Emulating real-world GLP-1 efficacy in type 2 diabetes through causal learning and virtual patients

Area of Science:

• Biomedical Informatics
• Artificial Intelligence in Medicine
• Clinical Trial Methodology

Background:

• Randomized controlled trials (RCTs) are gold standard for treatment effects but have limited populations.
• Extrapolating RCT findings to diverse real-world populations remains a challenge.
• Generative AI offers a novel approach to emulate clinical trials and expand generalizability.

Purpose of the Study:

• To introduce and validate a generative artificial intelligence (AI) driven emulation method for virtual clinical trials.
• To infer treatment effect sizes by emulating the RCT process and potentially extrapolating to wider populations.
• To compare the AI emulation's prediction of glucagon-like peptide-1 (GLP-1) agonist efficacy in type-2 diabetes mellitus (T2DM) with established trial data.

Main Methods:

• Developed a generative AI and causal learning approach to train an emulation model on real-world evidence data.
• Utilized pre- and post-treatment outcomes for 5,476 individuals with T2DM across GLP-1, basal insulin, and placebo arms.
• Conducted virtual trials by sampling patients and predicting outcomes, using difference-in-differences (DiD) for pairwise comparisons.

Main Results:

• The AI emulation successfully predicted significant HbA1c reductions for GLP-1 agonists compared to basal insulin and placebo in virtual trials.
• Virtual trial results for GLP-1 vs. basal insulin (-1.21 mmol/mol) and GLP-1 vs. placebo (-2.58 mmol/mol) aligned with the LEAD-5 trial outcomes.
• The AI model demonstrated the ability to emulate key measurements from a real-world clinical trial (LEAD-5).

Conclusions:

• Causal AI-powered virtual clinical trials can effectively emulate established RCTs for T2DM treatment efficacy.
• This AI method shows potential for exploring counterfactuals and assessing the generalizability of RCT results.
• The approach supports clinical decision-making and policy recommendations by bridging RCT findings with real-world applicability.