Using AI-generated digital twins to boost clinical trial efficiency in Alzheimer's disease

Area of Science:

• Neurology
• Artificial Intelligence
• Clinical Trials

Background:

• Machine learning models create AI-generated digital twins (DTs) for individualized clinical outcome predictions.
• DTs can potentially increase statistical power and reduce sample sizes in Phase 2/3 trials, enhancing efficiency in Alzheimer's disease (AD) studies.
• This study demonstrates these properties using data from an AD Phase 2 trial (AWARE).

Purpose of the Study:

• To evaluate the impact of AI-generated digital twins (DTs) on statistical power and sample size reduction in Alzheimer's disease (AD) clinical trials.
• To assess the prognostic value of DTs as covariates in clinical trial analysis.
• To validate the use of DTs in improving clinical development efficiency.

Main Methods:

• A conditional restricted Boltzmann machine (CRBM) model was trained on harmonized historical clinical trial and observational data from 6736 subjects.
• DTs were generated for 453 participants in the AWARE AD Phase 2 trial (mild cognitive impairment or mild AD).
• DTs were incorporated as prognostic covariates to assess gains in variance and potential sample size reduction.

Main Results:

• Positive correlations (0.30-0.39 at 96 weeks) were observed between DTs and cognitive assessment changes, consistent across validation trials (0.30-0.46).
• DTs reduced total residual variance by approximately 9% to 15%.
• DTs demonstrated the potential to reduce total sample size by 9%-15% and control arm sample size by 17%-26% while maintaining statistical power.

Conclusions:

• Incorporating AI-generated prognostic digital twins (DTs) into statistical analysis models significantly improves Alzheimer's disease (AD) clinical trial efficiency.
• This methodology aligns with regulatory guidance and is suitable for pivotal trial data analysis.
• Validated DTs offer a promising approach to enhance clinical development efficiency for AD and other neurological conditions.