Impact of Pharmacist-led Discharge Counseling on Hospital Readmission and Emergency Department Visits: A Systematic Review and Meta-analysis

Data Extraction and Quality Assessment

We used a standardized form to collect data on the following general characteristics of the studies: baseline data (author names, year of publication, study design, country, and sample size), methodological aspects, and outcomes of interest (ie, number of hospital readmission or emergency department visits). When outcomes were assessed in different time periods, the last period was considered for the overall analysis.

The methodological quality of the included studies was evaluated using the Cochrane Collaboration’s tool for risk of bias assessment that classifies each study as having a low, unclear, or high risk of bias.¹⁴

Data Analysis

Pairwise meta-analyses of the included RCTs were performed using the Comprehensive Meta-Analysis v 2.2 software (Biostat, Englewood, New Jersey). For each meta-analysis, we used the random effects model with the inverse variance method (DerSimonian and Laird) to estimate the pooled risk ratio (RR) with a 95% confidence interval (CI). With this method, a weight is given to each study that is the inverse of the variance of the effect estimate giving larger studies more weight than smaller studies. P values <.05 (two-tailed) were considered indicative of a statistically significant difference between groups.

The between-trial heterogeneity was estimated using the inconsistency relative index I² (I² > 50% indicates high and significant heterogeneity). Tau and Tau² measures were used to estimate the distribution of the true effect sizes and to compute the prediction intervals (PIs).^16-18 The calculation of PI was done in preformatted sheets in Excel considering the number of studies, the mean effect (random effect weights), the upper effect of mean effect, and tau-square in log units (normal approximation).¹⁷ PIs allow more informative inferences in meta-analyses (eg, true treatment effects that can be expected in future settings), especially when there is large variation in the strength of the effect (high heterogeneity between studies). This results in PIs generally having a wider range of expected treatment effects than CIs.¹⁹

We also conducted sensitivity analyses to test the robustness of the results and to evaluate the effect of individual studies on data heterogeneity. The sensitivity analysis consisted of the hypothetical sequential removal of studies from the meta-analysis. In addition, to verify the influence of small-study effects on the results of a meta-analysis with between-trial heterogeneity (I² > 0), we compared the results obtained in the random effect model with those obtained from fixed effects models.

When possible, subgroup analyses were performed considering (1) how discharge counseling was delivered (ie, alone or combined with other interventions) and (2) time of evaluation of the outcomes (weeks, months, or years postdischarge). The visual representation of the estimated treatment effect versus the standard error (funnel plots) was also performed to assess the potential role of publication bias.

A total of 2,656 records were retrieved from the electronic databases and manual searches. During the screening phase, 276 records were considered for full-text analysis, of which 21 were included in the qualitative analysis^20-40 and 18 were suitable for quantitative analyses^{21,22,24-36,38-40} (Figure 1). The references of excluded studies, with the reasons for exclusion, are mentioned in the Supplemental Material.