A Machine Learning Approach to Predict Cognitive Decline in Alzheimer's Disease Clinical Trials

BackgroundOf persons randomized to the placebo arm of Alzheimers Disease (AD) treatment trials, 40% do not show cognitive decline over 80 weeks of follow-up. Identifying and excluding these individuals from both arms of randomized clinical trials (RCTs) of AD has the potential to increase power to detect treatment effects. ObjectivesWe aimed to develop machine learning-based predictive models to identify persons unlikely to show decline on placebo treatment over 80 weeks. MethodWe used the data from 1072 patients with mild dementia and biomarker evidence of amyloid burden from the placebo arm of EXPEDITION3 trial. Participants were identified as those who demonstrated Clinically Meaningful Cognitive Decline (CMCD, change in ADAS-Cog[&ge;]3) or Cognitive Stable (CS, change in ADAS-Cog<3) at final visit of the trial (week 80). Machine learning-based classifiers were trained to classify participants into CMCD vs. CS groups using combinations of demographics, neuropsychological tests (NP) and biomarkers, including APOE4 genotype and volumetric MRI. The results were developed in 70% of the EXPEDITION3 placebo sample (EXPtrain) using 5-fold cross-validation. Trained models were then used to classify the participants in an internal validation sample (EXPvalid) and an external matched sample from the Alzheimers Disease Neuroimaging Initiative (ADNI) study. ResultParticipants selected from the EXPEDITION3 trial were on average 72.7({+/-}7.7) years old, 59% were female. CMCD was observed in 55.8% of participants of EXPEDITION3 at final visit. In the independent validation sample within the EXPEDITION3 data, all the models showed high sensitivity and modest specificity. Positive predictive values (PPVs) of models were at least 11% higher than base prevalence of CMCD observed at the end of the trial. The subset of matched ADNI participants (ADNIAD) were on average 74.5({+/-}6.4) years old and 46% female. The models that were validated in ADNIAD also showed high sensitivity, modest specificity and PPVs of at least 15% higher than the base prevalence in ADNIAD. ConclusionOur results indicate that predictive models have the potential to improve the design of AD trials through selective inclusion criteria for participants expected to decline and exclusion of those expected to remain stable.