Deprescribing is associated with reduced readmission to hospital: An updated meta-analysis of randomized controlled trials

A recent systematic review and meta-analysis of inpatient studies¹ found that medication reviews coupled with deprescribing interventions were associated with a small but significant reduction in rehospitalizations. The authors reported a hazard ratio (HR) of 0.92 (95% CI 0.85 to 0.99) derived from 19 randomized controlled trials (RCTs) and observational studies with outcomes measured between 1 and 12 months. That analysis provided much needed evidence in support of inpatient deprescribing initiatives. However, there are two key points for discussion. First, some important RCT data were not included by the search and selection strategy. Second, the inclusion of non-randomized data may increase bias. To increase the strength of the findings, we performed a sensitivity analysis restricted to RCTs and incorporated additional data.

The MedSafer Study² was an 11 center cluster RCT of deprescribing decision support paired with medication reconciliation versus medication reconciliation alone. The primary outcome was adverse drug events within 30 days post-discharge and a key secondary outcome was hospital readmission. A total of 5698 participants were enrolled with 4989 surviving to hospital discharge who were followed for readmission within 30-days.

Although the search strategy by Carollo et al. captured this study, it was subsequently excluded (reason not specified). We reanalyzed the 30-day hospital readmission data from the MedSafer Study using a time-to-event approach. Patient survival time was censored at 30 days, readmission to hospital, or death (whichever occurred first). Patients who were readmitted on the day of discharge were excluded (n = 14). In keeping with the trial's statistical analysis plan, we used a mixed-effects exponential proportional hazards regression model with adjustment for age, biological sex, the presence of moderate or severe frailty, residing in a nursing home at the time of admission, the number of potentially inappropriate medications at baseline, and the temporal period. Cluster was included as a random effect.

Carollo et al.¹ analyzed 10,136 RCT patients including duplicated control patients. Our updated analysis contained 14,201 unique patients including 4975 from McDonald et al. and 695 from Franchi et al.⁶ (Figure 1). Though the effect size was more modest, deprescribing remained statistically associated with a reduced hazard of readmission at 1–3 months (HR 0.84; 95% CI 0.73 to 0.97). When all durations of follow-up were included, the probability that readmission was reduced remained high, but this was no longer statistically significant (HR 0.94; 95%CI 0.87 to 1.01).

A meta-analysis restricted to RCTs and adding 5670 unique patients of additional data added substantial credibility to the finding of a reduction in the hazard of short-term readmission from inpatient deprescribing interventions. Because an adequate number of RCTs exist, we suggest that observational data can either be omitted from future meta-analyses or analyzed separately. The inclusion of patients from cluster and stepped-wedge randomized trials in a meta-analysis is permissible⁴ and should be encouraged. Limitations of our updated analysis include the inherent limitations in the source RCTs as well as the need to infer HRs for the seven studies not directly reporting them.

It is highly likely that inpatient deprescribing interventions reduce the risk of short-term readmission. A living systematic review approach is needed, with the addition of new RCT data as they emerge. Although it is possible that sustained reductions in rehospitalization will be demonstrated by additional deprescribing RCTs reporting longer term follow-up, it seems likely that a demonstrated early benefit may already justify implementation.

Conceptualization—TCL and EGM. Methodology—TCL and EGM. Software—TCL. Validation—TCL. Formal analysis—TCL. Investigation—all authors. Resources—TCL, EGM. Data Curation—TCL. Writing Original Draft—TCL, EBC, and EGM. Writing review and editing—all authors. Visualization—TCL. Supervision—TCL and EGM. Project administration—TCL. Funding acquisition – N/A.

Drs. McCarthy, McDonald, and Lee have operating grants from the Canadian Institutes of Health Research (CIHR). Drs. McDonald and Lee receive salary support from the Fonds de Recherche du Québec—Santé outside of this work. Drs. Lee and McDonald jointly hold the copyright to MedSafer, the deprescribing software used in the included randomized controlled trial.

This article did not receive any funding.

This article did not receive any funding.