,in.	4988144
15	or/13‐14	5415306
16	12 and 15	14195
17	randomized controlled trial.pt.	498732
18	controlled clinical trial.pt.	93522
19	multicenter study.pt.	264836
20	pragmatic clinical trial.pt.	1277
21	(randomis* or randomiz* or randomly).ti,ab,kf.	872592
22	groups.ab.	1998032
23	(trial or multicenter or multi center or multicentre or multi centre).ti.	250811
24	(intervention? or effect? or impact? or controlled or control group? or (pre adj5 post) or ((pretest or pre test) and (posttest or post test)) or quasiexperiment* or quasi experiment* or pseudo experiment* or pseudoexperiment* or evaluat* or time series or time point? or repeated measur*).ti,ab,kf.	9334407
25	non‐randomized controlled trials as topic/	610
26	interrupted time series analysis/	750
27	or/17‐26	10423000
28	exp animals/	22903001
29	humans/	18238668
30	28 not (28 and 29)	4664333
31	review.pt.	2600548
32	meta analysis.pt.	109833
33	news.pt.	198904
34	comment.pt.	824770
35	editorial.pt.	515077
36	cochrane database of systematic reviews.jn.	14808
37	comment on.cm.	824716
38	(systematic review or literature review).ti.	148362
39	or/30‐38	8493610

No.	Search terms	Results
1	*medication error/	8365
2	*inappropriate prescribing/	1423
3	*medication therapy management/	4166
4	((drug? or prescri* or medicat) adj3 alert).ti,ab,kw.	1889
5	((drug? or medication? or medicine? or dose or dosage? or dosing or prescri* or order?) adj2 wrong*).ti,ab,kw.	1502
6	(medication adj1 (review? or reconcil* or counsel* or error? or safety)).ti,ab,kw.	18748
7	((inappropriate* or appropriate) adj1 prescri).ti,ab,kw.	5693
8	(prescri* adj4 (safe* or error*)).ti,ab,kw.	7279
9	*clinical decision support system/	1316
10	*physician order entry system/	98
11	prescri*.ti,ab,kw,hw.	418777
12	or/9‐10	1401
13	11 and 12	246
14	or/1‐8,13	38313
15	(unit or units).ti,ab,kw,hw.	913799
16	hospital*.ti,ab,kw,hw,in.	8046625
17	or/15‐16	8545861
18	14 and 17	22561
19	randomized controlled trial/	586757
20	controlled clinical trial/	463266
21	quasi experimental study/	6357
22	pretest posttest control group design/	441
23	time series analysis/	24966
24	experimental design/	17983
25	multicenter study/	240605
26	(randomis* or randomiz* or randomly).ti,ab.	1223708
27	groups.ab.	2778000
28	(trial or multicentre or multicenter or multi centre or multi center).ti.	351927
29	(intervention? or effect? or impact? or controlled or control group? or (before adj5 after) or (pre adj5 post) or ((pretest or pre test) and (posttest or post test)) or quasiexperiment* or quasi experiment* or pseudo experiment* or pseudoexperiment* or evaluat* or time series or time point? or repeated measur*).ti,ab.	11970676
30	or/19‐29	13353223
31	(systematic review or literature review).ti.	178369
32	"cochrane database of systematic reviews".jn.	13931
33	exp animals/ or exp invertebrate/ or animal experiment/ or animal model/ or animal tissue/ or animal cell/ or nonhuman/	26825418
34	human/ or normal human/ or human cell/	20517865
35	33 not (33 and 34)	6370493
36	31 or 32 or 35	6561150
37	30 not 36	10292608
38	18 and 37	13764

No.	Search terms	Results
#1	[mh “medication errors”]	407
#2	[mh “inappropriate prescribing”]	130
#3	[mh “medication reconciliation”]	76
#4	((drug? or prescri* or medicat) near/3 alert):ti,ab,kw	247
#5	((drug? or medication? or medicine? or dose or dosage? or dosing or prescri* or order?) near/2 wrong*):ti,ab,kw	55
#6	(medication near/1 (review? or reconcil* or counsel* or error? or safety)):ti,ab,kw	1565
#7	((inappropriate* or appropriate) near/1 prescri):ti,ab,kw	518
#8	(prescri* near/4 (safe* or error*)):ti,ab,kw	574
#9	[mh “decision support systems, clinical”] or [mh “medical order entry systems”]	393
#10	prescri*	35146
#11	#9 and #10	87
#12	{or #1‐#8, #11}	2695
#13	(unit or units)	107814
#14	hospital*	335426
#15	#13 or #14	396828
#16	#12 and #15	1478

