Interventions to improve the appropriate use of polypharmacy for older people

Audrey Rankin; Cathal A Cadogan; Susan M Patterson; Ngaire Kerse; Chris R Cardwell; Marie C Bradley; Cristin Ryan; Carmel Hughes

doi:10.1002/14651858.CD008165.pub4

Interventions to improve the appropriate use of polypharmacy for older people

Authors' declarations of interest

Version published: 03 September 2018 Version history

https://doi.org/10.1002/14651858.CD008165.pub4

Collapse all Expand all

Abstract

available in

Background

Inappropriate polypharmacy is a particular concern in older people and is associated with negative health outcomes. Choosing the best interventions to improve appropriate polypharmacy is a priority, hence interest in appropriate polypharmacy, where many medicines may be used to achieve better clinical outcomes for patients, is growing. This is the second update of this Cochrane Review.

Objectives

To determine which interventions, alone or in combination, are effective in improving the appropriate use of polypharmacy and reducing medication‐related problems in older people.

Search methods

We searched CENTRAL, MEDLINE, Embase, CINAHL and two trials registers up until 7 February 2018, together with handsearching of reference lists to identify additional studies.

Selection criteria

We included randomised trials, non‐randomised trials, controlled before‐after studies, and interrupted time series. Eligible studies described interventions affecting prescribing aimed at improving appropriate polypharmacy in people aged 65 years and older, prescribed polypharmacy (four or more medicines), which used a validated tool to assess prescribing appropriateness. These tools can be classified as either implicit tools (judgement‐based/based on expert professional judgement) or explicit tools (criterion‐based, comprising lists of drugs to be avoided in older people).

Data collection and analysis

Two review authors independently reviewed abstracts of eligible studies, extracted data and assessed risk of bias of included studies. We pooled study‐specific estimates, and used a random‐effects model to yield summary estimates of effect and 95% confidence intervals (CIs). We assessed the overall certainty of evidence for each outcome using the GRADE approach.

Main results

We identified 32 studies, 20 from this update. Included studies consisted of 18 randomised trials, 10 cluster randomised trials (one of which was a stepped‐wedge design), two non‐randomised trials and two controlled before‐after studies. One intervention consisted of computerised decision support (CDS); and 31 were complex, multi‐faceted pharmaceutical‐care based approaches (i.e. the responsible provision of medicines to improve patient’s outcomes), one of which incorporated a CDS component as part of their multi‐faceted intervention. Interventions were provided in a variety of settings. Interventions were delivered by healthcare professionals such as general physicians, pharmacists and geriatricians, and all were conducted in high‐income countries. Assessments using the Cochrane 'Risk of bias' tool, found that there was a high and/or unclear risk of bias across a number of domains. Based on the GRADE approach, the overall certainty of evidence for each pooled outcome ranged from low to very low.

It is uncertain whether pharmaceutical care improves medication appropriateness (as measured by an implicit tool), mean difference (MD) ‐4.76, 95% CI ‐9.20 to ‐0.33; 5 studies, N = 517; very low‐certainty evidence). It is uncertain whether pharmaceutical care reduces the number of potentially inappropriate medications (PIMs), (standardised mean difference (SMD) ‐0.22, 95% CI ‐0.38 to ‐0.05; 7 studies; N = 1832; very low‐certainty evidence). It is uncertain whether pharmaceutical care reduces the proportion of patients with one or more PIMs, (risk ratio (RR) 0.79, 95% CI 0.61 to 1.02; 11 studies; N = 3079; very low‐certainty evidence). Pharmaceutical care may slightly reduce the number of potential prescribing omissions (PPOs) (SMD ‐0.81, 95% CI ‐0.98 to ‐0.64; 2 studies; N = 569; low‐certainty evidence), however it must be noted that this effect estimate is based on only two studies, which had serious limitations in terms of risk bias. Likewise, it is uncertain whether pharmaceutical care reduces the proportion of patients with one or more PPOs (RR 0.40, 95% CI 0.18 to 0.85; 5 studies; N = 1310; very low‐certainty evidence). Pharmaceutical care may make little or no difference in hospital admissions (data not pooled; 12 studies; N = 4052; low‐certainty evidence). Pharmaceutical care may make little or no difference in quality of life (data not pooled; 12 studies; N = 3211; low‐certainty evidence). Medication‐related problems were reported in eight studies (N = 10,087) using different terms (e.g. adverse drug reactions, drug‐drug interactions). No consistent intervention effect on medication‐related problems was noted across studies.

Authors' conclusions

It is unclear whether interventions to improve appropriate polypharmacy, such as reviews of patients’ prescriptions, resulted in clinically significant improvement; however, they may be slightly beneficial in terms of reducing potential prescribing omissions (PPOs); but this effect estimate is based on only two studies, which had serious limitations in terms of risk bias.

PICOs

Population

Intervention

Comparison

Outcome

The PICO model is widely used and taught in evidence-based health care as a strategy for formulating questions and search strategies and for characterizing clinical studies or meta-analyses. PICO stands for four different potential components of a clinical question: Patient, Population or Problem; Intervention; Comparison; Outcome.

See more on using PICO in the Cochrane Handbook.

Plain language summary

available in

A review of the ways that healthcare professionals can improve the use of suitable medicines for older people

What is the aim of this review?

The aim of this Cochrane Review was to find out which types of approaches can improve the use of suitable medicines in older people. Researchers collected and analysed all relevant studies to answer this question and included 32 trials in the review.

Key messages

Taking medicine to treat symptoms of chronic illness and to prevent worsening of disease is common in older people. However, taking too many medicines can cause harm.

What was studied in the review?

This review examines studies in which healthcare professionals have taken action to make sure that older people are receiving the most effective and safest medicines for their illness. Actions taken included providing a service, known as pharmaceutical care, which involves promoting the correct use of medicines by identifying, preventing and resolving medication‐related problems. Another strategy which we were interested in was using computerised decision support, which involves a programme on the doctor’s computer that aids the selection of appropriate treatment(s).

What are the main results of the review?

Review authors found 32 relevant trials from 12 countries that involved 28,672 older people. These studies compared interventions aiming to improve the appropriate use of medicines with usual care. It is uncertain whether the interventions improved the appropriateness of medicines (based on scores assigned by expert professional judgement), reduced the number of potentially inappropriate medicines (medicines in which the harms outweigh the benefits), reduced the proportion of patients with one or more potentially inappropriate medications, or reduced the proportion of patients with one or more potential prescribing omissions (cases where a useful medicine has not prescribed) because the certainty of the evidence is very low. The interventions may lead to little or no difference in hospital admissions or quality of life, however, the interventions may slightly decrease the number of potential prescribing omissions.

How up‐to‐date is this review?

Review authors searched for studies that had been published up to February 2018.

Authors' conclusions

available in

Implications for practice

The evidence obtained when results of these studies were combined is rather weak, and it is uncertain whether interventions provided to improve appropriate polypharmacy, such as pharmaceutical care, resulted in clinically significant improvement. Uncertainty surrounds the effects of such interventions on hospital admissions and medication‐related problems, and it could be argued that these are the critical outcomes for patients. However, the pooled effect estimates suggest some improvements in outcomes such as the number of potential prescribing omissions (PPOs) and potentially inappropriate prescriptions but due to limitations with the quality of evidence, uncertainty exists. There was a lack of certainty regarding the effects of pharmaceutical care interventions included in this review on inappropriate prescribing (medication appropriateness (as measured by an implicit tool), the number of potentially inappropriate medications (PIMs), the proportion of patients with one or more PIMs and the proportion of patients with one or more PPOs. Pharmaceutical care may slightly reduce the number of PPOs (however it must be noted that this effect estimate is based on only two studies, which had serious limitations in terms of high risk of bias), especially when a multi‐disciplinary element is included in the provision of care (Bucci 2003; Crotty 2004a;Crotty 2004b;Gallagher 2011; Garcia‐Gollarte 2014; Hanlon 1996; Schmader 2004; Spinewine 2007; Taylor 2003). In addition, although only two studies that involved CDS were included in this review, it would appear that computerised decision support (CDS) is a helpful component of interventions for improving appropriate polypharmacy (Clyne 2015; Tamblyn 2003).

Given the difficulties involved in applying the results of clinical studies to older people, physicians should carefully consider their sources of evidence and recommendations to find the right balance between avoiding the 'risk/treatment paradox' (high‐risk older patients denied safe medications capable of materially improving survival or quality of life) and avoiding inappropriate use of medications for which risks are likely to outweigh benefits (Scott 2010). It must also be noted that the intervention studies included in this review focused on reducing inappropriate prescribing of prescription medications and over‐the‐counter (OTC) medication use was often not assessed, nor was it specifically examined as part of this review. OTC medication use is common among older patients receiving prescription medications with the potential for drug interactions to occur (Agbabiaka 2017). This should not be overlooked by healthcare professionals when reviewing older patients’ medication use.

Based on the findings of our updated review, we are still uncertain about which elements of the intervention processes constitute success in improving appropriate polypharmacy, and a number of questions remain unanswered. For example, is it sufficient to provide the intervention during a single episode of care, or should patients be exposed to the intervention on a daily/weekly or monthly basis? What is the optimal duration of an intervention, and should interventions ideally be multi‐faceted or uni‐faceted? It is clear that control of processes to support fidelity and control of the chosen interventions is critical. Staff training is important to ensure consistency; the receptiveness of prescribers, patients and staff in various settings will have an impact on the uptake and effectiveness of interventions in older people.

Implications for research

Overall, the quality of the studies in this review was poor, and further research should attend to rigour in study design. More research is needed to test whether existing tools for comprehensive medication review (e.g. the hyperpharmacotherapy assessment tool (HAT tool) (Bushardt 2008) and other similar interventions) can improve appropriate polypharmacy. Since the last update of this review, a Scottish working group has published a guidance document on polypharmacy, which included a seven‐step process for standardised and structured medicines reviews that are holistic, patient‐centred and consider non‐pharmacological treatments (Scottish Government Model of Care 2018), as well as a review of the quality of development of available guidelines to promote appropriate polypharmacy (Stewart 2017). Further population‐based research is required to evaluate the implementation and effect of these resources on prescribing for older people.

Uncertainty about which elements of the intervention are critical to successful outcomes needs to be addressed. On the basis of the studies included in this review, this poses challenges, as details of intervention development and delivery were lacking. Methods of specifying and reporting complex interventions, as well as their implementation strategies, are necessary to strengthen the evidence base required for interventions to be more effective, implementable and replicable across different settings (Michie 2011; Proctor 2013). Future intervention studies targeting appropriate polypharmacy could benefit from guidance provided by the framework of the Medical Research Council (MRC) on the design of complex interventions (MRC 2008). This framework recognises the importance of the initial stage of intervention development, in which evidence and theory are used to inform the selection of relevant components before the intervention is piloted, and the feasibility of delivering it in practice is assessed. These initial stages precede formal evaluations seeking to establish the effectiveness of the intervention. Despite the potential availability of the MRC guidelines before the start of the new studies highlighted in this update, only one included study (Clyne 2015) and two ongoing studies (Anrys 2016; Sinnott 2017) referred to using the MRC guidelines when developing and evaluating their interventions.

Adequate documentation of intervention development and intervention content as well as the training and background of providers that may be critical to intervention effectiveness is essential for facilitating replication of successful interventions in practice. However, no studies included in this review referred to using available intervention tools reporting, such as TIDieR (Template for Intervention Description and Replication) checklist (Hoffmann 2014).

The framework of the MRC 2008 also outlines the potential application of qualitative methodologies, such as semi‐structured interviews, to involve users and to gain insights into the processes of change that underlie the intervention. For example, establishing the reasons why not all interventions are accepted may be enlightening and may support research into the development of more successful interventions. There appears to be a ceiling effect (approximately 75%), whereby inappropriate prescribing continues despite the evidence base of interventions (Furniss 2000; Zermansky 2006). Qualitative interviews of prescribers may uncover reasons as to why they did not accept interventions (e.g. timing or appropriateness of provision of the intervention, the expertise of providers). Additionally, poor prescribing practice must be explored and understood with the goal of learning how to improve it and how to enhance patient safety by reducing the need for intervention. The importance of these investigations extends beyond the research context alone. Given the high financial expenditure that has been attributed to potentially inappropriate prescribing (PIP) in older people (Bradley 2012; Cahir 2010), it is likely that policy makers will continue to be interested in the costs of these types of interventions.

In the previous version of this review (Patterson 2014), we recommended that future studies should utilise clearer definitions of appropriate polypharmacy because the term 'polypharmacy' can be both negative and positive, and this duality of meaning makes objective research difficult (Bushardt 2008). Reports by the King’s Fund in the UK (King's Fund 2013) and Scottish Guidance on polypharmacy (Scottish Government Model of Care 2018), discussed the need to reconsider current definitions of polypharmacy on account of the increasing numbers of medications being prescribed to patients and recommended that polypharmacy should be defined as appropriate (i.e. medicine use has been optimised and medicines prescribed according to best evidence) or problematic (i.e. medicines have been prescribed inappropriately or intended benefits have not been realised). Although the potential benefit of having a simple means of identifying patients at particular risk for inappropriate prescribing and adverse effects was acknowledged, the authors of the King's Fund report noted that existing thresholds used to define polypharmacy, such as four or five or more medicines, may be too low. A pragmatic approach was proposed to identify patients warranting medication review, which focused on particular patient groups (e.g. patients receiving ≥ 10 regular medicines, patients receiving four to nine medicines with other risk factors).

For the purpose of this update, the definition of polypharmacy was not changed from that used in the original review. Although a threshold of four or more medicines may now be considered to be low in the context of older people with multimorbidity, it is important to recognise that the number of medicines used to define polypharmacy is arbitrary. Furthermore, conceptualising polypharmacy solely on the basis of the number of medicines prescribed is often unhelpful as this approach fails to recognise that the appropriate number of medicines varies according to individual patients’ clinical needs and, moreover, may overlook the omission of potentially beneficial medications, which can equally have a negative impact on clinical outcomes (Cadogan 2016). Hence, for the purpose of the current update, our focus was on interventions targeting the appropriateness of the medications prescribed for older people. However, future updates of this review may reconsider the criteria used to define polypharmacy were validated tools to assess potentially inappropriate prescribing in older people, such as Beers criteria, are not specifically designed to measure appropriate polypharmacy, it is important that future interventions should include assessments of potentially inappropriate omissions/under‐prescribing with the goal of improving appropriate polypharmacy.

The judgement as to whether many (appropriate polypharmacy) or too many (inappropriate polypharmacy) medications are used is difficult. The complexity of the clinical situation, patient attributes and wishes and the individuality of prescribing for older complex patients will remain a challenge in this regard. Development of a new, universal, easily applied, valid and reliable outcome measure of appropriate polypharmacy in primary care is currently underway (Burt 2016). Ideally, this measure should be globally applicable across various healthcare and cultural settings.

It is important that sufficient detail about the context in which complex interventions are conducted, such as those included in this review, is reported and understood, so the transferability of complex interventions can be assessed (Wells 2012). For example, heterogeneity among older people in relation to differing levels of frailty (Spinewine 2007a) means that translational research and retesting of successful interventions may be necessary in dissemination to new populations, as a population of quite healthy 70‐year‐old people may respond differently to an intervention compared with a group of very frail 92‐year‐old individuals.

It is worth noting that only one of the included studies followed participants for longer than 12 months (Frankenthal 2014). The lack of evidence of effectiveness of pharmaceutical care interventions may be due in part to inadequate length of follow‐up. Future studies should be longer in duration to address this issue and to evaluate the longer‐term sustainability of pharmaceutical care interventions in improving the appropriate use of polypharmacy for older people.

Perhaps most critically, the selection of clinical and humanistic outcomes appropriate for older people (e.g. hospital admissions, adverse drug events (ADEs)) will be important to consider in future studies. Strategies for improving the evidence base for older patient care have been reviewed by Scott 2010. Indeed, a key challenge for interventions aimed at improving appropriate polypharmacy for older people is the selection and reporting of consistent outcomes (i.e. patient‐related or medication‐related outcomes). The Core Outcome Measures for Effectiveness Trials (COMET) initiative was launched to develop and apply core outcome sets (COS), which have been proposed as one method of addressing this problem (Williamson 2017). A COS is an agreed and standardised set of outcomes or outcome domains which should be measured and reported, as a minimum, in all trials in a specific clinical area. Alongside the Core Outcome Set‐STAndards for Reporting (COS‐STAR) guidelines (Kirkham 2016), the development of COSs in a specific health area should facilitate more robust synthesis of evidence in the future. A COS for use in interventions to improve the appropriate use of polypharmacy for older people in primary care is now available (Rankin 2018). The adoption of this COS will streamline the outcomes routinely measured in trials investigating appropriate polypharmacy in older people in primary care. This will ultimately facilitate the comparison and synthesis of outcome data across studies, thereby helping to determine which interventions work and inform both clinical decision making and health policy.

