Senior Scientific Investigator Kaiser Permanente Washington Health Research Institute University of Washington, Seattle, Washington Seattle, United States
Background: Polypharmacy is common in older adults and associated with many adverse outcomes. Deprescribing inappropriate medications holds promise for addressing this problem. Electronic health record (EHR) data are increasingly used for studies of deprescribing, but little is known about how best to measure key deprescribing variables from EHR data.
Objectives: To develop, operationalize, and examine definitions for key variables including chronic medication use and medication discontinuation from EHR medication orders and dispensing data.
Methods: We conducted a methodologic exploration using data from 5 US healthcare systems. The population was older adults age 65+ engaged with the healthcare system in 2017-2019, without dementia or serious mental illness, who had chronic use of benzodiazepines or other sedative-hypnotics (Z-drugs). Chronic use was defined as 3 dispensings or orders/refills in a 100-day period. Definitions for discontinuation examined 1) gaps in use of varying lengths or 2) non-use (having no medication on hand) at a fixed time point during one year follow-up.
Results: Prevalence of chronic benzodiazepine/Z-drug use ranged from 1.5%-2.6% at these healthcare systems in 2018. Cohorts included 431-2,122 people; the mean age was 71.5-73.3 years, and the majority were White. The proportion of people with an apparent gap of ≥30 days in benzodiazepine/Z-drug use during follow-up was 32-52% at sites with dispensing data vs. 12% at a site using orders data. For a gap of ≥180 days, the prevalence was 6-15% at sites with dispensing data and 6% with orders. The proportion with no medication on hand at 180 days after cohort entry ranged from 19-36% at sites with dispensings vs. 16% at a site using orders. Requiring a 30-day “halo” with no dispensing/order around the fixed point yielded substantially lower discontinuation estimates. At one site with orders and dispensing data for the same population, estimates for chronic use were considerably higher and for discontinuation were lower when based on orders data compared to dispensings.
Conclusions: Using medication orders yielded lower estimates of medication discontinuation than dispensings. Estimates for discontinuation varied according to definitions used, especially when length of medication gap was involved. Future research should explore actual medication taking behavior in individuals with apparent discontinuation and examine the performance of these measures for other drug classes.
