Affiliations: Department of Neurology, Albert Einstein College of Medicine and Montefiore Medical Center, Bronx, NY, USA
Correspondence:
[*]
Correspondence to: Ali Ezzati, MD, Albert Einstein College of Medicine, 1225 Morris Park Avenue, Bronx, NY 10461, USA. Tel.: +1 718 430 3885; Fax: +1 718 430 3870; E-mail: aezzati@einstein.yu.edu.
Note: [1] Part of data utilized in these analyses are from the following Lilly trial: H6L-MC-LFAN.
Note: [2] Data used in preparation of this article were obtained from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) database (http://adni.loni.usc.edu). As such, the investigators within the ADNI contributed to the design and implementation of ADNI and/or provided data but did not participate in analysis or writing of this report. A complete listing of ADNI investigators can be found at: http://adni.loni.usc.edu/wp-content/uploads/how_to_apply/ADNI_Acknowledgement_List.pdf
Abstract: 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 also would be responsive to the therapeutic intervention being studied (i.e., drug arm). One strategy to boost the power of trials is to enroll individuals who are more likely to progress targeted using data-driven predictive models. Objective:To investigate if machine learning (ML) models can effectively predict clinical disease progression (cognitive decline) in mild-to-moderate AD patients during the timeframe of a phase III clinical trial. Methods:Data from 202 participants with a diagnosis of AD at baseline from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) was used to train ML classifiers that can differentiate between individuals who had declining cognitive function (DC) and individuals with stable cognitive function (SC). DC was defined as any downward change in the Alzheimer’s Disease Assessment Scale cognitive subscale (ADAS-cog) score over 12 months of follow-up. SC was defined by the absence of decline in ADAS-cog. Trained models were applied to data from 77 participants from the placebo arm of the phase III trial of Semagacestat (LFAN study) to identify subgroups of SC versus DC. Results:Only 74.8% of ADNI participants and 63.6% of LFAN participants had cognitive decline after one year of follow up. K-nearest neighbors (kNN) classifier had an accuracy of 68.3%, sensitivity of 80.1%, and specificity of 33.3% for identifying decliners in ADNI (training sample). In LFAN (validation sample), the model showed an overall accuracy of 61.3%, sensitivity of 65.5%, and specificity of 47.0% in identifying decliners at the 12 months of follow-up. The model had a positive predictive value of 80.8%, which was 17.2% more than the base prevalence of decliners. Conclusions:Machine learning predictive models can be effectively used to boost the power of clinical trials by reducing the sample size.
