Targeting the Correct Population for Trials: A Post‐hoc Analysis of Trial of Solanezumab for Mild Dementia Due to Alzheimer’s disease.

Abstract

AbstractBackgroundUp to 50% of placebo treated participants in AD clinical trials do not show clinically meaningful cognitive decline (CMCD) over the course of study. Inclusion of these individuals makes it more difficult to demonstrate the benefits of treatments intended to slow cognitive decline. Here, we used data from the trial of solanezumab for mild AD (EXPEDITION3) to investigate effectiveness of predictive models for enrichment of trials.MethodA total of 1072 patients with mild dementia and biomarker evidence of amyloid by PET or CSF studies were enrolled in the placebo arm of EXPEDITION3 trial. 893 patients completed the trial and had baseline structural MRIs making them eligible for this analysis. We identified cognitive decliners, individuals with CMCD, based on the score of the 14‐item cognitive subscale of the Alzheimer’s Disease Assessment Scale (ADAS‐cog14; Range 0‐90; change of ≥4 from baseline to week 80) versus stable cognition, individuals without CMCD (change of <4 from baseline to week 80). Random forest classifiers were trained to classify participants into cognitive decliners vs. stable cognition groups using 3 feature‐sets: Model‐1) Demographics + Neuropsychological test (NP); Model‐2) Demographics + biomarkers (APOE4 alleles and volumetric MRI); Model‐3) All available measures. 70% of data was used for training and 30% for validation.ResultSample characteristics are summarized in Table 1. CMCD was observed in 55.8% of participants at week 80. In the validation sample, model 2 had the best performance in differentiating the two subgroups (AUC = 0.70), followed by model 3 (AUC = 0.68). Model 1 had poor performance in differentiating the two subgroups with AUC = 0.56. Table 2 summarizes performance metrics of models. Positive predictive value of the best performing model (model 2) was approximately 12% higher than the base prevalence of cognitive decline, and negative predictive value of it was approximately 20% higher than the base prevalence of participants who had stable cognition.ConclusionA large portion of biomarker confirmed AD cases enrolled in this trial did not show CMCD on placebo treatment. 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.