(C-128) Prevalence of Key Confounding Covariates in a Commercial Electronic Health Record Platform

Background: Integrating claims and electronic health record (EHR) data can help address uncertainties in claims-only analyses by better characterizing confounding factors insufficiently captured in claims but captured in EHR, and correcting effect estimates from claims-only analyses using confounder information from EHR. The availability of large scale commercial EHR platforms may provide an opportunity to address some of the challenges related to confounding adjustment in medical product safety evaluations.

Objectives: To describe the capture of key confounding covariates within a large commercial EHR platform to explore its usefulness as a supplementary data source for medical product safety surveillance.

Methods: The annual prevalence of obesity, body mass index (BMI), smoking, diastolic and systolic blood pressure (BP) records across individuals of all ages was calculated using a subset of the TriNetX Live™ USA Network EHR database without date shifting from 2017 to 2023. BMI and BP data were sourced from clinical data, while obesity and smoking data were derived from diagnosis, procedure, and medication codes. The denominator cohort included individuals with a visit in the year of interest and at least one other visit in the prior year. Individuals were eligible to be in a numerator if they had evidence of the characteristic of interest in the measurement year of interest, and evidence of a visit in the year prior to their first instance of the characteristic in the window of interest.

Results: Over the study period, the number of eligible individuals showed a general upward trend, peaking in 2022 at 17,240,150 before declining in 2023. The annual prevalence remained generally consistent over time for BMI (50%-54%), and systolic BP records (54%-57%) up until 2022, followed by a notable increase in 2023 (BMI=61%, systolic BP=63%). The prevalence of diastolic and systolic BP records showed slight discrepancies, with systolic BP record exhibiting higher prevalence compared to diastolic BP record starting 2019 (diastolic vs. systolic BP, 54% vs. 56% in 2019 and 58% vs. 63% in 2023). The prevalence of obesity and smoking stayed relatively consistent at 11%-14% and 10%-12%, respectively.

Conclusions: The study showed the capture level of key confounding covariates in a commercial EHR platform. Additional evaluation and comparison of different data sources, including mechanisms of missing data, are warranted to assess their ability to capture these confounding covariates for use in medical product safety surveillance. Moreover, characteristics of these data sources should be carefully evaluated when planning future safety surveillance to address the specific needs of each study question.