EHR-based vs population-based CVD risk predictions for older patients with diabetes

点击翻译按钮获取中文摘要

Our application entitled “Fair Risk Predictions for Underrepresented Populations using Electronic Health Records” responds to NOT-OD-21-094: “Administrative Supplements to Support Collaborations to Improve the Artificial Intelligence/Machine Learning [AI/ML]-Readiness of NIH-Supported Data”, and supplements our parent NIA R01 grant (R01 AG065330) entitled “EHR-based vs population-based CVD risk predictions for older patients with diabetes”. The overarching goal of the parent grant is to develop individualized absolute risk predictions of Cardiovascular diseases for older patients with diabetes using EHRs. We propose the supplement study, in parallel, to 1) investigate under-representations (bias) of racial/ethnic minorities and patients with disadvantaged Social Determinants of Health (SDOHs) in EHRs, 2) develop fairness-aware EHR prediction methods; and 3) share the simulated EHR datasets, linked SDOH datasets along with the developed fair risk prediction tools to inspire and enable the AI/ML research community for further investigations of fair EHR algorithms. During the implementation of our parent R01 project, we found emerging evidence of under- sampling bias in EHRs for racial/ethnic minorities and patients with disadvantaged SDOHs. Such patients are more likely to visit multiple institutions to receive care, and often receive fewer diagnostic tests and medications in the EHR data of a single institution. We hypothesize that racial/ethnic minorities and patients with disadvantaged SDOHs are under-represented with smaller sample sizes, insufficient diagnostics and laboratory information, and less frequent encounters in EHRs. Consequently, we hypothesize that EHR-based risk prediction models (including conventional linear models and modern AI/ML methods) ignoring the unbalanced samplings will have less-accurate predictions for these under-represented patient populations. Little or no work has been done to systematically investigate the impact of these biases. We then propose to develop fairness improvement prediction approaches for EHRs. Upon the supplement project completion, the developed fair predictions will lay the groundwork and provide resources for the broader AI/ML research community for developing fair predictions to advance disease predictions and detections for racial/ethnic minorities and patients with disadvantaged SDOHs.