RECONSTRUCTING TRIAL-SPECIFIC TIME-ZERO IN REAL-WORLD DATA: A GENETIC ALGORITHM APPROACH TO VALIDATE INDIGO TRIAL OUTCOMES
Author(s)
Anaïs Ragon, PharmD1, Celine Aubin, PharmD1, Tiphaine Obara, MD, PhD2, Luc Taillandier, MD, PhD2, Marie Blonski, MD, PhD2, Andy Clark, PhD3, Ash Bullement, PhD3.
1Servier International, Suresnes, France, 2Department of Neuro-Oncology, Nancy Hospital, Nancy, France, 3Petauri Evidence, Nottingham, United Kingdom.
1Servier International, Suresnes, France, 2Department of Neuro-Oncology, Nancy Hospital, Nancy, France, 3Petauri Evidence, Nottingham, United Kingdom.
OBJECTIVES: INDIGO (NCT04164901), a phase III study in IDH-mutant (mIDH) gliomas, defined time-to-next-intervention (TTNI) relative to a unique trial entry timing, requiring patients to be 1-to-5 years post-surgery. This varying "time-zero" is undefined in real-world settings, complicating validation and interpretation of TTNI in the INDIGO placebo arm. This study aimed to assess the external validity of this endpoint by estimating real-world TTNI using a genetic algorithm (GA) to align time-zero with the trial design.
METHODS: Real-world data from the Centre Hospitalier Régional Universitaire de Nancy were analysed and patients with grade 2 mIDH glioma were included. To replicate the INDIGO-specific entry timing, a GA was used to identify time-zeros within a 1-to-5-year window after surgery, aligning summary statistics for time since surgery (mean, median, standard deviation) with the INDIGO placebo arm. The GA was applied across five independently simulated cohorts, each targeting 300 solutions and generating multiple valid time-zero sets. Kaplan-Meier estimates of TTNI were derived for each solution, and average, least, and most extreme scenarios were identified using medians.
RESULTS: A total of 244 patients were included. Across simulated cohorts, median TTNI ranged from 12.2 to 22.0 months, spanning least- to most-extreme scenarios. Central estimates were more consistent, ranging from 15.9 to 17.7 months. Notably, 95% confidence intervals around these central estimates consistently encompassed the median TTNI observed in the INDIGO placebo arm (20.1 months).
CONCLUSIONS: Using a GA to align time-zero, this analysis demonstrates that real-world TTNI estimates overlap with those observed in the INDIGO placebo arm. These findings support the external validity of INDIGO placebo arm TTNI and demonstrate the importance of considering trial-specific time-zero when comparing clinical and real-world evidence. Further research should explore alternative methodologies and datasets to strengthen validation across settings.
METHODS: Real-world data from the Centre Hospitalier Régional Universitaire de Nancy were analysed and patients with grade 2 mIDH glioma were included. To replicate the INDIGO-specific entry timing, a GA was used to identify time-zeros within a 1-to-5-year window after surgery, aligning summary statistics for time since surgery (mean, median, standard deviation) with the INDIGO placebo arm. The GA was applied across five independently simulated cohorts, each targeting 300 solutions and generating multiple valid time-zero sets. Kaplan-Meier estimates of TTNI were derived for each solution, and average, least, and most extreme scenarios were identified using medians.
RESULTS: A total of 244 patients were included. Across simulated cohorts, median TTNI ranged from 12.2 to 22.0 months, spanning least- to most-extreme scenarios. Central estimates were more consistent, ranging from 15.9 to 17.7 months. Notably, 95% confidence intervals around these central estimates consistently encompassed the median TTNI observed in the INDIGO placebo arm (20.1 months).
CONCLUSIONS: Using a GA to align time-zero, this analysis demonstrates that real-world TTNI estimates overlap with those observed in the INDIGO placebo arm. These findings support the external validity of INDIGO placebo arm TTNI and demonstrate the importance of considering trial-specific time-zero when comparing clinical and real-world evidence. Further research should explore alternative methodologies and datasets to strengthen validation across settings.
Conference/Value in Health Info
2026-11, ISPOR Europe 2026, Vienna, Austria
Value in Health, Volume 29, Issue 12S
Code
CO155
Topic
Clinical Outcomes, Methodological & Statistical Research, Real World Data & Information Systems
Topic Subcategory
Clinical Outcomes Assessment
Disease
Neurological Disorders, Oncology