This simulation study used 201,216 US adults from linked health record and survey data in All of Us, 2017-2019. Logistic regression, random forest and XGBoost models were tested, varying threshold, feature set or training objective one at a time. Fair or poor health exceeded high utilisation among Black participants (33% vs 22%), but not White participants (19% vs 25%). Disparities persisted at every selection threshold; at baseline, uninsured people had the largest equal opportunity difference (-0.148). Outpatient visits dominated the Black-White gap; dropping them gave an EOD of 0.001 (AUC 0.77), a health status objective 0.008 (AUC 0.69). Need rested only on self-reported general health, and results come from simulation, not deployed programmes.
Why it is interesting: Algorithms that allocate scarce care often learn who used care, not who needed it. This study isolates which design choice drives that gap, and finds the training target matters most.
Cite this item
BibTeX
Free sign-up. A digest of the new papers twice a week, grouped by domain.
Other drivers in Data, models and systems
-
Forecasts and their record
How the forecasts are made, how well they explained the past, and how well mortality can be forecast — and by whom.
2 this week -
What linked data reveals
What linked registries make it possible to know about whole populations
14 this week -
AI in medicine
Whether machine learning predicts, detects and decides better than the tools we have, and whether that has yet changed an outcome for a population.
6 this week -
Health systems and access
What the health system can provide and who gets it.
5 this week -
Value in health
What a treatment, health programme or policy is worth against what it costs, and how far the models that decide it can be trusted.
5 this week -
Revisions and new sources
What the figures are built from, what is newly available, and how much the numbers move when they are revised.
1 this week