No.	Search terms	Results
S1	(MH "Medication Errors+")	15,070
S2	(MH "Medication Reconciliation")	1,518
S3	TI ((drug? or prescri* or medicat) N3 alert) OR AB ((drug? or prescri* or medicat) N3 alert)	888
S4	TI ((drug? or medication? or medicine? or dose or dosage? or dosing or prescri* or order?) N2 wrong) OR AB ((drug? or medication? or medicine? or dose or dosage? or dosing or prescri or order?) N2 wrong*)	446
S5	TI ((inappropriate* or appropriate) N1 prescri) OR AB ((inappropriate* or appropriate) N1 prescri)	2,092
S6	(MH "Decision Support Systems, Clinical")	4,682
S7	(MH "Electronic Order Entry")	2,978
S8	S6 OR S7	7,207
S9	TX prescri*	115,470
S10	S8 AND S9	1,299
S11	S1 OR S2 OR S3 OR S4 OR S5 OR S10	18,919
S12	TX (unit or units)	348,687
S13	TX hospital*	1,473,988
S14	S12 OR S13	1,629,464
S15	S11 AND S14	8,059
S16	PT randomized controlled trial	86,198
S17	PT clinical trial	85,810
S18	PT research	1,949,263
S19	(MH "Randomized Controlled Trials")	89,376
S20	(MH "Clinical Trials")	151,432
S21	(MH "Intervention Trials")	7,249
S22	(MH "Nonrandomized Trials")	455
S23	(MH "Experimental Studies")	23,867
S24	(MH "Pretest‐Posttest Design+")	40,092
S25	(MH "Quasi‐Experimental Studies+")	14,063
S26	(MH "Multicenter Studies")	158,080
S27	(MH "Health Services Research")	13,903
S28	TI ( randomis* or randomiz* or randomly) OR AB ( randomis* or randomiz* or randomly)	273,171
S29	TI (trial or effect* or impact* or intervention* or pre N5 post or ((pretest or "pre test") and (posttest or "post test")) or quasiexperiment* or quasi W0 experiment* or pseudo experiment* or pseudoexperiment* or evaluat* or "time series" or time W0 point* or repeated W0 measur) OR AB (trial or effect or impact* or intervention* or pre N5 post or ((pretest or "pre test") and (posttest or "post test")) or quasiexperiment* or quasi W0 experiment* or pseudo experiment* or pseudoexperiment* or evaluat* or "time series" or time W0 point* or repeated W0 measur*)	1,830,248
S30	S16 OR S17 OR S18 OR S19 OR S20 OR S21 OR S22 OR S23 OR S24 OR S25 OR S26 OR S27 OR S28 OR S29	2,883,722
S31	S15 AND S30	5,097

Organisational changes: reduced versus unreduced working hours for reducing medication errors in adults in hospital settings
Patient or population: adults Setting: hospitals (intensive care unit) Intervention: reduced working hours Comparison: unreduced working hours
Outcomes^#	Relative effect (95% CI)	Absolute effect (95% CI)	№ of participants (studies)	Certainty of the evidence (GRADE)	Comments
Medication errors (Follow‐up not available)	RR 0.83 (0.63 to 1.09)	17 fewer per 1000 (from 37 fewer to 9 more)	634 (1 RCT)	⊕⊕⊝⊝ Low^a,b	Serious medication errors per patient‐days Analysis 12.1
CI: confidence interval; RR: risk ratio
GRADE Working Group grades of evidence High certainty: we are very confident that the true effect lies close to that of the estimate of the effect. Moderate certainty: we are moderately confident in the effect estimate. The true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different. Low certainty: our confidence in the effect estimate is limited. The true effect may be substantially different from the estimate of the effect. Very low certainty: we have very little confidence in the effect estimate. The true effect is likely to be substantially different from the estimate of effect.

Feedback on prescribing errors versus no feedback for reducing medication errors in adults in hospital settings
Patient or population: adults Setting: hospitals Intervention: feedback on prescribing errors Comparison: no feedback
Outcomes^#	Relative effect (95% CI)	Absolute effect (95% CI)	№ of participants (studies)	Certainty of the evidence (GRADE)	Comments
Medication errors (Follow‐up 1 to 4 months)	OR 0.47 (0.33 to 0.67)	Not estimable	384^ƚ (4 RCTs)	⊕⊕⊝⊝ Low^a	Grouped outcomes Analysis 13.1 ^ƚOnly 1 out of 4 RCTs reported participants.
CI: confidence interval; OR: odds ratio
GRADE Working Group grades of evidence High certainty: we are very confident that the true effect lies close to that of the estimate of the effect. Moderate certainty: we are moderately confident in the effect estimate. The true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different. Low certainty: our confidence in the effect estimate is limited. The true effect may be substantially different from the estimate of the effect. Very low certainty: we have very little confidence in the effect estimate. The true effect is likely to be substantially different from the estimate of effect.

Education versus no education on prescribing or administration for reducing medication errors in adults in hospital settings
Patient or population: adults Setting: hospitals Intervention: education on prescribing or administration Comparison: no education on prescribing or administration
Outcomes^#	Relative effect (95%CI)	Absolute effect (95% CI)	№ of participants (studies)	Certainty of the evidence (GRADE)	Comments
Medication errors (Follow‐up 1 to 4 months)	OR 1.21 (0.93 to 1.58)	Not estimable	30^ƚ (4 RCTs)	⊕⊝⊝⊝ Very low^a,b,c	Grouped outcomes Analysis 15.1 ^ƚOnly 1 out of 4 RCTs reported participants. Education on prescriptions (physicians) OR 1.11 (95% CI 0.88 to 1.39) Education on administration (nurses) OR 1.64 (95% CI 0.88 to 3.08)
CI: confidence interval; OR: odds ratio
GRADE Working Group grades of evidence High certainty: we are very confident that the true effect lies close to that of the estimate of the effect. Moderate certainty: we are moderately confident in the effect estimate. The true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different. Low certainty: our confidence in the effect estimate is limited. The true effect may be substantially different from the estimate of the effect. Very low certainty: we have very little confidence in the effect estimate. The true effect is likely to be substantially different from the estimate of effect.

