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

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.

APPRAISE-AI Tool for Quantitative Evaluation of AI Studies for Clinical Decision Support

“…For example, model explainability remains a highly controversial topic among clinical and AI experts, with no universally accepted method for providing robust explanations for individual-level predictions. 33 Similarly, there is no clear consensus on the best strategy to incorporate algorithmic fairness considerations 34 , 35 ; therefore, APPRAISE-AI does not assign scores to any particular approach. Instead, the emphasis is placed on conducting bias assessments (item 17) so that researchers can examine the efficacy of their fairness strategies, regardless of the approach used.…”

Decision-Making under Miscalibration

“…The effect of discrimination on the Net Benefit was explored in [Van Calster et al, 2013, Vickers andElkin, 2006]. Recently, [Pfohl et al, 2022b, Pfohl et al, 2022a consider the impact of different fairness interventions on clinical utility.…”