Paper Type
Complete
Abstract
Healthcare organizations increasingly rely on predictive analytics to guide intervention decisions, yet fairness risks often emerge not in prediction accuracy but in how risk scores are translated into operational action. This study introduces CARE–Fair+, a decision-time governance artifact embedded within hospital readmission workflows. Rather than modifying predictive models, the artifact operates at the decision layer, monitoring subgroup disparities as intervention thresholds are applied and recommending constrained adjustments that preserve minimum clinical detection performance. Via a simulation experiment using approximately 13,000 admissions derived from MIMIC-III, we demonstrate that unconstrained fairness optimization collapses operational usefulness, while constrained threshold configuration substantially reduces disparity without undermining detection sensitivity. The results show that fairness can be implemented as a bounded decision parameter rather than treated as a static model property. By reframing fairness as configurable governance embedded within workflow execution, this research advances digital transformation in healthcare.
Paper Number
1730
Recommended Citation
Irani, Mina and Hevner, Alan, "CARE–Fair+: Decision-Time Governance for Just Clinical Interventions" (2026). AMCIS 2026 Proceedings. 13.
https://aisel.aisnet.org/amcis2026/sig_health/sig_health/13
CARE–Fair+: Decision-Time Governance for Just Clinical Interventions
Healthcare organizations increasingly rely on predictive analytics to guide intervention decisions, yet fairness risks often emerge not in prediction accuracy but in how risk scores are translated into operational action. This study introduces CARE–Fair+, a decision-time governance artifact embedded within hospital readmission workflows. Rather than modifying predictive models, the artifact operates at the decision layer, monitoring subgroup disparities as intervention thresholds are applied and recommending constrained adjustments that preserve minimum clinical detection performance. Via a simulation experiment using approximately 13,000 admissions derived from MIMIC-III, we demonstrate that unconstrained fairness optimization collapses operational usefulness, while constrained threshold configuration substantially reduces disparity without undermining detection sensitivity. The results show that fairness can be implemented as a bounded decision parameter rather than treated as a static model property. By reframing fairness as configurable governance embedded within workflow execution, this research advances digital transformation in healthcare.
When commenting on articles, please be friendly, welcoming, respectful and abide by the AIS eLibrary Discussion Thread Code of Conduct posted here.
Comments
SIGHealth