Impact of Participation Bias on Disease Prevalence Estimation in the All of Us Research Program: A Case Study of Ischemic Heart Disease and Stroke
Listed in
This article is not in any list yet, why not save it to one of your lists.Abstract
Importance
Disease prevalence estimation is highly sensitive to sample characteristics shaped by recruitment and data collection strategies. Using follow-up study modules that require active participant engagement may introduce participation bias, affecting the accuracy of disease prevalence estimation.
Objective
To estimate the prevalence of ischemic heart disease (IHD) and stroke using electronic health records (EHR) and the self-reported Personal Medical History (PMH) survey collected in the All of Us Research Program.
Design and settings
Cross-sectional study aimed at estimating the prevalence of IHD and stroke among 266,472 participants with EHR in the latest release of the All of Us Registered Tier Curated Data Repository (R2022Q4R9).
Main outcomes and measures
Primary outcomes were IHD and stroke, ascertained using expert-curated diagnostic and procedure codes recorded in EHR. Secondary outcomes were IHD and stroke, ascertained using responses from the PMH survey. To mitigate the impact of participation bias in the PMH survey responses, we applied poststratification weighting based on annual household income and education.
Results
Of the 266,472 participants with EHR, 17,054 (6.4%) were identified as having IHD and 7,461 (2.8%) as having stroke based on the EHR definitions. Among PMH survey respondents, the EHR-based prevalence was lower at 5.6% (95% CI, 5.4-5.7) for IHD and 2.2% (95% CI, 2.1-2.3) for stroke, compared to 7.2% (95% CI, 7.0-7.3) for IHD and 3.3% (95% CI, 3.2-3.4) for stroke among non-respondents. The PMH survey-based prevalence among respondents was 5.9% (95% CI, 5.7-6.0) for IHD and 3.6% (95% CI, 3.5-3.7) for stroke, with higher estimates among non-Hispanic White participants after applying poststratification weights. Conclusion and relevance: Our findings suggest that while the current All of Us cohort with EHR reflects the general US population for IHD and stroke prevalence, participants completing the PMH survey are skewed toward higher socioeconomic status and medical literacy. Future research should refine bias mitigation strategies when using voluntary follow-up data to estimate disease prevalence in this cohort.
Key Points
Question
What is the prevalence of ischemic heart disease (IHD) and stroke in the All of Us Research Program cohort with electronic health records (EHR), and in the subset of these participants who also completed the Personal Medical History (PMH) survey?
Findings
The EHR-based prevalence estimates of IHD and stroke were 6.4% and 2.8%, respectively. They were significantly lower among PMH survey respondents but higher among non-respondents.
Meaning
Participants who complete follow-up study modules, such as the PMH survey, may disproportionately represent those with higher socioeconomic status and better health, potentially leading to an underestimation of IHD and stroke prevalence.