Dispensing system versus control for reducing medication errors in adults in hospital settings
Patient or population: adults
Outcomes^#	Relative effect (95% CI)	Absolute effect (95% CI)	№ of participants (studies)	Certainty of the evidence (GRADE)	Comments
Medication errors (surgical wards) (Follow‐up 1 month)	OR 0.61 (0.47 to 0.79)	Not estimable	1775^ƚ (2 RCTs)	⊕⊕⊝⊝ Low^a	Grouped outcomes Analysis 16.1 ^ƚ1 out of 2 RCTs did not report participants.
Medication errors (operating rooms) (Follow‐up 5 to 12 months)	OR 0.92 (0.75 to 1.13)	Not estimable	2310 (2 RCTs)	⊕⊝⊝⊝ Very low^b,c,d	Grouped outcomes Analysis 16.2
CI: confidence interval; OR: odds ratio
GRADE Working Group grades of evidence High certainty: we are very confident that the true effect lies close to that of the estimate of the effect. Moderate certainty: we are moderately confident in the effect estimate. The true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different. Low certainty: our confidence in the effect estimate is limited. The true effect may be substantially different from the estimate of the effect. Very low certainty: we have very little confidence in the effect estimate. The true effect is likely to be substantially different from the estimate of effect.

Outcomes reported by authors of included studies	Outcome group	Type of medication error
Administration errors (per monthly administered doses)	Medication errors	Administration errors
Administration (per prescription)	Medication errors	Administration errors
Administration errors (per weekly anaesthetised patients)	Medication errors	Administration errors
Administration errors (per opportunities for error, by nurse)	Medication errors	Administration errors
Medication errors (per administered doses)	Medication errors	Administration errors
Medication errors (per administration)	Medication errors	Administration errors
Medication errors (per monthly administered doses)	Medication errors	Administration errors
All errors	Medication errors	All type of medication errors
Medication errors (per prescriptions)	Medication errors	All type of medication errors
Medication errors (per weekly prescriptions)	Medication errors	All type of medication errors
Medication errors per patient at discharge	Medication errors	All type of medication errors
Pharmacist only perform MR (all errors)	Medication errors	All type of medication errors
Serious medication errors per patient‐days by interns	Medication errors	All type of medication errors
Dispensing errors (per monthly administered doses)	Medication errors	Dispensing error
Dispensing errors (per prescriptions)	Medication errors	Dispensing error
Potential ADEs (≥ 1 per patient)	Medication errors	Potential ADEs
Potential ADEs (≥ 1 per patient)	Medication errors	Potential ADEs
Potential ADEs (per patient)	Medication errors	Potential ADEs
Potential ADEs (per prescriptions)	Medication errors	Potential ADEs
Potential ADEs + ADEs (≥ 1 per patient)	Medication errors	Potential ADEs
Serious PADEs (≥ 1 per patient)	Medication errors	Potential ADEs
Serious PADEs (per prescriptions)	Medication errors	Potential ADEs
Discrepancies errors (≥ 1 per patient)	Medication errors	Prescribing errors
Discrepancy errors	Medication errors	Prescribing errors
Discrepancy errors (per prescriptions)	Medication errors	Prescribing errors
Discrepancy errors per patient	Medication errors	Prescribing errors
Duplication errors (per prescription)	Medication errors	Prescribing errors
ID‐reentry function	Medication errors	Prescribing errors
ID‐verify alert	Medication errors	Prescribing errors
Medications omitted (per prescriptions)	Medication errors	Prescribing errors
Only omissions	Medication errors	Prescribing errors
Prescribing errors (per doctor)	Medication errors	Prescribing errors
Pharmacist history and supplementary prescribing	Medication errors	Prescribing errors
Pharmacist only perform MR (dosing errors)	Medication errors	Prescribing errors
Pharmacist only perform MR (frequency dosing errors)	Medication errors	Prescribing errors
Pharmacists perform MR + prescribing (dosing errors)	Medication errors	Prescribing errors
Pharmacists perform MR + prescribing (frequency dosing errors)	Medication errors	Prescribing errors
Prescribing error (per order session)	Medication errors	Prescribing errors
Prescribing errors (per prescriptions)	Medication errors	Prescribing errors
Serious discrepancy errors per patient	Medication errors	Prescribing errors
Serious prescribing errors (per prescriptions)	Medication errors	Prescribing errors
Unintended discrepancies (≥ 1 per patient)	Medication errors	Prescribing errors
ADEs	Adverse drug events (ADEs)
ADEs (≥ 1 per patient)	Adverse drug events (ADEs)
ADEs (per monthly administered doses)	Adverse drug events (ADEs)
ADEs (per prescriptions)	Adverse drug events (ADEs)
ADEs due to discrepancies per patient	Adverse drug events (ADEs)
ADEs due to medication error	Adverse drug events (ADEs)
ADEs per admissions	Adverse drug events (ADEs)
Medication error + ADEs (per monthly administered medication doses)	Adverse drug events (ADEs)
Preventable ADEs (≥ 1 per patient)	Adverse drug events (ADEs)
Preventable ADEs (per monthly patients admitted)	Adverse drug events (ADEs)
Preventable ADEs per admissions	Adverse drug events (ADEs)
Serious ADEs	Adverse drug events (ADEs)
Serious ADEs per admissions	Adverse drug events (ADEs)
Mortality	Mortality
Mortality at 6 months	Mortality
Mortality during hospitalisation	Mortality
Hospitalisations due to ADEs	Hospitalisations
Readmisson at 1 month	Hospitalisations
Length of stay (days)	LoS
Length of stay (days)	LoS
Quality of life (visual analogue scale (VAS) 0‐10; EQ‐5D‐3L)	QoL
Discrepancies resolutions (≥ 1 per patient)	Discrepancies resolutions
Discrepancy resolution (≥ 1 per patient)	Discrepancies resolutions
Discrepancy resolutions (per discrepancies at discharge)	Discrepancies resolutions
Discrepancy resolutions (per discrepancies)	Discrepancies resolutions
Resolved Potential ADEs (per prescriptions)	Resolution of MEs
Identified discrepancies (≥ 1 per patient)	Identified discrepancies
Identified discrepancies per patient	Identified discrepancies

