A joint analytic framework towards the problem of survival curves crossing and the application in patients with NSCLC

Abstract Background: When the problem of survival-curve-crossing arises, the traditional proportional hazard (PH) Cox model usually fails to accurately estimate the effect of prognostic factors. This problem has not been completely resolved in the prognostic studies of non-small cell lung cancer. Therefore, a joint analytic framework towards the SCC problem should be constructed for a better understanding of NSCLC prognosis. Method: The framework includes segmentation, model selection, and evaluation. It presents a clear roadmap on how to choose an appropriate model given different data conditions. The performance of the framework was evaluated through the data of NSCLC patients from SEER database. We compared its goodness-of-fit with that of the PH Cox model by the Akaike Information Criterion and Cox-Snell Residuals. Besides, we also conducted subgroup analyses to enhance the validity of our framework. Results: There was a crossing point in the survival curves of chemotherapy. The model selected by the joint analytic framework had a better goodness-of-fit performance. In contrast, the non-segmented model failed to find out that radiotherapy shortened the survival time of patients with survival time >14 months. The subgroup analyses showed specific patients who benefited from radiotherapy or chemotherapy. Conclusion: The joint analytic framework could help clinicians avoid pitfalls such as SCC during the prognosis study of NSCLC. The prognostic effect of radiotherapy and chemotherapy in NSCLC patients with survival time >14 months needs more attention in the future.