Medication review in hospitalised patients to reduce morbidity and mortality
Abstract
Background
Pharmacotherapy in the elderly population is complicated by several factors that increase the risk of drug‐related harms and less favourable effectiveness. The concept of medication review is a key element in improving the quality of prescribing and in preventing adverse drug events. Although there is no generally accepted definition of medication review, it can be broadly defined as a systematic assessment of pharmacotherapy for an individual patient that aims to optimise patient medication by providing a recommendation or by making a direct change. Medication review performed in adult hospitalised patients may lead to better patient outcomes.
Objectives
We examined whether delivery of a medication review by a physician, pharmacist or other healthcare professional leads to improvement in health outcomes of hospitalised adult patients compared with standard care.
Search methods
We searched the Specialised Register of the Cochrane Effective Practice and Organisation of Care (EPOC) Group; the Cochrane Central Register of Controlled Trials (CENTRAL); MEDLINE; EMBASE; and the Cumulative Index to Nursing and Allied Health Literature (CINAHL) to November 2014, as well as International Pharmaceutical Abstracts and Web of Science to May 2015. In addition, we searched reference lists of included trials and relevant reviews. We searched trials registries and contacted experts to identify additional published and unpublished trials. We applied no language restrictions.
Selection criteria
We included randomised controlled trials (RCTs) of medication review in hospitalised adult patients. We excluded trials of outclinic and paediatric patients. Our primary outcome was all‐cause mortality, and secondary outcomes included hospital readmissions, emergency department contacts and adverse drug events.
Data collection and analysis
Two review authors independently included trials, extracted data and assessed trials for risk of bias. We contacted trial authors for clarification of data and for additional unpublished data. We calculated risk ratios for dichotomous data and mean differences for continuous data (with 95% confidence intervals (CIs)). The GRADE (Grades of Recommendation, Assessment, Development and Evaluation) approach was used to assess the overall certainty of evidence for the most important outcomes.
Main results
We identified 6600 references (4647 references in our initial review) and included 10 trials (3575 participants). Follow‐up ranged from 30 days to one year. Nine trials provided mortality data (3218 participants, 466 events), with a risk ratio of 1.02 (95% CI 0.87 to 1.19) (low‐certainty evidence). Seven trials provided hospital readmission data (2843 participants, 1043 events) with a risk ratio of 0.95 (95% CI 0.87 to 1.04) (high‐certainty evidence). Four trials provided emergency department contact data (1442 participants, 244 events) with a risk ratio of 0.73 (95% CI 0.52 to 1.03) (low‐certainty evidence). The estimated reduction in emergency department contacts of 27% (with a CI ranging from 48% reduction to 3% increase in contacts) corresponds to a number needed to treat for an additional beneficial outcome of 37 for a low‐risk population and 12 for a high‐risk population over one year. Subgroup and sensitivity analyses did not significantly alter our results.
Authors' conclusions
We found no evidence that medication review reduces mortality or hospital readmissions, although we did find evidence that medication review may reduce emergency department contacts. However, because of short follow‐up ranging from 30 days to one year, important treatment effects may have been overlooked. High‐quality trials with long‐term follow‐up (i.e. at least up to a year) are needed to provide more definitive evidence for the effect of medication review on clinically important outcomes such as mortality, readmissions and emergency department contacts, and on outcomes such as adverse events. Therefore, if used in clinical practice, medication reviews should be undertaken as part of a clinical trial with long‐term follow‐up.
PICOs
Plain language summary
Reassessment of drugs given to hospitalised adult patients to improve patients’ health
Review question
This updated Cochrane systematic review studies the evidence for performing in‐hospital medication review (defined as a systematic reassessment by a healthcare professional of an individual patients's medication with suggestions for improvement). We aimed to assess whether medication review may improve the health of adult patients.
Background
Elderly patients are often prescribed several drugs despite a generally higher risk of adverse events and sometimes lesser treatment effectiveness in this population.
Search date
To find relevant trials, we searched electronic medical literature databases up to May 2015.
Study characteristics
We included 10 randomised controlled trials with a total of 3575 participants.
Key results
We found that medication review does not seem to prevent death and hospital readmissions, but that it might reduce emergency department contacts.
Certainty of the evidence
Our confidence in results across studies ranged from low to high. We found no evidence that medication review in hospitalised patients makes a difference towards preventing mortality (low‐certainty evidence) or hospital readmissions (high‐certainty evidence), but we found that medication review may have a preventive effect on reducing the number of emergency department contacts (low‐certainty evidence). In the included trials, participants were followed for a short time (ranging from 30 days to one year). Therefore, important long‐term treatment effects may have been overlooked. We suggest that further research with long‐term patient follow‐up and examination of specific methods of medication review should be undertaken before this intervention is implemented in clinical practice.
Authors' conclusions
Summary of findings
| Medication review compared with standard care for hospitalised adult patients | ||||||
| Patient or population: hospitalised adult patients Intervention: medication review Comparison: standard care | ||||||
| Outcomes | Illustrative comparative risks* (95% CI) | Relative effect | Number of participants | Certainty of the evidence | Comments | |
| Assumed risk | Corresponding risk | |||||
| Standard care | Medication review | |||||
| Mortality (all‐cause) 1 year | Low‐risk population | RR 1.02 (0.87 to 1.19) | 3218 | ⊕⊕⊝⊝ Lowb,c | NA | |
| 200 per 1000a | 204 per 1000 | |||||
| High‐risk population | ||||||
| 400 per 1000a | 408 per 1000 | |||||
| Hospital readmission (all‐cause) 1 year | Low‐risk population | RR 0.95 (0.87 to 1.04) | 2843 (7 trials) | ⊕⊕⊕⊕ High | NA | |
| 300 per 1000a | 285 per 1000 | |||||
| High‐risk population | ||||||
| 600 per 1000a | 570 per 1000 | |||||
| Hospital emergency department contacts (all‐cause) 1 year | Low‐risk population | RR 0.73 (0.52 to 1.03) | 1442 | ⊕⊕⊝⊝ | Equal to number needed to treat of 12 for the high‐risk population and 37 for the low‐risk population | |
| 100 per 1000a | 73 per 1000 | |||||
| High‐risk population | ||||||
| 300 per 1000a | 219 per 1000 | |||||
| *The basis for the assumed risk (e.g. median control group risk across studies) is provided in footnotes. The corresponding risk (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). | ||||||
| GRADE Working Group grades of evidence | ||||||
| aRisk for the high‐risk population of the control group was based on data from Gillespie 2009, one of the 3 trials with 12 months of follow‐up (all outcomes) and with greatest risk in the control group. For the low‐risk population, the 12 months of follow‐up for Scullin 2007 (mortality) and the 3 months of follow‐up for Lisby 2010 (hospital emergency department contacts) were extrapolated to determine 12‐month risk. All trials had almost similar control group risks for hospital readmissions, and the low‐risk group was based on half the risk of 12 months of follow‐up for Gillespie 2009 bSubgroup analysis comparing 'high'‐ and 'low'‐risk populations revealed a small trend toward increased mortality in the low‐risk group compared with the high‐risk group (P value = 0.08) (downgraded 1 category) cFollow‐up ranged from 3 to 12 months for mortality. Short follow‐up may be inadequate to detect the effect on changes in prophylactic medication (indirectness of evidence) (downgraded 1 category) d'Risk of bias' assessments determined that 3 of 4 trials in the analysis (Gillespie 2009; Lisby 2010; Lisby 2015) had overall 'high risk' of bias (downgraded 1 category) eThe confidence interval overlapped 1 (downgraded 1 category) | ||||||
Background
Evidence links polypharmacy (defined as the use of many drugs) to increased risk of preventable interactions and adverse drug events (e.g. falls) (Bourgeois 2010; Hallas 1996; Obreli‐Neto 2012; Rothschild 2000; Ziere 2006), poorer drug adherence (Pasina 2014), use of inappropriate medications (Beers 1997; Hanlon 2004), a greater economic burden (Classen 1997), emergency department visits and hospital admissions (Kongkaew 2008; Schneeweiss 2002; Zed 2008), drug‐related deaths and overall mortality (Ebbesen 2001; Gnjidic 2012). Thus, it is critical in the patient with multi‐morbidity to discern between appropriate polypharmacy (e.g. as often the case treating conditions such as hypertension, diabetes and chronic pain) and inappropriate polypharmacy leading to unfavourable health and economic consequences (Aronson 2006; Hajjar 2007; Page 2010; Routledge 2004; Spinewine 2007b). The existence and recognition of inappropriate polypharmacy are of particular concern for the elderly population, for whom age‐related physiological changes, a greater degree of frailty and multiple coexisting conditions have been associated with increased risk of adverse drug events (ElDesoky 2007; Mangoni 2004). Additionally, adherence to and efficacy of drug treatment are generally reduced in elderly patients (Hughes 2004; Zulman 2011). The problem of inappropriate pharmacotherapy is expected to grow in the future as new drugs are introduced, as new uses for old drugs are found and as individuals in most parts of the world live longer and have increased risk of chronic medical conditions (CDC 2011; Christensen 2009; European Communities 2006; Pefoyo 2015).