Field	Search terms
Other terms	medication error AND hospital
Study type	interventional studies
Age	adult, older adult

EPOC group taxonomy categories	Intervention included (comparison #)
Delivery arrangements
Who receives care and when	MR: before versus at admission (#5)
Who provides care	MR: pharmacist versus other professionals (#2)
Who provides care/Co‐ordination of care	MR by pharmacist: team/highly trained pharmacist versus standard pharmacist (#4)
Health Information and communication technology	CPOE/CDSS (#8, #9, #10); barcoding (#11); dispensing systems (#16); database‐assisted medication reconciliation conducted by pharmacists (#3); one to two charts versus four charts open simultaneously for MR (#6)
Working conditions of health workers	Organisational changes: reduced versus unreduced working hours (#12)
Coordination of care / Integration	Multimodal intervention (#7)
Implementation strategies
Interventions targeted at healthcare worker practice	Feedback on prescribing errors; education (#13, #14, #15)
Types of problems targeted at healthcare worker practice	Medication reconciliation (#1)
Multifaceted interventions	Multimodal intervention (#7)
Financial arrangements
‐	No included interventions
Governance arrangements
‐	No included interventions

Study ID	EPOC taxonomy	Comparison #
Greengold 2003	Implementation strategy	15
Schneider 2006	Implementation strategy	15
Aag 2014	Delivery arrangement	2
Adelman 2019	Delivery arrangement	6
Al‐Hashar 2018	Implementation strategy	1
Becerra‐Camargo 2015	Delivery arrangement	2
Beckett 2012	Delivery arrangement	2
Bell 2016	Delivery arrangement	2
Cadman 2017	Implementation strategy	1
Chiu 2018	Implementation strategy	1
De winter 2011	Delivery arrangement	2
George 2011	Delivery arrangement	2
Graabaek 2019	Delivery arrangement	2
Heselmans 2015	Delivery arrangement	2
Hickman 2018	Delivery arrangement	4
Khalil 2016	Delivery arrangement	2
Scullin 2007	Delivery arrangement	2
Lind 2017	Delivery arrangement	2
Piqueras 2015	Implementation strategy	1
Pevnick 2018	Delivery arrangement	2, 4
Schmader 2004	Delivery arrangement	2
Nielsen 2017	Implementation strategy	1
Bolas 2004	Implementation strategy	1
Juanes 2018	Implementation strategy	1
Leung 2017	Implementation strategy	13, 14, 15
Gursanscky 2018	Implementation strategy	13, 14, 16
Farris 2014	Delivery arrangement	2
Kwan 2007	Delivery arrangement	2
Hale 2013	Delivery arrangement + Implementation strategy	2, 13
Marotti 2011	Delivery arrangement	2
Quach 2015	Delivery arrangement	5
Tong 2016	Delivery arrangement	2
Willoch 2012	Implementation strategy	1
SUPERPILL 2015	Delivery arrangement	2
Vega 2016	Implementation strategy	1
Tompson 2012	Delivery arrangement	7
Merry 2011	Delivery arrangement	16
McCoy 2012	Delivery arrangement	9
Landrigan 2004	Delivery arrangement	12
Wang 2017	Delivery arrangement	16
Boockvar 2017	Delivery arrangement	3
Fernandes 2011	Delivery arrangement	3
Tamblyn 2018	Delivery arrangement	3
OSullivan 2015	Delivery arrangement	8
Schnipper 2009	Delivery arrangement	9
Ding 2012	Delivery arrangement	16
Barker 1984	Delivery arrangement	16
Colpaert 2006	Delivery arrangement	8
Redwood 2013	Delivery arrangement	8
Adelman 2013	Delivery arrangement	9
Gordon 2017	Implementation strategy	13
Seibert 2014	Delivery arrangement	11
Schnipper 2018	Implementation strategy	7
Agrawal 2009	Delivery arrangement	9
Narang 2013	Delivery arrangement	11
Bhakta 2019	Delivery arrangement	10
Green 2015	Delivery arrangement	9
Kannampallil 2018	Delivery arrangement	6
Ongering 2019	Delivery arrangement	8
Furuya 2013	Delivery arrangement	9
van Doormaal 2009	Delivery arrangement	8
Bowdle 2018	Delivery arrangement	11
Higgins 2010	Delivery arrangement	11
Burkoski 2019	Delivery arrangement	8, 11
Thompson 2018	Implementation strategy	11
Comparison # Medication reconciliation (MR) versus no MR MR: pharmacist versus other professionals MR by pharmacist: database‐assisted versus unassisted MR by pharmacist: team/highly trained pharmacist versus standard pharmacist MR: before versus at admission MR: one or two charts versus 4 charts open simultaneously MR: multimodal intervention versus usual care Computerised physician order entry (CPOE)/clinical decision support systems (CDSS) versus control/paper‐based systems CPOE/CDSS: improved versus standard CPOE/CDSS CPOE/CDSS: prioritised versus no prioritised alerts Barcoding versus no barcoding Organisational changes: reduced versus unreduced working hours Feedback on prescribing errors versus no feedback Feedback on prescribing errors versus education Education versus no education on prescribing Dispensing system versus control

