Abstract 4367357: Enrollment strategies that yield a representative and generalizable trial population for hypertension research.

Abstract

Background: Clinical trial participants often do not represent patients with hypertension in the real world, which limits the generalizability of findings. In our pragmatic multicenter hybrid effectiveness implementation randomized controlled trial, we used electronic health records (EHR) to automate the identification of patients aged 65 years and older with uncontrolled hypertension. Hypothesis: We hypothesized that the enrolled participants would be similar to the clinic populations due to the use of EHR, broader inclusion criteria, remote nature of the study, and low trial burden on participants. Methods: We compared enrolled participants with those who were eligible but did not enroll in the study. Socio-demographic and clinical characteristics were extracted from the EHR. Univariate ANOVA tests were performed to compare characteristics across three groups: enrolled, eligible but declined participation, and eligible but unreachable. We conducted a multiple logistic regression with stepwise feature selection against the enrollment indicator to demonstrate representativeness. Results: From the pool of >60,000 patients who had clinic visits during enrollment, 10,526 patients were automatically identified, and 5,311 were eligible and included in these analyses. Among those eligible, 18.8% enrolled, 53.5% declined to participate, and 27.6% were unreachable. The mean age of the enrolled patients was slightly less than that of patients who were not enrolled (73(5.6) vs 75 (6.2) years). Sex distribution was similar across all three eligible groups. The proportion of non-white race/ethnicity among those who declined was slightly higher. The qualifying systolic blood pressure (clinic blood pressure used for inclusion criteria) was lower in the enrolled compared to the unenrolled groups (151 vs 154 mmHg). The enrolled group had a lower proportion of patients with chronic kidney disease, stroke, or a history of cigarette smoking compared to the unenrolled group. When analyzing the entire unenrolled cohort (i.e., declined and unreachable) using multivariable logistic regression, a higher number of antihypertensive medications was associated with trial enrollment. Conclusion: A pragmatic approach and automated identification of potential participants through EHR enrollment yielded a cohort of participants very similar to the eligible but unenrolled clinic population, which will enhance the generalizability of the trial’s findings.