Target specification bias, counterfactual prediction, and algorithmic fairness in healthcare

TL;DR

This paper highlights target specification bias as a key, often overlooked source of bias in healthcare machine learning, emphasizing its impact on prediction accuracy and decision-making, and proposes solutions inspired by measurement science.

Contribution

It identifies target specification bias as a pervasive issue in healthcare ML and suggests methods from metrology to mitigate its effects.

Findings

Target specification bias causes overestimated accuracy and suboptimal decisions.

Addressing this bias improves resource utilization and patient outcomes.

Proposed solutions draw from measurement science to correct bias.

Abstract

Bias in applications of machine learning (ML) to healthcare is usually attributed to unrepresentative or incomplete data, or to underlying health disparities. This article identifies a more pervasive source of bias that affects the clinical utility of ML-enabled prediction tools: target specification bias. Target specification bias arises when the operationalization of the target variable does not match its definition by decision makers. The mismatch is often subtle, and stems from the fact that decision makers are typically interested in predicting the outcomes of counterfactual, rather than actual, healthcare scenarios. Target specification bias persists independently of data limitations and health disparities. When left uncorrected, it gives rise to an overestimation of predictive accuracy, to inefficient utilization of medical resources, and to suboptimal decisions that can harm patients. Recent work in metrology - the science of measurement - suggests ways of counteracting target specification bias and avoiding its harmful consequences.