Sex-specific clinical history improves multimorbidity prediction
In about 155,000 PLCO cohort participants, adding sex-specific clinical history to models improved disease prediction and calibration, especially at higher multimorbidity burdens.
International Journal of Medical Informatics
Researchers evaluated baseline data from approximately 155,000 human participants in the PLCO cohort to determine whether sex-specific clinical history improves disease prediction in multimorbid populations. They developed XGBoost models to identify prevalent diseases, comparing a reference model based on shared predictors and biological sex against an augmented model incorporating sex-specific clinical records. Sex-specific history improved discrimination and calibration, with greater gains at higher multimorbidity burdens. In the three-disease group, the augmented model achieved an AUC of 0.814, a PR-AUC of 0.624, and a Brier score of 0.172. Clinical utility was negligible for single diseases but rose with multimorbidity. Ablation showed that restoring original values drove the gain (ΔAUC = 0.01019) over missingness patterns alone (ΔAUC = 0.00017). Performance gains did not consistently reduce between-sex disparities.
Why it matters
Multimorbidity increases substantially with advancing age, making accurate risk assessment essential for managing older populations. Incorporating sex-specific biology into clinical algorithms can better capture the complex, divergent disease patterns that emerge as women and men age.
Caveats
The study relied on baseline data to model prevalent disease rather than prospective disease onset. Additionally, gains in predictive accuracy did not reliably eliminate disparities between sexes across different disease phenotypes.
The paper
Barcelona Institute for Global Health · Barcelona Supercomputing Center
International Journal of Medical Informatics · 18 Sep 2026