Comparison level ↓____________________ Comparison # →			1	2	3	4	5	6	7	8
Study designs			RCT	RCT	RCT	RCT	RCT	RCT ‐ ITS	RCT ‐ ITS	RCT ‐ ITS
Target population			Adults. Older adults	Adults. Older adults	Adults	Adults	Old adults with high alert medications	Adults	RCT Adults, Adults with 2 two chronic conditions ITS Adults	Adults
Setting			Wards, ED	Wards, ED, Surgery units, pre‐admission clinic	General hospital, Surgery units, Anesthesia units	Hospital/ED	Emergency department (ED)	RCT Hospital ‐ ITS ED	General hospital	RCT General hospital ‐ ITS General hospital, ICU
Countries			China, Denmark, Norway, Oman, Spain(3), UK (2)	Australia (5), Belgium (2), Canada, Colombia, Denmark (2), Netherlands, Norway, United Kingdom, USA (5)	Canada (2), USA (1)	Netherlands, USA	USA	RCT USA ‐ ITS USA	RCT New Zealand, USA ‐ ITS USA	RCT Belgium, Ireland, UK ‐ ITS Canada, Japan Netherlands (2)
Study level			Comparisons description
Study ID	Study design	Unit of analysis	MR vs No MR	MR: pharmacist vs other professionals	MR: database assisted vs not‐assisted.	MR by pharmacist: team/highly trainedpharmacist vs standard pharmacist	MR: before vs after admission	MR: 1‐2 vs 4 charts open	MR: Multimodal intervention vs Usual care	CPOE/CDSS vs control/paper based
Aag 2014	RCT‐ individual	Patients		x
Adelman 2013	RCT‐ individual	Patients
Adelman 2019	RCT‐ individual	Order session						x
Agrawal 2009	ITS	Unintended discrepancy per admission
Al‐Hashar 2018	RCT‐ individual	Patients	x
Barker 1984	RCT‐ individual	Prescriptions
Becerra‐Camargo 2015	RCT‐ individual	Patients		x
Beckett 2012	RCT‐ individual	Patients		x
Bell 2015	RCT‐ individual	Patients		x
Bhakta 2019	ITS	Weekly prescription
Bolas 2004	RCT‐ individual	Patients	x
Boockvar 2017	RCT‐ cluster	Patients			x
Bowdle 2018	ITS	Patients receiving anaesthesia
Burkoski 2019	ITS	Monthly medication doses administered								x
Cadman 2017	RCT‐ individual	Patients/Unintended discrepancies	x
Chiu 2018	Quasi‐RCT	Patients	x
Colpaert 2006	RCT‐ individual	Prescription								x
De winter 2011	Quasi‐RCT individual	Patients		x
Ding 2012	RCT‐ cluster	Prescriptions
Farris 2014	RCT‐ individual	Patients		x
Fernandes 2011	RCT‐ individual	Patients			x
Furuya 2013	ITS	Patient‐days								x
George 2011	RCT‐ individual	Patients		x
Gordon 2017	RCT‐ cluster	Prescriptions
Graabaek 2019	RCT‐ individual	Patients		x
Green 2015	ITS	Prescriptions
Greengold 2003	RCT‐ individual	Administered doses
Gursanscky 2018	RCT‐ cluster	Prescriptions
Hale 2013	RCT‐ individual	Prescriptions		x
Heselmans 2015	RCT‐ individual	Patients/Prescriptions		x
Hickman 2018	RCT‐ individual	Prescriptions				x
Higgins 2010 (Heelon)	ITS	Monthly administered doses
Juanes 2018	RCT‐ individual	Patients	x
Kannampallil 2018	ITS	Order session						x
Khalil 2016	RCT‐ individual	Patients		x
Kwan 2007	RCT‐ individual	Patients		x
Landrigan 2004	RCT‐ cluster	Patients
Leung 2017	RCT‐ individual	Prescriptions
Lind 2017	RCT‐ individual	Patients		x
Marotti 2011	RCT‐ individual	Patients		x
McCoy 2012	RCT‐ individual	Patients/Prescriptions
Merry 2011	Quasi‐RCT	Patients
Narang 2013	ITS	Probably monthly administered doses (it is unclear we cannot discard that were patients)
Nielsen 2017	RCT‐ cluster	Patients	x
Ongering 2019	ITS	Prescription								x
OSullivan 2016	RCT‐ cluster	Patients								x
Pevnick 2018	RCT‐ individual	Patients		x		x
Piqueras 2015	RCT‐ individual	Prescriptions	x
Quach 2015	RCT‐ individual	Patients					x
Redwood 2013	RCT‐ individual	Doctors								x
Schmader 2004	RCT‐ individual	Patients		x
Schneider 2006	RCT‐ individual	Opportunities for error by nurse
Schnipper 2009	RCT‐ cluster	Patients
Schnipper 2018	CITS	Patients							x
Scullin 2007	RCT‐ individual	Patients		x
Seibert 2014	ITS	Monthly administered doses
SUREPILL 2015	RCT‐ cluster	Patients		x
Tamblyn 2018	RCT‐ cluster	Patients			x
Thompson 2018	ITS	Monthly administered doses
Tompson 2012	RCT‐ individual	Patients							x
Tong 2016	RCT‐ cluster	Patients		x
van Doormaal 2009	ITS	Prescriptions (MO) Patients								x
Vega 2016	RCT‐ individual	Patients	x
Wang 2017	RCT‐ individual	Prescriptions
Willoch 2012	RCT‐ individual	Patients	x

