Premium
Is it time to use predictive models to boost power of Alzheimer’s disease clinical trials? A post‐hoc analysis of phase 3 solanezumab trials
Author(s) -
Ezzati Ali,
Davatzikos Christos,
Wolk David A.,
Aisen Paul S.,
Lipton Richard B.
Publication year - 2020
Publication title -
alzheimer's and dementia
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 6.713
H-Index - 118
eISSN - 1552-5279
pISSN - 1552-5260
DOI - 10.1002/alz.043022
Subject(s) - clinical trial , medicine , placebo , post hoc analysis , cognition , neuropsychology , disease , cognitive decline , physical therapy , physical medicine and rehabilitation , dementia , psychiatry , pathology , alternative medicine
Background The ideal participants for Alzheimer’s disease (AD) clinical trials would show cognitive decline in the absence of treatment (i.e. placebo arm) and would also respond to the therapeutic intervention. However, up to 50% of placebo treated participants enrolled in the AD clinical trials do not show cognitive or functional decline during the timeframe of trials. One strategy to boost the power of trials is to exclude individuals who are not likely to progress using data‐driven predictive models. In this study, we aimed to investigate if using machine learning (ML) predictive models can be an effective tool for enriching enrollment in AD trials. Method We used data from 769 patients from the placebo arm of two phase 3, double‐blind trials of EXPEDITION (N=367) and EXPEDITION2 (N=402). Patients had 18 months of follow‐up. Ensemble linear discriminant ML models were developed using EXPEDITION data to classify participants as 1) cognitive decliners, change of >0 from baseline in the Alzheimer’s Disease Assessment Scale cognitive subscale (ADAS‐cog) score, versus stable cognition, change of £0 from baseline ADAS‐cog score; or 2) functional decliners, change of <0 from baseline in the Alzheimer's Disease Cooperative Study ‐ Activities of Daily Living (ADCS‐ADL) versus stable function, change of ≥0 from baseline ADCS‐ADL, at both 15 and 18 months of follow‐up. Data from EXPEDITION2 were used to validate models. Features used in ML models included baseline demographics, Apo4 genotype, neuropsychological scores, functional scores and volumetric MRI. Result In EXPEDITION and EXPEDITION2 trials, 45.2% and 42.0% of participants in the placebo arm did not show cognitive decline, respectively (Table 1). Models had high sensitivity and modest specificity in both the training (EXPEDITION) and (EXPEDITION2) samples for detecting the stable group. Positive predictive value of models was higher than the base prevalence of cognitive or functional decline, and negative predictive value of models were higher than the base rate of participants who had stable cognition or function (Tables 2 and 3). Conclusion Predictive models may boost the power of AD trials through selective inclusion of participants expected to decline and exclusion of those expected to remain stable.