A combined cox random survival forest model with SHAP analysis predicts prognosis of peripheral lung adenocarcinoma using SEER data and external validation

Significant progress has been made in the prognosis of patients with lung adenocarcinoma. The objective of this retrospective study was to investigate risk factors and survival outcomes in patients with lung adenocarcinoma, and more importantly, to establish a survival stratification model based on the joint weighting of Cox and random forest models. A total of 1,851 eligible lung adenocarcinoma patients from the SEER database (2010–2022) were randomly divided into training and validation sets in a 7:3 ratio using stratified sampling. Initially, LASSO-Cox identified seven candidate variables. Multivariate Cox regression then retained histological grade, clinical stage, and chemotherapy as independent predictors. These three were used as final parameters in both the Cox nomogram and Random Survival Forest (RSF) model. Model performance was evaluated in the validation set using time-dependent AUC and decision curve analysis (DCA). A combined Cox-RF risk score was derived by weighting the two models based on their AUC performance. Patients were stratified into low-, intermediate-, and high-risk groups using optimal cutoffs. Kaplan–Meier analysis and log-rank tests were used for survival comparison. Additionally, we performed external validation of the model using survival differences in hospital patient case data and SHAP analysis. LASSO-Cox analysis identified histologic grade, clinical stage, and chemotherapy as independent prognostic factors, which were incorporated into a nomogram. Both the Cox and Random Survival Forest (RSF) models demonstrated comparable and moderate-to-good predictive accuracy for 1-, 2-, and 3-year survival (Cox AUCs: 0.741, 0.748, 0.703; RSF AUCs: 0.745, 0.736, 0.699). By integrating standardized predictions from both models, we developed a composite risk score that effectively stratified patients into three distinct prognostic groups (low-, medium-, high-risk) with significantly different survival outcomes, demonstrating clinical utility for risk stratification.By assigning risk scores to the external validation cohort based on Cox regression risk factors, the comparison revealed that the survival benefit in the high-risk group was significantly lower than that in the low-risk group. Further analysis using SHAP values characterized this cohort, identifying tumor grade, stage, and chemotherapy status as the top three critical factors influencing patient prognosis.