Using exponential decay to model survival curves in immunotherapy trials.

person Tarquin Dean Opperman Western University, Windsor, ON, Canada info_outline Tarquin Dean Opperman, David J. Stewart, Abdullah Nasser

Authors person Tarquin Dean Opperman Western University, Windsor, ON, Canada info_outline Tarquin Dean Opperman, David J. Stewart, Abdullah Nasser Organizations Western University, Windsor, ON, Canada, The Ottawa Hospital Cancer Centre, Ottawa, ON, Canada Abstract Disclosures Research Funding No funding sources reported Background: In oncology trials, survival outcomes, specifically overall survival (OS) and progression-free survival (PFS), are commonly assessed using the Kaplan–Meier estimator. Survival curves in immunotherapy trials for advanced solid tumors frequently display first-order kinetics, suitable for exponential decay modeling. This study utilizes such models on early trial data to predict median OS and PFS, aiming to yield earlier actionable insights. Methods: we searched PubMed to identify immunotherapy trials focusing on advanced solid tumors. Survival curves, along with actual OS and PFS data, were extracted and segmented into immunotherapy and control groups. An exponential decay model was fitted to the initial quartile of events, from which median OS and PFS were extrapolated. These forecasts were then benchmarked against documented trial results. Results: Our analysis included 347 subsets from 122 immunotherapy trials. We observed strong correlation coefficients between actual and predicted OS (0.882 for the immunotherapy group, 0.784 for the control group) and PFS (0.746 for the immunotherapy group, 0.791 for the control group). The predictive models were formulated as follows: Immunotherapy: OS = 0.867 × predicted OS (pOS) + 2.623 ⋅ Control: OS = 0.623 × pOS + 3.731; and Immunotherapy: PFS = 0.583 × predicted PFS (pPFS) + 1.827 ⋅ Control: PFS = 0.519 × pPFS + 1.901. Conclusions: The notable correlation between the actual and predicted survival outcomes supports the efficacy of exponential decay models in forecasting median OS and PFS from preliminary trial data. The consistency observed across trials underscores the model's potential as a predictive tool in clinical research. Further prospective studies are warranted to validate these predictive models and explore their implications for trial design and therapeutic decision-making in oncology.