Comparison level ↓_______________Comparison # →			9	10	11	12	13	14	15	16
Study designs			RCT ‐ ITS	ITS	ITS	RCT	RCT	RCT	RCT	RCT
Target population			Adults	Adults	Adults	Adults	Adults	Adults	Adults	Adults
Setting			RCT Hospital ‐ ITS Hospital/ED	Hospital	Hospital, Anesthesia units	ICU	Hospital	Hospital	Hospital	Hospital, Surgery units
Countries			RCT USA (3) ‐ ITS USA (2)	USA	Canada, USA (5)	USA	Australia (2), UK	Australia (2)	Australia (2), USA (2)	China (2), USA
Study level			Comparisons description
Study ID	Study design	Unit of analysis	CPOE/CDSS: Improved vs standard CPOE/CDSS	CPOE/CDSS: prioritised vs no prioritised alerts	Barcoding vs no barcoding S1 No CPOE/CDSS S2: with CPOE/CDSS + CDSS	Organisational changes: reduced vs not reduced work hours	Feedback on prescribing errors vs no feedback	Feedback vs prescribing education	Education vs no education	Dispensing system vs control
Aag 2014	RCT‐ individual	Patients
Adelman 2013	RCT‐ individual	Patients	x
Adelman 2019	RCT‐ individual	Order session
Agrawal 2009	ITS	Unintended discrepancy per admission	x
Al‐Hashar 2018	RCT‐ individual	Patients
Barker 1984	RCT‐ individual	Prescriptions								x
Becerra‐Camargo 2015	RCT‐ individual	Patients
Beckett 2012	RCT‐ individual	Patients
Bell 2015	RCT‐ individual	Patients
Bhakta 2019	ITS	Weekly prescription		x
Bolas 2004	RCT‐ individual	Patients
Boockvar 2017	RCT‐ cluster	Patients
Bowdle 2018	ITS	Weekly anaesthestized patients			x
Burkoski 2019	ITS	Monthly medication doses administered			x
Cadman 2017	RCT‐ individual	Patients/Unintended discrepancies
Chiu 2018	Quasi‐RCT	Patients
Colpaert 2006	RCT‐ individual	Prescription
De winter 2011	Quasi‐RCT individual	Patients
Ding 2012	RCT‐ cluster	Prescriptions								x
Farris 2014	RCT‐ individual	Patients
Fernandes 2011	RCT‐ individual	Patients
Furuya 2013	ITS	Patient‐days
George 2011	RCT‐ individual	Patients
Gordon 2017	RCT‐ cluster	Prescriptions					x
Graabaek 2019	RCT‐ individual	Patients
Green 2015	ITS	Prescriptions	x
Greengold 2003	RCT‐ individual	Administered doses							x
Gursanscky 2018	RCT‐ cluster	Prescriptions					x	x	x
Hale 2013	RCT‐ individual	Prescriptions
Heselmans 2015	RCT‐ individual	Patients/Prescriptions
Hickman 2018	RCT‐ individual	Prescriptions
Higgins 2010 (Heelon)	ITS	Monthly administered doses			x
Juanes 2018	RCT‐ individual	Patients
Kannampallil 2018	ITS	Order session
Khalil 2016	RCT‐ individual	Patients
Kwan 2007	RCT‐ individual	Patients
Landrigan 2004	RCT‐ cluster	Patients				x
Leung 2017	RCT‐ individual	Prescriptions					x	x	x
Lind 2017	RCT‐ individual	Patients
Marotti 2011	RCT‐ individual	Patients
McCoy 2012	RCT‐ individual	Patients/Prescriptions	x
Merry 2011	Quasi‐RCT	Patients								x
Narang 2013	ITS	Probably monthly administered doses (it is unclear we cannot discard that were patients)			x
Nielsen 2017	RCT‐ cluster	Patients
Ongering 2019	ITS	Prescription
OSullivan 2016	RCT‐ cluster	Patients
Pevnick 2018	RCT‐ individual	Patients
Piqueras 2015	RCT‐ individual	Prescriptions
Quach 2015	RCT‐ individual	Patients
Redwood 2013	RCT‐ individual	Doctors
Schmader 2004	RCT‐ individual	Patients
Schneider 2006	RCT‐ individual	Opportunities for error by nurse							x
Schnipper 2009	RCT‐ cluster	Patients	x
Schnipper 2018	CITS	Patients
Scullin 2007	RCT‐ individual	Patients
Seibert 2014	ITS	Monthly administered doses			x
SUREPILL 2015	RCT‐ cluster	Patients
Tamblyn 2018	RCT‐ cluster	Patients
Thompson 2018	ITS	Monthly administered doses			x
Tompson 2012	RCT‐ individual	Patients
Tong 2016	RCT‐ cluster	Patients
van Doormaal 2009	ITS	Prescriptions (MO) Patients
Vega 2016	RCT‐ individual	Patients
Wang 2017	RCT‐ individual	Prescriptions								x
Willoch 2012	RCT‐ individual	Patients

