RaceMED: A Race-Aware Approach to Accurate and Fair Medication Recommendation

FairGlucose: A CGM Fairness Benchmark Reveals Subgroup Disparities Hidden in Population-Level Validation

It is established that population-level validation alone is insufficient for equity assessment of digital health AI, motivating subgroup-disaggregated reporting as a default standard, and subgroup-disaggregated reporting as a default standard for personalized configurations.

Multidimensional Resource Disparities by Race During COVID-19: The Role of Chronic Illness, Disability, and Mental Health

Reducing Racial Disparities in Stroke Thrombolysis Using Fairness-Aware Machine Learning.

Racial and ethnic disparities in the administration of intravenous thrombolysis (IVT) for acute ischemic stroke (AIS) remain persistent, raising concerns that bias-both systemic and algorithmic-may influence treatment decisions. While the NIH Stroke Scale (NIHSS) is intended as an objective tool for assessing stroke severity, its application may not be uniform across patient groups. Using differential item functioning (DIF) analysis on a national inpatient cohort of 983,785 patients, we found that minority patients required higher NIHSS scores than White patients to receive IVT, indicating potential bias in score interpretation. To address this, we applied a fairness-aware machine learning framework using counterfactual adjustment to account for historical inequities embedded in clinical data. This approach reduced treatment disparities by 64.8% without compromising predictive accuracy (AUC = 0.89). Our findings illustrate how standardized tools can contribute to inequity and demonstrate the potential of ethical AI to mitigate disparities in real-world stroke care.

FairMoE-Health: Fairness-Aware Mixture of Experts for Equitable Multimodal Clinical Prediction.

Mixture-of-Experts (MoE) models for multi modal clinical prediction route patients to specialized ex pert networks based on input modalities, but we show that this routing mechanism introduces a previously un recognized source of demographic bias: because modality availability (e.g., whether a chest X-ray exists) correlates with race, gender, and insurance status, the gating network learns routing patterns that systematically differ across demographic groups. We propose FairMoE Health, a fairness-aware MoE framework that intervenes at three architectural levels: adversarial debiasing of encoded representations reduces demographic information before it reaches the gating network; adversarial debiasing of gating weights further discourages demographic leakage into routing decisions; and equalized odds regularization provides a prediction-level safety net. We introduce the Routing Disparity (RD) metric to quantify demographic imbalance in expert assignments. On three clinical prediction tasks from MIMIC-IV (mortality, length-of-stay, readmission), FairMoE-Health consistently reduces racial Equalized Odds Difference (EOD) and RD across all three tasks, with only small AUROC decreases. The Routing Disparity metric and multi-level debiasing framework introduced here generalize to MoE systems operating on demographically heterogeneous populations, providing both an audit tool and an architectural intervention for a bias mechanism that existing fairness methods leave unaddressed.

A framework for fair and robust clinical risk prediction through collaborative learning and localized uncertainty quantification

Clinical prediction models trained on heterogeneous populations may exhibit disparities in predictive performance across patient subgroups, potentially reflecting differences in data availability, feature completeness, and representation in training data. Existing fairness-aware approaches can introduce performance-fairness trade-offs and commonly optimize group-level disparities without explicitly accounting for their subgroup-specific sources, limiting both their effectiveness and their clinical interpretability. This dissertation makes two methodological contributions. The first is a collaborative learning framework that treats demographic subgroups as clients and aggregates their model parameters using strategies that encode different assumptions about group contribution. Rather than optimizing an explicit group-fairness constraint, the framework preserves and integrates subgroup-specific information to improve equity while largely maintaining subgroup predictive performance. The second contribution is the localized conformal classifier, a neighborhood-adaptive conformal prediction framework for instance-level uncertainty quantification. By calibrating prediction sets according to the density and composition of a patient's local neighborhood in the feature space, the method communicates both the model's prediction and the uncertainty associated with that individual prediction. This information is particularly consequential when predictions inform clinical decision-making. Both contributions are evaluated across three clinical prediction tasks: 30-day mortality in ICU patients with sepsis, treatment non-completion in patients with substance use disorder, and hospital readmission after surgical resection in patients with colorectal cancer. Compared with reweighting and constrained optimization, collaborative learning achieves more favorable fairness-performance trade-offs with less degradation in predictive performance, and its variants consistently appear on the Pareto frontier across all three prediction tasks. A novel difficulty decomposition framework, derived from the localized conformal classifier's calibration scores, distinguishes sources of uncertainty arising from the local neighborhood from those specific to an individual prediction. SHAP attribution reveals how collaborative learning and fairness interventions alter the feature-attribution patterns in the underlying prediction models. Surrogate regression trees trained on the localized conformal classifier's difficulty scores relate neighborhood and instance difficulty to clinical features observable at the point of care. Clinical features remain strongly associated with instance-level difficulty under collaborative learning, whereas these associations weaken substantially after reweighting and are absent in some analyses. Across all three clinical contexts, local outcome heterogeneity is the most consistent driver of neighborhood-level predictive difficulty, while proximity to the decision boundary is the principal driver of instance-level difficulty. Together, these contributions provide a framework for developing and evaluating clinical prediction models that are not only accurate, but also more equitable across patient groups and more informative about the reliability of individual predictions.

Quantifying Error Disparities in Population Health Models

A large-scale audit of sociodemographic error disparities in both lexical- and transformer-based models of county-level health outcomes, over a dataset cover billions of community-mapped messages highlights a critical accuracy–fairness trade-off in community-level models for public health tasks.