Substantial efforts have been made to characterise and improve the appropriateness of prescribing for the elderly (Patterson 2012; Spinewine 2007b). Medication review constitutes such an attempt to improve the quality of prescribing and to prevent adverse drug events. There is no generally accepted definition of medication review, but it can be defined as a systematic assessment of the pharmacotherapy of an individual patient that aims to evaluate and optimise medication by providing a recommendation or by making a direct change. Medication review involves evaluating the therapeutic efficacy and harms of each drug in relation to the individual patient and conditions being treated. Other issues, such as adherence, interactions, biochemical monitoring and patient preferences and understanding of the condition, should also be considered and addressed when appropriate (Zermansky 2001). It is also important to include medication reconciliation (i.e. identifying the most accurate list of medications a patient is taking and using that list to provide correct pharmacotherapy), especially during transitions in care (Joint Commission 2012; Rogers 2006; Steurbaut 2010). To aid the process of reviewing patient medication, several criteria have been formulated to identify potentially inappropriate medications, especially for elderly people (Beers 1991; Beers 1997; Fick 2003; Gallagher 2008a; Hanlon 1992; Holt 2010; Laroche 2007a; McLeod 1997; Naugler 2000; O'Mahony 2015; Samsa 1994). However, the applicability and effects in clinical practice for these various measures remain uncertain (Bregnhøj 2009; Gallagher 2008b; Lozano‐Montoya 2015; Lund 2010; Ryan 2009; Spinewine 2007b).
Randomised trials of medication review have been summarised in recent systematic reviews (Holland 2008; Nkansah 2010; Royal 2006). The systematic reviews investigating medication review most often included trials with elderly people in primary care and failed to show effects on morbidity or mortality. Trials often involved pharmacist‐led medication reviews that ranged from 'hands‐on' clinical evaluation of hospital inpatient medication to informational approaches to physicians in outpatient clinics or primary care (Holland 2008; Nkansah 2010; Royal 2006). It is important to note that some pharmacist‐led medication reviews may be restricted because they are not directly linked to changes in clinical care (Spinewine 2007b). Physicians do not always implement pharmacists' suggestions (Chen 2007; Mannheimer 2006; Spinewine 2007b), and older patients may be reluctant to accept pharmacists' suggestions (Salter 2007) or may prefer to have their medications reviewed by a physician (Jones 1997). However, some evidence indicates that inpatient medication reviews by pharmacists in close contact with physicians might lead to fewer readmissions and lower morbidity (Gillespie 2009). Hospitalised patients likely represent a more frail patient group compared with primary care patients (Laroche 2007b); therefore, we investigated whether medication reviews affect hard clinical endpoints in hospitalised patients. This is an update of a previous meta‐analysis on this subject (Christensen 2013).
Description of the condition
Inappropriate pharmacotherapy is a major cause of patient morbidity and mortality. Inappropriate pharmacotherapy includes situations where medicines are prescribed without correct indication or dosage, in unfavorable combination with certain patient conditions, or combined with other interacting medicines that may increase the risk of treatment failure or adverse effects. Also included in the term 'inappropriate pharmacotherapy' is the presence of unacceptable adverse effects, lack of necessary biochemical monitoring of pharmacotherapy and poor adherence to pharmacotherapy, as well as underprescribing (i.e. not prescribing despite indication for pharmacotherapy).
Description of the intervention
Any medication review of a patient's list of medicines delivered by a healthcare professional with the aim of improving the pharmacotherapy (i.e. optimising effectiveness, minimising harms and/or costs of the prescribed medication).
How the intervention might work
More appropriate prescribing could improve effectiveness, reduce adverse events and improve adherence to, and thereby appropriateness of, drug therapy (i.e. ensure that treatment is properly indicated and monitored, and that the individual patient receives the right drug and dosage), possibly leading to reduced morbidity and mortality.
Why it is important to do this review
Medication reviews are performed in many parts of the world, but it is unclear whether medication reviews for hospitalised adult patients reduce patient morbidity and mortality. In addition, the best method for medication review is at present unknown. Through analysis of collective scientific evidence from randomised controlled trials (RCTs), we will clarify whether medication review can reduce mortality, hospital readmissions, emergency department contacts or adverse drug events among patients. We will also examine whether some methods of medication review are more effective than others. The results of this systematic review could encourage optimisation of current practices in this complex and important area. In addition, future research could be pointed in a more favourable direction.
Objectives
We examined whether delivery of a medication review by a physician, pharmacist or other healthcare professional leads to greater improvement in health outcomes of hospitalised adult patients compared with standard care.
Methods
Criteria for considering studies for this review
Types of studies
We included randomised controlled trials (RCTs) in any language, published or unpublished, with randomisation on an individual level or on an aggregated level (i.e. cluster‐RCTs).
Types of participants
We included hospitalised patients (i.e. patients admitted to a hospital).
We excluded outpatients and patients seen in the emergency department but not admitted to a hospital, as well as patients admitted to a paediatric department.
Types of interventions
We included any medication review of a patient's medicines delivered by a healthcare professional with the aim of improving pharmacotherapy for the patient (i.e. optimising the balance of effectiveness, harms and costs of the prescribed medication). We defined medication review as any systematic assessment of the pharmacotherapy of an individual patient that aims to evaluate and optimise patient medication by providing a recommendation or by making a direct change in prescriptions. We included trials comparing medication review with usual care or comparing two or more types of medication review.
We excluded:
-
trials aimed solely at increasing the patient's knowledge about current medication, improving adherence or reducing costs;
-
trials in which the results of medication review were to be implemented after discharge (e.g. letter to patient's general practitioner); and
-
trials reviewing only portions of a patient's medication related to a specific condition or to a single class of drugs (e.g. dealing only with diabetes or heart failure medication).
Types of outcome measures
Primary outcomes
-
Mortality (all‐cause).
Secondary outcomes
-
Hospital readmission (all‐cause).
-
Hospital readmission (due to adverse drug events).
-
Hospital emergency department contacts (all‐cause).
-
Hospital emergency department contacts (due to adverse drug events).
-
Mortality (due to adverse drug events).
-
Adverse drug events.
We included any trial that reported data on either primary or secondary outcomes.
Search methods for identification of studies
Electronic searches
We searched the following electronic databases for trials.
