Case studies
From research to practice: AI innovations implemented and validated in real clinical trial settings.
Improving phase 2 Trial Precision in Parkinson's Disease Using Digital Twin-Derived Prognostic Covariates
Joint development with Sanofi
Executive Summary
Clinical trials in Parkinson’s disease (PD) are challenged by marked inter-individual variability in disease progression, which reduces statistical power and complicates interpretation of treatment effects—particularly in Phase 2 studies. Qairnel and Sanofi collaborated to evaluate whether proprietary digital twin technology could improve trial precision through prognostic covariate adjustment.
Using data from Sanofi’s Phase 2 randomized controlled trial, Qairnel retrospectively applied the PD Course Map, a predictive digital twin model of disease progression, to derive patient-specific prognostic covariates. Incorporating these covariates into standard statistical models resulted in narrower confidence intervals and meaningful gains in effective sample size.
Clinical Context
Parkinson’s disease is a progressive neurodegenerative disorder with highly heterogeneous clinical trajectories. This heterogeneity presents a major challenge for clinical trials, especially in Phase 2, where limited sample sizes and modest disease progression can obscure treatment effects.
The MOVES-PD study was a Phase 2 randomized controlled trial evaluating venglustat in patients with GBA1-associated PD. As in many PD trials, outcome variability in motor and functional endpoints limited statistical precision, increasing uncertainty around efficacy signals and development decisions.
Digital Twin-Derived Prognostic Covariates
Qairnel applied its proprietary CourseMap AI technology, to build a predictive digital twin model of Parkinson’s disease progression, to retrospectively analyze MOVES-PD trial data.
The PD Course Map estimated expected disease trajectories for individual participants based on their baseline data. Prognostic covariates derived from the predicted trajectories were integrated into mixed models for repeated measurements (MMRM), alongside standard baseline covariates. We compared 4 different combinations of covariates.
Key Results
Narrower Confidence Intervals
Inclusion of PD Course Map covariates reduced uncertainty around the treatment effect estimate. Confidence interval widths decreased progressively compared with models using no covariates or standard baseline covariates alone, improving interpretability of trial outcomes; confirming in this case a negative effect of the treatment compared to placebo.
Increased Effective Sample Size
Variance reduction achieved through digital twin–derived covariates corresponded to an estimated ~15–16% gain in effective sample size, substantially exceeding gains obtained with standard covariates (~5%), decreasing number of participants for a hypothetical phase 3 from 361 to 304. These results suggest that such an approach could enable more efficient trials or increase power at fixed sample sizes.
Conclusion
This case study demonstrates how digital twin technology can be integrated into standard clinical trial analyses to address a central challenge in drug development for neurodegenerative diseases: variability in disease progression.
By generating patient-specific prognostic covariates, Qairnel’s PD CourseMap AI improved trial precision and effective sample size in a Phase 2 setting. Digital twins represent a practical, regulatory-compatible tool to support more efficient and informative clinical development in PD and other neurodegenerative diseases.