Summary of findings

Open in table viewer

Summary of findings for the main comparison. Pharmaceutical care compared with usual care for older people receiving polypharmacy

Patient or population: older people receiving polypharmacy Settings: community, nursing home, hospital Intervention: pharmaceutical care Comparison: usual care
Outcomes	Effect estimate		Relative Risk effect (95% CI)	No. of participants (studies)	Certainty of evidence (GRADE)	Comments
Outcomes	Usual care	Pharmaceutical care	Relative Risk effect (95% CI)	No. of participants (studies)	Certainty of evidence (GRADE)	Comments
Medication appropriateness (as measured by an implicit tool) From baseline to follow‐up Follow‐up: 0 to 6 months	Medication appropriateness (as measured by an implicit tool) across control groups ranged from ‐0.49 to 2.86	Medication appropriateness (as measured by an implicit tool) in the intervention groups was 4.76lower (0.33 to 9.20 lower)		517 (5 studies)	⊕⊝⊝⊝ very low ^a,b,c,d	MAI used as implicit tool in the pooled studies A sensitivity analysis showed that medication appropriateness (as measured by an implicit tool) in the intervention group was 0.50 lower (2.27 lower to 1.28 higher)^e Heterogeneity: I² = 57%, P = 0.10
Potentially inappropriate medications
The number of potentially inappropriate medications (PIMs) Follow‐up: 0 to 12 months	The number of PIMs (Standardised mean^§) across control groups ranged from 0.04 to 1.29	The number of PIMs (Standardised mean^§) in the intervention groups was 0.22lower (0.05 to 0.38 lower)		1832 (7 studies)	⊕⊝⊝⊝ very low^a,b,c	STOPP and Beers criteria used as explicit tools in the pooled studies
The proportion of patients with one or more potentially inappropriate medications (PIMs) Follow‐up: 0 to 12 months	421 per 1000	333 per 1000 (257 to 430)	RR 0.79 (0.61 to 1.02)	3079 (11 studies)	⊕⊝⊝⊝ very low^a,b,c	STOPP and Beers criteria used as explicit tools in the pooled studies A sensitivity analysis showed that the proportion of patients with one or more potentially inappropriate medications in the intervention group was lower (333 per 1000)^f Heterogeneity: I² = 75%, P = 0.24
Potential prescribing omissions
The number of potential prescribing omissions (PPOs) Follow‐up: 0 to 12 months	The number of PPOs (Standardised mean^§) across control groups ranged from 0.63 to 0.85	The number of PPOs (Standardised mean^§) in the intervention groups was 0.81 lower (0.64 to 0.98 lower)		569 (2 studies)	⊕⊕⊝⊝ low^a	START and ACOVE used as explicit tools in the pooled studies
The proportion of patients with one or more potential prescribing omissions (PPOs) Follow‐up: 0 to 24 months	387 per 1000	155 per 1000 (70 to 329)	RR 0.40 (0.18 to 0.85)	1310 (5 studies)	⊕⊝⊝⊝ very low^a,c	START and ACOVE used as explicit tools in the pooled studies
Hospital admissions Follow‐up: 0 to 12 months	Pharmaceutical care may make little or no difference in hospital admissions			4052 (12 studies)	⊕⊕⊝⊝ low^a
Quality of Life Follow‐up: 0 to 12 months	Pharmaceutical care may make little or no difference in quality of life			3211 (12 studies)	⊕⊕⊝⊝ low^a
GRADE Working Group grades of evidence High: This research provides a very good indication of the likely effect. The likelihood that the effect will be substantially different^‡ is low. Moderate: This research provides a good indication of the likely effect. The likelihood that the effect will be substantially different^‡ is moderate. Low: This research provides some indication of the likely effect. However, the likelihood that it will be substantially different^‡ is high. Very low: This research does not provide a reliable indication of the likely effect. The likelihood that the effect will be substantially different^‡ is very high. ^‡Substantially different = a large enough difference that it might affect a decision
ACOVE: Assessing Care of the Vulnerable Elderly, CI: confidence interval, MAI: Medication Appropriateness Index, PIMs: Potentially Inappropriate Medications, PPOs: Potential prescribing omissions, RR: risk ratio, STOPP: Screening Tool of Older People’s potentially inappropriate Prescriptions, START: Screening Tool to Alert to Right Treatment ^§ Standardised mean was used in cases where a range of tools were used to generate the pooled effect estimate. ^a We downgraded the evidence due to risk of bias. ^b We downgraded the evidence due to indirectness of the evidence. ^c We downgraded the evidence due to inconsistency in the results that could not be fully explained. ^d We downgraded the evidence due to imprecision. CIs were wide and/or crossed the line of no effect. ^e Two studies were excluded from the analysis because of a unit of analysis error (Crotty 2004a) and an outlying effect estimate with a high risk of bias (Spinewine 2007). ^f Two studies were excluded from the analysis because of a large effect size and high risk of bias (Spinewine 2007) and a small effect size (Gallagher 2011).

Background

available in

Prescribing for older people is complex because of factors such as age‐related changes in body composition and multiple pathologies. Finding the balance between aggressively treating diseases and avoiding medication‐related harm is a critical objective for healthcare professionals, yet has proven challenging to achieve in clinical practice (Steinman 2007). This review updates the previous Cochrane Review ofInterventions to improve the appropriate use of polypharmacy for older people (Patterson 2014), which concluded that despite the potential to reduce inappropriate prescribing, it was unclear whether interventions to improve appropriate polypharmacy in older people resulted in clinically significant improvements such as reduced hospital admissions or improved quality of life.

Polypharmacy refers to the use of multiple medicines. The term itself has been the subject of much discussion but no standard definition is used consistently (Cadogan 2016a; King's Fund 2013; Stewart 1990). A simple definition has been used ("the administration of more medicines than are clinically indicated, representing unnecessary drug use" Montamat 2004). For the purpose of this update of the review, we defined it as 'the concomitant ingestion of four or more medicines', however, in recognition of the fact that the number of medicines used to define polypharmacy is arbitrary, the focus of the interventions of interest to this review is the appropriateness of the medications prescribed for older people.

Polypharmacy is common in older people, conventionally defined as those aged 65 years and older, as this age group is often subject to multimorbidity (defined as two or more chronic conditions) (Barnett 2012), such as cardiovascular disease and diabetes that require multiple medicines for treatment and prophylaxis. In the USA, the prevalence of polypharmacy in older people has increased over time, and the most recent available data indicate that approximately 39% of older people in the USA take five or more medicines (Kantor 2015). Data from The Irish Longitudinal Study on Ageing have reported polypharmacy in 27% of the older population using the same definition (McGarrigle 2017). Although prevalence estimates in older people vary across countries, polypharmacy in older people is recognised as a widespread global issue (Stewart 2017). Consequently, older people use a disproportionate quantity of health service resources. For example, in terms of medicines, in 2016, patients aged 60 and older accounted for 23% of the population in England and were dispensed 61.0% of all prescription items (Information Centre 2017).

Multiple factors contribute to the occurrence of polypharmacy in older people including an increase in life expectancy and the resultant growth in the prevalence of multimorbidity, the wider availability of effective drug treatments, and prescribing guidelines that recommend the use of more than one medicine in the prevention and management of various health conditions (Cadogan 2016). It is widely recognised that prescribing guidelines typically focus on single diseases and when applied to complex multimorbid patients often fail to provide information on how to prioritise treatment recommendations and can act as a driving force for polypharmacy (Hughes 2012). In light of this, the National Institute for Health and Care Excellence (NICE) has recently developed guidelines for the clinical treatment of patients with multiple morbidities, highlighting the importance of appropriate prescribing in this population (NICE 2016).

Inappropriate prescribing in the context of older people can be defined as the prescribing of "medications or medication classes that should generally be avoided in persons 65 years or older because they are either ineffective or they pose unnecessarily high risk for older persons and a safer alternative is available" (Beers 1991). The term ‘potentially inappropriate prescribing (PIP)’ encompasses potentially inappropriate medicines (PIMs) and potential prescribing omissions (PPOs). A PIM is a medicine that could potentially lead to a significant risk of adverse drug events (ADEs) and arises from prescribing practices such as continuing therapy for longer than necessary or recommended in prescribing guidelines. A PPO involves the omission of a medication that is clinically indicated for disease treatment or prevention (O' Connor 2012).

Although polypharmacy is often clinically indicated and beneficial in specific conditions (e.g. hypertension, diabetes mellitus) and patient populations (e.g. patients with multimorbidity), it also poses risks of medication‐related harm and safety risks to patients. A medication‐related problem is described as “an event or circumstance involving a patient’s drug treatment that actually, or potentially, interferes with the achievement of an optimal outcome” and includes adverse drug reactions and drug interactions (Simonson 2005). Polypharmacy in older people has been associated with PIP and negative health outcomes including an increased risk of hospital admissions, adverse drug events and mortality (Cahir 2010). The chance of medication‐related problems (such as adverse drug reactions and drug‐drug interactions) occurring increases in older age, in part, because the ageing process reduces the efficiency of the body’s organs in eliminating drugs (Mangoni 2003). A large study of community‐dispensed prescribing in Scotland (between 1995 and 2010) showed that the proportion of older adults prescribed more than five medicines and with potentially serious drug‐drug interactions had more than doubled to 13% in 2010 (Guthrie 2015). It is known that the number of medicines prescribed is predictive of the number of drug interactions likely to occur (Gallagher 2001). Poor understanding of causes of certain disorders makes prescribing drug combinations more difficult and treating poorly understood diseases may increase the risk for inappropriate prescribing (Werder 2003).

Despite the recognised potential for medication safety risks in older people, recent cohort studies have challenged previous assumptions that polypharmacy is hazardous and associated with poor clinical outcomes (Appleton 2014; Guthrie 2015). For example, an analysis of Scottish primary care data linked to hospital discharge data highlighted the limitations of crude measures of polypharmacy (i.e. the number of medicines prescribed) as quality indicators or predictors of hospital admissions when patients’ clinical context is not taken into consideration (Appleton 2014). The findings showed that patients prescribed an increased number of cardiovascular medicines were more likely to experience unplanned hospital admissions. However, when the analysis was adjusted to account for clinical factors such as non‐cardiovascular morbidity and drug burden, no evidence of an increase in non‐cardiovascular admissions with increasing numbers of cardiovascular medicines was found.

Consequently, greater use of the term ‘appropriate polypharmacy’, has been advocated which refers to ‘prescribing for an individual with complex or multiple conditions where medicine use has been optimised and prescribing is in accordance with best evidence’ (Cadogan 2016; King's Fund 2013). In assessing older patients’ prescriptions, it is important to consider whether each drug has been prescribed appropriately or inappropriately, both individually and in the context of the whole prescription (Aronson 2006). Improving appropriate polypharmacy involves encouraging use of the correct drugs under appropriate conditions to treat the right diseases. In certain circumstances, this may include the removal of unnecessary drugs or those with no valid clinical indication and the addition of useful ones. Thus, interventions that seek solely to reduce the number of prescribed medicines fail to consider polypharmacy in its entirety. PPOs are also highly prevalent in older populations and have been shown to be associated with polypharmacy, whereby the probability of under‐prescription increases with the number of medicines prescribed (Galvin 2014).

These findings may be explained by the unwillingness of general practitioners (GPs) to prescribe additional drugs for patients with polypharmacy (for reasons such as complexity of drug regimens, fear of ADEs and drug‐drug interactions and poor adherence) (Kuijpers 2007). This so‐called treatment/risk paradox or risk/treatment mismatch is seen when patients with the highest risk of complications are determined to have the lowest probability of receiving the recommended medications (Ko 2004; Lee 2005).

Differentiating between 'many' medicines (appropriate polypharmacy) and 'too many' medicines (inappropriate polypharmacy) is a prescriber's dilemma, and choosing the best interventions aimed at ensuring appropriate polypharmacy remains a challenge for healthcare practitioners and organisations.

Description of the condition

The causes of inappropriate polypharmacy are multifactorial (Stewart 2017), and for the purpose of this review we have focused on interventions that have targeted PIM, PPO, or both, using validated instruments or screening tools such as a validated list of medicines considered inappropriate for older people (AGS 2012; Beers 1991; Fick 2003; King's Fund 2013), a list of clinically significant criteria for potentially inappropriate prescribing in older people (Gallagher 2008) or the Medication Appropriateness Index (MAI) (Hanlon 1992). These screening tools can be classified as either implicit (judgement‐based) or explicit (criterion‐based) tools (Kaufmann 2014; O' Connor 2012). Implicit tools, such as MAI (Appendix 1) and the Assessment of Underutilization of Medication (AOU) tool (Jeffery 1999), are judgement‐based indicators of prescribing quality that are applied by clinicians to a patient’s prescription. Explicit tools such as Beers’ criteria (Appendix 1) and Screening Tool of Older Person's Prescriptions (STOPP)/Screening Tool to Alert doctors to the Right Treatment (START) criteria (Gallagher 2008), are usually developed from literature reviews, expert opinion and consensus exercises. The criteria typically comprise lists of drugs to be avoided or added in older people.

Description of the intervention

Improvement in appropriate polypharmacy can be achieved through a wide range of interventions (e.g. educational programmes for prescribers or consumers; medication review clinics and specific prescribing audits; prescribing incentive schemes and regulatory interventions). Interventions that reduce the risk of medication‐related problems are important to consider (Fick 2008). These may be provided by healthcare professionals, educators, policy makers and healthcare service planners. Previously, interventions targeting polypharmacy in older people have often focused on reducing the number of medicines prescribed (Rollason 2003), based on the assumption that polypharmacy is harmful. However, by focusing solely on the number of prescribed medicines, these interventions have failed to consider inappropriate prescribing in its entirety. As noted above, inappropriate prescribing is not restricted to over‐prescribing, but also encompasses mis‐prescribing (i.e. incorrect prescribing of a necessary drug) and under‐prescribing (i.e. prescribing omissions).

Methods recommended in previous intervention studies include use of computer data entry and feedback procedures, which have been shown to decrease polypharmacy and drug‐drug interactions (Werder 2003); visual identification of medicines; continuous medication review and thorough patient education to optimise polypharmacy (Fulton 2005).

This review seeks to identify evidence regarding which types of interventions can improve appropriate polypharmacy in older people.

How the intervention might work

Interventions to improve appropriate polypharmacy are likely to achieve the following outcomes.

Improvement in medication appropriateness (as measured by an implicit tool).
Reduction of inappropriately prescribed medication (as measured by an explicit tool).
Reduction of prescribing omissions (as measured by an explicit tool) by promoting prescribing of evidence‐based therapy where clinically indicated.

Computerised decision support (CDS) aimed at prescribers, whereby electronic alerts are produced to guide the prescriber to the right treatment, has been successful in reducing inappropriate prescribing for older people.

Pharmaceutical care is the responsible provision of drug therapy for the purpose of achieving definitive outcomes that improve a patient’s quality of life (Hepler 1990). Pharmaceutical care reflects a systematic approach that ensures patients receive the correct medicines, at an appropriate dose, for appropriate indications. It involves pharmacists moderating drug management in collaboration with physician, patient and carer (Hepler 1990). Pharmacist‐led interventions such as medication review, co‐ordinated transition from hospital to long‐term care facility and pharmacist consultations with patients and physicians have been shown to effectively reduce inappropriate prescribing and ADEs (Hanlon 1996; Kaur 2009). Multi‐disciplinary case conferences involving GPs, geriatricians, pharmacists and residential care staff, wherein individual patient cases are discussed, have reduced the use of inappropriate medications in residential care (Crotty 2004a).

Why it is important to do this review

A systematic review may help to identify how we can improve appropriate polypharmacy in older people. Inappropriate prescribing for older people is both highly prevalent and commonly associated with polypharmacy (Bradley 2012; Cahir 2010). It is important that the current available evidence be identified and appraised, so that interventions that are effective in managing disease with appropriate polypharmacy may be identified and put into practice. This is an update of the Cochrane Review (Patterson 2014).

Objectives

available in

To determine which interventions, alone or in combination, are effective in improving the appropriate use of polypharmacy and reducing medication‐related problems in older people.

Methods

available in

Criteria for considering studies for this review

Types of studies

We included randomised trials and cluster‐randomised trials, non‐randomised trials, controlled before‐after studies (CBAs) and interrupted time series (ITS) studies meeting the Effective Practice and Organisation of Care (EPOC) specification (EPOC 2017).

We classified trials eligible for inclusion according to the degree of certainty that random allocation was used to form comparison groups in the trial. If study author(s) stated explicitly that groups compared in the trial were established by random allocation, we classified the trial as a randomised trial. If study author(s) did not state explicitly that the trial was randomised, but randomisation could not be ruled out, we classified the report as a non‐randomised trial.

Types of participants

The review included studies of people aged 65 years and older, who had more than one long‐term medical condition and were receiving polypharmacy (classified as four or more medicines. This included a prescribed medication (one that is scheduled or part of a repeat prescription, and does not include over‐the‐counter and herbal products) and included studies targeting patient groups in which polypharmacy was common practice, such as patients with Parkinson’s disease or diabetes. We considered trials for inclusion if they included a majority (80% or more) of participants aged 65 years and older, or if the mean age of study participants was over 65 years. If studies included both older and younger people, we included them if we were able to extract relevant data. We contacted study authors to check the availability of relevant data.

We excluded studies in which the intervention focused on people with a single long‐term medical condition or who were receiving short‐term polypharmacy, for example, those who were terminally ill or were receiving cancer chemotherapy.

Types of interventions

We examined all types of interventions aimed at improving appropriate polypharmacy in any setting (such as pharmaceutical care) compared with usual care (as defined by the study). We included all uni‐faceted interventions, for example, those targeted solely at drug prescriptions, and multi‐faceted interventions, for example, specialist clinics involving comprehensive geriatric assessment. We included studies of interventions for which the target was polypharmacy across all ages, provided results for those aged 65 years and older were available separately. We examined all types of interventions as set out by the most recent EPOC taxonomy of health systems interventions (EPOC 2015; EPOC 2016) that directly or indirectly affected prescribing and were aimed at improving appropriate polypharmacy. These included the following.

Implementation strategies (previously categorised as professional interventions), defined as interventions designed to bring about changes in healthcare organisations, the behaviour of healthcare professionals or the use of health services by healthcare recipients, such as educational programmes aimed at prescribers.
Delivery arrangements (previously categorised as organisational interventions) defined as changes in how, when and where healthcare is organised and delivered, and who delivers healthcare, such as skill‐mix changes, pharmacist‐led medication review services or specialist clinics, information and communication technology (ICT) interventions such as clinical decision support systems or use of risk screening tools.
Financial arrangements (previously categorised as financial interventions) defined as changes in how funds are collected, insurance schemes, how services are purchased, and the use of targeted financial incentives or disincentives, such as incentive schemes for changes in prescribing practice.
Governance arrangements (previously categorised as regulatory interventions) defined as rules or processes that affect the way in which powers are exercised, particularly with regard to authority, accountability, openness, participation, and coherence, such as changes in government policy or legislation affecting prescribing.

Types of outcome measures

Validated measures of inappropriate prescribing (such as Beers criteria (Fick 2003), MAI (Hanlon 1992), STOPP/START criteria (Gallagher 2008) or Assessing Care of Vulnerable Elderly (ACOVE) (Wenger 2001)) were the main outcome measures considered in the review. We excluded studies in which medication appropriateness was determined solely by expert opinion (i.e. no measures/tools were used).

Primary outcomes

The primary outcomes of interest for this review were the following.

Medication appropriateness (as measured by an implicit tool), e.g. MAI (Hanlon 1992) or a defined subset of criteria from a validated instrument.
Potentially inappropriate medications (as defined by a validated explicit tool (e.g. STOPP criteria (Gallagher 2008)), which could consist of the number of potentially inappropriate medications and/or the proportion of patients with one or more potentially inappropriate medications.
Potential prescribing omissions (as defined by a validated explicit tool (e.g. START criteria (Gallagher 2008)), which could consist of the number of potential prescribing omissions and/or the proportion of patients with one or more potential prescribing omissions.
Hospital admissions (including all‐cause hospital admissions and unplanned hospital readmissions).

Secondary outcomes

Secondary outcomes included the following.

