Validity of pharmaceutical claims data for determining medication exposure in long-term care facilities

Background: Although pharmaceutical claims data are commonly used to determine medication exposure in pharmacoepidemiological studies, little is known about the concordance of claims data with medicines administration charts in long-term care facilities (LTCFs).

Objectives: To compare the accuracy of pharmaceutical claims data with data extracted from medication administration charts for individuals in LTCFs.

Methods: Secondary analyses of the Generating evidence on Resistant bacteria in the Aged Care Environment (GRACE) study (n=279 residents from 5 South Australian LTCFs) were conducted. Details of medications charted for regular administration at study entry were extracted from medication administration charts and coded using the Anatomical Therapeutic Chemical (ATC) classification system. Pharmaceutical claims for government-subsidized medications dispensed in the 0-30, 0-60, 0-90, 0-120 and 0-180 days prior to study entry were compared to medication chart data (considered the reference standard). Sensitivity, specificity, and positive predictive value (PPV) and 95% confidence intervals (CIs) were determined at the 3- and 7-digit ATC code levels.

Results: 149 residents (median age 89.2 years (IQR 83.9-93.4), n=109 (73.2%) female) with both medication charts and claims data available were included. The proportion of PPVs ≥75% for medications at the 3-digit ATC level was 91.4% and 87.4% using a lookback of 30 and 180 days for pharmaceutical claims, respectively. Sensitivity varied considerably by medication class and generally increased with greater lookback periods. Using a 180-day lookback for claims, sensitivity was highest for diabetes agents, antihypertensives, anti-Parkinson drugs, psychoanaleptics and drugs for airways disease (all ≥90%) and was lowest for vitamins (1.4%, 95%CI 0-7.7) and mineral supplements (10.3%, 95%CI 2.9-24.2). Specificity was ≥90% for all medications other than antibacterials, analgesics and psycholeptics. Using a 120-day lookback for claims, sensitivity/PPV at the 7-digit ATC level was highest for warfarin, clopidogrel, frusemide and buprenorphine (all ≥90%) and sensitivity was lowest for docusate/senna and cholecalciferol (both 0%). Sensitivity/PPV for risperidone was 63.6%/77.8%. Sensitivity increased for thyroxine, digoxin, and allopurinol as the claims lookback period increased to 180 days but remained similar for other medications.

Conclusions: Pharmaceutical claims data has good accuracy for determining exposure to prescription medications in LTCFs. Our findings suggest lookback periods of 90 to 120-days are generally sufficient for determining medication exposure using pharmaceutical claims, although longer lookbacks may be required for medications supplied in large pack sizes.