An On-Treatment Analysis of the MARQUIS Study: Interventions to Improve Inpatient Medication Reconciliation
It is unclear which medication reconciliation interventions are most effective at reducing inpatient medication discrepancies. Five United States hospitals’ interdisciplinary quality improvement (QI) teams were virtually mentored by QI-trained physicians. Sites implemented one to seven evidence-based interventions in 791 patients during the 25-month implementation period. Three interventions were associated with significant decreases in potentially harmful discrepancy rates: (1) defining clinical roles and responsibilities, (2) training, and (3) hiring staff to perform discharge medication reconciliation. Two interventions were associated with significant increases in potentially harmful discrepancy rates: training staff to take medication histories and implementing a new electronic health record (EHR). Hospitals should focus first on hiring and training pharmacy staff to assist with medication reconciliation at discharge and delineating roles and responsibilities of clinical staff. We caution hospitals implementing a large vendor EHR, as medication discrepancies may increase. Finally, the effect of medication history training on discrepancies needs further study.
© 2019 Society of Hospital Medicine
Unintentional medication discrepancies in the hospital setting are common and contribute to adverse drug events, resulting in patient harm.1 Discrepancies can be resolved by implementing high-quality medication reconciliation, but there are insufficient data to guide hospitals as to which interventions are most effective at improving medication reconciliation processes and reducing harm.2 We recently reported that implementation of a best practices toolkit reduced total medication discrepancies in the Multi-Center Medication Reconciliation Quality Improvement Study (MARQUIS).3 This report describes the effect of individual toolkit components on rates of medication discrepancies with the potential for patient harm.
METHODS
Detailed descriptions of the intervention toolkit and study design of MARQUIS are published.4,5 Briefly, MARQUIS was a pragmatic, mentored, quality improvement (QI) study in which five hospitals in the United States implemented interventions from a best practices toolkit to improve medication reconciliation on noncritical care medical and surgical units from September 2011 to July 2014. We used a mentored implementation approach, in which each site identified the leaders of their local quality improvement team (ie, mentees) who received mentorship from a trained physician with QI and medication safety experience.6 Mentors conducted monthly calls with their mentees and two site visits. Sites adapted and implemented one or more components from the MARQUIS toolkit, a compilation of evidence-based best practices in medication reconciliation.5,7
The primary outcome was unintentional medication discrepancies in admission and discharge orders with the potential for causing harm, as previously described.4 Trained study pharmacists at each site took “gold standard” medication histories on a random sample of up to 22 patients per month. These medications were then compared with admission and discharge medication orders, and all unintentional discrepancies were identified. The discrepancies were then adjudicated by physicians blinded to the treatment arm, who confirmed whether discrepancies were unintentional and carried the potential for patient harm.
We employed a modification of a stepped wedge methodology to measure the incremental effect of implementing nine different intervention components, introduced at different sites over the course of the study, on the number of potentially harmful discrepancies per patient. These analyses were restricted to the postimplementation period on hospital units that implemented at least one intervention. All interventions conducted at each site were categorized by component, including dates of implementation. Each intervention component could be applied more than once per site (eg, when involving a new group of providers) or implemented on a new hospital unit or service, in which case, all dates were included in the analysis. We conducted a multivariable Poisson regression (with time divided into months) adjusted for patient factors, season, and site, with the number of potentially harmful discrepancies as the dependent variable, and the total number of gold standard medications as a model offset. The model was designed to analyze changes in the y-intercept each time an intervention component was either implemented or spread and assumed the change in the y-intercept was the same for each of these events for any given component. The model also assumes that combinations of interventions had independent additive effects.