Medication‐related problems, for example, adverse drug reactions and drug‐drug interactions.
Adherence to medication.
Quality of life (as assessed by a validated method).

Search methods for identification of studies

The Information Specialist for the EPOC group updated the searches and searched the Cochrane Database of Systematic Reviews and the Database of Abstracts of Reviews of Effects (DARE) for related systematic reviews, as well as the databases listed below for primary studies. Searches were conducted in May 2016, with an updated search conducted in February 2018; exact search dates for each database are included with the search strategies, which are provided in Appendix 2 and Appendix 3.

Databases

Cochrane Central Register of Controlled Trials (CENTRAL; 2018, Issue 1) in the Cochrane Library
Health Technology Assessment Database (HTA; 2016, Issue 4) in the Cochrane Library
NHS Economic Evaluation Database (NHSEED; 2015, Issue 2) in the Cochrane Library
MEDLINE Ovid (including Epub Ahead of Print, In‐Process & Other Non‐Indexed Citations) (1946 to 31 January 2018)
Embase Ovid (1974 to 6 February 2018)
CINAHL EBSCO (Cumulative Index to Nursing and Allied Health Literature; 1980 to 7 February 2018)

Trial registries

Two trials registers were searched on 7 February 2018.

International Clinical Trials Registry Platform (ICTRP), Word Health Organization (WHO) www.who.int/ictrp/en
ClinicalTrials.gov, US National Institutes of Health (NIH) clinicaltrials.gov

Search strategies comprised keywords and, when available, controlled vocabulary such as MeSH (medical subject headings). All databases were searched for articles indexed between Nov 2013 and February 2018. Two methodological search filters were used to limit retrieval to appropriate study designs. No language restrictions were applied.

Searching other resources

We screened selected issues of the Journal of the American Geriatrics Society (e.g. handsearching).
We reviewed reference lists of relevant systematic reviews (Appendix 4).
We contacted authors of relevant studies and reviews to ask that they clarify reported published information or to seek unpublished results/data.
We contacted researchers with expertise relevant to the review topic or to EPOC interventions.
We conducted cited reference searches on studies selected for inclusion in this review, related reviews and other relevant citations as listed on the Institute for Scientific Information (ISI) Web of Science/Web of Knowledge.

Data collection and analysis

Selection of studies

For this update, three reviewers (AR, CAC and JC) independently screened titles and abstracts identified in searches to assess which studies met the inclusion criteria of the review. At this stage, we excluded papers that did not meet the inclusion criteria. If uncertainty or disagreement arose at this stage, we obtained full‐text articles and assessed them independently to determine whether they met previously defined inclusion criteria. Any remaining disagreement or uncertainty was resolved by consensus through discussion with another review author (CH).

Data extraction and management

Three reviewers (AR, CAC and JC) independently extracted details of articles included in this update, including study design, study population, intervention, usual care, outcome measures used and length of follow‐up data, using a specially designed data extraction form based on the EPOC template (EPOC 2017). We contacted study authors to ask for missing information or clarification. We used information from data extraction forms to guide the extraction of numerical data for meta‐analysis in Review Manager 5.3 (RevMan 2014).

We presented data from randomised trials and controlled before‐after studies (CBA) studies using the format suggested in the EPOC Working Paper on presentation of data (EPOC 2017). We extracted outcome at the last time point reported to assess enduring effects of the intervention.

Assessment of risk of bias in included studies

Three reviewers (AR, CAC and JC) independently assessed the internal validity of each study included in this update and resolved discrepancies by discussion.

We used the Cochrane tool for assessing risk of bias (Higgins 2011), based on six standard criteria: adequate sequence generation, concealment of allocation, blinding of participants and personnel, blinded or objective assessment of primary outcome(s), adequately addressed incomplete outcome data, freedom from selective reporting and freedom from other risks of bias. We used three additional criteria specified by EPOC (EPOC 2017): similarity of baseline characteristics, reliable primary outcome measures and adequate protection against contamination. We reported all included studies in the 'Risk of bias' tables.

Measures of treatment effect

We measured the effect of the intervention by referencing published tools (e.g. implicit, judgement‐based tools such as the MAI (Hanlon 1992) and/or explicit, criterion‐based tools such as 'Beers' (Fick 2003)) used to assess inappropriate prescribing as outlined above. We reported outcomes for each study in natural units. When baseline results were available from studies, means and standard deviation (SD) values for the change from baseline for study and control groups were reported. When baseline results were not available, we reported postintervention means and SD values and/or the proportion of patients with one or more PIMs or PPOs for study and control groups. We analysed data using RevMan 5.3.

In previous versions of this review, we pooled data according to the specific screening tool used. As a modification to the original review protocol, we pooled outcome data on the basis of whether included studies had used an implicit (judgement‐based) or explicit (criterion‐based) tool to measure inappropriate prescribing. The reason for this change to the protocol was that, with an ever increasing number of screening tools being used, it would not be feasible to continue to categorise trial outcome data according to specific screening tools or generate meaningful summary effect estimates. When possible, we presented results with 95% CIs, and estimates when different scales were used to report the same dichotomous outcomes (e.g. the proportion of patients with one or more potentially inappropriate prescriptions) as risk ratios (RRs). We used standardised mean differences (SMDs) in meta‐analyses when different scales were used to report the same continuous outcome.

Unit of analysis issues

We critically examined the methods of analysis of all study types. When studies with a unit of analysis error were identified, we re‐analysed the data excluding such studies (sensitivity analysis).

Dealing with missing data

We assessed the methods used in each included study to deal with missing data. Any study with a differential loss to follow‐up between groups greater than 20% was excluded from meta‐analysis.

Assessment of reporting biases

We assessed reporting bias by scrutinising study results using the 'Risk of bias' tables provided in RevMan 5.3. We examined funnel plots corresponding to meta‐analysis of the primary outcome to assess the potential for small‐study effects such as publication bias.

Data synthesis, subgroup analysis and investigation of heterogeneity

Methods utilised to synthesise the studies depended on their quality, design and heterogeneity. We pooled the results of studies if at least two studies were homogeneous regarding participants, interventions and outcomes. We grouped studies and described them according to type of intervention, setting and study design, and we planned to perform an assessment of evidence on the theoretical basis underpinning the interventions. For example, if studies reported that interventions were based on the Theory of Diffusion (Rogers 2003), then we planned to pool data across these studies, where appropriate, in order to develop a cumulative evidence base for the theory in question. Where possible, instead of subgrouping outcomes according to the specific tool (i.e. STOPP versus Beers), we pooled studies under the broad descriptions of medication appropriateness (as measured by an implicit tool), potentially inappropriate medications (which consists of the number of potentially inappropriate medications and/or the proportion of patients with one or more potentially inappropriate medications), and potential prescribing omissions (which consists of the number of potential prescribing omissions and/or the proportion of patients with one or more potential prescribing omissions).

In the presence of statistical heterogeneity (greater than 50%, as estimated by the I² statistic), we applied a random‐effects model for meta‐analysis. For pooling, we considered only groups of studies of the same design (randomised trials and non‐randomised trials). When it was not possible to combine outcome data because of differences in reporting or substantive heterogeneity, we provided a narrative summary.

Sensitivity analysis

We performed a sensitivity analysis for pooled results based on methodological quality to assess the overall effect. Studies with a unit of analysis error or high risk of bias were excluded from the meta‐analysis.

'Summary of findings' table

We graded our confidence in the evidence by creating a 'Summary of findings' table, using the approach recommended by the GRADE Working Group and guidance developed by EPOC (EPOC 2017b; Guyatt 2008). We included the most important outcomes, which were: medication appropriateness (as measured by an implicit tool), the number of potentially inappropriate medications, the proportion of patients with one or more potentially inappropriate medications, the number of potential prescribing omissions, the proportion of patients with one or more potential prescribing omissions, hospital admission, and quality of life. We used methods and recommendations described in Section 8.5 and Chapter 12 of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011), along with GRADE worksheets, to assess the certainty of evidence (GRADEpro GDT 2015). Two review authors (AR, CC) independently assessed the certainty of evidence for each outcome. We have presented certainty of evidence for each outcome in GRADE tables (summary of findings Table for the main comparison, Appendix 5).

Results

Description of studies

See Characteristics of included studies; Characteristics of excluded studies; Characteristics of ongoing studies; and Characteristics of studies awaiting classification.

Results of the search

We updated the electronic searches and identified 7526 potentially relevant citations (Figure 1). Following review of titles and abstracts, we retrieved 432 full‐text publications for more detailed assessment. We identified 11 additional potentially relevant citations through searches of other sources, such as relevant reviews (Appendix 4), including the list of ongoing studies provided in the previous review (Patterson 2014), and the Clinical Trials Registry, as well as through contact with study authors. From this updated search, 20 studies met all other inclusion criteria (including study design, study population, types of interventions examined) and were added to the review. There were 27 ongoing studies (see Characteristics of ongoing studies).

Figure 1

Study flow diagram.

Included studies

In total, we identified 32 eligible studies, of which 20 were included for this update. The North Carolina Long‐Term Care Polypharmacy Initiative was published as three separate studies (Christensen 2004; Trygstad 2005; Trygstad 2009), but only two of these studies (Trygstad 2005; Trygstad 2009) met the inclusion criteria. Where data from the studies that were added to the review could not be included in any form of meta‐analysis, narrative descriptions of results are presented. Details are provided in the Characteristics of included studies table and are briefly summarised below.

Study design

Included studies consisted of 18 randomised trials (Basger 2015; Bladh 2011; Bucci 2003; Campins 2017; Crotty 2004b; Dalleur 2014; Frankenthal 2014; Fried 2017; Gallagher 2011; Haag 2016; Hanlon 1996; Michalek 2014; Milos 2013; Olsson 2012; Schmader 2004; Spinewine 2007; Taylor 2003; Wehling 2016), 10 cluster‐randomised trials (Clyne 2015; Crotty 2004a; Garcia‐Gollarte 2014; Franchi 2016; Koberlein‐Neu 2016; Muth 2016; Muth 2018; Pitkala 2014; Tamblyn 2003; Thyrian 2017), one of which was a stepped‐wedge design (Koberlein‐Neu 2016), two non‐randomised trials (Chiu 2018; Van der Linden 2017) and two controlled before‐after studies (Trygstad 2005; Trygstad 2009).

Settings

Of the 16 studies conducted in hospital settings (3779 participants), three were conducted in hospital outpatient clinics (Hanlon 1996; Bucci 2003; Schmader 2004), one at the hospital/homecare interface (Crotty 2004b). and 12 in an inpatient setting (Basger 2015; Bladh 2011; Chiu 2018; Dalleur 2014; Franchi 2016; Gallagher 2011; Haag 2016;Michalek 2014; Olsson 2012; Spinewine 2007; Wehling 2016; Van der Linden 2017). Ten studies were conducted in primary care settings (14,969 participants) (Campins 2017; Clyne 2015; Fried 2017; Koberlein‐Neu 2016; Milos 2013; Muth 2016; Muth 2018; Tamblyn 2003; Taylor 2003; Thyrian 2017). Six studies took place in nursing homes (9924 participants) (Crotty 2004a; Frankenthal 2014; Garcia‐Gollarte 2014; Pitkala 2014; Trygstad 2005; Trygstad 2009). All studies reported trials which were confined to a single setting.

The included studies were carried out in 12 high‐income countries: Australia (three studies), Belgium (three studies), Canada (two studies), Finland (one study), Germany (six studies), Hong Kong (one study), Ireland (two studies), Israel (one study), Italy (one study), Spain (two studies) and Sweden (three studies), and the USA (seven studies).

Participants

A total of 28,672 participants were included in this review, most of whom were female (64.4%) and had a mean age of 72.8 years. In those studies where ethnicity was reported (five studies, N = 8710), most participants were white. All study participants had more than one long‐term medical condition, which included asthma, diabetes, dyslipidaemia, hypertension, cardiovascular disease (including congestive heart failure) and dementia. On average, participants were receiving more than four medicines at baseline. In 31 of the 32 studies for which data were available (16,112 participants), participants were prescribed on average 8.9 medicines at baseline.

Interventions

In all cases, interventions were classified as either delivery arrangements (Basger 2015; Bladh 2011; Bucci 2003; Chiu 2018; Crotty 2004b; Fried 2017; Haag 2016; Koberlein‐Neu 2016; Michalek 2014; Milos 2013; Muth 2016; Muth 2018; Olsson 2012; Schmader 2004; Spinewine 2007; Thyrian 2017; Van der Linden 2017), implementation strategies (Franchi 2016; Garcia‐Gollarte 2014), or both (Campins 2017; Clyne 2015; Crotty 2004a; Dalleur 2014; Frankenthal 2014; Gallagher 2011; Hanlon 1996; Pitkala 2014; Tamblyn 2003; Taylor 2003; Trygstad 2005; Trygstad 2009; Wehling 2016) (see Types of interventions for definitions).

Thirty‐one studies examined complex, multi‐faceted interventions of pharmaceutical care in a variety of settings. One uni‐faceted study (Tamblyn 2003) examined computerised decision support (CDS) provided to GPs in their own practices. Pharmaceutical care was commonly provided by pharmacists working closely with other healthcare professionals in a variety of settings. In hospital settings, pharmacists worked as part of a multi‐disciplinary team in outpatient clinics (Bucci 2003; Hanlon 1996; Schmader 2004), in inpatient services on hospital wards as a clinical pharmacy service (Basger 2015; Bladh 2011; Chiu 2018; Dalleur 2014; Franchi 2016; Gallagher 2011; Haag 2016; Michalek 2014; Olsson 2012; Spinewine 2007; Van der Linden 2017; Wehling 2016), or as part of the hospital discharge process (Crotty 2004b). In community settings, pharmaceutical care services, including medication reviews, patient interviews and counselling, were provided by different healthcare professionals. This included pharmacists working in community‐based family medicine clinics (Taylor 2003), or within primary care centres (Campins 2017; Milos 2013), GP (Clyne 2015;Fried 2017; Koberlein‐Neu 2016) and nurses/healthcare assistants (Muth 2016; Muth 2018; Thyrian 2017). In nursing homes, interventions involved multi‐disciplinary case conferences combined with staff education provided by pharmacists (Crotty 2004a), medication reviews by the study pharmacists and discussed with the chief physician (Frankenthal 2014), training sessions for staff (Garcia‐Gollarte 2014; Pitkala 2014), and a drug therapy management service (Trygstad 2005; Trygstad 2009).

Physicians delivered the intervention via a computerised support programme in one study (Tamblyn 2003), whereas in all other studies, structured processes were used to develop recommendations for improving the appropriateness of prescribing to prescribers.

Models of pharmaceutical care provided in the included studies were complex and variable. In 17 studies, the pharmacist(s) conducted an independent medication review using participant notes (Bladh 2011; Campins 2017; Crotty 2004a; Crotty 2004b;Koberlein‐Neu 2016; Milos 2013; Van der Linden 2017), together with participants during a face‐to‐face encounter (Basger 2015; Bucci 2003; Chiu 2018; Frankenthal 2014; Hanlon 1996; Schmader 2004;Spinewine 2007; Tamblyn 2003; Taylor 2003), or during an medication therapy management (MTM) consultation over the telephone (Haag 2016). Following medication reviews, recommendations were discussed with a multi‐disciplinary team during case conferences (Crotty 2004a; Crotty 2004b), sent to patient's own GPs or consultants (Basger 2015; Bladh 2011; Campins 2017; Frankenthal 2014;Milos 2013;Van der Linden 2017), or discussed with prescribers and followed up by written recommendations (Hanlon 1996) from multi‐disciplinary team members at the same outpatient clinic (Bucci 2003), or during inpatient ward rounds (Spinewine 2007). In five studies, medicine reviews were undertaken by the doctor (Clyne 2015; Fried 2017; Muth 2016; Muth 2018; Wehling 2016). In three studies, nurses were asked to identify potential medication‐related problems and bring these to the attention of the consulting physician (Pitkala 2014), or conduct prescription reviews (Thyrian 2017), which were sent to the study physician (Olsson 2012). In one study, the pharmacist was an integral member of the multi‐disciplinary team (Schmader 2004) and contributed to the pharmaceutical care aspect of participants' care plans at the point of decision making. In two studies, consultant pharmacists performed a comprehensive profile review of the computerised drug profiles of selected participants using a range of tools such as the Beers criteria and made recommendations to prescribers in nursing homes by fax, telephone or written communication (Trygstad 2005; Trygstad 2009).

In four studies, participants' medication lists were screened by a geriatrician (Dalleur 2014), or by the primary research physician (Gallagher 2011; Garcia‐Gollarte 2014; Michalek 2014) upon admission to hospital, and oral and written recommendations outlining appropriate prescribing changes were then provided to the attending physicians. In the Dalleur 2014 study, no pharmacist was available to collaborate with the inpatient geriatric consultation team owing to lack of resources within the hospital.

Participant education was provided as part of the pharmaceutical care intervention in four of six studies in which the intervention was conducted face‐to‐face, and these participants were given 'directive guidance' and specialised medication scheduling tools (e.g. monitored dosage systems) to encourage adherence to their prescribed medication regimens (Bucci 2003;Hanlon 1996; Spinewine 2007; Taylor 2003). Directive guidance describes pharmaceutical care activities such as provision of information about medications, their administration and their adverse effects (Bucci 2003). In one study, patients received information leaflets during the medicines reviews, describing potentially inappropriate prescribing (PIP) and alternative treatment options (pharmacological and non‐pharmacological) (Clyne 2015).

Education was provided to prescribers and other healthcare professionals included in the multi‐disciplinary team as part of the intervention in 10 studies (Bucci 2003; Clyne 2015; Crotty 2004a; Crotty 2004b; Franchi 2016; Garcia‐Gollarte 2014; Hanlon 1996;Pitkala 2014; Spinewine 2007; Wehling 2016); this occurred at case conferences, during ward rounds, as part of workshops, or when evidence‐based information and answers to specific medication‐related queries were presented. In two studies in which the pharmacist was part of a multi‐disciplinary team, no educational intervention was specified in the methodology (Schmader 2004; Taylor 2003).

