Predicting postoperative chronic opioid use with fair machine learning models integrating multi-modal data sources: a demonstration of ethical machine learning in healthcare DOI
Nidhi Soley, Ilia Rattsev, Traci J. Speed

et al.

Journal of the American Medical Informatics Association, Journal Year: 2025, Volume and Issue: unknown

Published: March 27, 2025

Abstract Objective Building upon our previous work on predicting chronic opioid use using electronic health records (EHR) and wearable data, this study leveraged the Health Equity Across AI Lifecycle (HEAAL) framework to (a) fine tune previously built model with genomic data evaluate performance in (b) apply IBM’s AIF360 pre-processing toolkit mitigate bias related gender race various fairness metrics. Materials Methods Participants included approximately 271 All of Us Research Program subjects EHR, wearable, data. We fine-tuned 4 machine learning models new dataset. The SHapley Additive exPlanations (SHAP) technique identified best-performing predictors. A preprocessing boosted by race. Results genetic enhanced from prior model, area under curve improving 0.90 (95% CI, 0.88-0.92) 0.95 0.89-0.95). Key predictors Dopamine D1 Receptor (DRD1) rs4532, general type surgery, time spent physical activity. reweighing applied stacking algorithm effectively improved model’s across racial groups without compromising performance. Conclusion 2 dimensions HEAAL build a fair artificial intelligence (AI) solution. Multi-modal datasets (including data) applying mitigation strategies can help more fairly accurately assess risk diverse populations, promoting healthcare.

Language: Английский

Predicting postoperative chronic opioid use with fair machine learning models integrating multi-modal data sources: a demonstration of ethical machine learning in healthcare DOI
Nidhi Soley, Ilia Rattsev, Traci J. Speed

et al.

Journal of the American Medical Informatics Association, Journal Year: 2025, Volume and Issue: unknown

Published: March 27, 2025

Abstract Objective Building upon our previous work on predicting chronic opioid use using electronic health records (EHR) and wearable data, this study leveraged the Health Equity Across AI Lifecycle (HEAAL) framework to (a) fine tune previously built model with genomic data evaluate performance in (b) apply IBM’s AIF360 pre-processing toolkit mitigate bias related gender race various fairness metrics. Materials Methods Participants included approximately 271 All of Us Research Program subjects EHR, wearable, data. We fine-tuned 4 machine learning models new dataset. The SHapley Additive exPlanations (SHAP) technique identified best-performing predictors. A preprocessing boosted by race. Results genetic enhanced from prior model, area under curve improving 0.90 (95% CI, 0.88-0.92) 0.95 0.89-0.95). Key predictors Dopamine D1 Receptor (DRD1) rs4532, general type surgery, time spent physical activity. reweighing applied stacking algorithm effectively improved model’s across racial groups without compromising performance. Conclusion 2 dimensions HEAAL build a fair artificial intelligence (AI) solution. Multi-modal datasets (including data) applying mitigation strategies can help more fairly accurately assess risk diverse populations, promoting healthcare.

Language: Английский

Citations

0