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

Calum Robert MacLellan; Hristo Petkov; Conor McKeag; Feng Dong; David John Lowe; Roma Maguire; Sotiris Moschoyiannis; Jo Armes; Simon Skene; Alastair Finlinson; Christopher Sainsbury; Krasimira Tsaneva-Atanasova

doi:10.1371/journal.pdig.0000927

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

Calum Robert MacLellan, Hristo Petkov, Conor McKeag, Feng Dong^*, David John Lowe, Roma Maguire, Sotiris Moschoyiannis, Jo Armes, Simon Skene, Alastair Finlinson, Christopher Sainsbury, Krasimira Tsaneva-Atanasova (Editor)

^*Corresponding author for this work

Research output: Contribution to journal › Article › peer-review

4 Downloads (Pure)

Abstract

Randomized controlled trials (RCTs) remain the benchmark for assessing treatment effects but are limited to phenotypically narrow populations by design. We introduce a novel generative artificial intelligence (AI) driven emulation method that infers effect size through virtual clinical trials, which can emulate the RCT process and potentially extrapolate into wider populations. We validate the virtual trials by comparing the predicted impact of glucagon-like peptide-1 (GLP-1) agonists on HbA1c in type-2 diabetes (T2DM) with its true efficacy established in the LEAD-5 trial. Our emulation model learns treatment effects from real-world evidence data by a combined generative AI and causal learning approach. Training data comprised pre- and post-treatment outcomes for 5,476 people with T2DM. We considered three treatment arms: GLP-1 (Liraglutide), basal insulin (glargine), and placebo. After training, virtual trials were conducted by sampling 232 virtual patients per arm (according to the LEAD-5 inclusion criteria) and predicting post-treatment outcomes. We used difference-in-differences (DiD) for pairwise comparisons between arms. Our goal was to emulate LEAD-5 by demonstrating a significant DiD in post-treatment HbA1c reduction for GLP-1 compared to basal insulin and placebo. We found significant differences in HbA1c reduction for GLP-1 vs basal insulin (-1.21 mmol/mol (-0.11%); p < 0.001) and GLP-1 vs placebo (-2.58 mmol/mol (-0.24%); p < 0.001) in our virtual populations, consistent with LEAD-5 (Liraglutide vs glargine: -2.62mmol/mol (-0.24%); p = 0.0015, Liraglutide vs placebo: -11.91 mmol/mol (-1.09%); p < 0.0001). The causal AI-powered clinical trials can emulate LEAD-5 in important measurements for T2DM. Our algorithm is specialty agnostic and can explore counterfactual questions, making it suitable for further study in the generalizability of RCT results in real-world populations to support clinical decision-making and policy recommendations.

Original language	English
Article number	e0000927
Number of pages	16
Journal	PLOS Digital Health
Volume	4
Issue number	7
DOIs	https://doi.org/10.1371/journal.pdig.0000927
Publication status	Published - 21 Jul 2025

Funding

This work was supported by the Medical Research Council (MRC), UK Research and Innovation (UKRI), (Grant No. MR/X005925/1 to CRM, HP, FD, CM, DJL, RM, SM, JA, SS, AF, and CS). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Keywords

virtual patients
GenAI
emulation method

UN SDGs

This output contributes to the following UN Sustainable Development Goals (SDGs)

Access to Document

10.1371/journal.pdig.0000927Licence: CC BY 4.0

MacLellan-etal-PLOSDH-2025-Emulating-real-world-GLP-1-efficacy-in-type-2-diabetes-throughFinal published version, 576 KBLicence: CC BY 4.0

Cite this

MacLellan, C. R., Petkov, H., McKeag, C., Dong, F., Lowe, D. J., Maguire, R., Moschoyiannis, S., Armes, J., Skene, S., Finlinson, A., Sainsbury, C., & Tsaneva-Atanasova, K. (Ed.) (2025). Emulating real-world GLP-1 efficacy in type 2 diabetes through causal learning and virtual patients. PLOS Digital Health, 4(7), Article e0000927. https://doi.org/10.1371/journal.pdig.0000927

@article{51e4930a40f246f0bbb97b59e465e702,

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

abstract = "Randomized controlled trials (RCTs) remain the benchmark for assessing treatment effects but are limited to phenotypically narrow populations by design. We introduce a novel generative artificial intelligence (AI) driven emulation method that infers effect size through virtual clinical trials, which can emulate the RCT process and potentially extrapolate into wider populations. We validate the virtual trials by comparing the predicted impact of glucagon-like peptide-1 (GLP-1) agonists on HbA1c in type-2 diabetes (T2DM) with its true efficacy established in the LEAD-5 trial. Our emulation model learns treatment effects from real-world evidence data by a combined generative AI and causal learning approach. Training data comprised pre- and post-treatment outcomes for 5,476 people with T2DM. We considered three treatment arms: GLP-1 (Liraglutide), basal insulin (glargine), and placebo. After training, virtual trials were conducted by sampling 232 virtual patients per arm (according to the LEAD-5 inclusion criteria) and predicting post-treatment outcomes. We used difference-in-differences (DiD) for pairwise comparisons between arms. Our goal was to emulate LEAD-5 by demonstrating a significant DiD in post-treatment HbA1c reduction for GLP-1 compared to basal insulin and placebo. We found significant differences in HbA1c reduction for GLP-1 vs basal insulin (-1.21 mmol/mol (-0.11\%); p < 0.001) and GLP-1 vs placebo (-2.58 mmol/mol (-0.24\%); p < 0.001) in our virtual populations, consistent with LEAD-5 (Liraglutide vs glargine: -2.62mmol/mol (-0.24\%); p = 0.0015, Liraglutide vs placebo: -11.91 mmol/mol (-1.09\%); p < 0.0001). The causal AI-powered clinical trials can emulate LEAD-5 in important measurements for T2DM. Our algorithm is specialty agnostic and can explore counterfactual questions, making it suitable for further study in the generalizability of RCT results in real-world populations to support clinical decision-making and policy recommendations.",