The timing of provision of the intervention was variable. Interventions were delivered over a period of time, for example, during the hospital inpatient stay and at discharge (Bladh 2011; Chiu 2018; Franchi 2016; Haag 2016; Michalek 2014; Schmader 2004; Spinewine 2007; Van der Linden 2017), or over several clinic visits and over several months on an ongoing basis (Tamblyn 2003). Interventions were also delivered at the time of an event, for example, following hospital admission (Dalleur 2014; Gallagher 2011), at discharge from hospital (Basger 2015), during attendance at outpatient clinics (Bucci 2003; Hanlon 1996; Schmader 2004; Taylor 2003), at nursing home visits (Crotty 2004a; Trygstad 2005; Trygstad 2009), at hospital discharge to a nursing home (Crotty 2004b), home visit by a nurse (Olsson 2012), or GP visit (Campins 2017; Clyne 2015; Fried 2017;Muth 2016;Muth 2018). In studies for which details of intervention administration were provided, interventions were most commonly administered during a single episode of care (Bucci 2003; Crotty 2004a; Hanlon 1996; Tamblyn 2003; Taylor 2003; Trygstad 2005; Trygstad 2009). Interventions were implemented over varying durations, ranging from five or six months (Bucci 2003; Trygstad 2005), one year (Frankenthal 2014; Koberlein‐Neu 2016), to three years and three months (Schmader 2004). Further details of the interventions are detailed in the Characteristics of included studies tables.

Outcomes

The first primary outcomes of interest in this review were medication appropriateness (as measured by an implicit tool), potentially inappropriate medications (PIMs) and potential prescribing omissions (PPOs). Validated assessments of appropriateness reported in all included studies were measured independently by pharmacists, geriatricians or the research team, who had access to participants' charts and medication records, except in Trygstad 2005 and Trygstad 2009, where the Medicaid dispensed prescription claims database was used. Time between delivery of the intervention and follow‐up outcome measurement varied from immediately postintervention (e.g. post hospital discharge or clinic visit) (Michalek 2014; Schmader 2004; Spinewine 2007; Tamblyn 2003; Wehling 2016) to at least one month (Bucci 2003), eight weeks (Crotty 2004b), three months (Basger 2015; Crotty 2004a; Garcia‐Gollarte 2014; Trygstad 2005; Trygstad 2009), six months (Clyne 2015; Gallagher 2011), up to one year (Dalleur 2014; Franchi 2016; Hanlon 1996; Pitkala 2014; Taylor 2003), and up to two years (Frankenthal 2014).

Eleven studies measured medication appropriateness (as measured by an implicit tool); the only implicit tool (judgement‐based) used was the Medication Appropriateness Index (MAI) (Bucci 2003; Chiu 2018; Crotty 2004a; Crotty 2004b; Gallagher 2011; Hanlon 1996; Muth 2016; Muth 2018; Schmader 2004; Spinewine 2007; Taylor 2003). Six studies reported MAI as a change from baseline and nine studies reported postintervention scores. One study reported the MAI score in terms of the number of prescriptions with inappropriate medications; this was unsuitable for inclusion in the meta‐analysis (Taylor 2003).

Twenty‐one studies measured PIMs (Bladh 2011; Campins 2017; Clyne 2015; Dalleur 2014; Franchi 2016; Frankenthal 2014; Fried 2017; Gallagher 2011; Garcia‐Gollarte 2014; Haag 2016; Koberlein‐Neu 2016; Milos 2013; Olsson 2012; Pitkala 2014; Schmader 2004; Spinewine 2007; Tamblyn 2003; Thyrian 2017; Trygstad 2005; Trygstad 2009;Van der Linden 2017). These studies used a range of explicit (criterion‐based) tools, including Beers criteria (Franchi 2016; Pitkala 2014; Schmader 2004; Spinewine 2007; Trygstad 2005; Trygstad 2009), Screening Tool of Older Person’s Prescriptions (STOPP) criteria (Campins 2017; Clyne 2015; Dalleur 2014; Frankenthal 2014; Gallagher 2011; Garcia‐Gollarte 2014; Haag 2016), Tool to Reduce Inappropriate Medication (TRIM) recommendations (Fried 2017), the drug‐specific quality indicators established by the Swedish National Board of Health and Welfare (Bladh 2011; Milos 2013; Olsson 2012), the PRISCUS criteria (Koberlein‐Neu 2016; Thyrian 2017) and the Rationalization of home medication by an Adjusted STOPP in older Patients (RASP) list (Van der Linden 2017), which were measured at varying time points ranging from at the point of inpatient discharge to 24‐months follow‐up. Seven studies reported the number of PIMs, as identified using Beers criteria (Pitkala 2014; Schmader 2004; Spinewine 2007) and STOPP criteria (Clyne 2015; Garcia‐Gollarte 2014), the PRISCUS criteria (Koberlein‐Neu 2016), and the RASP list (Van der Linden 2017). Thirteen studies reported the proportion of patients with one or more PIMs, as identified using Beers criteria (Pitkala 2014; Spinewine 2007), the STOPP criteria (Clyne 2015; Dalleur 2014; Frankenthal 2014; Gallagher 2011; Garcia‐Gollarte 2014; Haag 2016), the drug‐specific quality indicators established by the Swedish National Board of Health and Welfare (Milos 2013), TRIM recommendations (Fried 2017) or the PRISCUS criteria (Thyrian 2017).

One study used the McLeod criteria and reported the rate of inappropriate medications prescribed per physician visit postintervention (Tamblyn 2003).

Potential prescribing omissions (PPOs) or under‐use of medication were reported in six studies (Frankenthal 2014; Gallagher 2011; Garcia‐Gollarte 2014; Haag 2016; Schmader 2004; Spinewine 2007), and both were reported as postintervention scores. The only implicit tool used was the Assessment of Under‐utilisation of Medication (AUM) instrument (Jeffery 1999; Gallagher 2011; Schmader 2004). Five studies used explicit tools including the seven process measures from the full range of Assessing Care of Vulnerable Elderly (ACOVE) criteria (Spinewine 2007) and the Screening Tool to Alert doctors to the Right Treatment (START) criteria (Frankenthal 2014; Gallagher 2011; Garcia‐Gollarte 2014;Haag 2016). All five studies using an explicit tool reported the proportion of patients with one or more PPOs, which were measured at varying time points ranging from at the point of inpatient discharge to 24‐months follow‐up.

Three other studies reported results in the form of combined PIM and PPO indicators/scores (Basger 2015; Michalek 2014; Wehling 2016). One study measured appropriateness using the prescribing appropriateness criteria‐set for application in older Australians (Basger 2012) and reported changes in the number of criteria met (Basger 2015). This method uses a combination of both explicit and implicit tools to measure appropriateness. Two studies used the Fit for The Aged (FORTA) criteria (Kuhn‐Thiel 2014), to evaluate the appropriateness of medications in terms of unnecessary, inappropriate or harmful medications and drug omissions (Michalek 2014 Wehling 2016). In the Michalek 2014 study, the number of drugs within each FORTA classification (i.e. FORTA drug labels range from A (indispensable), B (beneficial), C (questionable) to D (avoid)), while the Wehling 2016 study reported the summated FORTA score postintervention along with the change in FORTA score postintervention.

No other validated criteria (e.g. Zhan criteria) were reported.

The other primary outcome of interest in this review was hospital admissions (including unplanned hospital readmissions). Twelve studies measured hospital admissions by examining hospital records at varying time points postintervention (Campins 2017; Chiu 2018; Crotty 2004b; Franchi 2016; Frankenthal 2014; Gallagher 2011; Haag 2016; Muth 2018; Spinewine 2007; Taylor 2003; Trygstad 2005; Van der Linden 2017) ranging from eight weeks (Crotty 2004b; Spinewine 2007), one to three months (Chiu 2018; Haag 2016; Trygstad 2009; Van der Linden 2017) and six months to one year (Campins 2017; Franchi 2016; Frankenthal 2014; Gallagher 2011; Muth 2018; Taylor 2003).

The secondary outcomes of interest in this review were medication‐related problems (i.e. drug interactions, adverse drug reactions (ADRs)), adherence to medication and quality of life. Medication‐related problems, were measured in eight studies and were reported as medication misadventures (defined as iatrogenic incidents that occur as a result of error, immunological response or idiosyncratic response and are always unexpected or undesirable to the participant) (Taylor 2003), potential drug therapy problems (Trygstad 2005; Trygstad 2009), potential drug–drug interaction (DDI) and potentially severe DDI (Franchi 2016) or postintervention adverse drug events (ADEs) (Crotty 2004b; Hanlon 1996; Schmader 2004; Wehling 2016). Adherence to medication was measured in five studies (Campins 2017; Haag 2016; Muth 2016; Muth 2018; Taylor 2003), three studies used Morisky‐Green test (Campins 2017; Muth 2016; Muth 2018), one study used an adapted Morisky Medication Adherence Scale (MMAS) (Haag 2016), and one study assessed adherence to medication via participant self‐report (Taylor 2003). Adherence to medications was assessed at varying time points postintervention ranging from 30 days (Haag 2016), six to nine months (Campins 2017; Muth 2018) and one year (Muth 2016; Taylor 2003). Quality of life (QoL) was assessed in 12 studies using the Medical Outcomes Study 36‐item Short Form health survey (SF‐36) in three studies (Basger 2015; Hanlon 1996; Taylor 2003), the Medical Outcomes Study 12‐item Short‐Form Health Survey (SF‐12) in one study (Frankenthal 2014), the EuroQol‐ED (EQ‐5D) in six studies (Bladh 2011; Campins 2017; Muth 2016; Muth 2018; Olsson 2012; Van der Linden 2017) the 15 dimensional instrument of health‐related quality of life (15D) in one study (Pitkala 2014), and the Quality of Life in Alzheimer Disease instrument in one study (Thyrian 2017). Quality of life was assessed at varying time points postintervention ranging from three months (Basger 2015; Van der Linden 2017), six to nine months (Bladh 2011; Campins 2017; Muth 2018) and one year (Frankenthal 2014; Hanlon 1996; Muth 2016; Olsson 2012; Pitkala 2014; Taylor 2003; Thyrian 2017).

Excluded studies

Excluded publications that were read in full are summarised along with the reasons for exclusion in the Characteristics of excluded studies table.

Studies awaiting classification

Studies for which sufficient information was not available to determine eligibility for inclusion in this review have been allocated to the Studies awaiting classification section.

Ongoing studies

We described ongoing studies identified during completion of the review and provided details such as primary author, research question(s) and methods and outcome measures, together with an estimate of the reporting date in the Characteristics of ongoing studies table appended to this review.

Risk of bias in included studies

Details of the risk of bias are presented in Figure 2 and Figure 3 and in the Characteristics of included studies tables.

Figure 2

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

Figure 3

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

Allocation

Fourteen trials reported adequate sequence generation (Bucci 2003; Campins 2017; Clyne 2015; Crotty 2004a; Crotty 2004b; Gallagher 2011; Garcia‐Gollarte 2014; Haag 2016; Hanlon 1996; Milos 2013; Muth 2018; Pitkala 2014; Schmader 2004; Thyrian 2017), and 13 reported concealment of allocation (Bladh 2011; Campins 2017; Clyne 2015; Crotty 2004a; Crotty 2004b; Frankenthal 2014; Gallagher 2011; Haag 2016; Koberlein‐Neu 2016; Michalek 2014; Milos 2013; Pitkala 2014; Wehling 2016).

Blinding

In 14 studies, blinded measurement of outcomes had taken place to ensure that primary outcome assessors had no knowledge of the intervention received by participants (Bucci 2003; Clyne 2015; Crotty 2004b;Dalleur 2014; Franchi 2016; Frankenthal 2014; Gallagher 2011; Haag 2016; Hanlon 1996; Muth 2016; Pitkala 2014; Schmader 2004; Tamblyn 2003; Wehling 2016). Blinding of participants and personnel had taken place to ensure there was no performance bias in five studies (Garcia‐Gollarte 2014; Michalek 2014; Muth 2016; Olsson 2012; Pitkala 2014).

Incomplete outcome data

Incomplete outcome data were adequately addressed in 21 studies. In one study (Schmader 2004), 864 participants were randomly assigned but only 834 were included in the analysis, and no intention‐to‐treat analysis was reported. Therefore, it was unclear whether all outcome data were included.

Selective reporting

Three studies (Koberlein‐Neu 2016; Spinewine 2007; Thyrian 2017) were considered at high risk of reporting bias. In the Spinewine 2007 study, the authors failed to report one of the secondary outcomes, medications taken.

Similarity of baseline characteristics

In eight studies, baseline demographic differences existed between intervention and control groups and there was no reported adjustment of results to account for baseline differences in analyses.

Other potential sources of bias

The primary outcome measures used were reliable instruments in all studies, for example, MAI kappa value = 0.84.

Participants in six studies were protected from contamination (Clyne 2015; Crotty 2004a; Michalek 2014; Muth 2018; Pitkala 2014,Thyrian 2017). In 14 studies it was unclear whether protection against contamination had been provided (Basger 2015; Dalleur 2014; Franchi 2016; Frankenthal 2014; Fried 2017; Gallagher 2011; Garcia‐Gollarte 2014; Milos 2013; Muth 2016; Olsson 2012; Schmader 2004; Tamblyn 2003; Trygstad 2005; Trygstad 2009), and 12 studies were determined to have high risk of contamination (Bladh 2011; Bucci 2003;Campins 2017; Chiu 2018;Crotty 2004b;Haag 2016;Hanlon 1996; Koberlein‐Neu 2016; Spinewine 2007; Taylor 2003; Van der Linden 2017; Wehling 2016). Contamination bias occurs when members of the control group are inadvertently exposed to the intervention, thus potentially minimising differences in outcomes between the two groups (Higgins 2011). This is an important limitation for this review, where, in some studies, for example, a pharmacist involved in the provision of pharmaceutical care to members of the intervention group may have inadvertently modified the treatment of those in the control group as a result of having knowledge of the intervention. The possible influence of contamination bias should be considered when the results of this review are interpreted.

Funnel plots of postintervention estimates of medication appropriateness (as measured by an implicit tool), the number of potentially inappropriate medications, the proportion of patients with one or more potentially inappropriate medications and the proportion of patients with one or more potential prescribing omissions showed little evidence of publication bias (Figure 4; Figure 5; Figure 6).

Figure 4

Funnel plot of comparison: 1 Postintervention analysis, outcome: 1.1 Medication appropriateness (as measured by an implicit tool).

Figure 5

Funnel plot of comparison: 1 Postintervention analysis, outcome: 1.4 The number of potentially inappropriate medications.

Figure 6

Funnel plot of comparison: 1 Postintervention analysis, outcome: 1.5 The proportion of patients with one or more potentially inappropriate medications.

Effects of interventions

See: Summary of findings for the main comparison Pharmaceutical care compared with usual care for older people receiving polypharmacy

There was a lack of certainty regarding the effects of pharmaceutical care interventions included in this review on inappropriate prescribing (medication appropriateness (as measured by an implicit tool), the number of potentially inappropriate medications (PIMs), the proportion of patients with one or more PIMs and the proportion of patients with one or more potential prescribing omissions (PPOs)). Pharmaceutical care may reduce the number of PPOs, however it must be noted that this effect estimate is based on only two studies, which had serious limitations in terms of risk bias. Hospital admissions, as reported in 12 studies, were reduced in four studies (Chiu 2018; Crotty 2004b; Taylor 2003; Trygstad 2009) (in one cohort, but not in the remaining nine cohorts), and eight studies (Campins 2017; Franchi 2016; Frankenthal 2014; Gallagher 2011; Haag 2016; Muth 2018; Spinewine 2007;Van der Linden 2017) found little or no difference.

No consistent intervention effect on medication‐related problems was observed across studies (eight studies); these problems were reported in terms of adverse drug events (ADEs) (Crotty 2004b; Hanlon 1996; Schmader 2004; Wehling 2016), medication misadventures (Taylor 2003), potential drug therapy problems (Trygstad 2005; Trygstad 2009), and potential drug–drug interactions (DDIs) or potentially severe DDIs (Franchi 2016). Improvement in adherence to medication was demonstrated in one study (Taylor 2003), while the other four studies (Campins 2017; Haag 2016; Muth 2016; Muth 2018) found little or no difference. In the Van der Linden 2017 study, analysis showed that participants in the intervention group experienced an increased quality of life (QoL), in the Pitkala 2014 study, there was a decline in QoL in both the intervention and control groups, although the decline was significantly lower in the intervention group (‐0.038 in the intervention group versus ‐0.072 in the control group), and no changes in QoL were detected in 10 studies (Bladh 2011; Basger 2015; Campins 2017; Frankenthal 2014; Hanlon 1996; Muth 2016; Muth 2018; Olsson 2012; Taylor 2003;Thyrian 2017).

Based on the GRADE approach (Guyatt 2008), the overall certainty of the body of evidence for each primary outcome for which data were included in a meta‐analysis was deemed to be low or very low, which means that the confidence in the effect estimates is very limited. Although each study included in the meta‐analyses was of a randomised design, and, where assessed, no evidence of publication bias was found (Figure 4; Figure 5; Figure 6), the certainty of the body of evidence was downgraded for each outcome based on other GRADE considerations (i.e. study limitations, consistency of effect, imprecision, indirectness) (Appendix 5).

Primary outcome results

Medication appropriateness (as measured by an implicit tool)

It is uncertain whether pharmaceutical care improves medication appropriateness (as measured by an implicit tool) because the certainty of this evidence is very low (5 studies, N = 517). Three studies reported medication appropriateness using an implicit (judgement‐based) assessment tool (Bucci 2003; Crotty 2004a; Muth 2016), and further unpublished data were received from the authors of two studies (Crotty 2004b; Spinewine 2007). All of these studies used the Medication Appropriateness Index (MAI) as the implicit tool. Comparison of medication appropriateness (as measured by an implicit tool) from baseline to follow‐up between the intervention group and the control group is shown in Analysis 1.1. Overall, a greater improvement in medication appropriateness (as measured by an implicit tool) postintervention was seen in the intervention group compared with the control group (mean difference (MD) ‐4.76, 95% confidence interval (CI) ‐9.20 to ‐0.33; I² = 95%; 5 studies; N = 517, Analysis 1.1). Marked heterogeneity between studies was noted (95%). Crotty 2004a reported a unit of analysis error; nursing homes were the unit of randomisation, but the analysis was conducted at the participant level. A sensitivity analysis excluding Crotty 2004a showed a similar improvement in medication appropriateness (as measured by an implicit tool) (MD ‐5.16, 95% CI ‐11.04 to 0.72; I² = 96%; N = 446, Analysis 1.2) in favour of the intervention group. A further sensitivity analysis removing both Crotty 2004a and Spinewine 2007,an outlying study with a large effect size that had a high risk of bias with respect to selection bias (allocation concealment), performance bias, detection bias, contamination bias and selective reporting, also showed a greater improvement in medication appropriateness (as measured by an implicit tool) in the intervention group, but the magnitude of the difference was smaller compared with previous analyses (MD ‐0.50, 95% CI ‐2.27 to 1.28; I² = 57%; N = 260, Analysis 1.3). The level of heterogeneity between studies was also found to have reduced.

