Fairness Correction in COVID-19 Predictive Models Using Demographic Optimization: Algorithm Development and Validation Study
Published in Online Journal of Public Health Informatics, 2026
Overview
This paper introduces Demographic Optimization (DemOpts), an in-processing debiasing method for deep learning–based COVID-19 county case forecasting models.
Key Contributions
- Shows that state-of-the-art COVID-19 forecasting models exhibit unequal errors across racial and ethnic groups.
- Proposes DemOpts, which modifies the loss function to penalize statistically significant associations between prediction error and county racial and ethnic composition.
- Demonstrates that DemOpts improves both hard and soft error parity compared with several existing debiasing baselines.
Recommended Citation
Awasthi, N., Abrar, S., Smolyak, D., & Frias-Martinez, V. (2026). Fairness Correction in COVID-19 Predictive Models Using Demographic Optimization: Algorithm Development and Validation Study. Online Journal of Public Health Informatics, 18, e78235. https://doi.org/10.2196/78235
Recommended citation: Awasthi, N., Abrar, S., Smolyak, D., & Frias-Martinez, V. (2026). Fairness Correction in COVID-19 Predictive Models Using Demographic Optimization: Algorithm Development and Validation Study. Online Journal of Public Health Informatics, 18, e78235. https://doi.org/10.2196/78235
Download Paper