Evaluating Long-term Health Disparity Impacts of Clinical Algorithms Using a Patient-level Simulation Framework.

Abstract

Objective: This study applies a simulation framework to evaluate the long-term effects of omitting race from a colon cancer decision algorithm for adjuvant chemotherapy, assessing impacts on health outcomes, costs, and disparities while accounting for measurement errors across racial groups.

Methods: We developed a patient-level state-transition model using electronic health records from a large Southern California health system to project outcomes for 4,839 adults with stage II and III colon cancer post-surgery. We compared 30-year quality-adjusted life-years (QALYs), healthcare costs, and QALY distribution among racial groups under three chemotherapy treatment scenarios: 1) current practice, 2) treatment guided by an algorithm that includes race, and 3) the same algorithm with race omitted. An additional health state addressed racial bias in cancer recurrence ascertainment, and probabilistic sensitivity analysis (PSA) assessed uncertainty.

Results: The clinical algorithm, compared to current practice, could improve average health by 0.048 QALYs and reduce racial health disparity by 0.20 QALYs at an incremental cost of $3,221, with the disparity gap decreasing in 96% of PSA iterations. Omitting race showed minimal effects on overall health or costs but resulted in 13% fewer Black patients receiving treatment, decreasing their QALYs by 0.07 and widening the disparity gap by 0.13 QALY. Health disparity increased in 94% of PSA iterations.

Conclusions: A cancer decision algorithm can improve population health and reduce health disparities, but omitting race may harm disadvantaged groups and limit reductions in disparities. Patient-level simulations can be routinely used to evaluate the potential health disparity impacts of algorithms before implementation.