Algorithmic design choices as determinants of health equity: a causal simulation study of healthcare resource allocation

Training objective is the most impactful modifiable design choice for care management targeting, driving near-zero Black–White equal opportunity difference (EOD: 0.008) in the All of Us cohort.

• Why it matters: Prediction models used for allocating limited healthcare resources may miss patients whose utilization is suppressed by structural barriers, creating equity gaps that are hard to correct after deployment.
• What they did: In a causal simulation using 201,216 adults (NIH All of Us, 2017–2019), logistic regression, random forest, and XGBoost were retrained while varying selection threshold, feature set, and training objective; equity was measured with FNR, EOD, and allocation rate conditional on need.
• The result: Changing the training objective produced the most consistent equity shifts, including a reversal of disparity direction for most groups under the survey-enriched health status condition, while threshold changes left disparities largely unchanged.

AI summary of the abstract · may contain errors · not medical advice · read the paper.