Identification of pre-clinical Alzheimer’s disease (AD) is necessary for the development of future disease-modifying treatments, which would ideally target pre-clinical stages to mitigate functional loss. Despite advanced in biomarker development, clinical trials are still without a non-invasive and cost-effective means of identifying pre-symptomatic subjects who are at high risk for eventual conversion to AD. In previous work, we developed a machine learning algorithm using neuropsychological test scores and health history to identify subjects at high risk for eventual conversion. Here, we examine the performance of a similar algorithm in predicting the timing of that conversion in years.

Data were obtained from the National Alzheimer’s Coordination Center (NACC) Uniform Data Set (UDS) version 3.0. Subjects with normal cognition at baseline were used to train a multi-class Random Forest classifier to predict conversion to AD. Each subject could be classified as a short-, mid-, or long-term converter (0-3 years, 4 to 6 years, and 7 to 9 years, respectively) or as a non-converter, if no dementia diagnosis was given within ten years of baseline. Predictors included baseline demographics, basic medical history, and neuropsychological test results. Algorithms were evaluated using standard, cross-validated performance metrics.

Multi-class Matthews correlation coefficient between predicted time to diagnosis and the ground truth averaged 0.26 +/- 0.06 across 100 cross validation splits. Prediction accuracy exceeded 0.67 in all cases, when computed for each class individually, and was greatest for the short-term (0.75) and nonconverter (0.78) classes.

Machine-learning algorithms applied to neuropsychological, demographic, and medical history information were able to predict the eventual timing of conversion to dementia in cognitively healthy adults significantly better than chance. Results were most accurate when predicting shorter time to conversion. Results illustrate the potential of this data analytic approach for targeted recruitment in clinical trials.