Forecasting Alzheimer's disease progression using non-imaging clinical features and XGBoost
- 1Lane Department of Computer Science and Electrical Engineering, West Virginia University, Morgantown, WV, USA
Abstract
Identifying progressive dementia during its prodromal stage is a challenging yet critical task for healthcare providers. Traditional clinical assessments are often not sensitive enough to detect warning signs of cognitive decline. Machine learning (ML) methods have shown promise in modeling clinical trajectories associated with neurodegenerative disorders such as Alzheimers disease (AD). Validation of these models forecasting abilities, however, is widely lacking in the literature. To address this gap, we established a longitudinal cohort from the National Alzheimers Coordinating Center (n = 15,727) comprising individuals who were cognitively normal (CN) or had mild cognitive impairment (MCI) at baseline. Cognitive status and clinical observations collected over up to 10 visits were used to train several ML classifiers. We employed the gradient-boosted tree ensemble method XGBoost with a custom feature-engineering pipeline. On non-imputed data, our models achieved an AUC-ROC of 0.9447 [0.9417-0.9476] and PR-AUC of 0.8554 [0.8497-0.8611] for the CN[->]MCI task, and an AUC-ROC of 0.9059 [0.8983-0.9134] and PR-AUC of 0.9214 [0.9156-0.9272] for the MCI[->]AD task. In a lead-time forecasting analysis spanning mean follow-up periods of 56.1 months (range, 9.1-214.1) for CN[->]MCI and 42.7 months (range, 9.1-126.1) for MCI[->]AD, the non-imputed models detected future progression before diagnosis in 29.58% [28.29%-30.87%] of CN[->]MCI progressors and 22.41% [21.36%-23.46%] of MCI[->]AD progressors. Among these individuals, the mean lead time was 33.04 months [32.52-33.56] and 17.95 months [17.71-18.19], respectively. False alarm rates among non-progressors were 12.6% [11.5%-13.7%] and 20.0% [18.4%-21.6%], respectively. These findings suggest that longitudinal machine learning models can identify future cognitive decline before diagnosis, indicating clinical utility for early dementia risk stratification.
Author summaryAlzheimers disease (AD) is a progressive and currently incurable neurodegenerative disorder that remains highly difficult to diagnose during its early stages. While modern early-stage dementia research relies on advanced neuroimaging, the high cost of these tests excludes many vulnerable communities. We designed our study to advance scalable, low-cost diagnostic alternatives that can identify individuals at significant risk of cognitive decline, ensuring faster access to emerging therapies.
We examined trajectories of cognitive decline in 15,727 adults (mean age = 71.45 years, SD = 9.06) who had between 2 and 10 follow-up visits. Using 37 demographic, behavioral, clinical, and genetic variables, we modeled cognitively stable and declining trajectories using the XGBoost algorithm. Additionally, we evaluated the models ability to forecast future diagnoses and the impact of longitudinal imputation.
Our models performed well in both classification tasks (AUC-ROC [≥] 0.90).
Imputation did not substantively alter the classifiers performance. In forecasting diagnosis conversion, the models raised alarms before the recorded conversion visit in a subset of progressors, with non-imputed models yielding longer net lead times than imputed models in AD forecasting. This work describes a strategy for handling mixed-length temporal dimensions in longitudinal cohorts and evaluates its prognostic ability.
Related
- No evidence that serum metabolite slopes improve held-out prediction of clinical progression in the Alzheimer’s disease neuroimaging initiative
- Incident versus prevalent progressive supranuclear palsy: analyses of U.S. Alzheimer's Disease Research Centers data indicate clinical and neuropathological features of a relatively common and underrecognized tauopathy in the aging brain.
