Prospective validation of in silico clinical trials prediction using an EGFR-mutated NSCLC mechanistic computational disease model.

Abstract

8614 Background: Advanced lung adenocarcinoma (aLUAD) is divided into multiple molecularly defined subsets, each with specific biological characteristics and responses to targeted therapy. Consequently, conducting practice-changing and cost-effective randomized trials is challenging. In silico clinical trials, utilizing mechanistic computational models, offer a potential solution, yet their ability to predict clinical trial results has not been proven. Methods: We updated a previously published version of the computational mechanistic model of EGFR-mutated aLUAD, notably by incorporating metastases. Disease model was integrated with physiologically based pharmacokinetic (PBPK) models of cisplatin, pemetrexed, and osimertinib, offering a more comprehensive understanding of the disease's dynamics. These models are founded on causal connections between biological entities and phenomena, establishing a link between the specific biological behaviors of EGFR-mutated aLUAD and the evolution of tumor size and progression over time. We have independently and prospectively simulated the outcomes of the FLAURA2 and MARIPOSA randomized trials dedicated to EGFR-mutated aLUAD. This was achieved by replicating the experimental protocols and generating cohorts of virtual twins of patients using publicly available data. Results: In silico results were released before the actual outcomes of the FLAURA2 and MARIPOSA trials were publicly available. The simulations yielded predictions with overlapping confidence intervals, similar hazard ratios, median survivals, and shapes of curves compared to the actual trial data. Bootstrapped weighted log-rank tests were used to compare the simulated and observed Kaplan-Meier curves for the two arms of FLAURA2 and for the osimertinib arm of MARIPOSA. Between 94 and 98% of the tests were statistically non-significant (α=0.05), confirming the accuracy of the predictions. Conclusions: This first-of-its-kind prediction of clinical trial outcomes demonstrates that in silico trials, when based on robust mechanistic models, can be a reliable tool for enhancing the design of actual trials, particularly for EGFR-mutated aLUAD. They offer a promising avenue for overcoming the challenges of comparator arms in single-arm trials and may serve as a new standard for formulating statistical hypotheses in future studies.