We downgraded the certainty of the body of evidence for medication appropriateness (as measured by an implicit tool) to very low. Very serious design limitations with implications in terms of selection bias, performance bias, reporting bias and risk of contamination bias were identified in several studies. Spinewine 2007 was deemed to have high risk of bias in terms of selection bias (allocation concealment), performance bias, detection bias, contamination bias and selective reporting, which resulted in the downgrading of the certainty of evidence. The certainty of evidence was downgraded due to indirectness, some studies answered a restricted version of the research question, as a validated assessment of under‐prescribing was not included as part of the overall assessment of inappropriate prescribing. Therefore, interventions did not directly target appropriate polypharmacy. Additionally, evidence of inconsistency (I² = 95%) was identified, as well as imprecision in the effect estimate, whereby the 95% CI was wide and/or crossed the line of no effect. These observations resulted in the downgrading of the certainty of evidence.

Potentially inappropriate medications (PIMs) (including the number of potentially inappropriate medications and the proportion of patients with one or more PIMs)

Pooled data from seven studies (Bladh 2011; Clyne 2015; Garcia‐Gollarte 2014; Koberlein‐Neu 2016; Pitkala 2014; Schmader 2004; Spinewine 2007) showed that the number of potentially inappropriate medications was lower in the intervention group participants compared with control group participants postintervention (standardised mean difference (SMD) ‐0.22, 95% CI ‐0.38 to ‐0.05; I² = 67%; 7 studies; N = 1832, Analysis 1.4). The numbers of PIMs were determined using explicit (criterion‐based) assessment tools, including Screening Tool of Older Person’s Prescriptions (STOPP) (version 1: Gallagher 2008), and Beers (1997 version: Beers 1997 and 2003 version: Fick 2003), PRISCUS criteria (Holt 2010), and the drug‐specific quality indicators established by the Swedish National Board of Health and Welfare (Fastbom 2015). However, it is uncertain whether pharmaceutical care reduces the number of potentially inappropriate medications because the certainty of this evidence is very low. The Trygstad 2009 study, which also reported the number of Beers list drugs, comprised 10 cohorts. It was not included in the meta‐analysis, as the study design, analysis and reporting (e.g. using propensity matching, reporting results as difference‐in‐difference) differed from the others, resulting in estimates that were not sufficiently similar to support inclusion. The Trygstad 2009 study, also reported no statistically significant reductions in Beers list alerts, which is not inline with the meta‐analysis results. The Olsson 2012 study reported number of drug‐risk indicators per patient according to the drug‐specific quality indicators established by the Swedish National Board of Health and Welfare and the Campins 2017 study reported the proportion of patients with at least one drug discontinuation based on STOPP criteria. These studies were not included in the meta‐analyses as the analysis and reporting differed from the other. We were also unable to ascertain the standard deviation of the results for two studies (Trygstad 2005;Van der Linden 2017), which were also not included in the meta‐analysis.

We downgraded the certainty of the body of evidence for the number of potentially inappropriate medications to very low due to very serious design limitations in both studies that were included in the meta‐analysis, with implications in terms of risk of selection bias, performance bias and contamination bias. Evidence of inconsistency (I² = 67%) was identified possibly due to some of the studies answering a restricted version of the research question, as a validated assessment of under‐prescribing was not included as part of the overall assessment of inappropriate prescribing. Therefore, all of the interventions did not directly target appropriate polypharmacy.

Eleven studies reported the proportions of patients with one or more potentially inappropriate medications (Clyne 2015; Dalleur 2014; Franchi 2016; Frankenthal 2014; Fried 2017; Gallagher 2011; Garcia‐Gollarte 2014; Haag 2016; Milos 2013; Spinewine 2007; Thyrian 2017) before and after intervention. The proportions of patients with one or more PIMs were determined using explicit (criterion‐based) assessment tools, including STOPP (version 1: Gallagher 2008), and Beers (1997 version: Beers 1997 and 2012 version: AGS 2012 ) (Appendix 1), the Tool to Reduce Inappropriate Medication (TRIM) recommendations based on Beers (2012 version: AGS 2012) and STOPP criteria (version 1: Gallagher 2008), PRISCUS (Holt 2010) and the drug‐specific quality indicators established by the Swedish National Board of Health and Welfare (Fastbom 2015). Pooled data from 11 studies showed that improvements were reported in the proportion of intervention patients with one or more PIMs, compared to the control group participants, between baseline and discharge (risk ratio (RR) 0.79, 95% CI 0.61 to 1.02; I² = 85%; 11 studies; N = 3079, Analysis 1.5). There was considerable heterogeneity among the 11 trials (heterogeneity: Tau² = 0.14; Chi² = 64.90, df = 10 (P < 0.00001); I² = 85%). A sensitivity analysis excluding Spinewine 2007, a study with a large effect size that had a high risk of bias with respect to selection bias (allocation concealment), performance bias, detection bias, contamination bias and selective reporting, showed similar improvements in the proportion of intervention patients with one or more PIMs, compared to the control group participants, between baseline and discharge (RR 0.79, 95% CI 0.61 to 1.02; I² = 86%; 10 studies; N = 2893, Analysis 1.6). A further sensitivity analysis removing both Spinewine 2007 and Gallagher 2011, which had a smaller treatment effect compared to the other studies, also showed similar improvements in the proportion of intervention patients with one or more PIMs, compared to the control group participants, between baseline and discharge (RR 0.88, 95% CI 0.72 to 1.09; I² = 75%; 9 studies; N = 2535, Analysis 1.7). It is uncertain whether pharmaceutical care reduces the proportion of patients with one or more potentially inappropriate medications because the certainty of this evidence is very low.

We downgraded the certainty of the body of evidence for the proportion of patients with one or more potentially inappropriate medications to very low. Very serious design limitations with implications in terms of selection bias, performance bias and risk of contamination bias were identified in several studies. Spinewine 2007 was deemed to have high risk of bias in terms of selection bias (allocation concealment), performance bias, detection bias, contamination bias and selective reporting which resulted in the downgrading the certainty of evidence. The certainty of evidence was downgraded due to indirectness, as some studies answered a restricted version of the research question, as a validated assessment of under‐prescribing was not included as part of the overall assessment of inappropriate prescribing. Therefore, interventions did not directly target appropriate polypharmacy. Additionally, evidence of inconsistency (I² = 85%) as well as imprecision in the effect estimate, whereby the 95% CI was wide and/or crossed the line of no effect was identified which resulted in the downgrading of the certainty of evidence.

Potential prescribing omissions (PPOs) (including the number of potential prescribing omissions and the proportion of patients with one or more PPOs)

Pooled data from two studies (Garcia‐Gollarte 2014; Spinewine 2007) showed that the number of PPOs was lower in the intervention group participants compared with control group participants postintervention (SMD ‐0.81, 95% CI ‐0.98 to ‐0.64; 2 studies; N = 569, Analysis 1.8). The number of PPOs was determined using explicit (criterion‐based) assessment tools, including Assessing Care of the Vulnerable Elderly (ACOVE) (version 1: Wenger 2001) and START (version 1: Gallagher 2008). Pharmaceutical care may slightly reduce the number of potential prescribing omissions (low‐certainty evidence).

We downgraded the certainty of the body of evidence for the number of potential prescribing omissions to low. Very serious design limitations with implications in terms of selection bias, performance bias and risk of contamination bias were high or unclear in both studies. Spinewine 2007 was deemed to have high risk of bias in terms of selection bias (allocation concealment), performance bias, detection bias, contamination bias and selective reporting which resulted in the downgrading of the certainty of evidence.

Five studies (Frankenthal 2014; Gallagher 2011; Garcia‐Gollarte 2014; Haag 2016; Spinewine 2007), also reported the proportion of patients with one or more potential prescribing omissions. The proportions of patients with one or more PPOs were determined using explicit (criterion‐based) assessment tools, including START (version 1: Gallagher 2008), and ACOVE (version 1: Wenger 2001). The proportion of patients in the intervention group with one or more potential prescribing omissions was lower than for those in the control group (RR 0.40, 95% CI 0.18 to 0.85; I² = 90%; 5 studies; N = 1310, Analysis 1.9). There was considerable heterogeneity among the four trials (heterogeneity: Tau² = 0.67; Chi² = 41.82, df = 4 (P < 0.00001); I² = 90%). It is uncertain whether pharmaceutical care reduces the proportion of patients with one or more potential prescribing omissions because the certainty of this evidence is very low.

We downgraded the quality of the body of evidence for the proportion of patients with one or more PPOs due to very serious design limitations with implications in terms of selection bias, performance bias and risk of contamination bias in several studies. Spinewine 2007 was deemed to have high risk of bias in terms of selection bias (allocation concealment), performance bias, detection bias, contamination bias and selective reporting which resulted in downgrading the certainty of evidence. Evidence of inconsistency (I² = 90%) was identified which resulted in the downgrading of the certainty of evidence.

As only one uni‐faceted study was included (Tamblyn 2003), a subgroup analysis was not possible.

Hospital admissions

Twelve studies measured hospital admissions postintervention (Campins 2017; Chiu 2018; Crotty 2004b; Franchi 2016; Frankenthal 2014; Gallagher 2011; Haag 2016; Muth 2018; Spinewine 2007; Taylor 2003, Trygstad 2009;Van der Linden 2017). Eight studies (Campins 2017; Franchi 2016; Frankenthal 2014; Gallagher 2011; Haag 2016; Muth 2018; Spinewine 2007;Van der Linden 2017) (N = 3041) reported similar hospital admissions between intervention and control group participants postintervention, and the remaining studies reported some overall reductions in hospital admissions using a variety of measurements, as detailed below.

Taylor 2003 reported a reduction in both the number of hospital admissions (P value = 0.003) and the number of emergency department visits (P value = 0.044) during the intervention year compared with preintervention. Crotty 2004b reported less hospital usage among participants who received the intervention and were still alive at eight weeks postintervention compared with control group participants (risk ratio (RR) 0.38, 95% CI 0.15 to 0.99). However, analysis of all participants including deaths and losses to follow‐up showed similar hospital usage in the intervention and control groups (‐9 (16.7%) with intervention versus ‐15 (26.8%) with control; RR 0.58, 95% CI 0.28 to 1.21). Trygstad 2009 showed a reduction in the RR of hospital admissions in one cohort of nursing home residents receiving retrospective‐only–type medication reviews (RR 0.84, 95% CI 0.71 to 1.00; P value = 0.04). The remaining eight cohorts also had an RR below 1.0; however, confidence intervals for the individual point estimates crossed the line of no effect. Inappropriate prescribing was also reported by these studies. In the study by Trygstad 2009, the Beers list was used to measure inappropriate medication, but no reductions were observed in the cohorts receiving retrospective medication review. In the remaining four studies, inappropriate prescribing was reduced, as shown by reductions in PIMs, but the association with hospital admissions was inconsistent. Chiu 2018 reported that the unplanned hospital readmission rate one month after discharge was significantly lower in the intervention group than that in the control group (13.2% versus 29.1%; P = 0.005).

Because of differences in methods used to measure hospital admissions and the expression of results, a meta‐analysis was not possible for studies reporting hospital admissions. Overall, pharmaceutical care may make little or no difference in hospital admissions (low‐certainty evidence). We downgraded the certainty of the body of evidence for hospital admissions to very low due to very serious design limitations with implications in terms of selection bias, performance bias and risk of contamination bias in several studies.

Secondary outcome results

Medication‐related problems (e.g. adverse drug reactions (ADRs), drug‐drug interactions (DDIs))

Medication‐related problems were reported in eight studies (Crotty 2004b; Franchi 2016; Hanlon 1996; Schmader 2004; Taylor 2003; Trygstad 2005; Trygstad 2009; Wehling 2016, N = 10,087) using different terms. In the studies which gave details, medication‐related problems were measured via hospital records (Wehling 2016), patient self‐report during closeout telephone interviews (Hanlon 1996), reviewing the adverse event narrative using Naranjo’s algorithm (Schmader 2004), and using the INTERcheck® software to detect DDIs (Franchi 2016).

No consistent intervention effect on medication‐related problems was noted across studies. Four studies reported medication‐related problems as adverse drug events (ADEs) (Crotty 2004b; Hanlon 1996; Schmader 2004; Wehling 2016). Schmader 2004 showed that the risk of a serious ADE was reduced (RR 0.65, 95% CI 0.45 to 0.93; P value = 0.02) in a geriatric outpatient clinic compared with usual outpatient care; however, little or no difference in the risk of an ADE was noted when all types of ADEs were considered (RR 1.03, 95% CI 0.86 to 1.23; P value = 0.75). Wehling 2016 showed that the total number of adverse drug reactions (ADRs) of specific geriatric relevance (incidence of falls, confusion, nausea, dizziness, obstipation, diarrhoea, dyspnoea, cardiac decompensation, angina pectoris and renal failure) were significantly reduced by implementation of the FORTA‐based intervention (P value < 0.05). The other two studies (Crotty 2004b; Hanlon 1996), showed little or no difference between proportions of intervention and control group participants with ADEs at follow‐up. Franchi 2016 also reported no decrease in the prevalence of at least one potential DDI (odds ratio (OR) 0.67, 95% CI 0.34 to 1.28) and potentially severe DDI (OR 0.86, 95% CI 0.63 to 1.15) at discharge. Taylor 2003 reported medication‐related problems as medication misadventures. Proportions of intervention group (2.8%) and control group (3.0%) participants with at least one medication misadventure at 12 months were similar (P value = 0.73).

Potential medication problems categorised as 'consider duration' (of therapy), 'clinical initiatives' and 'therapeutic duplication' were reported in the two North Carolina initiative studies (see Characteristics of included studies tables; Trygstad 2005; Trygstad 2009). At three months, duration alert rates were reduced by 6.3% in the intervention group (N = 5160) and by 16.7% in the control group (N = 2202); clinical initiatives were reduced by 10.8% in the intervention group and 0.7% in the control group, and therapeutic duplication was reduced in the intervention group by 9.4% and in the control group by 8.8% (Trygstad 2005). Control group results were not reported separately in Trygstad 2009. At three months, duration of therapy alerts were reduced by 27.8% (mean difference in the difference (mDID) = ‐0.023; P value > 0.05); clinical initiative alerts were reduced by 13.9% (mDID = ‐0.24; P < 0.05); and therapeutic duplication alerts were reduced by 5.6% (mDID = ‐0.087; P value > 0.05) (Trygstad 2009).

Adherence to medication

Five studies reported adherence to medication. Four studies reported little or no differences in adherence scores between intervention and control groups at follow‐up (Campins 2017; Haag 2016; Muth 2016; Muth 2018) based on the Morisky‐Green test and adapted Morisky Medication Adherence Scale. One study (Taylor 2003) (N = 69) reported adherence to medication in terms of compliance scores, calculated through assessment of participants' reports of missed doses. Those with medication compliance scores of 80% to 100% increased by 15% at 12 months from a mean (± standard deviation (SD)) of 84.9 ± 6.7% to 100% in the intervention group (N = 33), but the control group (N = 36) did not change from 88.9% ± 5.8% at baseline to 88.9% ± 6.3% at 12 months (P value = 0.115). Because of differences in methodology in the measurement of adherence and the expression of results, a meta‐analysis was not possible for studies reporting adherence to medication.

Quality of life (QoL) (as assessed by a validated method)

Twelve studies (Basger 2015; Bladh 2011; Campins 2017; Frankenthal 2014; Hanlon 1996; Muth 2016; Muth 2018; Olsson 2012; Pitkala 2014; Taylor 2003; Thyrian 2017; Van der Linden 2017, N = 3211) assessed QoL using four different scales (EQ‐5D, SF‐36, SF‐12 and 15D). In the Van der Linden 2017 study, analysis showed that participants in the intervention group experienced an increased QoL when compared to the control group. In the Pitkala 2014 study, there was a decline in QoL (using the 15D) in both the intervention and control groups, although the decline was significantly lower in the intervention group (‐0.038 in the intervention group versus ‐0.072 in the control group). Little or no differences in QoL scores (SF‐36, EQ‐5D and SF‐12) were observed between groups at baseline or at endpoint in ten studies (Basger 2015; Bladh 2011; Campins 2017; Frankenthal 2014; Hanlon 1996; Muth 2016; Muth 2018; Olsson 2012; Taylor 2003; Thyrian 2017). Pharmaceutical care may make little or no difference in QoL (low‐certainty evidence). The certainty of the body of evidence for QoL was downgraded to low. Very serious design limitations with implications in terms of selection bias, performance bias and risk of contamination bias were identified in several studies. Because of differences in methodology in the measurement of quality of life and the expression of results, a meta‐analysis was not possible for studies reporting quality of life.

Discussion

available in

Summary of main results

The addition of 20 studies to this updated review, which now includes 32 studies, highlights a notable increase in intervention studies that have been conducted to date aimed at improving appropriate polypharmacy in older people. However, these additional 20 studies had little impact on the overall findings of the review. The included studies were limited by their small sample sizes and poor certainty of evidence (as assessed using GRADE).

The presentation of primary outcome data in this update differed to previous versions of the review. The review authors considered that with the ever‐increasing number of tools/indicators being developed and used in studies to assess inappropriate prescribing, it may not be helpful to continue subgrouping outcomes according to the specific tool (i.e. STOPP versus Beers). Instead, the outcomes were classified under the broad categorisation of medication appropriateness (as measured by an implicit tool), potentially inappropriate medications (PIMs) (which consists of the number of PIMs and/or the proportion of patients with one or more PIMs) and potential prescribing omissions (PPOs) (which consists of the number of PPOs and/or the proportion of patients with one or more PPOs). For example, rather than looking at explicit tools like STOPP and Beers individually, the current review has focused on the number of PIMs and pooled relevant data (using appropriate statistical methods), assessed by different tools. The standardised mean difference (SMD) is used as a summary statistic in meta‐analyses when the studies all assess the same continuous outcome but measure it in a variety of ways (for example, the studies measuring the numbers of PIMs using different explicit tools). In this circumstance, it is necessary to standardise the results of the studies to a uniform scale before they can be combined. The SMD expresses the size of the intervention effect in each study relative to the variability observed in that study. This would also therefore ameliorate any differences between revised versions of the same scale (i.e. Beers criteria: 1997, 2003 and 2012 versions).

