Pharmacoeconomic evaluation of first-line tislelizumab for extensive-stage small cell lung cancer using a comparative validation of traditional survival and machine learning models.

Objective: To evaluate the cost-effectiveness of first-line tislelizumab plus chemotherapy for extensive-stage small cell lung cancer (ES-SCLC) within the Chinese healthcare context. Furthermore, this study aims to comparatively validate the impact of traditional survival models vs. advanced machine learning models on the robustness of research findings, thereby providing refined empirical evidence for healthcare decision-making. Methods: Based on data from the Phase III RATIONALE-312 clinical trial, a partitioned survival model (PSM) was constructed with a 10-year time horizon. From the perspective of the Chinese healthcare system, only direct medical costs were included. A willingness-to-pay (WTP) threshold was set at three times the 2025 per capita GDP of China (298,995 CNY/QALY). A dual-validation strategy was employed for survival extrapolation: a base-case analysis using the optimal parametric model (Log-logistic distribution) selected via Akaike Information Criterion/Bayesian Information Criterion (AIC/BIC) criteria, followed by a comparative validation using DeepSurv deep learning and random survival forest (RSF) models utilizing a simplified feature set. Uncertainty was assessed through one-way sensitivity analysis and probabilistic sensitivity analysis (PSA) using 1,000 Monte Carlo simulations. Results: The base-case analysis (Log-logistic model) revealed that, compared with chemotherapy alone, the tislelizumab group yielded an incremental cost of 101,506.9 CNY and incremental quality-adjusted life years (QALYs) of 0.4044, resulting in an incremental cost-effectiveness ratio (ICER) of 251,030.5 CNY/QALY, which remains below the WTP threshold. Machine learning validation demonstrated exceptional consistency, with ICERs of 261,718.45 CNY/QALY for the DeepSurv model and 248,299.41 CNY/QALY for the RSF model. Sensitivity analysis identified the utility of progression-free survival (PFS) and the unit price of tislelizumab as the primary drivers of the model. PSA confirmed that the probability of tislelizumab being cost-effective exceeded 85% across all three survival-fitting logics, reaching a maximum of 94.90%. Conclusion: First-line tislelizumab plus chemotherapy demonstrates a significant cost-effectiveness advantage for ES-SCLC in China. The integration of machine learning survival models effectively reduced the inherent uncertainties of non-linear extrapolation in immunotherapy data, confirming the robustness of the conclusions. These findings provide a solid empirical basis for national health insurance negotiations and precision clinical applications.