Addressing algorithmic bias in lung cancer screening eligibility
Menée aux Etats-Unis à partir de données portant sur une cohorte de 52 667 adultes, cette étude analyse plusieurs stratégies de sélection pour le dépistage du cancer du poumon et examine comment les biais des algorithmes de sélection peuvent être résolus pour améliorer l'équité au niveau de l'éligibilité
Background : The United States Preventive Services Task Force (USPSTF) lung cancer screening eligibility guidelines and proposed risk models have been developed using data predominantly from White populations. Studies show that these eligibility strategies perform inconsistently across racially diverse populations, suggesting evidence of algorithmic bias. We assessed several lung cancer screening eligibility strategies and explored how algorithmic bias can be resolved to improve equity in eligibility.
Methods : Using the Southern Community Cohort Study, a large US study of predominantly Black/African American (AA) individuals, we evaluated the performance of eight existing lung cancer screening eligibility strategies (USPSTF 2021, American Cancer Society 2023 recommendations, USPSTFSmokeDuration, PLCOm2012, PLCOm2012NoRace, PLCOm2012Update, LCRAT, and LCDRAT) and two new race-aware strategies proposed by our team (USPSTFRaceSpecific and PLCOm2012RaceSpecific).
Results : Among 52,667 adults (65% Black/AA, 31% White, 4% Multiracial/Other) with a smoking history, 1,689 developed lung cancer over 15 years. Almost all screening strategies identified fewer Black/AA participants who developed lung cancer as eligible for screening versus their White counterparts (SensitivityBlack/AA 0.46-0.73 vs SensitivityWhite: 0.72-0.80). Racial eligibility disparities were not resolved by removing race, removing the ‘years since quit’ criterion, or using uniform risk-thresholds. Replacing pack-years with smoke duration improved equity but overinflated the false positive rate (FPRBlack/AA: 0.71; FPRWhite: 0.61). Instead, race-aware approaches that tailored eligibility thresholds by race yielded the best sensitivity-specificity trade-off and minimized inequities (SensitivityBlack/AA: 0.71-0.73 vs SensitivityWhite: 0.72-0.74; FPRBlack/AA: 0.49-0.50; FPRWhite: 0.50-0.53).
Conclusion : Our findings suggest that race-aware approaches are necessary to address algorithmic bias and ensure equitable opportunities for lung cancer screening.
Journal of the National Cancer Institute , article en libre accès, 2026