-
The Cochrane Effective Practice and Organisation of Care (EPOC) Group Specialised Register, November 2014.
-
CENTRAL (The Cochrane Library, November 2014).
-
MEDLINE, 1946 to November 2014, In‐Process & Other Non‐indexed Citations, Ovid.
-
EMBASE, 1980 to November 2014, Ovid SP.
-
CINAHL (Cumulative Index to Nursing and Allied Health Literature), 1980 to November 2014, EbscoHost.
-
International Pharmaceutical Abstracts, 1970 to May 2015, Ovid.
The search strategies (Appendix 1) were developed for Ovid MEDLINE and were adapted for the other databases. We used the Cochrane RCT Sensitivity/Precision‐Maximizing Filter to limit our search to RCTs (Lefebvre 2011).
Searching other resources
We searched the reference lists of all included trials and relevant review articles for additional trials. We searched MEDLINE (PubMed, May 2015) for relevant papers by authors (first and last) of included trials, and Web of Science (ISI Web of Knowledge, May 2015) for papers that cited any of the included trials. We contacted content experts in the field and corresponding authors of included trials to identify additional trials.
Unpublished trials
To identify conference abstracts of unpublished trials, we searched EMBASE and International Pharmaceutical Abstracts as described above.
In addition, we searched the following clinical trial registries (May 2015) to identify unpublished and ongoing trials.
-
ClinicalTrials.gov, US National Institutes of Health (NIH), www.clinicaltrials.gov.
-
Current Controlled Trials, www.controlled‐trials.com.
-
International Clinical Trials Registry Platform (ICTRP), World Health Organization (WHO), www.who.int/trialsearch.
Data collection and analysis
Selection of studies
We (MC, AL) independently selected all trials for inclusion in two rounds. First, we screened titles and abstracts for potentially includable articles. Then we screened the full text of all potential articles for inclusion. We resolved disagreements by discussion.
Data extraction and management
We (MC, AL) independently and unblinded extracted data for all included trials . We resolved disagreements by discussion.
Data included:
-
study characteristics: author name, publication year, journal name, methods of randomisation;
-
participants: number of participants, country, age, gender, type of department, morbidities, medication history, inclusion and exclusion criteria;
-
intervention: description of medication review, profession of reviewer (pharmacist, physician, other), explanation of how medication could be changed (recommendation by letter to patient's attending physician, meeting between pharmacist and physician, assessment by physician with direct change of prescription);
-
control: any co‐interventions that could influence the change in prescription;
-
outcome: outcome assessor, timing of outcomes; and
-
other characteristics: funding source.
Assessment of risk of bias in included studies
We (MC, AL) independently and unblinded assessed each trial and outcome for risk of bias using the 'Risk of bias assessment' of The Cochrane Collaboration (Higgins 2011). In addition, we evaluated contamination bias (EPOC 2015) and assessed the following domains: random sequence generation, allocation concealment, blinding of participants and personnel, blinding of outcome assessment, incomplete outcome data, selective reporting and other biases. We resolved disagreements by discussion.
Measures of treatment effect
For dichotomous data, we used risk ratios (RRs), and for continuous data, we used mean differences (MDs).
Dealing with missing data
We contacted authors of all included trials by email requesting missing data. One study author provided us with raw data (Gillespie 2009), and one provided tabulated data for two trials (Lisby 2010; Lisby 2015).
Assessment of heterogeneity
We assessed statistical heterogeneity by using I2.
Assessment of reporting biases
We assessed publication bias by using a funnel plot for our primary outcome (all‐cause mortality).
Data synthesis
We analysed all data by performing intention‐to‐treat analysis using available case analysis. In some trial reports, patients who died in‐hospital were excluded from the analysed population, but we retained these patients in our analysis. With Review Manager 5 (RevMan 2012), we calculated pooled RRs and estimated 95% confidence intervals (CIs) by using the random‐effects model with the Mantel‐Haenszel method for dichotomous data. In our original review, we used a fixed‐effect model, but because of large clinical heterogeneity in settings, patient populations and methodology of medication reviews in identified trials, we used a random‐effects model for this update. We calculated absolute risk reduction and number needed to treat for an additional beneficial outcome for outcomes with a clinically significant effect for low‐risk and high‐risk populations (see summary of findings Table for the main comparison). For continuous data, we calculated pooled MDs and estimated 95% CIs using the random‐effects model with the inverse variance method.
We used the GRADE (Grades of Recommendation, Assessment, Development and Evaluation) approach to assess the overall certainty of evidence (Guyatt 2008). We constructed a summary of findings Table for the main comparison for mortality (all‐cause), hospital readmissions (all‐cause) and hospital emergency department contacts (all‐cause), as these were the most reliable and patient‐relevant outcomes.
Subgroup analysis and investigation of heterogeneity
We planned to explore our findings by performing the following prespecified subgroup analyses.
-
Trials including only patients with high risk of medication errors and adverse drug events (study inclusion and exclusion criteria defined patient population as a high‐risk population (e.g. elderly patients, patients with multiple co‐medications)).
-
Trials in which the medication review was performed by a person or team with the capability to change the patient's medication directly (as opposed to a medication review carried out by healthcare professionals who were not allowed to change the patient's medications, but who recommended changes to an in‐hospital tending physician).
-
Trials in which the medication review was done using a validated method (e.g. Beers’ criteria (Beers 1997), START/STOPP criteria (Gallagher 2008a)).
To avoid multiplicity issues, we restricted these analyses to the dichotomous outcomes of mortality (all‐cause), hospital readmissions (all‐cause) and hospital emergency department contacts (all‐cause).
Originally we planned to investigate the intervention effect by performing a sensitivity analysis of trials at low risk of bias. However, in keeping with recent recommendations provided in the Cochrane Handbook for Systematic Reviews of Interventions (Deeks 2011), and with the goal of testing for subgroup differences, we instead conducted a subgroup analysis to compare low risk of bias trials with high risk of bias trials. We defined low risk of bias trials as trials with low risk of selection bias, detection bias and selective reporting, and all other trials as having high risk of bias.
Sensitivity analysis
We intended to perform a sensitivity analysis of only cluster‐RCTs, but none of the identified trials were cluster‐randomised. For some trials and outcomes, the reported denominator was different from what was to be expected from available case analysis. We therefore performed another sensitivity analysis while assuming that data were available for all patients if otherwise not directly stated. To test the robustness of our findings, we performed a sensitivity analysis in which we reanalysed all outcomes by using a fixed‐effect instead of a random‐effects model.
Results
Description of studies
Results of the search
In this update, we identified 1953 references (Figure 1) through our searches. By reading titles and abstracts, we eliminated 1943 references, as they were not relevant to the review. We obtained full papers for 10 references, excluded seven studies (Ahmad 2012; Bondesson 2013; Bonnet‐Zamponi 2013; Connor 2012; Frankenthal 2014; Marusic 2013; Sjoberg 2013) and included three new trials (Bladh 2011; Dalleur 2014; Farris 2014). We included in this update one of the trials (Scullin 2007) excluded from our original review (Christensen 2013) on the basis of comments received after its publication.
Study flow diagram.
We contacted the authors of two unpublished trials identified in our original review (ISRCTN08043800; NCT00844025). One trial author responded that the trial had not yet been submitted for publication (ISRCTN08043800), and the other trial author did not respond to our message. We identified four unpublished trials (Bonnerup 2014; Loffler ongoing; NCT01467128; NCT01504672) by searching trial registries. We contacted authors of all four trials and received a reply from all trial authors. One trial had finished enrolment (NCT01504672) but follow‐up data were still being collected, one trial was ongoing and completion was planned for November 2016 (Loffler ongoing) and one trial was finished and could be included, as the author supplied us with data that had been published in the form of a PhD thesis (Bonnerup 2014). As only participants with high risk of prescription error received a medication review intervention in the trial, we included only data from the subgroup of high‐risk participants in control and intervention groups. The last trial was finished and the manuscript had been submitted for publication (NCT01467128), but because of lack of resources, study authors had not collected follow‐up data and so the trial was excluded.