PERMALINK

Reducing medication errors for adults in hospital settings

Agustín Ciapponi

Simon E Fernandez Nievas

Mariana Seijo

María Belén Rodríguez

Valeria Vietto

Herney A García-Perdomo

Sacha Virgilio

Ana V Fajreldines

Josep Tost

Christopher J Rose

Ezequiel Garcia-Elorrio

Abstract

Background

Objectives

Search methods

Selection criteria

Data collection and analysis

Main results

Authors' conclusions

Plain language summary

Summary of findings

Background

Description of the condition

1.

Description of the intervention

How the intervention might work

Why it is important to do this review

Objectives

Methods

Criteria for considering studies for this review

Types of studies

Types of participants

Setting

Healthcare professionals

Types of interventions

Delivery arrangements

Financial arrangements

Governance arrangements

Implementation strategies

Types of outcome measures

Primary outcomes

Secondary outcomes

Search methods for identification of studies

Electronic searches

Searching other resources

Trials Registries

Data collection and analysis

Selection of studies

Data extraction and management

Assessment of risk of bias in included studies

Measures of treatment effect

Reporting

Analytical approach

Primary analyses

Secondary analyses

Methods for reanalysis

Unit of analysis issues

Dealing with missing data

Assessment of heterogeneity

Assessment of reporting biases

Data synthesis

Subgroup analysis and investigation of heterogeneity

Sensitivity analysis

Summary of findings and assessment of the certainty of the evidence

Results

Description of studies

Results of the search

2.

Included studies

Population

Setting

Interventions and comparisons

Ongoing Studies

Excluded studies

Risk of bias in included studies

3.

4.

5.

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1.1 Medication errors	3		Odds Ratio (IV, Random, 95% CI)	0.55 [0.17, 1.74]
1.2 ADEs	3		Odds Ratio (IV, Random, 95% CI)	0.38 [0.18, 0.80]
1.3 Mortality during hospitalisation	1	212	Risk Ratio (M‐H, Random, 95% CI)	3.85 [0.44, 33.89]
1.4 Length of Stay (days)	3	527	Mean Difference (IV, Random, 95% CI)	‐0.30 [‐1.93, 1.33]
1.5 QoL (VAS 0‐10 ‐ EQ‐5D‐3L ‐ high score better)	1	131	Mean Difference (IV, Random, 95% CI)	‐1.51 [‐10.04, 7.02]
1.6 Discrepancy resolutions (per discrepancies at discharge)	1	564	Risk Ratio (M‐H, Random, 95% CI)	7.48 [5.62, 9.95]

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
2.1 Medication errors	8		Odds Ratio (IV, Random, 95% CI)	0.21 [0.09, 0.48]
2.2 ADEs	3		Odds Ratio (IV, Random, 95% CI)	1.34 [0.73, 2.44]
2.3 Mortality during hospitalisation	2	1000	Risk Ratio (M‐H, Random, 95% CI)	0.99 [0.57, 1.73]
2.4 Readmisson at 1 month	2	997	Risk Ratio (M‐H, Random, 95% CI)	0.93 [0.76, 1.14]
2.5 Length of stay (days)	6	3983	Mean Difference (IV, Random, 95% CI)	‐0.25 [‐1.05, 0.56]
2.5.1 General ward inpatients	5	3383	Mean Difference (IV, Random, 95% CI)	‐0.25 [‐1.09, 0.59]
2.5.2 Inpatients coming from ICU	1	600	Mean Difference (IV, Random, 95% CI)	‐0.30 [‐6.71, 6.11]
2.6 QoL (VAS 0‐10 ‐ EQ‐5D‐3L, high score is better)	1	724	Mean Difference (IV, Random, 95% CI)	0.00 [‐14.09, 14.09]
2.7 Discrepancy resolution	3		Odds Ratio (IV, Random, 95% CI)	4.80 [1.81, 12.76]

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
3.1 Potential ADEs (≥1 per patient)	2	3326	Odds Ratio (M‐H, Random, 95% CI)	0.26 [0.10, 0.64]
3.2 Lenght of stay (days)	1	311	Mean Difference (IV, Random, 95% CI)	1.00 [‐0.17, 2.17]
3.3 Discrepancy resolution (higher number is better)	2		Odds Ratio (IV, Random, 95% CI)	1.37 [0.97, 1.93]

