Using Explainable AI to Cross-Validate Socio-economic Disparities Among Covid-19 Patient Mortality

Socioeconomic status significantly affects COVID-19 survival: The study found that Medicare financial class was among the highest-impact factors in predicting patient mortality, indicating that a patient's insurance type and economic circumstances played a crucial role in COVID-19 outcomes—a disparity that deserves policy attention.

AI transparency tools can reveal hidden biases in healthcare: By using explainable AI methods (SHAP and LIME), researchers were able to identify which factors the prediction model relied on most heavily, making it possible to spot and investigate potential inequities that might otherwise remain hidden in traditional "black box" algorithms.

Multiple explanation methods strengthen confidence in findings: The paper demonstrates that using two different explainability techniques (SHAP and LIME) that reached similar conclusions about feature importance provides stronger evidence that the identified disparities are real patterns rather than artifacts of a single analytical approach.

Age and gender also emerged as significant mortality factors: Beyond socioeconomic status, the model identified older age and gender as having high impact on COVID-19 mortality predictions, suggesting these demographic factors warrant further investigation to understand whether they reflect biological differences, healthcare access gaps, or other systemic inequities.

Explainable AI enables accountability in healthcare algorithms: This work shows how XAI methods can make AI decision-making transparent enough for healthcare providers and policymakers to understand why predictions are made and identify whether algorithms are perpetuating or could help address existing health disparities.