Medication appropriateness (as measured by an implicit tool) were normally distributed and were more suitable for meta‐analysis, but greater heterogeneity was noted among the included studies (I² = 95%), largely because of the influence of the results of one study (Spinewine 2007). Overall, medication appropriateness (as measured by an implicit tool) in the intervention group postintervention was greater than that in the control group and indicated an improvement in the appropriateness of the medications prescribed. A sensitivity analysis in which Crotty 2004a was removed because of a unit of analysis error showed further improvement in the effect estimate when compared with the meta‐analysis. Furthermore, removal of an outlying study with a large effect size (Spinewine 2007), reduced heterogeneity but also reduced the effect estimate. This may have been related to the small sample size for this meta‐analysis (82 intervention participants and 85 control participants). However, it is uncertain whether pharmaceutical care improves medication appropriateness (as measured by an implicit tool) because the certainty of this evidence is very low.

When the studies measuring PIMs (i.e. based on the number of PIMs and/or the proportion of patients with one or more PIMs), as determined using explicit tools (criterion‐based), were combined the number of PIMs: Bladh 2011; Clyne 2015; Garcia‐Gollarte 2014;Koberlein‐Neu 2016; Pitkala 2014; Schmader 2004; Spinewine 2007; the proportion of patients with one or more PIMs: Clyne 2015; Dalleur 2014; Franchi 2016; Frankenthal 2014; Fried 2017; Gallagher 2011; Garcia‐Gollarte 2014; Haag 2016; Milos 2013; Spinewine 2007; Thyrian 2017), differences between intervention and control groups in the number of PIMs favoured the intervention group. A sensitivity analysis excluding Spinewine 2007, a study with a large effect size that had a high risk of bias showed similar improvements in the proportion of intervention patients with one or more PIMs, compared to the control group participants, between baseline and discharge. A further sensitivity analysis removing both Spinewine 2007 and Gallagher 2011, which had a smaller treatment effect compared to the other studies, also showed similar improvements in the proportion of intervention patients with one or more PIMs, compared to the control group participants, between baseline and discharge. It is uncertain whether pharmaceutical care reduces the number of PIMs or the proportion of patients with one or more PIMs because the certainty of this evidence is very low.

When the studies measuring PPOs (i.e. based on the number of PPOs and/or the proportion of patients with one or more potential prescribing omissions), as determined using explicit tools (criterion‐based), were combined (The number of PPOs: Garcia‐Gollarte 2014; Spinewine 2007; the proportion of patients with one or more PPOs: Frankenthal 2014; Gallagher 2011; Garcia‐Gollarte 2014; Haag 2016; Spinewine 2007), there was a reduction in the proportion of patients with one or more PPOs in the interventions group compared to the control groups. The heterogeneity present in the meta‐analysis may have been due to the fact that the studies employed a number of different measurement instruments (Analysis 1.8; Analysis 1.9). Furthermore, differences between intervention and control groups in the number of PPOs also favoured the intervention group. It is uncertain whether pharmaceutical care reduces the proportion of patients with one or more PPOs because the certainty of this evidence is very low. Yet, pharmaceutical care may slightly reduce the number of PPOs (low‐certainty evidence). However, the clinical significance of these changes is unclear due to the fact that this effect estimate is based on only two studies, which had serious limitations in terms of high risk of bias.

The various tools used to assess inappropriate prescribing in the included studies are surrogate markers of appropriate polypharmacy. As was observed in previous versions of this review, few studies examined clinical outcomes, and this should be addressed in future studies. For example, only 12 studies reported on hospital admissions and quality of life. However, we were unable to combine these results, as the reporting styles were different across studies. Based on available evidence, pharmaceutical care may make little or no difference in hospital admissions or quality of life (low‐certainty evidence).

Overall completeness and applicability of evidence

The types of interventions included in the review were limited. Few trials aimed to improve the skills of the prescriber. Most interventions were pharmaceutical care interventions, which included outreach by pharmacists, screening of automated drug alerts by consultant pharmacists visiting nursing homes and clinical pharmacist interventions in various settings. Only two trials involving computerised decision support (CDS) (one of which had incorporated CDS as a component of a multi‐faceted intervention) were identified. The interventions were complex and most were multi‐faceted. The observed heterogeneity observed in the pooled estimates means that the results of the meta‐analyses should be treated cautiously as the interventions did not seem to work consistently across all studies. There was also a lack of studies which have explored implementation at the population level. In addition, study‐specific factors, such as variation in the quality of studies, may have played a role. The methods sections of studies provided little detail on how complex interventions were developed, how trials were designed and how staff were trained in delivery of the intervention. Other information pertinent to the success of pharmaceutical care interventions including background practice and culture, documentation, communication and sharing of information and extent of access to clinical records given to intervention pharmacists was not stated clearly in the papers.

Although the effect of interventions on potentially inappropriate prescribing (PIP) was potentially promising and suggested that some of the interventions described in this review may have helped to improve the appropriateness of polypharmacy, despite observed limitations in the available evidence, the clinical impact of these reductions in inappropriate prescribing is not known. For example, the clinical impact of a mean difference of 0.22 PIMs between intervention and control group patients is unclear. This is partly due to the fact that the predictive validity of many assessment tools has not been established (Cahir 2014). In addition, we were unable to pool data from included studies for clinical outcomes such as hospital admissions due to heterogeneity in terms of outcome assessment and reporting across studies

Furthermore, few rigorously conducted studies have tested interventions and examined clinically relevant outcomes such as hospital admissions or ADEs. Twelve studies in this review reported hospital admissions postintervention (Campins 2017; Chiu 2018; Crotty 2004b; Franchi 2016; Frankenthal 2014; Gallagher 2011; Haag 2016; Muth 2018; Spinewine 2007; Taylor 2003;Trygstad 2009; Van der Linden 2017), and four studies (Crotty 2004b; Gallagher 2011; Spinewine 2007; Taylor 2003) reported that the appropriateness of prescribing improved, as was shown by reductions in PIMs, although the association with hospital admissions was inconsistent. In Trygstad 2009, little or no difference was found in the number of Beers list alerts postintervention, but the relative risk of hospital admissions was reduced. Use of different appropriateness scales in the included studies made it difficult to assess the impact of any change of medication appropriateness on hospital admissions. Similarly, some associations between measures of medication appropriateness and medication‐related problems appeared to exist but were difficult to assess because of variation in scales used to measure outcomes and in reporting methods.

Evidence of potential bias was found in numerous studies. For example, only 13 studies reported adequate concealment of allocation, and only six reported appropriate protection from contamination, both of which may have influenced the effect estimate in these studies and therefore the overall pooled estimate.

The aim of many of the intervention studies included in this review was to reduce harm resulting from inappropriate prescribing and to ensure that older people were prescribed appropriate medications that enhance their quality of life. In previous iterations of this review, several studies focused on reducing the number of medications, rather than improving overall appropriateness of prescribing, including under‐prescribing, that is, recommending medications that are clinically indicated yet are currently missing. An increasing number of studies meeting the inclusion criteria included a validated assessment of under‐prescribing; three studies in the updated review assessed under‐prescribing adding to the three studies reported in the previous version. Furthermore, an increasing number of studies meeting the inclusion criteria also included a measure of quality of life, however only one of the 12 studies reported a benefit; this may be due to the fact that the follow‐up period ranged from three months to 12‐months follow‐up.

Certainty of the evidence

Although we identified 32 studies, pooled analyses remain limited. For example, the meta‐analysis based on the number of PPOs per participant comprised just two studies. This limits the value of any pooled effect estimate. Furthermore, as shown in the summary of findings Table for the main comparison, the certainty of evidence presented in this review, as described by the GRADE approach, remains low or very low. Despite inclusion of data from randomised trial designs in the meta‐analyses, the certainty of the body of evidence was subsequently downgraded when each of the GRADE considerations (i.e. study limitations, consistency of effect, imprecision, indirectness, publication bias) was taken into account. This limits our confidence in the pooled effect estimates.

Based on observed heterogeneity in the pooled effect estimates, the findings of meta‐analyses [medication appropriateness (as measured by an implicit tool), the number of PIMs and proportion of patients with one or more PIMs or PPOs) should be treated cautiously, as the interventions did not seem to work consistently across all studies. Factors contributing to this heterogeneity could have included variation in type, intensity and duration of interventions, as well as differences in the timing of follow‐up assessments. In addition, study‐specific factors such as variation in study quality may have played a role. However, no systematic approach was used to ensure a consistent level of detail in published reports of the interventions. For example, the methods sections of the studies provided little detail on the development of complex interventions, trial design or staff training in delivery of interventions. Other information pertinent to intervention success, such as documentation, communication and intervention pharmacists' level of access to clinical records, was not clearly reported in the papers. The specific processes that constituted successful interventions were often unclear, which may have contributed to heterogeneity in effect estimates.

Potential biases in the review process

No language restrictions were placed on the search strategy, but all of the included trials were published in English and were conducted in high‐income countries. Despite the limited number of studies included in the meta‐analyses, funnel plots of studies reporting medication appropriateness (as measured by an implicit tool), the number of PIMs, the proportion of patients with one or more PIMs, the number of PPOs and the proportion of patients with one or more PPOs, outcomes revealed no apparent publication bias.

Agreements and disagreements with other studies or reviews

Other systematic reviews have reported that the most influential factor affecting the results of pharmaceutical care interventions is the way that interventions were conducted, for example, face‐to‐face consultations with physicians achieved a greater reduction in the number of medications taken than was achieved by written recommendations (Rollason 2003). Another narrative review reported that timely provision of the intervention, that is, prospective advice at the time of prescription rather than at the time of dispensing of medication, is more effective (Spinewine 2007a). A recent and related Cochrane Review of interventions to optimise prescribing for older people in care homes (Alldred 2016), found no evidence of an intervention effect on any of the primary outcomes, which included ADEs and hospital admissions. Other studies of interventions conducted across a variety of settings have also been unable to detect the effects of pharmaceutical care on these outcome measures (Holland 2007; Spinewine 2007a; Johansson 2016). One systematic review (Kaur 2009), revealed that the most successful types of interventions used to reduce inappropriate prescribing in older people were those that had multi‐disciplinary involvement including a geriatrician, utilised CDS or had mandatory pharmaceutical services or drug restriction policies in place. Results of this current review largely support the findings described above, as most of the pharmaceutical care interventions involved a multi‐disciplinary component, and the CDS intervention study (Tamblyn 2003) reported a positive result. A Cochrane Review of interventions to improve outcomes in patients with multimorbidity in primary care and community settings (Smith 2016), found that there may have been small improvements in provider behaviour (in terms of prescribing behaviour) and patient‐reported outcomes (i.e. quality of life). Additionally, a systematic review and meta‐analysis (Meid 2015) found that pharmaceutical care interventions, including medication reviews, can significantly reduce medication underuse in older people.

Figure 1

Study flow diagram.

Navigate to figure in ReviewOpen in new tab

Figure 2

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

Navigate to figure in ReviewOpen in new tab

Figure 3

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

Navigate to figure in ReviewOpen in new tab

Figure 4

Funnel plot of comparison: 1 Postintervention analysis, outcome: 1.1 Medication appropriateness (as measured by an implicit tool).

Navigate to figure in ReviewOpen in new tab

Figure 5

Funnel plot of comparison: 1 Postintervention analysis, outcome: 1.4 The number of potentially inappropriate medications.

Navigate to figure in ReviewOpen in new tab

Figure 6

Funnel plot of comparison: 1 Postintervention analysis, outcome: 1.5 The proportion of patients with one or more potentially inappropriate medications.

Navigate to figure in ReviewOpen in new tab

Analysis 1.1

Comparison 1 Postintervention analysis, Outcome 1 Medication appropriateness (as measured by an implicit tool).

Navigate to figure in ReviewOpen in new tab

Analysis 1.2

Comparison 1 Postintervention analysis, Outcome 2 Medication appropriateness (as measured by an implicit tool) (excl Crotty 2004a).

Navigate to figure in ReviewOpen in new tab

Analysis 1.3

Comparison 1 Postintervention analysis, Outcome 3 Medication appropriateness (as measured by an implicit tool) (excl Crotty 2004a and Spinewine 2007).

Navigate to figure in ReviewOpen in new tab

Analysis 1.4

Comparison 1 Postintervention analysis, Outcome 4 The number of potentially inappropriate medications.

Navigate to figure in ReviewOpen in new tab

Analysis 1.5

Comparison 1 Postintervention analysis, Outcome 5 The proportion of patients with one or more potentially inappropriate medications.

Navigate to figure in ReviewOpen in new tab

Analysis 1.6

Comparison 1 Postintervention analysis, Outcome 6 The proportion of patients with one or more potentially inappropriate medications (excl Spinewine 2007).

Navigate to figure in ReviewOpen in new tab

Analysis 1.7

Comparison 1 Postintervention analysis, Outcome 7 The proportion of patients with one or more potentially inappropriate medications (excl Spinewine 2007 and Gallagher 2011).

Navigate to figure in ReviewOpen in new tab

Analysis 1.8

Comparison 1 Postintervention analysis, Outcome 8 The number of potential prescribing omissions.

Navigate to figure in ReviewOpen in new tab

Analysis 1.9

Comparison 1 Postintervention analysis, Outcome 9 The proportion of patients with one or more potential prescribing omissions.

Navigate to figure in ReviewOpen in new tab

Summary of findings for the main comparison. Pharmaceutical care compared with usual care for older people receiving polypharmacy

Patient or population: older people receiving polypharmacy Settings: community, nursing home, hospital Intervention: pharmaceutical care Comparison: usual care
Outcomes	Effect estimate		Relative Risk effect (95% CI)	No. of participants (studies)	Certainty of evidence (GRADE)	Comments
Outcomes	Usual care	Pharmaceutical care	Relative Risk effect (95% CI)	No. of participants (studies)	Certainty of evidence (GRADE)	Comments
Medication appropriateness (as measured by an implicit tool) From baseline to follow‐up Follow‐up: 0 to 6 months	Medication appropriateness (as measured by an implicit tool) across control groups ranged from ‐0.49 to 2.86	Medication appropriateness (as measured by an implicit tool) in the intervention groups was 4.76lower (0.33 to 9.20 lower)		517 (5 studies)	⊕⊝⊝⊝ very low ^a,b,c,d	MAI used as implicit tool in the pooled studies A sensitivity analysis showed that medication appropriateness (as measured by an implicit tool) in the intervention group was 0.50 lower (2.27 lower to 1.28 higher)^e Heterogeneity: I² = 57%, P = 0.10
Potentially inappropriate medications
The number of potentially inappropriate medications (PIMs) Follow‐up: 0 to 12 months	The number of PIMs (Standardised mean^§) across control groups ranged from 0.04 to 1.29	The number of PIMs (Standardised mean^§) in the intervention groups was 0.22lower (0.05 to 0.38 lower)		1832 (7 studies)	⊕⊝⊝⊝ very low^a,b,c	STOPP and Beers criteria used as explicit tools in the pooled studies
The proportion of patients with one or more potentially inappropriate medications (PIMs) Follow‐up: 0 to 12 months	421 per 1000	333 per 1000 (257 to 430)	RR 0.79 (0.61 to 1.02)	3079 (11 studies)	⊕⊝⊝⊝ very low^a,b,c	STOPP and Beers criteria used as explicit tools in the pooled studies A sensitivity analysis showed that the proportion of patients with one or more potentially inappropriate medications in the intervention group was lower (333 per 1000)^f Heterogeneity: I² = 75%, P = 0.24
Potential prescribing omissions
The number of potential prescribing omissions (PPOs) Follow‐up: 0 to 12 months	The number of PPOs (Standardised mean^§) across control groups ranged from 0.63 to 0.85	The number of PPOs (Standardised mean^§) in the intervention groups was 0.81 lower (0.64 to 0.98 lower)		569 (2 studies)	⊕⊕⊝⊝ low^a	START and ACOVE used as explicit tools in the pooled studies
The proportion of patients with one or more potential prescribing omissions (PPOs) Follow‐up: 0 to 24 months	387 per 1000	155 per 1000 (70 to 329)	RR 0.40 (0.18 to 0.85)	1310 (5 studies)	⊕⊝⊝⊝ very low^a,c	START and ACOVE used as explicit tools in the pooled studies
Hospital admissions Follow‐up: 0 to 12 months	Pharmaceutical care may make little or no difference in hospital admissions			4052 (12 studies)	⊕⊕⊝⊝ low^a
Quality of Life Follow‐up: 0 to 12 months	Pharmaceutical care may make little or no difference in quality of life			3211 (12 studies)	⊕⊕⊝⊝ low^a
GRADE Working Group grades of evidence High: This research provides a very good indication of the likely effect. The likelihood that the effect will be substantially different^‡ is low. Moderate: This research provides a good indication of the likely effect. The likelihood that the effect will be substantially different^‡ is moderate. Low: This research provides some indication of the likely effect. However, the likelihood that it will be substantially different^‡ is high. Very low: This research does not provide a reliable indication of the likely effect. The likelihood that the effect will be substantially different^‡ is very high. ^‡Substantially different = a large enough difference that it might affect a decision
ACOVE: Assessing Care of the Vulnerable Elderly, CI: confidence interval, MAI: Medication Appropriateness Index, PIMs: Potentially Inappropriate Medications, PPOs: Potential prescribing omissions, RR: risk ratio, STOPP: Screening Tool of Older People’s potentially inappropriate Prescriptions, START: Screening Tool to Alert to Right Treatment ^§ Standardised mean was used in cases where a range of tools were used to generate the pooled effect estimate. ^a We downgraded the evidence due to risk of bias. ^b We downgraded the evidence due to indirectness of the evidence. ^c We downgraded the evidence due to inconsistency in the results that could not be fully explained. ^d We downgraded the evidence due to imprecision. CIs were wide and/or crossed the line of no effect. ^e Two studies were excluded from the analysis because of a unit of analysis error (Crotty 2004a) and an outlying effect estimate with a high risk of bias (Spinewine 2007). ^f Two studies were excluded from the analysis because of a large effect size and high risk of bias (Spinewine 2007) and a small effect size (Gallagher 2011).