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
4.1 Medication errors	2		Odds Ratio (IV, Random, 95% CI)	0.65 [0.25, 1.70]
4.2 Length of stay (days)	1	183	Mean Difference (IV, Random, 95% CI)	‐0.30 [‐2.12, 1.52]

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
7.1 Unintended discrepancies (≥1 per patient)	1		Risk Ratio (IV, Random, 95% CI)	0.92 [0.87, 0.97]
7.2 Potential ADEs (≥ 1 per patient)	1		Risk Ratio (IV, Random, 95% CI)	0.97 [0.86, 1.09]
7.3 Discrepancies resolutions (≥1 per patient, higher number is better)	1	487	Risk Ratio (M‐H, Random, 95% CI)	2.14 [1.81, 2.53]

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
8.1 Medication error	1		Odds Ratio (IV, Fixed, 95% CI)	0.74 [0.31, 1.79]
8.2 ADEs	2		Odds Ratio (IV, Random, 95% CI)	0.24 [0.04, 1.50]
8.3 Mortality	1	737	Risk Ratio (M‐H, Random, 95% CI)	1.04 [0.54, 2.01]
8.4 Length of stay (days)	1	737	Mean Difference (IV, Random, 95% CI)	‐1.00 [‐2.05, 0.05]

Outcome or subgroup title	No. of studies	Statistical method	Effect size
9.1 Medication errors	4	Odds Ratio (IV, Random, 95% CI)	Subtotals only
9.1.1 RCTs	2	Odds Ratio (IV, Random, 95% CI)	0.84 [0.73, 0.97]
9.1.2 ITSs	2	Odds Ratio (IV, Random, 95% CI)	0.77 [0.37, 1.62]
9.2 ADEs	2	Odds Ratio (IV, Random, 95% CI)	0.82 [0.71, 0.94]

Outcome or subgroup title	No. of studies	Statistical method	Effect size
15.1 Medication errors	4	Odds Ratio (IV, Random, 95% CI)	1.21 [0.93, 1.58]
15.1.1 Education on prescriptions (physicians)	2	Odds Ratio (IV, Random, 95% CI)	1.11 [0.88, 1.39]
15.1.2 Education on administration (nurses)	2	Odds Ratio (IV, Random, 95% CI)	1.64 [0.88, 3.08]

Outcome or subgroup title	No. of studies	Statistical method	Effect size
16.1 Medication errors	4	Odds Ratio (IV, Random, 95% CI)	Subtotals only
16.1.1 Surgical wards	2	Odds Ratio (IV, Random, 95% CI)	0.61 [0.47, 0.79]
16.1.2 Operating rooms	2	Odds Ratio (IV, Random, 95% CI)	0.92 [0.75, 1.13]
16.2 Medication errors (per prescriptions)	1	Mean Difference (IV, Random, 95% CI)	‐8.66 [‐12.77, ‐4.55]

*Study characteristics*
Methods	RCT. Non‐blinded, two‐armed, randomised controlled trial conducted by the Department of Cardiology at the University Hospital of North Norway. An expert team comprising a ward resident in cardiology and two clinical pharmacists retrospectively rated the clinical relevance of the identified medications discrepancies (MDs) using the classification system for clinical relevance described by Scullin and colleagues(Scullin 2007), where 1 = no relevance to patient care, 2 = relevant but does not lead to an improvement in patient care, 3 = relevant and results in an improvement in the standard of care, 4 = very relevant and prevents major organ failure or adverse reaction of similar importance and 5 = potentially lifesaving [2]. Unit of allocation: patients Unit of analysis: patients
Participants	People aged 18 years or more admitted to the ward during a five‐week period. IP adults (Department of Cardiology) (N = 206) Oncological patients (tertiary care centre)
Interventions	Intervention Human resources, medication reconciliation. Intervention: reconciliation: medication reconciliation (MR) performed by pharmacists Control: reconciliation: medication reconciliation performed by nurses
Outcomes	Mean errors (medications discrepancies) per patient Mean time spent during MR, minutes.
Notes	No funding information No trial number
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	"Patients were allocated to PG or NG in a 1:1 relationship, block randomized in block sizes six to ten, and stratified on gender only" "An online randomization procedure was applied to randomize eligible patients in two groups: PG (clinical pharmacist performing MR) and NG (nurse performing MR)" (randomization service from the Norwegian University of Science and Technology. https://www.ntnu.no/dmf/akf/randomisering. Accessed 25 March 2014)
Allocation concealment (selection bias)	Low risk	An online randomisation procedure
Blinding of participants and personnel (performance bias) All outcomes	High risk	Non‐blinded, two‐armed, randomised controlled trial. "The expert team was blinded to the patients' group allocation."
Blinding of outcome assessment (detection bias) All outcomes	Low risk	An expert team retrospectively rated the clinical relevance of the identified medications discrepancies (MDs) using the classification system for clinical relevance described by Scullin et al. "The expert team was blinded to the patients' group allocation."
Incomplete outcome data (attrition bias) All outcomes	Low risk	It is unlikely that missing data (PG 1% and NG 6%) had a great impact on outcomes.
Selective reporting (reporting bias)	Low risk	The study protocol is not available but it is clear that the publications include all the expected results, including those that were prespecified.
Conflict of interest	Low risk	None detected
Other bias	Low risk	The study appears to be free of other sources of bias.