In summary, with five trials included in our original review, four newly identified trials and inclusion of one previously excluded trial, this review now includes 10 trials (see Characteristics of included studies).
Included studies
Setting
The ten trials included 3575 participants in total and reported follow‐up from 30 days to one year. Trial reports were published between 2006 and 2015; two studies were conducted in the USA (Farris 2014; Schnipper 2006) and the remaining eight in Europe (Belgium, Denmark, Ireland Northern Ireland and Sweden). Six trials included participants admitted to departments of internal medicine (Bladh 2011; Bonnerup 2014; Dalleur 2014; Gillespie 2009; Lisby 2010; Scullin 2007); one to departments of internal medicine, family medicine, cardiology and orthopaedics (Farris 2014); one to a tertiary medical centre admitted via the emergency department (Gallagher 2011); one to an orthopaedic ward (Lisby 2015); and one to the general medicines service (Schnipper 2006).
Participants
Five trials listed age as an inclusion criterion (two trials (Gallagher 2011; Lisby 2015) used 65 years or older, one used 70 years or older (Lisby 2010), one used 75 years or older (Dalleur 2014) and one used 80 years or older (Gillespie 2009). In general, participants were elderly with a mean age around 80 years in all trials except three, in which participant age was 59, 61 and 70 years, respectively (Farris 2014; Schnipper 2006; Scullin 2007). The proportion of women among included participants ranged from 53% to 71%, and the mean number of drugs per participant ranged from seven to 11.
Types of interventions
The medication review was performed by a pharmacist in four trials (Bladh 2011; Farris 2014; Gillespie 2009; Schnipper 2006), by a team of both pharmacists and pharmacy technicians in one (Scullin 2007), by a physician in two (Dalleur 2014; Gallagher 2011), by a pharmacist or a physician specialised in clinical pharmacology in one (Bonnerup 2014) and by a team of both pharmacists and physicians specialised in clinical pharmacology in two (Lisby 2010; Lisby 2015). In two trials the medication review was done using the validated Screening Tool of Older Persons’ potentially inappropriate Prescriptions (STOPP) (Dalleur 2014; Gallagher 2011); the latter trial also used the Screening Tool to Alert to Right Treatment (START). One trial described that the medication review was done via a computer decision support system (MiniQ) (Bladh 2011). In the remaining seven trials, the medication review was performed primarily by a pharmacist, who performed medication reconciliation and systematically reviewed the medication. In four trials, the medication review ended with a written recommendation to the prescribing physicians (Bonnerup 2014; Dalleur 2014; Lisby 2010; Lisby 2015); in three it was discussed with the prescribing physicians (Bladh 2011; Gallagher 2011; Schnipper 2006); and three provided no description of how recommendations were communicated to prescribing physicians (Farris 2014; Gillespie 2009; Scullin 2007). Seven trials provided additional interventions besides medication review for the intervention group. One trial included drug counselling (Lisby 2010); one included a discharge letter to the general practitioner (GP) (Dalleur 2014); one included drug counselling and a discharge letter to the GP (Bladh 2011); two included patient education, drug counselling, a discharge letter to the GP and telephone follow‐up (Farris 2014; Gillespie 2009); one included telephone follow‐up (Schnipper 2006); and one included a comprehensive integrated medicines management service including drug counselling and in‐patient monitoring (Scullin 2007). Two trials reported that the medication review resulted in a recommendation for drug changes for 58% (Gallagher 2011) and 60% of patients (Schnipper 2006). The proportion of suggested medication review recommendations that were followed by the prescribing physicians varied between trials: 18% (Lisby 2015), 39% (Lisby 2010), 41% (Bladh 2011), 65% (Bonnerup 2014),75% (Gillespie 2009) and 94% (Gallagher 2011).
Excluded studies
See Characteristics of excluded studies for the complete list of excluded studies with reasons.
Risk of bias in included studies
Risk of bias in the included trials is described in the Characteristics of included studies section (see Figure 2 and Figure 3 for graphical displays).
Risk of bias graph: review authors' judgements about each risk of bias item presented as percentages across all included studies. White spaces in this figure represent instances where it was not possible to make a judgement regarding objective or non‐objective outcomes.
Risk of bias summary: review authors' judgements about each risk of bias item for each included study. White spaces in this figure represent instances where it was not possible to make a judgement regarding objective or non‐objective outcomes.
Allocation
All 10 trials were randomised at an individual level; two trials used fixed blocks of 10 in each group (Gillespie 2009; Scullin 2007), and one trial used random blocks of maximum 20 (Bonnerup 2014). Seven trials had adequate randomisation sequence generation, of which five also had adequate allocation concealment and thus low risk of selection bias (Bladh 2011; Bonnerup 2014; Farris 2014; Gallagher 2011; Schnipper 2006) and two did not describe who included participants; risk of selection bias was therefore unclear (Lisby 2010; Lisby 2015). Of the remaining three trials, two did not describe how the randomisation sequence was generated or who included participants (Gillespie 2009; Scullin 2007) but used a fixed block size, which could have led to deciphering of the sequence, thus conferring unclear risk of selection bias. The final trial did not adequately describe how the randomisation sequence was generated, but as a study nurse both generated the sequence and included participants, risk of selection bias was high (Dalleur 2014).
Blinding
Nine trials described directly or indirectly that participants or personnel were not blinded, leading to high risk of performance bias in all trials. The last trial randomised patients to two geriatric teams, with one doing medication reviews using STOPP criteria. As participants and personnel were unaware of which team used STOPP, risk of performance bias was low (Dalleur 2014).
Three trials were described as providing blinded assessment for readmissions (Farris 2014; Gillespie 2009; Schnipper 2006), and two for hospital emergency department contacts (Farris 2014; Schnipper 2006). However, we judged it unlikely that awareness of group assignments would lead to risk of detection bias for these objective outcomes. Both trials assessing hospital readmissions due to adverse drug events provided blinded outcome assessment (Gillespie 2009; Schnipper 2006), as did one trial assessing hospital emergency department contacts due to adverse drug events (Schnipper 2006); one study did not describe this (Gillespie 2009). Of the three trials assessing adverse drug events, two provided blinded outcome assessment (Farris 2014; Schnipper 2006) and the remaining one (Gallagher 2011) was not blinded and had high risk of detection bias. For trials with blinded assessment of hospital readmissions due to adverse drug events, hospital emergency department contacts due to adverse drug events and adverse events, we judged risk as unclear, as participants were aware of group assignment and had knowledge of the drug adverse event profile; this could have led to differences in reporting of adverse events.
Incomplete outcome data
Six of nine trials reporting on mortality had low risk of attrition bias: One trial (Gillespie 2009) described no loss to follow‐up, five (Bladh 2011; Bonnerup 2014; Gallagher 2011; Lisby 2010; Lisby 2015) did not describe loss to follow‐up and all data seem to have been available from registries. In contrast, one trial (Dalleur 2014) reported only mortality follow‐up data for 66 out of 158 participants, leading to high risk of attrition bias, and for two trials (Farris 2014; Scullin 2007), it was unclear whether mortality data for participants lost to follow‐up were available, leading to unclear risk of attrition bias. Nine trials reported other outcomes; of these, one trial (Gillespie 2009) described no loss to follow‐up for all outcomes, four trials measured outcomes that should be available in registry data (Bladh 2011; Bonnerup 2014; Lisby 2010; Lisby 2015), two trials (Farris 2014; Schnipper 2006) showed discrepancies between reported participants lost to follow‐up and participants excluded from analysis; thus, we judged risk as unclear. One trial (Scullin 2007) described loss to follow‐up of around 1% in the manuscript but around 9% on the basis of reported tabular data; thus it was judged as high risk. Finally, one trial (Gallagher 2011) measured outcomes by using registry data and contact with general practitioners and participants without reporting how often data were not available. As adverse events, such as falls, could lead to loss to follow‐up, we judged this outcome as unclear.
