A Machine Learning-Based Model to Predict In-Hospital Mortality of Lung Cancer Patients: A Population-Based Study of 523,959 Cases

Abstract

Background: Stratify new lung cancer patients based on the risk of in-hospital mortality rate after diagnosis. Methods: 522,941 lung cancer cases with available data on the Surveillance, Epidemiology, and End Results (SEER) were analyzed for the predicted probability based on six fundamental variables including age, gender, tumor size, T, N, and AJCC stages. The patients were randomly assigned to the training (n = 115,145) and validation datasets (n = 13,017). The remaining cohort with missing values (n = 394,779) was then combined with the primary lung tumour datasets (n = 1018) from The Cancer Genome Atlas, Lung Adenocarcinoma and Lung Squamous Cell Carcinoma projects (TCGA-LUAD & TCGA-LUSC) for external validation and sensitivity analysis. Results: Receiver Operating Characteristic (ROC) analyses showed high discriminatory power in the training and internal validation cohorts (Area under the curve [AUC] of 0.78 (95%CI = 0.78–0.79) and 0.78 (95%CI = 0.77–0.79), respectively), whereas that of the model on external validation data was 0.759 (95%CI = 0.757–0.761). We developed a static nomogram, a web app, and a risk table based on a logistic regression model using algorithm-selected variables. Conclusions: Our model can stratify lung cancer patients into high- and low-risk of in-hospital mortality to assist clinical further planning.