(B-226) Characterization of Confounding Covariate Data Capture in the U.S. Food and Drug Administration’s Sentinel Distributed Database

Background: The U.S. Food and Drug Administration’s Sentinel Distributed Database (SDD) is a collection of electronic healthcare datasets comprising mainly administrative claims data. A known limitation of claims data is its inadequate capture of key confounding covariates, which could lead to biased study estimates potentially negatively impacting study finding validity.

Objectives: To estimate the incidence and prevalence of selected confounding covariates and vitals data in the SDD to characterize the capture of those variables over time.

Methods: Fourteen Sentinel Data Partners (DPs) contributed data from 2010 to 2023. Healthcare encounters among members of all ages (18+ years for calculatable Body Mass Index (BMI) cohorts) with at least 365 days of prior enrollment in medical and drug plans, were eligible to contribute to cohorts. Cohorts with prevalent and incident confounding conditions (smoking, obesity, overweight, alcohol and drug abuse/dependence, etc.) were described using diagnosis and procedure codes, measurements from vitals data (height and weight for BMI, blood pressure (BP), tobacco use) and demographic data (race). Incident cohort episodes were excluded if the defining condition was present in the prior 365 days. The number of qualifying occurrences of the cohort-defining variables were stratified by year and sex.

Results: About 260 million members from 14 DPs overall, including 8.3 million members from seven DPs contributing vitals data, were eligible for cohort entry. Incidence and prevalence of obesity, overweight, smoking, and BMI from diagnosis codes steadily increased over time with slight dips occurring in 2015-2016 and 2020-2021. The highest values observed were incidence of 923 and prevalence of 2,649 per 10,000 eligible members for obesity and BMI, respectively, in 2023. Incidence and prevalence of drug and alcohol abuse/dependence remained below 204 per 10,000 eligible members, with the highest value being prevalence of alcohol abuse/dependence in 2023. Prevalence of diastolic/systolic BP and BMI calculated from vitals data peaked in 2013 at 5,735 and 5,003 per 10,000 eligible members, respectively, declining afterward. A slightly higher prevalence of alcohol and drug abuse/dependence was observed in males compared to females. Race was unknown in 35% of members with claims data from January 2020-December 2021.

Conclusions: Capture of key confounding covariates in the SDD has improved over time; however, low prevalence from this study compared to national estimates suggests that these covariates remain insufficiently captured. Incorporation of additional data sources with novel statistical methods may enhance the SDD’s capability for conducting drug safety studies.