- Plasma eMTBR-tau as a marker of clinical stage and progression across the Alzheimer's disease continuum
- Sex‐specific brain reserve and progression from mild cognitive impairment to Alzheimer's disease dementia
- Higher plasma oleic acid is associated with slower progression from mild cognitive impairment to Alzheimer’s disease in men and women
BibTeX
@misc{gautier2026forecasting,
title = {{Forecasting Alzheimer's disease progression using non-imaging clinical features and XGBoost}},
author = {A. Gautier and L. Zheng and B. G. Schreurs},
howpublished = {medRxiv},
year = {2026},
month = oct,
doi = {10.64898/2026.10.05.26364229},
url = {https://www.medrxiv.org/content/10.64898/2026.10.05.26364229v1},
note = {CC BY},
}
RIS
TY - UNPB
TI - Forecasting Alzheimer's disease progression using non-imaging clinical features and XGBoost
AU - A. Gautier
AU - L. Zheng
AU - B. G. Schreurs
PY - 2026
DA - 2026/10/05
PB - medRxiv
DO - 10.64898/2026.10.05.26364229
UR - https://www.medrxiv.org/content/10.64898/2026.10.05.26364229v1
AB - Identifying progressive dementia during its prodromal stage is a challenging yet critical task for healthcare providers. Traditional clinical assessments are often not sensitive enough to detect warning signs of cognitive decline. Machine learning (ML) methods have shown promise in modeling clinical trajectories associated with neurodegenerative disorders such as Alzheimers disease (AD). Validation of these models forecasting abilities, however, is widely lacking in the literature. To address this gap, we established a longitudinal cohort from the National Alzheimers Coordinating Center (n = 15,727) comprising individuals who were cognitively normal (CN) or had mild cognitive impairment (MCI) at baseline. Cognitive status and clinical observations collected over up to 10 visits were used to train several ML classifiers. We employed the gradient-boosted tree ensemble method XGBoost with a custom feature-engineering pipeline. On non-imputed data, our models achieved an AUC-ROC of 0.9447 [0.9417-0.9476] and PR-AUC of 0.8554 [0.8497-0.8611] for the CN[->]MCI task, and an AUC-ROC of 0.9059 [0.8983-0.9134] and PR-AUC of 0.9214 [0.9156-0.9272] for the MCI[->]AD task. In a lead-time forecasting analysis spanning mean follow-up periods of 56.1 months (range, 9.1-214.1) for CN[->]MCI and 42.7 months (range, 9.1-126.1) for MCI[->]AD, the non-imputed models detected future progression before diagnosis in 29.58% [28.29%-30.87%] of CN[->]MCI progressors and 22.41% [21.36%-23.46%] of MCI[->]AD progressors. Among these individuals, the mean lead time was 33.04 months [32.52-33.56] and 17.95 months [17.71-18.19], respectively. False alarm rates among non-progressors were 12.6% [11.5%-13.7%] and 20.0% [18.4%-21.6%], respectively. These findings suggest that longitudinal machine learning models can identify future cognitive decline before diagnosis, indicating clinical utility for early dementia risk stratification. Author summaryAlzheimers disease (AD) is a progressive and currently incurable neurodegenerative disorder that remains highly difficult to diagnose during its early stages. While modern early-stage dementia research relies on advanced neuroimaging, the high cost of these tests excludes many vulnerable communities. We designed our study to advance scalable, low-cost diagnostic alternatives that can identify individuals at significant risk of cognitive decline, ensuring faster access to emerging therapies. We examined trajectories of cognitive decline in 15,727 adults (mean age = 71.45 years, SD = 9.06) who had between 2 and 10 follow-up visits. Using 37 demographic, behavioral, clinical, and genetic variables, we modeled cognitively stable and declining trajectories using the XGBoost algorithm. Additionally, we evaluated the models ability to forecast future diagnoses and the impact of longitudinal imputation. Our models performed well in both classification tasks (AUC-ROC [≥] 0.90). Imputation did not substantively alter the classifiers performance. In forecasting diagnosis conversion, the models raised alarms before the recorded conversion visit in a subset of progressors, with non-imputed models yielding longer net lead times than imputed models in AD forecasting. This work describes a strategy for handling mixed-length temporal dimensions in longitudinal cohorts and evaluates its prognostic ability.
ER -