Selective reporting
We judged one trial (Schnipper 2006) as having unclear risk for selective reporting of all‐cause mortality, as it did not report on the outcome, although the data seem to have been available. Also one trial (Dalleur 2014) was unclear for selective reporting of all‐cause readmissions and emergency department contacts, as the co‐authors had previously assessed these outcomes in similar trials.
Other potential sources of bias
The funnel plot for all‐cause mortality showed no sign of publication bias (Figure 4). Nine trials were judged as having high risk of contamination bias, as they were not cluster‐randomised. One trial (Dalleur 2014), which was also randomised at the individual level, was judged as having unclear risk of contamination bias. The medication review intervention (i.e. applying STOPP criteria) was blinded to personnel, and study authors stated that contamination bias was thus avoided. However, intervention group recommendations may also have been applied to control patients at the same hospital. From one trial (Bonnerup 2014), we included only a subgroup of participants with high risk of prescription errors. Including only this subgroup confers unclear risk of bias, as assessment of risk score for prescription errors was performed after allocation to groups (i.e. may introduce unbalance). Finally, one trial (Scullin 2007) was judged as having high risk of bias in relation to data analysis. First, a difference of 20 participants was noted between treatment arms, and this should not have been possible because the trial randomised in blocks of 10 in each arm. Second, data from a surgical ward were excluded from the analysis without an explanation.
Funnel plot of comparison: 1 Primary outcome: 1.1 Mortality (all‐cause).
Effects of interventions
See summary of findings Table for the main comparison for main comparisons.
Mortality (all‐cause)
See Analysis 1.1. Nine trials with data from 3218 participants and follow‐up from three to 12 months reported all‐cause mortality. During follow‐up, 238 participants in the medication review group died as did 228 in the control group (RR 1.02, 95% CI 0.87 to 1.19) (low‐certainty evidence). We did not observe any heterogeneity.
Hospital readmissions (all‐cause)
See Analysis 2.1; Analysis 2.2; and Analysis 2.3. Seven trials with data from 2843 participants and follow‐up from three to 12 months reported on hospital readmissions. During follow‐up, 529 participants in the medication review group and 514 in the control group had one or more hospital readmissions (RR 0.95, 95% CI 0.87 to 1.04) (high‐certainty evidence). We did not observe any heterogeneity. Three trials reported continuous data from 330 participants with three months of follow‐up. There was no difference in the number of readmissions per participant (mean difference (MD) ‐0.05, 95% CI ‐0.31 to 0.21; I2 = 35%). Two trials reported continuous data from 1063 participants with 12 months of follow‐up. There was no difference in the number of readmissions per participant (mean difference (MD) ‐0.12, 95% CI ‐0.29 to 0.05). We did not observe any heterogeneity.
One trial (Schnipper 2006) with data from 176 participants and 30 days of follow‐up reported hospital readmissions and hospital emergency department contacts as a composite. During follow‐up, 28 participants in the medication review group and 25 in the control group had one or more hospital readmissions or hospital emergency department contacts (RR 1.02, 95% CI 0.65 to 1.61).
Hospital readmissions (due to adverse drug events)
One trial (Gillespie 2009) with data from 368 participants and 12 months of follow‐up reported on hospital readmissions due to adverse drug events. During follow‐up, nine participants in the medication review group and 33 in the control group had one or more hospital readmissions due to adverse drug events, yielding a relative risk reduction of 72% favouring medication review (RR 0.28, 95% CI 0.14 to 0.57). The trial also reported continuous data and found fewer readmissions due to adverse drug events per participant in the medication review group (MD ‐0.19, 95% CI ‐0.28 to ‐0.10).
One trial (Schnipper 2006) with data from 176 participants and 30 days of follow‐up reported hospital readmissions and hospital emergency department contacts due to adverse drug events as a composite. During follow‐up, four participants in the medication review group and seven in the control group had one or more hospital readmissions or hospital emergency department contacts due to adverse drug events (RR 0.52, 95% CI 0.16 to 1.72).
Hospital emergency department contacts (all‐cause)
See Analysis 2.4 and Analysis 2.5. Four trials with data from 1442 participants and follow‐up from three to 12 months reported on hospital emergency department contacts. During follow‐up, 126 participants in the medication review group and 118 in the control group had one or more contacts, yielding a relative risk reduction of 27% favouring medication review (RR 0.73, 95% CI 0.52 to 1.03) (low‐certainty evidence; I2= 35%). Three trials reported continuous data from 330 participants with three months of follow‐up. Data show no difference in the number of emergency department contacts per participant (MD ‐0.05, 95% CI ‐0.16 to 0.06). We did not observe any heterogeneity. One trial (Gillespie 2009) reported continuous data from 368 participants with 12 months of follow‐up. There were fewer emergency department contacts per participant in the medication review group (MD ‐0.23, 95% CI ‐0.43 to ‐0.03).
Hospital emergency department contacts (due to adverse drug events)
One trial (Gillespie 2009) of 368 participants with 12 months of follow‐up reported on emergency department contacts due to adverse drug events. During follow‐up, four participants in the medication review group and nine in the control group had one or more contacts due to adverse drug events (RR 0.45, 95% CI 0.14 to 1.45). The trial also reported continuous data and found no differences in emergency department contacts due to adverse drug events (MD ‐0.03, 95% CI ‐0.07 to 0.01).
Mortality (due to adverse drug events)
No trials reported data for this outcome.
Adverse drug events
One trial (Schnipper 2006) with data from 152 participants and 30 days of follow‐up reported adverse drug events. During follow‐up, 14 participants in the medication review group and 12 in the control group had one or more adverse drug events (RR 1.08, 95% CI 0.53 to 2.18). One trial (Gallagher 2011) with data from 382 participants and six months of follow‐up reported on falls as an adverse drug event. During follow‐up, 11 participants in the medication review group and 16 in the control group had one or more falls (RR 0.69, 95% CI 0.33 to 1.46). One trial (Farris 2014) reported on adverse events and adverse drug events as a composite. We were unable to get separate data on adverse drug events from the author and did not include data for this outcome.
Subgroup analysis and investigation of heterogeneity
Six trials used age as an inclusion criterion (three older than 65, one older than 70, one older than 75 and one older than 80 years), two used number of drugs as an inclusion criterion (minimum four drugs) and one included patients for medication review if they had high risk of prescription errors based on an algorithm. Three trials (Bladh 2011; Farris 2014; Schnipper 2006) did not have any risk factors for medication errors and adverse drug events as part of the inclusion criteria. Reporting of the data in one trial (Schnipper 2006) precluded its inclusion in the subgroup analysis.
Comparison between subgroups revealed a small trend toward increased mortality in the low‐risk group compared with the high‐risk group (P value = 0.08) (Analysis 3.1), but no differences in effects on readmissions (Analysis 3.2) or on hospital emergency department contacts (Analysis 3.3).
In none of the trials was the person or team performing the medication review allowed to change the medication directly; therefore, we could not explore this in a separate subgroup analysis.
In two trials (Dalleur 2014; Gallagher 2011), medication review was performed through validated methods. Comparison between subgroups with and without validated methods revealed no difference in effect (Analysis 3.4; Analysis 3.5).
