(C52) Validating Medicare claims-based algorithms for deprescribing of benzodiazepines using a novel linkage with electronic health records data

Publication Number: 996

Background: Deprescribing is advised when the potential harms of continued medication use outweigh the potential benefits. Prior research using Medicare claims considered gaps in days’ supply and dose reductions as proxy measures for deprescribing.

Objectives: Evaluate the performance of claims-based definitions using gaps in days’ supply or dose reductions to identify deprescribing of benzodiazepines (BZDs) after a fall-related hospitalization (a clear deprescribing indication) against manual abstraction of electronic health records (EHRs).

Methods: We used a novel linkage between Medicare fee-for-service (FFS) and Part D claims with the University of North Carolina (UNC) Health EHRs. Using Medicare claims from 2015-2017, we identified patients aged 66+ years with a fall-related hospitalization and 6+ months of continuous enrollment in Medicare FFS and Part D prior to and following hospitalization. Patients had to have at least two BZD fills in the 6 months prior to hospitalization with one fill overlapping in days’ supply with the hospitalization. Among those with a UNC fall-related hospitalization, we used a standardized EHR abstraction tool to identify deprescribing using medication orders, discharge summaries, physician notes, medication lists, patient instructions, and problem lists and summarized our findings. We then evaluated the validity of three claims-based deprescribing definitions versus chart review: (def 1) 30-day gap in days’ supply, (def 2) 90-day gap in days’ supply, and (def 3) 30-day gap in days’ supply or reduction in BZD dose. Sensitivity (Se) and specificity (Sp) were computed.

Results: Among 257 eligible patients, 44% were aged 66-74 years, 35% had a Medicare low-income subsidy, 79% were female; and 74% had recorded diagnosis codes for anxiety. Among 47 patients hospitalized at UNC, 33% had BZDs deprescribed based on the EHR; evidence of deprescribing was often documented during inpatient encounters, in medication orders, and using terminology like “discontinue.” Claims definitions 1-3 resulted in deprescribing proportions of 27%, 15%, and 55%, respectively. The Se was highest for the 30-day gap or reduction (def 3, 60%) and lowest for the 90-day gap (def 2, 33%); while Sp was highest for the 90-day gap (def 2, 94%) and lowest for 30-day gap or reduction (def 3, 50%).

Conclusions: BZD deprescribing (based on EHR abstraction) after a fall-related hospitalization occurred in 1 out of 3 older adults. In general, Se of claims algorithms for detecting deprescribing was low ( < 60%); however, Sp of the 90-day gap was >90%. Replication of these findings in other EHRs and for other low-value medications is needed to further guide the use of claims data for deprescribing research.