Development, validation, and clinical utility of a risk prediction model to identify older women with dementia for...

Accurately estimating one-year mortality risk in older women with dementia can inform clinical decision-making, facilitate timely advanced care planning, and optimise palliative care delivery. This study aimed to develop, validate, and assess the clinical utility of a prediction model for one-year all-cause mortality in this population using a nationally representative Australian cohort.

This prognostic study utilised data from the 1921-26 cohort of the nationally representative, population-based Australian Longitudinal Study on Women's Health (ALSWH) and linked national and state-based administrative health records. Candidate predictors were identified through a systematic review and expert consultation, then refined using a data-driven statistical approach. A multivariable binary logistic regression model was developed and validated to predict one-year all-cause mortality.

The analysis included 1576 older women with dementia (mean age, 72.6 ± 1.5 years). The model demonstrated good discrimination (AUC: 75.1 %, 95 % CI: 72.7 %-77.5 %) and excellent calibration (slope = 1.00, 95 % CI: 0.87-1.13; intercept = 0.00, 95 % CI: 0.11 - 0.11). Model validation using both 10-fold cross-validation and 1000 bootstrap iterations showed minimal optimism in its predictive performance, with AUC optimism of 0.0047 and 0.0042, respectively. Decision curve analysis indicated a net benefit across probability thresholds from 0.24 to 0.88, supporting the model's clinical utility for guiding palliative care decisions.

This prediction model, incorporating readily available predictors, demonstrated compelling performance and clinical utility for identifying older women with dementia at high risk of one-year mortality. The model has the potential to facilitate timely palliative care interventions and is publicly accessible via a web-based calculator. Further external validation in diverse populations and healthcare settings is warranted to confirm its generalisability.