Abstract
BACKGROUND: Most of previous studies on predictive models for patients with small cell lung cancer (SCLC) were single institutional studies or showed relatively low Harrell concordance index (C-index) values. To build an optimal nomogram, we collected clinicopathological characteristics of SCLC patients from Surveillance, Epidemiology, and End Results (SEER) database. METHODS: 24,055 samples with SCLC from 2010 to 2016 in the SEER database were analyzed. The samples were grouped into derivation cohort (n=20,075) and external validation cohort (n=3,980) based on America's different geographic regions. Cox regression analyses were used to construct nomograms predicting cancer-specific survival (CSS) and overall survival (OS) using derivation cohort. The nomograms were internally validated by bootstrapping technique and externally validated by calibration plots. C-index was computed to compare the accuracy and discrimination power of our nomograms with the 8th of version AJCC TNM staging system and nomograms built in previous studies. Decision curve analysis (DCA) was applied to explore whether the nomograms had better clinical efficiency than the 8th version of AJCC TNM staging system. RESULTS: Age, sex, race, marital status, primary site, differentiation, T classification, N classification, M classification, surgical type, lymph node ratio, radiotherapy, and chemotherapy were chosen as predictors of CSS and OS for SCLC by stepwise multivariable regression and were put into the nomograms. Internal and external validations confirmed the nomograms were accurate in prediction. C-indexes of the nomograms were relatively satisfactory in derivation cohort (CSS: 0.761, OS: 0.761) and external validation cohort (CSS: 0.764, OS: 0.764). The accuracy of the nomograms was superior to that of nomograms built in previous studies. DCA showed the nomograms conferred better clinical efficiency than 8th version of TNM staging system. CONCLUSIONS: We developed practical nomograms for CSS (https://guowei2020.shinyapps.io/DynNom-CSS-SCLC/) and OS (https://drboidedwater.shinyapps.io/DynNom-OS-SCLC/) prediction of SCLC patients which may facilitate clinicians in individualized therapeutics.