We judged four trials (Bladh 2011; Bonnerup 2014; Farris 2014; Gallagher 2011) as having low overall risk of bias. Comparison between subgroups of trials with low overall risk of bias and with high overall risk of bias revealed no difference in effect (Analysis 3.6; Analysis 3.7; Analysis 3.8).
Sensitivity analysis
For some trials, we could calculate a different available case analysis for mortality and readmission outcomes, but our reanalysis was similar to our main analysis (Analysis 4.1; Analysis 4.2; Analysis 4.3). Our reanalysis of all outcomes based on a fixed‐effect model did not change our results (Analysis 4.4; Analysis 4.5; Analysis 4.6; Analysis 4.7; Analysis 4.8; Analysis 4.9).
Discussion
Summary of main results
We found no evidence suggesting that medication review reduces mortality (low‐certainty evidence) or hospital readmissions (high‐certainty evidence), but found that medication review may reduce the number of emergency department contacts compared with standard care (low‐certainty evidence). The estimated reduction in emergency department contacts of 27% (with a confidence interval ranging from 48% reduction to 3% increase in contacts) corresponds to a number needed to treat (to prevent one emergency department contact) of 37 for a low‐risk population and 12 for a high‐risk population over one year. The specific type of medication review provided did not seem to influence the results. Despite consistent results suggesting no effect, a beneficial or detrimental effect on mortality or readmissions cannot be ruled out because estimates were uncertain and follow‐up was short.
Overall completeness and applicability of evidence
This review focused primarily on patient‐relevant outcomes such as mortality, readmissions and emergency department contacts. Adverse drug events are often linked causally to all of these outcomes (Budnitz 2011; Hallas 1996), and an intervention reducing the inappropriateness of patient medication would, in contrast to our findings, be assumed to have a beneficial affect on all. Therefore, the possible intervention effect on emergency department contacts could be considered somewhat paradoxical, when no effect on readmissions was observed. This discrepancy might be explained by the large beneficial effect particularly in one trial (Gillespie 2009), which was crucial for the observed pooled estimate. However, it is possible that emergency department contacts represent a more sensitive outcome measure of adverse drug events than readmission or mortality, or that the duration of follow‐up simply was too short to reveal an effect on these outcomes. Along these lines, it is important to note that medication review as a general rule of thumb includes the addition of relevant prophylactic medicines (e.g. statins, antihypertensives), which mainly confer beneficial effects on hospital admissions and mortality after several years of treatment (Gutierrez 2012; Wright 2009). Thus, beneficial effects of a reduction in inappropriate underprescribing will likely not have an effect on admission and mortality in studies with shorter follow‐up. Whereas the occurrence or prevention of adverse drug events would be expected to have a shorter time frame, longer follow‐up in these trials is thus crucial for a true evaluation of the effects of medication review.
The size of the suggested effect on emergency department contacts depends on baseline risk in the population receiving the medication review. All trials included older participants receiving multiple medications and, based on the trial populations, had numbers needed to treat for an additional beneficial outcome of 12 for a high‐risk population and 37 for a low‐risk population to prevent one emergency department contact over one year. As medication review is time‐consuming, the question is whether the intervention is cost‐effective. In one trial, authors estimated the costs of the medication review as USD 170 per patient (Gillespie 2009). If their figure is used as an estimate of the costs, it would cost between USD 2040 and USD 6290 to avoid one emergency department contact in a year. However, in addition to possible cost‐savings from reducing the number of emergency department contacts, a reduction in unnecessary medications could reduce costs. A future cost‐effectiveness analysis based on data from our systematic review could clarify in which subpopulations medication review is cost‐effective.
The two trials (Gillespie 2009; Schnipper 2006) reporting emergency department contacts or readmissions attributable to adverse drug events stipulated that medication review conferred sizeable reductions in these outcomes. In the trial by Gillespie et al, participants receiving medication review had their risk of drug‐related readmissions lowered by 72% (nine vs 33 participants), but this was not reflected by a similar absolute reduction in participants with all‐cause readmissions (106 vs 112 participants). This difference perhaps illustrates some of the problems associated with the causality assessment of adverse drug events. Adverse events are rarely drug specific but are often general symptoms or illnesses that could have many causes (e.g. dizziness or gastric ulcer). In the context of polypharmacy, it is easy to associate medicines with symptoms, particularly if the drug is inappropriate and the symptom is a known adverse effect of that drug. It follows that any intervention that results in patients taking fewer inappropriate drugs may, solely by reducing the possibility of attributing a symptom to a drug, lead to fewer of these assessed drug‐related outcomes. Additionally, despite the fact that outcome assessors were blinded to group assignments, participants' knowledge of assignments could have resulted in unmasking during the participant interview, thereby introducing detection bias. Another complex issue is that the medication review resulting in discontinuation of medicines might lead to alleviation of adverse events at the expense of undertreatment of certain conditions. For example, less use of antihypertensive agents could lead to fewer readmissions due to dizziness but more readmissions due to stroke. The effect of medication review on readmissions due to adverse drug events (with no effect on all‐cause readmissions) should therefore be viewed with caution.
Previous admissions are a major risk factor for subsequent admissions (Epstein 2011; Hasan 2010; Marcantonio 1999). The decision to exclude trials of medication review performed in primary care, outpatient clinics, emergency departments or paediatric departments was taken to limit the study population to patients with demonstrated high risk of hospital admissions. In general, trial populations consisted of elderly participants receiving polypharmacy and with multimorbidity. The 10 included trials differed slightly with regards to the content of the medication review. Applying explicit criteria for reviewing medication could improve the applicability and reviewer independency of study findings (Dalleur 2014; Gallagher 2011), whereas interventions depending on few reviewers using unvalidated methods (Farris 2014; Gillespie 2009; Lisby 2010; Lisby 2015; Schnipper 2006) may introduce problems in relation to generalisability. Furthermore, co‐interventions such as telephone contact with patients or general practitioners (Bladh 2011; Dalleur 2014; Farris 2014; Gallagher 2011; Schnipper 2006) are resource demanding, and the added effect of including them as part of the medication review is not known. We chose to exclude trials of interventions aimed solely at increasing patient knowledge or adherence, interventions that were to be implemented after discharge, or interventions in which the medication review was related to only a portion of a participant's medication, because such interventions might have a lesser effect on clinical outcomes, thereby introducing heterogeneity.
Medication review may have an impact on other outcomes such as number of drugs prescribed, adherence, drug knowledge and/or patient satisfaction. However, these outcomes do not capture the potentially harmful effects of medication review resulting from undertreatment and because of their subjective nature are more prone to bias; therefore, we excluded them from our review.
Quality of the evidence
Our review was based on a very comprehensive literature search and was further strengthened by the inclusion of unpublished data. We included data from 10 trials with around 3600 participants, most of whom had a presumed high risk of adverse drug events.
However, some limitations must be considered. Most included trials had some problems related to risk of bias, and some had problems due to inadequate reporting. The nature of the intervention precluded blinding of participants, which may introduce performance bias. Similarly, as previously stated, it can be questioned whether detection bias for the drug‐related outcomes can actually be prevented. We judged outcomes to have low risk of bias when outcome assessors were unaware of group assignment, but whether this was sufficient to prevent detection bias is debatable. Two trials described drug counselling, which may in some form have taken place as part of the medication review in the other trials. Knowledge of adverse events from prescribed drugs, for example, dizziness from antihypertensive agents, may lead participants to focus on these problems when presenting a broader problem during an emergency department contact or readmission. This may result in underestimation of the intervention effect on adverse drug events. However, it could also work the other way around, as participants knowing the adverse events of a drug would not focus on those particular symptoms during the participant interview.