keywords = "virtual patients, GenAI, emulation method",

author = "MacLellan, \{Calum Robert\} and Hristo Petkov and Conor McKeag and Feng Dong and Lowe, \{David John\} and Roma Maguire and Sotiris Moschoyiannis and Jo Armes and Simon Skene and Alastair Finlinson and Christopher Sainsbury and Krasimira Tsaneva-Atanasova",

year = "2025",

month = jul,

day = "21",

doi = "10.1371/journal.pdig.0000927",

language = "English",

volume = "4",

number = "7",

}

TY - JOUR

T1 - Emulating real-world GLP-1 efficacy in type 2 diabetes through causal learning and virtual patients

AU - MacLellan, Calum Robert

AU - Petkov, Hristo

AU - McKeag, Conor

AU - Dong, Feng

AU - Lowe, David John

AU - Maguire, Roma

AU - Moschoyiannis, Sotiris

AU - Armes, Jo

AU - Skene, Simon

AU - Finlinson, Alastair

AU - Sainsbury, Christopher

A2 - Tsaneva-Atanasova, Krasimira

PY - 2025/7/21

Y1 - 2025/7/21

N2 - Randomized controlled trials (RCTs) remain the benchmark for assessing treatment effects but are limited to phenotypically narrow populations by design. We introduce a novel generative artificial intelligence (AI) driven emulation method that infers effect size through virtual clinical trials, which can emulate the RCT process and potentially extrapolate into wider populations. We validate the virtual trials by comparing the predicted impact of glucagon-like peptide-1 (GLP-1) agonists on HbA1c in type-2 diabetes (T2DM) with its true efficacy established in the LEAD-5 trial. Our emulation model learns treatment effects from real-world evidence data by a combined generative AI and causal learning approach. Training data comprised pre- and post-treatment outcomes for 5,476 people with T2DM. We considered three treatment arms: GLP-1 (Liraglutide), basal insulin (glargine), and placebo. After training, virtual trials were conducted by sampling 232 virtual patients per arm (according to the LEAD-5 inclusion criteria) and predicting post-treatment outcomes. We used difference-in-differences (DiD) for pairwise comparisons between arms. Our goal was to emulate LEAD-5 by demonstrating a significant DiD in post-treatment HbA1c reduction for GLP-1 compared to basal insulin and placebo. We found significant differences in HbA1c reduction for GLP-1 vs basal insulin (-1.21 mmol/mol (-0.11%); p < 0.001) and GLP-1 vs placebo (-2.58 mmol/mol (-0.24%); p < 0.001) in our virtual populations, consistent with LEAD-5 (Liraglutide vs glargine: -2.62mmol/mol (-0.24%); p = 0.0015, Liraglutide vs placebo: -11.91 mmol/mol (-1.09%); p < 0.0001). The causal AI-powered clinical trials can emulate LEAD-5 in important measurements for T2DM. Our algorithm is specialty agnostic and can explore counterfactual questions, making it suitable for further study in the generalizability of RCT results in real-world populations to support clinical decision-making and policy recommendations.

AB - Randomized controlled trials (RCTs) remain the benchmark for assessing treatment effects but are limited to phenotypically narrow populations by design. We introduce a novel generative artificial intelligence (AI) driven emulation method that infers effect size through virtual clinical trials, which can emulate the RCT process and potentially extrapolate into wider populations. We validate the virtual trials by comparing the predicted impact of glucagon-like peptide-1 (GLP-1) agonists on HbA1c in type-2 diabetes (T2DM) with its true efficacy established in the LEAD-5 trial. Our emulation model learns treatment effects from real-world evidence data by a combined generative AI and causal learning approach. Training data comprised pre- and post-treatment outcomes for 5,476 people with T2DM. We considered three treatment arms: GLP-1 (Liraglutide), basal insulin (glargine), and placebo. After training, virtual trials were conducted by sampling 232 virtual patients per arm (according to the LEAD-5 inclusion criteria) and predicting post-treatment outcomes. We used difference-in-differences (DiD) for pairwise comparisons between arms. Our goal was to emulate LEAD-5 by demonstrating a significant DiD in post-treatment HbA1c reduction for GLP-1 compared to basal insulin and placebo. We found significant differences in HbA1c reduction for GLP-1 vs basal insulin (-1.21 mmol/mol (-0.11%); p < 0.001) and GLP-1 vs placebo (-2.58 mmol/mol (-0.24%); p < 0.001) in our virtual populations, consistent with LEAD-5 (Liraglutide vs glargine: -2.62mmol/mol (-0.24%); p = 0.0015, Liraglutide vs placebo: -11.91 mmol/mol (-1.09%); p < 0.0001). The causal AI-powered clinical trials can emulate LEAD-5 in important measurements for T2DM. Our algorithm is specialty agnostic and can explore counterfactual questions, making it suitable for further study in the generalizability of RCT results in real-world populations to support clinical decision-making and policy recommendations.

KW - virtual patients

KW - GenAI

KW - emulation method

U2 - 10.1371/journal.pdig.0000927

DO - 10.1371/journal.pdig.0000927

M3 - Article

VL - 4

JO - PLOS Digital Health

JF - PLOS Digital Health

IS - 7

M1 - e0000927

ER -

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

Abstract

Funding

Keywords

UN SDGs

Access to Document

Fingerprint

Cite this