Summary of findings for the main comparison. Pharmaceutical care compared with usual care for older people receiving polypharmacy

Navigate to table in Review

Table 1. Medication Appropriateness Index

To assess the appropriateness of the drug, please answer the following questions and circle the applicable score.
1. Is there an indication for the drug? Comments:	1	2	3	9 DK
	Indicated		Not Indicated
2. Is the medication effective for the condition? Comments:	1	2	3	9 DK
	Effective		Ineffective
3. Is the dosage correct? Comments:	1	2	3	9 DK
	Correct		Incorrect
4. Are the directions correct? Comments:	1	2	3	9 DK
	Correct		Incorrect
5. Are the directions practical? Comments:	1	2	3	9 DK
	Practical		Impractical
6. Are there clinically significant drug‐drug interactions? Comments:	1	2	3	9 DK
	Insignificant		Significant
7. Are there clinically significant drug‐disease/condition interactions? Comments:	1	2	3	9 DK
	Insignificant		Significant
8. Is there unnecessary duplication with other drug(s)? Comments:	1	2	3	9 DK
	Necessary		Unnecessary
9. Is the duration of therapy acceptable? Comments:	1	2	3	9 DK
	Acceptable		Unacceptable
10. Is this drug the least expensive alternative compared with others of equal utility? Comments:	1	2	3	9 DK
	Least expensive		Most expensive
DK: Don't know

Table 1. Medication Appropriateness Index

Navigate to table in Review

Table 2. Updated Beers (2003) criteria for potentially inappropriate medication use in older adults: independent of diagnosis or condition

Drug	Concern	Severity rating (high or low)
Propoxyphene (Darvon) and combination products (Darvon with ASA, Darvon‐N and Darvocet‐N)	Offers few analgesic advantages over paracetamol (acetaminophen), yet is associated with the adverse effects of other narcotic drugs	Low
Indomethacin (Indocin and Indocin SR)	Of all available NSAIDs, this drug produces the most CNS adverse effects	High
Pentazocine (Talwin)	Narcotic analgesic that causes more CNS adverse effects, including confusion and hallucinations, more commonly than other narcotic drugs. Additionally, it is a mixed agonist and antagonist	High
Trimethobenzamide (Tigan)	One of the least effective antiemetic drugs, yet it can cause extrapyramidal adverse effects	High
Muscle relaxants and antispasmodics: methocarbamol (Robaxin), carisoprodol (Soma), chlorzoxazone (Paraflex), metaxalone (Skelaxin), cyclobenzaprine (Flexeril) and oxybutynin (Ditropan). Do not consider the extended‐release formulation of Ditropan XL	Most muscle relaxants and antispasmodic drugs are poorly tolerated by elderly patients because they cause anticholinergic adverse effects, sedation and weakness. Additionally, their effectiveness at doses tolerated by elderly patients is questionable	High
Flurazepam (Dalmane)	This benzodiazepine hypnotic has an extremely long half‐life in elderly patients (often days), producing prolonged sedation and increasing the incidence of falls and fracture. Medium‐ or short‐acting benzodiazepines are preferable	High
Amitriptyline (Elavil), chlordiazepoxide‐amitriptyline (Limbitrol) and perphenazine‐amitriptyline (Triavil)	Because of its strong anticholinergic and sedation properties, amitriptyline is rarely the antidepressant of choice for elderly patients	High
Doxepin (Sinequan)	Because of its strong anticholinergic and sedating properties, doxepin is rarely the antidepressant of choice for elderly patients	High
Meprobamate (Miltown and Equanil)	This is a highly addictive and sedating anxiolytic. Those using meprobamate for prolonged periods may become addicted and may need to be withdrawn slowly	High
Doses of short‐acting benzodiazepines: doses greater than lorazepam (Ativan) 3 mg; oxazepam (Serax) 60 mg; iprazolam (Xanax) 2 mg; temazepam (Restoril) 15 mg and triazolam (Halcion) 0.25 mg	Because of increased sensitivity to benzodiazepines in elderly patients, smaller doses may be effective and safer. Total daily doses should rarely exceed the suggested maximum	High
Long‐acting benzodiazepines: chlordiazepoxide (Librium), chlordiazepoxide‐amitriptyline (Limbitrol), clidinium‐chlordiazepoxide (Librax), diazepam (Valium), quazepam (Doral), halazepam (Paxipam) and chlorazepate (Tranxene)	These drugs have a long half‐life in elderly patients (often several days), producing prolonged sedation and increasing the risk of falls and fractures. Short‐ and intermediate‐acting benzodiazepines are preferred if a benzodiazepine is required	High
Disopyramide (Norpace and Norpace CR)	Of all antiarrhythmic drugs, this is the most potent negative inotrope and therefore may induce heart failure in elderly patients. It also has strong anticholinergic effects. Other antiarrhythmic drugs should be used as well	High
Digoxin (Lanoxin) (should not exceed 0.125 mg/d except when treating atrial arrhythmias)	Decreased renal clearance may lead to increased risk of toxic effects	Low
Short‐acting dipyridamole (Persantine). Do not consider the long‐acting dipyridamole (which has better properties than the short‐acting formulation in older adults) except with patients with artificial heart valves	May cause orthostatic hypotension	Low
Methyldopa (Aldomet) and methyldopa‐hydrochlorothiazide (Aldoril)	May cause bradycardia and exacerbate depression in elderly patients	High
Reserpine at doses > 0.25 mg	May induce depression, impotence, sedation and orthostatic hypotension	Low
Chlorpropamide (Diabinese)	It has a prolonged half‐life in elderly patients and could cause prolonged hypoglycaemia. Additionally, it is the only oral hypoglycaemic agent that causes SIADH	High
GI antispasmodic drugs: dicyclomine (Bentyl), hyoscyamine (Levsin and Levsinex), propantheline (Pro‐Banthine), belladonna alkaloids (Donnatal and others) and clidinium‐chlordiazepoxide (Librax)	GI antispasmodic drugs have potent anticholinergic effects and have uncertain effectiveness. These drugs should be avoided (especially for long‐term use)	High
Anticholinergics and antihistamines: chlorpheniramine (Chlor‐Trimeton), diphenhydramine (Benadryl), hydroxyzine (Vistaril and Atarax), cyproheptadine (Periactin), promethazine (Phenergan), tripelennamine, dexchlorpheniramine (Polaramine)	All non‐prescription and many prescription antihistamines may have potent anticholinergic properties. Non‐anticholinergic antihistamines are preferred in elderly patients for the treatment of allergic reactions	High
Diphenhydramine (Benadryl)	May cause confusion and sedation. Should not be used as a hypnotic, and when used to treat emergency allergic reactions, it should be used in the smallest possible dose	High
Ergot mesyloids (Hydergine) and cyclandelate (Cyclospasmol)	Have not been shown to be effective in the doses studied	Low
Ferrous sulphate > 325 mg/d	Doses > 325 mg/d do not dramatically increase the amount absorbed but greatly increase the incidence of constipation	Low
All barbiturates (except phenobarbital) except when used to control seizures	Are highly addictive and cause more adverse effects than most sedative or hypnotic drugs in elderly patients	High
Meperidine (Demerol)	Not an effective oral analgesic in doses commonly used. May cause confusion and has many disadvantages compared with other narcotic drugs	High
Ticlopidine (Ticlid)	Has been shown to be no better than aspirin in preventing clotting and may be considerably more toxic Safer, more effective alternatives exist	High
Ketorolac (Toradol)	Immediate and long‐term use should be avoided in older people, as a significant number have asymptomatic GI pathological conditions	High
Amphetamines and anorexic agents	These drugs have potential for causing dependence, hypertension, angina and myocardial infarction	High
Long‐term use of full‐dosage, longer half‐life, non–COX‐selective NSAIDs: naproxen (Naprosyn, Avaprox and Aleve), oxaprozin (Daypro) and piroxicam (Feldene)	Have the potential to produce GI bleeding, renal failure, hypertension and heart failure	High
Daily fluoxetine (Prozac)	Long half‐life of drug and risk of producing excessive CNS stimulation, sleep disturbances and increasing agitation. Safer alternatives are available	High
Long‐term use of stimulant laxatives: bisacodyl (Dulcolax), cascara sagrada and Neoloid except in the presence of opiate analgesic use	May exacerbate bowel dysfunction	High
Amiodarone (Cordarone)	Associated with QT interval problems and risk of provoking torsades de pointes. Lack of efficacy in older adults	High
Orphenadrine (Norflex)	Causes greater sedation and anticholinergic adverse effects than safer alternatives	High
Guanethidine (Ismelin)	May cause orthostatic hypotension. Safer alternatives are available	High
Guanadrel (Hylorel)	May cause orthostatic hypotension	High
Cyclandelate (Cyclospasmol)	Lack of efficacy	Low
Isoxsurpine (Vasodilan)	Lack of efficacy	Low
Nitrofurantoin (Macrodantin)	Potential for renal impairment. Safer alternatives are available	High
Doxazosin (Cardura)	Potential for hypotension, dry mouth and urinary problems	Low
Methyltestosterone (Android, Virilon and Testrad)	Potential for prostatic hyperplasia and cardiac problems	High
Thioridazine (Mellaril)	Greater potential for CNS and extrapyramidal adverse effects	High
Mesoridazine (Serentil)	CNS and extrapyramidal adverse effects	High
Short‐acting nifedipine (Procardia and Adalat)	Potential for hypotension and constipation	High
Clonidine (Catapres)	Potential for orthostatic hypotension and CNS adverse effects	Low
Mineral oil	Potential for aspiration and adverse effects. Safer alternatives are available	High
Cimetidine (Tagamet)	CNS adverse effects including confusion	Low
Ethacrynic acid (Edecrin)	Potential for hypertension and fluid imbalances. Safer alternatives are available	Low
Desiccated thyroid	Concerns about cardiac effects. Safer alternatives are available	High
Amphetamines (excluding methylphenidate hydrochloride and anorexic agents)	CNS stimulant adverse effects	High
Oestrogens only (oral)	Evidence of the carcinogenic (breast and endometrial cancer) potential of these agents and lack of cardioprotective effects in older women	Low
Source: Fick 2003. CNS: central nervous system; COX: cyclo‐oxygenase; CR: controlled release; GI: gastrointestinal; NSAID: non‐steroidal anti‐inflammatory drug; SIADH: syndrome of inappropriate antidiuretic hormone hypersecretion; SR: slow release.

Table 2. Updated Beers (2003) criteria for potentially inappropriate medication use in older adults: independent of diagnosis or condition

Navigate to table in Review

Table 3. Updated Beers (2003) criteria for potentially inappropriate medication use in older adults: considering diagnoses or conditions

Disease or condition	Drug	Concern	Severity rating (high or low)
Heart failure	Disopyramide (Norpace) and high‐sodium‐content drugs (sodium and sodium salts (alginate bicarbonate, biphosphate, citrate, phosphate, salicylate, and sulphate))	Negative inotropic effect. Potential to promote fluid retention and exacerbation of heart failure	High
Hypertension	Phenylpropanolamine hydrochloride (removed from the market in 2001), pseudoephedrine; diet pills and amphetamines	May produce elevation of blood pressure secondary to sympathomimetic activity	High
Gastric or duodenal ulcers	NSAIDs and aspirin (> 325 mg) (COXIBs excluded)	May exacerbate existing ulcers or produce new/additional ulcers	High
Seizures or epilepsy	Clozapine (Clozaril), chlorpromazine (Thorazine), thioridazine (Mellaril) and thiothixene (Navane)	May lower seizure thresholds	High
Blood clotting disorders or receiving anticoagulant therapy	Aspirin, NSAIDs, dipyridamole (Persantin), ticlopidine (Ticlid) and clopidogrel (Plavix)	May prolong clotting time and elevate INR values or inhibit platelet aggregation, resulting in increased potential for bleeding	High
Bladder outflow obstruction	Anticholinergics and antihistamines, gastrointestinal antispasmodics, muscle relaxants, oxybutynin (Ditropan), flavoxate (Urispas), anticholinergics, antidepressants, decongestants and tolterodine (Detrol)	May decrease urinary flow, leading to urinary retention	High
Stress incontinence	α‐Blockers (doxazosin, prazosin and terazosin), anticholinergics, tricyclic antidepressants (imipramine hydrochloride, doxepin hydrochloride and amitriptyline hydrochloride) and long‐acting benzodiazepines	May produce polyuria and worsening of incontinence	High
Arrhythmias	Tricyclic antidepressants (imipramine hydrochloride, doxepin hydrochloride and amitriptyline hydrochloride)	Concern due to proarrhythmic effects and ability to produce QT interval changes	High
Insomnia	Decongestants, theophylline (Theodur), methylphenidate (Ritalin), MAOIs and amphetamines	Concern due to CNS stimulant effects	High
Parkinson's disease	Metoclopramide (Reglan), conventional antipsychotics and tacrine (Cognex)	Concern due to their antidopaminergic/ cholinergic effects	High
Cognitive impairment	Barbiturates, anticholinergics, antispasmodics and muscle relaxants. CNS stimulants: dextroamphetamine (Adderall), methylphenidate (Ritalin), methamphetamine (Desoxyn) and pemolin	Concern due to CNS‐altering effects	High
Depression	Long‐term benzodiazepine use. Sympatholytic agents: methyldopa (Aldomet), reserpine and guanethidine (Ismelin)	May produce or exacerbate depression	High
Anorexia and malnutrition	CNS stimulants: dextroamphetamine (Adderall), methylphenidate (Ritalin), methamphetamine (Desoxyn), pemolin and fluoxetine (Prozac)	Concern due to appetite‐suppressing effects	High
Syncope or falls	Short‐ to intermediate‐acting benzodiazepine and tricyclic antidepressants (imipramine hydrochloride, doxepin hydrochloride and amitriptyline hydrochloride)	May produce ataxia, impaired psychomotor function, syncope and additional falls	High
SIADH/hyponatraemia	SSRIs: fluoxetine (Prozac), citalopram (Celexa), fluvoxamine (Luvox), paroxetine (Paxil) and sertraline (Zoloft)	May exacerbate or cause SIADH	Low
Seizure disorder	Bupropion (Wellbutrin)	May lower seizure threshold	High
Obesity	Olanzapine (Zyprexa)	May stimulate appetite and increase weight gain	Low
COPD	Long‐acting benzodiazepines: chlordiazepoxide (Librium), chlordiazepoxide‐amitriptyline (Limbitrol), clidinium‐chlordiazepoxide (Librax), diazepam (Valium), quazepam (Doral), halazepam (Paxipam) and chlorazepate (Tranxene). β‐Blockers: propranolol	CNS adverse effects. May induce respiratory depression. May exacerbate or cause respiratory depression	High
Chronic constipation	Calcium channel blockers, anticholinergics and tricyclic antidepressant (imipramine hydrochloride, doxepin hydrochloride and amitriptyline hydrochloride)	May exacerbate constipation	Low
Source: Fick 2003. COPD: chronic obstructive pulmonary disease; COXIB: cyclo‐oxygenase inhibitor; INR: international normalized ratio; MAOI: monoamine oxidase inhibitor; NSAID: non‐steroidal anti‐inflammatory drug; SIADH: syndrome of inappropriate antidiuretic hormone secretion; SSRIs: selective serotonin reuptake inhibitors.

Table 3. Updated Beers (2003) criteria for potentially inappropriate medication use in older adults: considering diagnoses or conditions

Navigate to table in Review

Table 4. Updated Beers (2012) criteria for potentially inappropriate medication usage in older adults: independent of diagnosis or condition

