Net benefit, calibration, threshold selection, and training objectives for algorithmic fairness in healthcare

TL;DR

This paper examines how fairness, calibration, and threshold choices affect healthcare predictive models, finding that fairness-aware training often does not outperform standard methods when considering net benefit and calibration.

Contribution

It provides empirical evidence that incorporating fairness into training does not necessarily improve model performance or net benefit compared to traditional threshold-based approaches in healthcare.

Findings

Fairness-aware training does not improve net benefit over standard methods.

Calibrated models predicting well for all populations are crucial.

Threshold selection aligned with patient preferences is effective regardless of fairness considerations.

Abstract

A growing body of work uses the paradigm of algorithmic fairness to frame the development of techniques to anticipate and proactively mitigate the introduction or exacerbation of health inequities that may follow from the use of model-guided decision-making. We evaluate the interplay between measures of model performance, fairness, and the expected utility of decision-making to offer practical recommendations for the operationalization of algorithmic fairness principles for the development and evaluation of predictive models in healthcare. We conduct an empirical case-study via development of models to estimate the ten-year risk of atherosclerotic cardiovascular disease to inform statin initiation in accordance with clinical practice guidelines. We demonstrate that approaches that incorporate fairness considerations into the model training objective typically do not improve model performance or confer greater net benefit for any of the studied patient populations compared to the use of standard learning paradigms followed by threshold selection concordant with patient preferences, evidence of intervention effectiveness, and model calibration. These results hold when the measured outcomes are not subject to differential measurement error across patient populations and threshold selection is unconstrained, regardless of whether differences in model performance metrics, such as in true and false positive error rates, are present. In closing, we argue for focusing model development efforts on developing calibrated models that predict outcomes well for all patient populations while emphasizing that such efforts are complementary to transparent reporting, participatory design, and reasoning about the impact of model-informed interventions in context.