As described previously, the effect on all‐cause outcomes is therefore preferred as the result of lower risk of bias. Although mortality was not the primary outcome in any of the trials, only one trial did not report on this outcome (Schnipper 2006). Our funnel plot showed no sign of publication bias. In all trials, medication review was delivered by a special team of dedicated persons, and participants in control groups were treated by the same healthcare providers as were those in the intervention groups. Although it seems unlikely that participants in control groups should have received a similar intervention, some contamination bias might have occurred (e.g. increasing physicians' and nurses' focus on appropriate pharmacotherapy), thereby introducing bias towards the null. Remarkably, we did not identify any cluster‐randomised controlled trials of medication review, which by their design could minimise contamination bias.
Trials were conducted in different settings and employed different co‐interventions. The longest follow‐up was one year and was assessed in only two of the trials (Gillespie 2009; Scullin 2007). The short duration of follow‐up should be a caveat when interpreting the results of this review, while bearing in mind that many drugs are used for preventive purposes to avoid long‐term events (e.g. cardiovascular mortality). Likewise, the confidence intervals included both possible beneficial and harmful effects of the intervention, making any conclusions about the effects uncertain. However, the narrow confidence intervals suggest that any effect is likely to be small.
Agreements and disagreements with other studies or reviews
We attempted to examine the effects of medication review on hospitalised adult patients. A recent review (Hohl 2015) failed to identify an effect of early in‐hospital pharmacist‐led medication review on health outcomes. In contrast to our analyses, the review included only trials in which medication review was initiated within 24 hours of emergency department presentation or within 72 hours of admission, and also included trials in which medication review was investigated using quasi‐randomised methods. A systematic review (Holland 2008) that included patients from both primary and secondary care, of younger age and receiving fewer drugs on average, found no effects of medication review on mortality and readmissions. Likewise, two systematic reviews (Nkansah 2010; Royal 2006) included trials with participants at lower risk, and reported no effects on mortality and readmissions. A recent systematic review (Wallerstedt 2014) assessed medication review for nursing home residents and found no beneficial effect on mortality nor on hospitalisation.
Risk of bias graph: review authors' judgements about each risk of bias item presented as percentages across all included studies. White spaces in this figure represent instances where it was not possible to make a judgement regarding objective or non‐objective outcomes.
Risk of bias summary: review authors' judgements about each risk of bias item for each included study. White spaces in this figure represent instances where it was not possible to make a judgement regarding objective or non‐objective outcomes.
Funnel plot of comparison: 1 Primary outcome: 1.1 Mortality (all‐cause).
Comparison 1 Primary outcome, Outcome 1 Mortality (all‐cause).
Comparison 2 Secondary outcomes, Outcome 1 Hospital readmissions (all‐cause).
Comparison 2 Secondary outcomes, Outcome 2 Hospital readmissions (all‐cause) ‐ 3 months.
Comparison 2 Secondary outcomes, Outcome 3 Hospital readmissions (all‐cause) ‐ 12 months.
Comparison 2 Secondary outcomes, Outcome 4 Hospital emergency department contacts (all‐cause).
Comparison 2 Secondary outcomes, Outcome 5 Hospital emergency department contacts (all‐cause) ‐ 3 months.
Comparison 3 Subgroup analysis, Outcome 1 Mortality (all‐cause).
Comparison 3 Subgroup analysis, Outcome 2 Hospital readmissions (all‐cause).
Comparison 3 Subgroup analysis, Outcome 3 Hospital emergency department contacts (all‐cause).
Comparison 3 Subgroup analysis, Outcome 4 Mortality (all‐cause).
Comparison 3 Subgroup analysis, Outcome 5 Hospital readmissions (all‐cause).
Comparison 3 Subgroup analysis, Outcome 6 Mortality (all‐cause).
Comparison 3 Subgroup analysis, Outcome 7 Hospital readmissions (all‐cause).
Comparison 3 Subgroup analysis, Outcome 8 Hospital emergency department contacts (all‐cause).
Comparison 4 Sensitivity analysis, Outcome 1 Mortality (all‐cause) ‐ alternative available case analysis.
Comparison 4 Sensitivity analysis, Outcome 2 Hospital readmissions (all‐cause) ‐ alternative available case analysis.
Comparison 4 Sensitivity analysis, Outcome 3 Hospital emergency department contacts (all‐cause) ‐ alternative available case analysis.
Comparison 4 Sensitivity analysis, Outcome 4 Mortality (all‐cause) ‐ fixed‐effect.
Comparison 4 Sensitivity analysis, Outcome 5 Hospital readmissions (all‐cause) ‐ fixed‐effect.
Comparison 4 Sensitivity analysis, Outcome 6 Hospital readmissions (all‐cause) ‐ 3 months ‐ fixed‐effect.
Comparison 4 Sensitivity analysis, Outcome 7 Hospital readmissions (all‐cause) ‐ 12 months ‐ fixed‐effect.
Comparison 4 Sensitivity analysis, Outcome 8 Hospital emergency department contacts (all‐cause) ‐ fixed‐effect.
Comparison 4 Sensitivity analysis, Outcome 9 Hospital emergency department contacts (all‐cause) ‐ 3 months ‐ fixed‐effect.
| Medication review compared with standard care for hospitalised adult patients | ||||||
| Patient or population: hospitalised adult patients Intervention: medication review Comparison: standard care | ||||||
| Outcomes | Illustrative comparative risks* (95% CI) | Relative effect | Number of participants | Certainty of the evidence | Comments | |
| Assumed risk | Corresponding risk | |||||
| Standard care | Medication review | |||||
| Mortality (all‐cause) 1 year | Low‐risk population | RR 1.02 (0.87 to 1.19) | 3218 | ⊕⊕⊝⊝ Lowb,c | NA | |
| 200 per 1000a | 204 per 1000 | |||||
| High‐risk population | ||||||
| 400 per 1000a | 408 per 1000 | |||||
| Hospital readmission (all‐cause) 1 year | Low‐risk population | RR 0.95 (0.87 to 1.04) | 2843 (7 trials) | ⊕⊕⊕⊕ High | NA | |
| 300 per 1000a | 285 per 1000 | |||||
| High‐risk population | ||||||
| 600 per 1000a | 570 per 1000 | |||||
| Hospital emergency department contacts (all‐cause) 1 year | Low‐risk population | RR 0.73 (0.52 to 1.03) | 1442 | ⊕⊕⊝⊝ | Equal to number needed to treat of 12 for the high‐risk population and 37 for the low‐risk population | |
| 100 per 1000a | 73 per 1000 | |||||
| High‐risk population | ||||||
