Identification of heterogeneity in Alzheimer’s disease clinical trials using computational modelling

Lead Investigator: Neil Oxtoby, University College London
Title of Proposal Research: Identification of heterogeneity in Alzheimer’s disease clinical trials using computational modelling
Vivli Data Request: 7232
Funding Source: UK Research and Innovation (UKRI) Medical Research Council grant number MR/S03546X/1
Potential Conflicts of Interest: None

Summary of the Proposed Research:

Alzheimer’s disease is the leading cause of dementia, which is becoming a global pandemic for the world’s ageing population, and currently affects 55 million people worldwide and will nearly triple by 2050. Alzheimer’s disease is devastating for patients, their families and carers, and the economy. Decades of experimental research into novel treatments have produced many new very promising drugs but none have been unequivocally proven to work in clinical trials.

Our computational research has revealed previously hidden subtypes and stages of Alzheimer’s disease progression. We believe that clinical trials should take this information into account when recruiting patients because it is likely that any given drug won’t work on all subtypes, nor at all stages of the disease. Here we propose to test this idea in data from a completed clinical trial known as TOMMORROW. If our idea works, we can then contribute to the success of future clinical trials to finding treatments for this devastating disease.

Statistical Analysis Plan:

 This is a straightforward post hoc subgroup analysis that follows the TOMMORROW trial protocol. We will write custom python scripts to interact with TOMMORROW data and with our pre-trained model.

Our model will be pre-trained on publicly available data such as from the Alzheimer’s Disease Neuroimaging Initiative. The model estimates subtypes of disease progression using a computational algorithm called SuStaIn (Young, et al. 2018). The trained model can then be applied to the TOMMORROW trial participants at screening/baseline to assign each individual a model subtype and stage. This information is used to stratify TOMMORROW for the longitudinal subgroup analysis of trial outcomes.

This process requires participant-level data from all arms of the TOMMORROW trial: all cognitive test/instrument scores, biomarkers, neuroimaging data (where available), demographics, treatment assignment, etc.

We will then assess all trial endpoints in each of these model-based subgroups. This requires all participant-level longitudinal clinical and biomarker data (including imaging) at each visit during the TOMMORROW trial.

The neuroimaging data is of interest but is not essential. Should it pose a barrier, we can proceed without it.

We selected the TOMMORROW study because it is a large phase 3 trial in Alzheimer’s disease with data available. Our general line of research is applicable to any and all late-phase (2 and above) trials, but participant-level data is very rarely shared.

Missing values are imputed probabilistically. For example, our modelling maps biomarker data to a severity probability where p=0 is maximally normal and p=1 is maximally abnormal. Missing data corresponds to p = 0.5 (i.e., maximally uncertain). Since we combine biomarker data in multimodal modelling, we can tolerate missing data in any given biomarker/feature so long as there is non-missing data in other biomarkers/features.

Requested Studies:

A Double Blind, Randomized, Placebo Controlled, Parallel Group Study to Simultaneously Qualify a Biomarker Algorithm for Prognosis of Risk of Developing Mild Cognitive Impairment Due to Alzheimer’s Disease (MCI Due to AD) and to Test the Safety and Efficacy of Pioglitazone (AD-4833 SR 0.8 mg QD) to Delay the Onset of MCI Due to AD in Cognitively Normal Subjects
Data Contributor: Takeda
Study ID: NCT01931566
Sponsor ID: AD-4833/TOMM40_301

Public Disclosures:

C. Shand, N.P. Oxtoby. INVESTIGATING TREATMENT EFFECT HETEROGENEITY IN DATA-DRIVEN SUBGROUPS OF TOMMORROW. Journal of Prevention of Alzheimer’s Disease. 2023; Abstract P004, pg. S57.