Authors
Sarah E Hegarty, Kristin A Linn, Hong Zhang, Stephanie Teeple, Paul S Albert, Ravi B Parikh, Katherine Courtright, David M Kent, Jinbo Chen
Published in
Biostatistics (Oxford, England). Volume 27. Issue 1. Jan 20, 2026.
Abstract
The increasing use of algorithm-assisted decision making has generated strong calls for systematic fairness assessment. Equal opportunity, a popular fairness criterion, demands parity in true positive rates (TPRs) across population subgroups. We identify a fundamental but under-recognized weakness: Even when a model perfectly captures underlying risk within each subgroup, TPRs will generally differ at any given threshold whenever the true risk distributions vary across subgroups. Neglecting these distributional differences risks erroneous claims of performance inequity. To correct for this, we propose the adjusted true positive rate (aTPR), which normalizes subgroup TPRs against the risk distribution of a reference subgroup. This corresponding adjusted fairness metric, the aTPR difference, better operationalizes equal treatment for equal risk by evaluating whether individuals with comparable true risks would receive comparable opportunities for high-risk classification, independent of subgroup membership. We validate the metric through simulation studies and demonstrate its application on an electronic health record cohort to evaluate a 6-mo mortality prediction model designed to enhance palliative care referral timing.
PMID:
43034051
Bibliographic data and abstract were imported from PubMed on 15 Nov 2026.
Read full publication at:
Please sign in
to see all details.
Advertisement
Stats
- Recommendations n/a n/a positive of 0 vote(s)
- Views 3
- Comments 0