| 300 per 1000a | 219 per 1000 | |||||
| *The basis for the assumed risk (e.g. median control group risk across studies) is provided in footnotes. The corresponding risk (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). | ||||||
| GRADE Working Group grades of evidence | ||||||
| aRisk for the high‐risk population of the control group was based on data from Gillespie 2009, one of the 3 trials with 12 months of follow‐up (all outcomes) and with greatest risk in the control group. For the low‐risk population, the 12 months of follow‐up for Scullin 2007 (mortality) and the 3 months of follow‐up for Lisby 2010 (hospital emergency department contacts) were extrapolated to determine 12‐month risk. All trials had almost similar control group risks for hospital readmissions, and the low‐risk group was based on half the risk of 12 months of follow‐up for Gillespie 2009 bSubgroup analysis comparing 'high'‐ and 'low'‐risk populations revealed a small trend toward increased mortality in the low‐risk group compared with the high‐risk group (P value = 0.08) (downgraded 1 category) cFollow‐up ranged from 3 to 12 months for mortality. Short follow‐up may be inadequate to detect the effect on changes in prophylactic medication (indirectness of evidence) (downgraded 1 category) d'Risk of bias' assessments determined that 3 of 4 trials in the analysis (Gillespie 2009; Lisby 2010; Lisby 2015) had overall 'high risk' of bias (downgraded 1 category) eThe confidence interval overlapped 1 (downgraded 1 category) | ||||||
| Outcome or subgroup title | No. of studies | No. of participants | Statistical method | Effect size |
| 1 Mortality (all‐cause) Show forest plot | 9 | 3218 | Risk Ratio (M‐H, Random, 95% CI) | 1.02 [0.87, 1.19] |
| Outcome or subgroup title | No. of studies | No. of participants | Statistical method | Effect size |
| 1 Hospital readmissions (all‐cause) Show forest plot | 7 | 2843 | Risk Ratio (M‐H, Random, 95% CI) | 0.95 [0.87, 1.04] |
| 2 Hospital readmissions (all‐cause) ‐ 3 months Show forest plot | 3 | 330 | Mean Difference (IV, Random, 95% CI) | ‐0.05 [‐0.31, 0.21] |
| 3 Hospital readmissions (all‐cause) ‐ 12 months Show forest plot | 2 | 1063 | Mean Difference (IV, Random, 95% CI) | ‐0.12 [‐0.29, 0.05] |
| 4 Hospital emergency department contacts (all‐cause) Show forest plot | 4 | 1442 | Risk Ratio (M‐H, Random, 95% CI) | 0.73 [0.52, 1.03] |
| 5 Hospital emergency department contacts (all‐cause) ‐ 3 months Show forest plot | 3 | 330 | Mean Difference (IV, Random, 95% CI) | ‐0.05 [‐0.16, 0.06] |
| Outcome or subgroup title | No. of studies | No. of participants | Statistical method | Effect size |
| 1 Mortality (all‐cause) Show forest plot | 9 | 3218 | Risk Ratio (M‐H, Random, 95% CI) | 1.02 [0.87, 1.19] |
| 1.1 High‐risk population | 7 | 1922 | Risk Ratio (M‐H, Random, 95% CI) | 0.96 [0.81, 1.14] |
| 1.2 Low‐risk population | 2 | 1296 | Risk Ratio (M‐H, Random, 95% CI) | 1.49 [0.95, 2.34] |
| 2 Hospital readmissions (all‐cause) Show forest plot | 7 | 2843 | Risk Ratio (M‐H, Random, 95% CI) | 0.95 [0.87, 1.04] |
| 2.1 High‐risk population | 5 | 1627 | Risk Ratio (M‐H, Random, 95% CI) | 0.92 [0.83, 1.03] |
| 2.2 Low‐risk population | 2 | 1216 | Risk Ratio (M‐H, Random, 95% CI) | 1.04 [0.87, 1.24] |
| 3 Hospital emergency department contacts (all‐cause) Show forest plot | 4 | 1442 | Risk Ratio (M‐H, Random, 95% CI) | 0.73 [0.52, 1.03] |
| 3.1 High‐risk population | 3 | 574 | Risk Ratio (M‐H, Random, 95% CI) | 0.61 [0.36, 1.03] |
| 3.2 Low‐risk population | 1 | 868 | Risk Ratio (M‐H, Random, 95% CI) | 0.90 [0.64, 1.25] |
| 4 Mortality (all‐cause) Show forest plot | 9 | 3218 | Risk Ratio (M‐H, Random, 95% CI) | 1.02 [0.87, 1.19] |
| 4.1 Systematic medication review | 2 | 442 | Risk Ratio (M‐H, Random, 95% CI) | 0.91 [0.56, 1.46] |
| 4.2 Non‐systematic medication review | 7 | 2776 | Risk Ratio (M‐H, Random, 95% CI) | 1.03 [0.87, 1.22] |
| 5 Hospital readmissions (all‐cause) Show forest plot | 7 | 2843 | Risk Ratio (M‐H, Random, 95% CI) | 0.95 [0.87, 1.04] |
| 5.1 Systematic medication review | 1 | 358 | Risk Ratio (M‐H, Random, 95% CI) | 1.04 [0.79, 1.36] |
| 5.2 Non‐systematic medication review | 6 | 2485 | Risk Ratio (M‐H, Random, 95% CI) | 0.94 [0.86, 1.04] |
| 6 Mortality (all‐cause) Show forest plot | 9 | 3218 | Risk Ratio (M‐H, Random, 95% CI) | 1.02 [0.87, 1.19] |
| 6.1 Low risk of bias | 4 | 1796 | Risk Ratio (M‐H, Random, 95% CI) | 1.21 [0.89, 1.66] |
| 6.2 High risk of bias | 5 | 1422 | Risk Ratio (M‐H, Random, 95% CI) | 0.96 [0.79, 1.15] |
| 7 Hospital readmissions (all‐cause) Show forest plot | 7 | 2843 | Risk Ratio (M‐H, Random, 95% CI) | 0.95 [0.87, 1.04] |
| 7.1 Low risk of bias | 3 | 1574 | Risk Ratio (M‐H, Random, 95% CI) | 1.04 [0.89, 1.20] |
| 7.2 High risk of bias | 4 | 1269 | Risk Ratio (M‐H, Random, 95% CI) | 0.91 [0.81, 1.02] |
| 8 Hospital emergency department contacts (all‐cause) Show forest plot | 4 | 1442 | Risk Ratio (M‐H, Random, 95% CI) | 0.73 [0.52, 1.03] |
| 8.1 Low risk of bias | 1 | 868 | Risk Ratio (M‐H, Random, 95% CI) | 0.90 [0.64, 1.25] |
| 8.2 High risk of bias | 3 | 574 | Risk Ratio (M‐H, Random, 95% CI) | 0.61 [0.36, 1.03] |
| Outcome or subgroup title | No. of studies | No. of participants | Statistical method | Effect size |
| 1 Mortality (all‐cause) ‐ alternative available case analysis Show forest plot | 9 | 3379 | Risk Ratio (M‐H, Random, 95% CI) | 1.02 [0.87, 1.20] |
| 2 Hospital readmissions (all‐cause) ‐ alternative available case analysis Show forest plot | 7 | 3039 | Risk Ratio (M‐H, Random, 95% CI) | 0.95 [0.87, 1.05] |
| 3 Hospital emergency department contacts (all‐cause) ‐ alternative available case analysis Show forest plot | 4 | 1510 | Risk Ratio (M‐H, Random, 95% CI) | 0.85 [0.68, 1.07] |
| 4 Mortality (all‐cause) ‐ fixed‐effect Show forest plot | 9 | 3218 | Risk Ratio (M‐H, Fixed, 95% CI) | 1.03 [0.88, 1.21] |
| 5 Hospital readmissions (all‐cause) ‐ fixed‐effect Show forest plot | 7 | 2843 | Risk Ratio (M‐H, Fixed, 95% CI) | 0.96 [0.87, 1.05] |
| 6 Hospital readmissions (all‐cause) ‐ 3 months ‐ fixed‐effect Show forest plot | 3 | 330 | Mean Difference (IV, Fixed, 95% CI) | ‐0.05 [‐0.24, 0.14] |
| 7 Hospital readmissions (all‐cause) ‐ 12 months ‐ fixed‐effect Show forest plot | 2 | 1063 | Mean Difference (IV, Fixed, 95% CI) | ‐0.12 [‐0.29, 0.05] |
| 8 Hospital emergency department contacts (all‐cause) ‐ fixed‐effect Show forest plot | 4 | 1442 | Risk Ratio (M‐H, Fixed, 95% CI) | 0.76 [0.60, 0.96] |
| 9 Hospital emergency department contacts (all‐cause) ‐ 3 months ‐ fixed‐effect Show forest plot | 3 | 330 | Mean Difference (IV, Fixed, 95% CI) | ‐0.05 [‐0.16, 0.06] |
