Abstract
INTRODUCTION
Mixed neuropathology is common in Alzheimer’s disease (AD) and related dementias (ADRD), yet joint prediction using multidomain predictors remains limited.
METHODS
We analyzed 1463 autopsied participants from the Religious Orders Study and Rush Memory and Aging Project (ROSMAP) and externally validated the models in 2613 participants from the National Alzheimer’s Coordinating Center (NACC). The primary model was an explainable boosting machine classifier chain predicting seven neuropathologic outcomes. Complementary epidemiologic analyses examined four neuropathologic pathways, isolated versus mixed multi-pathway profiles, and inter-pathology dependence.
RESULTS
Mixed neuropathology was common and heterogeneous. Outcome prevalence ranged from 9.6% to 65.1% in ROSMAP and from 15.6% to 68.3% in NACC. In ROSMAP, the macro-averaged area under the receiver operating characteristic curve (AUROC) and area under the precision-recall curve (AUPRC) were 0.80 (95% confidence interval [CI], 0.74–0.82) and 0.70 (95% CI, 0.65–0.77), respectively; outcome-specific AUROCs ranged from 0.78 to 0.83, AUPRCs from 0.56 to 0.87, and Brier scores from 0.07 to 0.12. In NACC, corresponding macro-averaged AUROC and AUPRC were 0.77 (95% CI, 0.74–0.79) and 0.68 (95% CI, 0.66–0.70), respectively; outcome-specific AUROCs ranged from 0.72 to 0.80, AUPRCs from 0.54 to 0.84, and Brier scores from 0.10 to 0.13. Predictor contributions showed shared and pathology-specific patterns. The 15-predictor model provided a lower burden alternative with reduced performance.
DISCUSSION
Interpretable multi-label prediction using multidomain predictors may help characterize the heterogeneity of mixed neuropathology in AD/ADRD. Prospective evaluation in diverse populations is warranted to assess generalizability.


If you do not see content above, kindly GO TO SOURCE.
Not all publishers encode content in a way that enables republishing at Neuro.vip.

This post is Copyright: | November 19, 2026
Neuro-Dementia