Organ System or Therapeutic Category or Drug	Rationale	Recommendation	Quality of Evidence	Strength of Recommendation
Anticholinergics (excludes TCAs)
First‐generation antihistamines (as single agent or as part of combination products) Brompheniramine Carbinoxamine Chlorpheniramine Clemastine Cyproheptadine Dexbrompheniramine Dexchlorpheniramine Diphenhydramine (oral) Doxylamine Hydroxyzine Promethazine Triprolidine	Highly anticholinergic; clearance reduced with advanced age, and tolerance develops when used as hypnotic; greater risk of confusion, dry mouth, constipation and other anticholinergic effects and toxicity Use of diphenhydramine in special situations such as short‐term treatment of severe allergic reaction may be appropriate	Avoid	Hydroxyzine and promethazine: high; all others: moderate	Strong
Antiparkinson agents Benztropine (oral) Trihexyphenidyl	Not recommended for prevention of extrapyramidal symptoms with antipsychotics; more effective agents available for treatment of Parkinson's disease	Avoid	Moderate	Strong
Antispasmodics Belladonna alkaloids Clidinium‐chlordiazepoxide Dicyclomine Hyoscyamine Propantheline Scopolamine	Highly anticholinergic, uncertain effectiveness	Avoid except in short‐term palliative care to decrease oral secretions	Moderate	Strong
Antithrombotics
Dipyridamole, oral short‐acting* (does not apply to extended‐release combination with aspirin)	May cause orthostatic hypotension; more effective alternatives available; intravenous form acceptable for use in cardiac stress testing	Avoid	Moderate	Strong
Ticlopidine*	Safer effective alternatives available	Avoid	Moderate	Strong
Anti‐infective
Nitrofurantoin	Potential for pulmonary toxicity; safer alternatives available; lack of efficacy in patients with CrCl < 60 mL/min due to inadequate drug concentration in the urine	Avoid for long‐term suppression; avoid in patients with CrCl < 60 mL/min	Moderate	Strong
Cardiovascular
Alpha₁‐blockers Doxazosin Prazosin Terazosin	High risk of orthostatic hypotension; not recommended as routine treatment for hypertension; alternative agents have superior risk/benefit profile	Avoid use as an antihypertensive	Moderate	Strong
Alpha‐agonists, central Clonidine Guanabenz* Guanfacine* Methyldopa* Reserpine (> 0.1 mg/d)*	High risk of adverse CNS effects; may cause bradycardia and orthostatic hypotension; not recommended as routine treatment for hypertension	Avoid clonidine as a first‐line antihypertensive Avoid others as listed	Low	Strong
Antiarrhythmic drugs (Class Ia, Ic, III) Amiodarone Dofetilide Dronedarone Flecainide Ibutilide Procainamide Propafenone Quinidine Sotalol	Data suggest that rate control yields better balance of benefits and harms than rhythm control for most older adults Amiodarone is associated with multiple toxicities, including thyroid disease, pulmonary disorders and QT interval prolongation	Avoid antiarrhythmic drugs as first‐line treatment of atrial fibrillation	High	Strong
Disopyramide*	Disopyramide is a potent negative inotrope and therefore may induce heart failure in older adults; strongly anticholinergic; other antiarrhythmic drugs preferred	Avoid	Low	Strong
Dronedarone	Worse outcomes have been reported in patients taking dronedarone who have permanent atrial fibrillation or heart failure. In general, rate control is preferred over rhythm control for atrial fibrillation	Avoid in patients with permanent atrial fibrillation or heart failure	Moderate	Strong
Digoxin > 0.125 mg/d	In heart failure, higher dosages are associated with no additional benefit and may increase risk of toxicity; slow renal clearance may lead to risk of toxic effects	Avoid	Moderate	Strong
Nifedipine, immediate release*	Potential for hypotension; risk of precipitating myocardial ischaemia	Avoid	High	Strong
Spironolactone > 25 mg/d	In heart failure, the risk of hyperkalaemia is higher in older adults, especially if taking > 25 mg/d or taking concomitant NSAID, angiotensin‐converting enzyme inhibitor, angiotensin receptor blocker or potassium supplement	Avoid in patients with heart failure or with a CrCl < 30 mL/min	Moderate	Strong
Central nervous system
Tertiary TCAs, alone or in combination: Amitriptyline Chlordiazepoxide‐amitriptyline Clomipramine Doxepin > 6 mg/d Imipramine Perphenazine‐amitriptyline Trimipramine	Highly anticholinergic, sedating and causing orthostatic hypotension; safety profile of low‐dose doxepin (≤ 6 mg/d) is comparable with that of placebo	Avoid	High	Strong
Antipsychotics, first (conventional) and second (atypical) generation (see AGS 2012 for full list)	Increased risk of cerebrovascular accident (stroke) and mortality in persons with dementia	Avoid use for behavioural problems of dementia unless non‐pharmacological options have failed and patient is threat to self or others	Moderate	Strong
Thioridazine Mesoridazine	Highly anticholinergic and risk of QT interval prolongation	Avoid	Moderate	Strong
Barbiturates Amobarbital* Butabarbital* Butalbital Mephobarbital* Pentobarbital* Phenobarbital Secobarbital*	High rate of physical dependence; tolerance to sleep benefits; risk of overdose at low dosages	Avoid	High	Strong
Benzodiazepines Short‐ and intermediate‐acting: Alprazolam Estazolam Lorazepam Oxazepam Temazepam Triazolam Long‐acting: Clorazepate Chlordiazepoxide Chlordiazepoxide‐amitriptyline Clidinium‐chlordiazepoxide Clonazepam Diazepam Flurazepam Quazepam	Older adults have increased sensitivity to benzodiazepines and slower metabolism of long‐acting agents. In general, all benzodiazepines increase risk of cognitive impairment, delirium, falls, fractures and motor vehicle accidents in older adults May be appropriate for seizure disorders, rapid eye movement sleep disorders, benzodiazepine withdrawal, ethanol withdrawal, severe generalized anxiety disorder, periprocedural anaesthesia and end‐of‐life care	Avoid benzodiazepines (any type) for treatment of insomnia, agitation or delirium	High	Strong
Chloral hydrate*	Tolerance occurs within 10 days, and risks outweigh benefits in light of overdose with doses only 3 times the recommended dose	Avoid	Low	Strong
Meprobamate	High rate of physical dependence; very sedating	Avoid	Moderate	Strong
Non‐benzodiazepine hypnotics Eszopiclone Zolpidem Zaleplon	Benzodiazepine‐receptor agonists that have adverse events similar to those of benzodiazepines in older adults (e.g. delirium, falls, fractures); minimal improvement in sleep latency and duration	Avoid long‐term use (> 90 days)	Moderate	Strong
Ergot mesylates* Isoxsuprine*	Lack of efficacy	Avoid	High	Strong
Endocrine
Androgens Methyltestosterone* Testosterone	Potential for cardiac problems and contraindicated in men with prostate cancer	Avoid unless indicated for moderate to severe hypogonadism	Moderate	Weak
Desiccated thyroid	Concerns about cardiac effects; safer alternatives available	Avoid	Low	Strong
Oestrogens with or without progestins	Evidence of carcinogenic potential (breast and endometrium); lack of cardioprotective effect and cognitive protection in older women Evidence that vaginal oestrogens for treatment of vaginal dryness are safe and effective in women with breast cancer, especially at dosages of estradiol < 25 μg twice weekly	Avoid oral and topical patch Topical vaginal cream: acceptable to use low‐dose intravaginal oestrogen for the management of dyspareunia, lower urinary tract infection and other vaginal symptoms	Oral and patch: high Topical: moderate	Oral and patch: strong Topical: weak
Growth hormone	Effect on body composition is small and is associated with oedema, arthralgia, carpal tunnel syndrome, gynaecomastia, impaired fasting glucose	Avoid, except as hormone replacement after pituitary gland removal	High	Strong
Insulin, sliding scale	Higher risk of hypoglycaemia without improvement in hyperglycaemia management regardless of care setting	Avoid	Moderate	Strong
Megestrol	Minimal effect on weight; increases risk of thrombotic events and possibly death in older adults	Avoid	Moderate	Strong
Sulphonylureas, long duration Chlorpropamide Glyburide	Chlorpropamide: prolonged half‐life in older adults; can cause prolonged hypoglycaemia; causes syndrome of inappropriate antidiuretic hormone secretion. Glyburide: greater risk of severe prolonged hypoglycaemia in older adults	Avoid	High	Strong
Gastrointestinal
Metoclopramide	Can cause extrapyramidal effects including tardive dyskinesia; risk may be even greater in frail older adults	Avoid, unless for gastroparesis	Moderate	Strong
Mineral oil, oral	Potential for aspiration and adverse effects; safer alternatives available	Avoid	Moderate	Strong
Trimethobenzamide	One of the least effective antiemetic drugs; can cause extrapyramidal adverse effects	Avoid	Moderate	Strong
Pain
Meperidine	Not an effective oral analgesic in dosages commonly used; may cause neurotoxicity; safer alternatives available	Avoid	High	Strong
Non–COX‐selective NSAIDs, oral Aspirin > 325 mg/d Diclofenac Diflunisal Etodolac Fenoprofen Ibuprofen Ketoprofen Meclofenamate Mefenamic acid Meloxicam Nabumetone Naproxen Oxaprozin Piroxicam Sulindac Tolmetin	Increase risk of GI bleeding and peptic ulcer disease in high‐risk groups, including those aged > 75 or taking oral or parenteral corticosteroids, anticoagulants or antiplatelet agents. Use of proton pump inhibitor or misoprostol reduces but does not eliminate risk. Upper GI ulcers, gross bleeding or perforation caused by NSAIDs occurs in approximately 1% of patients treated for 3 to 6 months and in approximately 2% to 4% of patients treated for 1 year. These trends continue with longer duration of use	Avoid long‐term use unless other alternatives are not effective and patient can take gastroprotective agent (proton pump inhibitor or misoprostol)	Moderate	Strong
Indomethacin Ketorolac, includes parenteral	Increase risk of GI bleeding and peptic ulcer disease in high‐risk groups (see above Non–COX‐selective NSAIDs) Of all the NSAIDs, indomethacin has the most adverse effects	Avoid	Indomethacin: moderate Ketorolac: high	Strong
Pentazocine*	Opioid analgesic that causes CNS adverse effects, including confusion and hallucinations, more commonly than other narcotic drugs; also a mixed agonist and antagonist; safer alternatives available	Avoid	Low	Strong
Skeletal muscle relaxants Carisoprodol Chlorzoxazone Cyclobenzaprine Metaxalone Methocarbamol Orphenadrine	Most muscle relaxants are poorly tolerated by older adults because of anticholinergic adverse effects, sedation, risk of fracture; effectiveness at dosages tolerated by older adults is questionable	Avoid	Moderate	Strong
Source: AGS 2012. CNS = central nervous system; COX = cyclo‐oxygenase; CrCl = creatinine clearance; GI = gastrointestinal; NSAID = non‐steroidal anti‐inflammatory drug; TCA = tricyclic antidepressant. *Infrequently used drugs.

Table 4. Updated Beers (2012) criteria for potentially inappropriate medication usage in older adults: independent of diagnosis or condition

Navigate to table in Review

Table 5. Updated Beers (2012) criteria for potentially inappropriate medication usage in older adults due to drug–disease or drug–syndrome interactions that may exacerbate the disease or syndrome

Disease or syndrome	Drug	Rationale	Recommendation	Quality of evidence	Strength of recommendation
Cardiovascular
Heart failure	NSAIDs and COX‐2 inhibitors Non‐dihydropyridine CCBs (avoid only for systolic heart failure) Diltiazem Verapamil Pioglitazone, rosiglitazone Cilostazol Dronedarone	Potential to promote fluid retention and exacerbate heart failure	Avoid	NSAIDs: moderate CCBs: moderate Thiazolidinediones (glitazones): high Cilostazol: low Dronedarone: moderate	Strong
Syncope	AChEIs Peripheral alpha‐blockers Doxazosin Prazosin Terazosin Tertiary TCAs Chlorpromazine, thioridazine and olanzapine	Increase risk of orthostatic hypotension or bradycardia	Avoid	Alpha‐blockers: high TCAs, AChEIs and antipsychotics: moderate	AChEIs and TCAs: strong Alpha‐blockers and antipsychotics: weak
Central nervous system
Chronic seizures or epilepsy	Bupropion Chlorpromazine Clozapine Maprotiline Olanzapine Thioridazine Thiothixene Tramadol	Lower seizure threshold; may be acceptable in patients with well‐controlled seizures in whom alternative agents have not been effective	Avoid	Moderate	Strong
Delirium	All TCAs Anticholinergics (see AGS 2012 for full list) Benzodiazepines Chlorpromazine Corticosteroids H₂‐receptor antagonist Meperidine Sedative‐hypnotics Thioridazine	Avoid in older adults with or at high risk of delirium because of inducing or worsening delirium in older adults; if discontinued drugs used long‐term, taper to avoid withdrawal symptoms	Avoid	Moderate	Strong
Dementia and cognitive impairment	Anticholinergics (see AGS 2012 for full list) Benzodiazepines H₂‐receptor antagonists Zolpidem Antipsychotics, long‐term and as‐needed use	Avoid because of adverse CNS effects Avoid antipsychotics for behavioural problems of dementia unless non‐pharmacological options have failed and patient is a threat to himself or others. Antipsychotics are associated with increased risk of cerebrovascular accident (stroke) and mortality in persons with dementia	Avoid	High	Strong
History of falls or fractures	Anticonvulsants Antipsychotics Benzodiazepines Non‐benzodiazepine hypnotics Eszopiclone Zaleplon Zolpidem TCAs and selective serotonin reuptake inhibitors	Ability to produce ataxia, impaired psychomotor function, syncope and additional falls; shorter‐acting benzodiazepines are not safer than long‐acting ones	Avoid unless safer alternatives are not available; avoid anticonvulsants except for seizure disorders	High	Strong
Insomnia	Oral decongestants Pseudoephedrine Phenylephrine Stimulants Amphetamine Methylphenidate Pemoline Theobromines Theophylline Caffeine	CNS stimulant effects	Avoid	Moderate	Strong
Parkinson's disease	All antipsychotics (see AGS 2012 for full list, except for quetiapine and clozapine) Antiemetics Metoclopramide Prochlorperazine Promethazine	Dopamine receptor antagonists with potential to worsen parkinsonian symptoms Quetiapine and clozapine appear to be less likely to precipitate worsening of Parkinson's disease	Avoid	Moderate	Strong
Gastrointestinal
Chronic constipation	Oral antimuscarinics for urinary incontinence Darifenacin Fesoterodine Oxybutynin (oral) Solifenacin Tolterodine Trospium Non‐dihydropyridine CCB Diltiazem Verapamil First‐generation antihistamines as single agent or part of combination products Brompheniramine (various) Carbinoxamine Chlorpheniramine Clemastine (various) Cyproheptadine Dexbrompheniramine Dexchlorpheniramine (various) Diphenhydramine Doxylamine Hydroxyzine Promethazine Triprolidine Anticholinergics and antispasmodics (see AGS 2012 for full list of drugs with strong anticholinergic properties) Antipsychotics Belladonna alkaloids Clidinium‐chlordiazepoxide Dicyclomine Hyoscyamine Propantheline Scopolamine Tertiary TCAs (amitriptyline, clomipramine, doxepin, imipramine and trimipramine)	Can worsen constipation; agents for urinary incontinence: Antimuscarinics overall differ in incidence of constipation; response variable; consider alternative agent if constipation develops	Avoid unless no other alternatives	For urinary incontinence: high All others: moderate to low	Weak
History of gastric or duodenal ulcers	Aspirin (> 325 mg/d) Non–COX‐2–selective NSAIDs	May exacerbate existing ulcers or cause new or additional ulcers	Avoid unless other alternatives are not effective and patient can take gastroprotective agent (proton pump inhibitor or misoprostol)	Moderate	Strong
Kidney and urinary tract
Chronic kidney disease Stages IV and V	NSAIDs Triamterene (alone or in combination)	May increase risk of kidney injury	Avoid	NSAIDs: moderate Triamterene: low	NSAIDs: strong Triamterene: weak
Urinary incontinence (all types) in women	Oestrogen oral and transdermal (excludes intravaginal oestrogen)	Aggravate incontinence	Avoid in women	High	Strong
Lower urinary tract symptoms, benign prostatic hyperplasia	Inhaled anticholinergic agents Strongly anticholinergic drugs, except antimuscarinics for urinary incontinence (see AGS 2012 for complete list)	May decrease urinary flow and cause urinary retention	Avoid in men	Moderate	Inhaled agents: strong All others: weak
Stress or mixed urinary incontinence	Alpha‐blockers Doxazosin Prazosin Terazosin	Aggravate incontinence	Avoid in women	Moderate	Strong
Source: AGS 2012. CCB = calcium channel blocker; AChEI = acetylcholinesterase inhibitor; CNS = central nervous system; COX = cyclo‐oxygenase; NSAID = non‐steroidal anti‐inflammatory drug; TCA = tricyclic antidepressant.

Table 5. Updated Beers (2012) criteria for potentially inappropriate medication usage in older adults due to drug–disease or drug–syndrome interactions that may exacerbate the disease or syndrome

Navigate to table in Review

Table 6. Updated Beers (2012) criteria for potentially inappropriate medications to be used with caution in older adults

Drug	Rationale	Recommendation	Quality of evidence	Strength of recommendation
Aspirin for primary prevention of cardiac events	Lack of evidence of benefit versus risk in individuals aged ≥ 80	Use with caution in adults aged ≥ 80	Low	Weak
Dabigatran	Greater risk of bleeding than with warfarin in adults aged ≥ 75; lack of evidence of efficacy and safety in individuals with CrCl < 30 mL/min	Use with caution in adults aged ≥ 75 or if CrCl < 30 mL/min	Moderate	Weak
Prasugrel	Greater risk of bleeding in older adults; risk may be offset by benefit in highest‐risk older adults (e.g. with prior myocardial infarction or diabetes mellitus)	Use with caution in adults aged ≥ 75	Moderate	Weak
Antipsychotics Carbamazepine Carboplatin Cisplatin Mirtazapine Serotonin–norepinephrine reuptake inhibitor Selective serotonin reuptake inhibitor Tricyclic antidepressants Vincristine	May exacerbate or cause syndrome of inappropriate antidiuretic hormone secretion or hyponatraemia; need to monitor sodium level closely when starting or changing dosages in older adults because of increased risk	Use with caution	Moderate	Strong
Vasodilators	May exacerbate episodes of syncope in individuals with history of syncope
Source: AGS 2012. CrCl = creatinine clearance.

Table 6. Updated Beers (2012) criteria for potentially inappropriate medications to be used with caution in older adults

Navigate to table in Review

Comparison 1. Postintervention analysis

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Medication appropriateness (as measured by an implicit tool) Show forest plot	5	517	Mean Difference (IV, Random, 95% CI)	‐4.76 [‐9.20, ‐0.33]

2 Medication appropriateness (as measured by an implicit tool) (excl Crotty 2004a) Show forest plot	4	446	Mean Difference (IV, Random, 95% CI)	‐5.16 [‐11.04, 0.72]

3 Medication appropriateness (as measured by an implicit tool) (excl Crotty 2004a and Spinewine 2007) Show forest plot	3	260	Mean Difference (IV, Random, 95% CI)	‐0.50 [‐2.27, 1.28]

4 The number of potentially inappropriate medications Show forest plot	7	1832	Std. Mean Difference (IV, Random, 95% CI)	‐0.22 [‐0.38, ‐0.05]

5 The proportion of patients with one or more potentially inappropriate medications Show forest plot	11	3079	Risk Ratio (M‐H, Random, 95% CI)	0.79 [0.61, 1.02]

6 The proportion of patients with one or more potentially inappropriate medications (excl Spinewine 2007) Show forest plot	10	2893	Risk Ratio (M‐H, Random, 95% CI)	0.79 [0.61, 1.02]

7 The proportion of patients with one or more potentially inappropriate medications (excl Spinewine 2007 and Gallagher 2011) Show forest plot	9	2535	Risk Ratio (M‐H, Random, 95% CI)	0.88 [0.72, 1.09]

8 The number of potential prescribing omissions Show forest plot	2	569	Std. Mean Difference (IV, Random, 95% CI)	‐0.81 [‐0.98, ‐0.64]

9 The proportion of patients with one or more potential prescribing omissions Show forest plot	5	1310	Risk Ratio (M‐H, Random, 95% CI)	0.40 [0.18, 0.85]

Comparison 1. Postintervention analysis

Navigate to table in Review

Cochrane Review language

Website language

Abstract

Background

Objectives

Search methods

Selection criteria

Data collection and analysis

Main results

Authors' conclusions

PICOs

PICOs

Population

Intervention

Comparison

Outcome

Plain language summary

A review of the ways that healthcare professionals can improve the use of suitable medicines for older people

Visual summary

Authors' conclusions

Implications for practice

Implications for research

Summary of findings

Background

Description of the condition

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

Types of interventions

Types of outcome measures

Primary outcomes

Secondary outcomes

Search methods for identification of studies

Databases

Trial registries

Searching other resources

Data collection and analysis

Selection of studies

Data extraction and management

Assessment of risk of bias in included studies

Measures of treatment effect

Unit of analysis issues

Dealing with missing data

Assessment of reporting biases

Data synthesis, subgroup analysis and investigation of heterogeneity

Sensitivity analysis

'Summary of findings' table

Results

Description of studies

Results of the search

Included studies

Study design

Settings

Participants

Interventions

Outcomes

Excluded studies

Studies awaiting classification

Ongoing studies

Risk of bias in included studies

Allocation

Blinding

Incomplete outcome data

Selective reporting

Similarity of baseline characteristics

Other potential sources of bias

Effects of interventions

Primary outcome results

Medication appropriateness (as measured by an implicit tool)

Potentially inappropriate medications (PIMs) (including the number of potentially inappropriate medications and the proportion of patients with one or more PIMs)

Potential prescribing omissions (PPOs) (including the number of potential prescribing omissions and the proportion of patients with one or more PPOs)

Hospital admissions

Secondary outcome results

Medication‐related problems (e.g. adverse drug reactions (ADRs), drug‐drug interactions (DDIs))

Adherence to medication