Accurate prediction of progression from mild cognitive impairment (MCI) to Alzheimer’s disease (AD) over a 24-month interval is important for prognosis, clinical management, and selection for therapeutic trials. Standard memory clinic practice routinely uses cognitive and functional assessments, but the relative predictive value of individual domains and instruments has been uncertain. This study evaluated a large ADNI cohort to compare the predictive performance of demographic/genetic variables, global cognitive scales, domain-specific cognitive measures, functional instruments, and Everyday Cognition (ECog) questionnaires for 24-month conversion from MCI to AD.
The analysis included 2,430 participants with baseline MCI from the Alzheimer’s Disease Neuroimaging Initiative (ADNI). Participants were classified by whether they progressed to AD within 24 months. Predictive modelling used Extreme Gradient Boosting (XGBoost) trained with repeated stratified 5-fold cross-validation and 10 repetitions.
Predictor sets compared in models included:
Model performance was assessed primarily by area under the receiver operating characteristic curve (AUC), with sensitivity and specificity reported for the best-performing multivariable combination.
The baseline clinical model including age, sex, education, and APOE ε4 status achieved an AUC of 0.692.
Domain-level performance:
Combination models:
Individual tests within domains:
Overall, episodic memory measures provided the strongest single-domain predictive power for 24-month progression from MCI to AD, while functional assessment added substantial complementary predictive information.
In this ADNI cohort analysis, episodic memory testing emerged as the most informative cognitive domain for predicting short-term (24-month) conversion from MCI to AD. Functional status—measured here by the Functional Activities Questionnaire—performed nearly as well alone and meaningfully improved prediction when combined with memory and executive measures.
These findings suggest that a streamlined assessment battery focused on sensitive episodic memory tests plus a brief functional instrument could enable efficient risk stratification in memory clinic settings and improve selection criteria for AD clinical trials. The high AUC achieved by the combined model (0.943) indicates strong discriminative ability in this cohort when multiple complementary domains are included.
This report is a medRxiv preprint and has not undergone peer review. Details beyond those reported in the source (for example, exact variable lists for every model, covariate handling, missing data methods, or external validation) were not presented in the abstract. The work used ADNI data; generalizability to other clinical populations or community samples may be limited by ADNI cohort characteristics. The authors reported methods for internal cross-validation (repeated stratified 5-fold with 10 repetitions) but external prospective validation was not described in the source abstract.
Within an ADNI cohort of 2,430 participants with baseline MCI, episodic memory measures provided the strongest predictive performance for 24-month progression to AD, and functional assessment (FAQ) offered substantial complementary value. Combining episodic memory, functional status, and executive function produced the best predictive accuracy (AUC = 0.943; sensitivity 0.857; specificity 0.889). The authors propose that streamlined assessment batteries emphasizing episodic memory and functional status could improve efficient risk stratification in memory clinics and for AD clinical trial enrollment.
The datasets analyzed are available from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) to qualified investigators upon application and approval (https://adni.loni.usc.edu). The authors declared no competing interests. The preprint states that relevant ethical guidelines were followed and that necessary IRB or oversight approvals were addressed via ADNI; patient consent procedures and data protections were reported as completed in the source.