Factors Associated With Survival in Patients With Non–small Cell Lung Cancer from a Single Institution Study of 3569 Patients

Abstract

Non-small cell lung cancer (NSCLC) constitutes more than one quarter of cancer deaths in the United States. The primary aim of this study is to identify risk factors and build a model for overall survival prediction based on patient and tumor factors from a large single institution database. The study population was taken from our cancer registry of 8620 lung cancer patients treated in the Indiana University Health System between 2000 and 2016. Patient and tumor factors were tested for their correlation with survival. The patient records with loss to follow up were censored. The survival analysis was performed using a multivariable cox proportional hazards regression model with R. The accuracy of the models were assessed by 8 folder cross validation. After excluding other histologies and duplicate registered cases, 3569 patients with confirmed NSCLC and complete staging data were eligible for this analysis. Median follow-up was 16.7 months in all patients and 41.8 months for alive patients. The median/5-year OS were 34.3 months/51.3%, 27.1 months/44.0% (HR = 1.2, p = 0.03), 14.2 months/18.0% (HR = 2.4, p = 2e-16) and 6.3 months/6.7% (HR = 4.9, p = 2e-16) for stage I, II, III and IV, respectively. Age, gender, race, marital status, smoking status, tumor size, stage, year of diagnosis and treating hospital were significant factors for survival (all p<0.001). Using these significant factors, a multivariate cox proportional hazard regression model was built, with the stage adjusted by gender (HR = 0.8 for female, p = 9.52e-7), smoking status (HR = 1.2 for current smoker, p = 0.001), age (HR = 1.03, p = 2e-16), marital status (HR = 1.2 for single, p = 0.001), tumor size (HR = 1.008, p = 2e-16), year of diagnosis (HR = 0.97 for more recent year, p = 1.3e-9) and hospital (HR = 0.8 for university hospital, p = 0.0001). This model was further tested with 8 folders cross validation, to have 66% accuracy which was slightly improved from 63% of using staging group alone. Many patient factors contribute to survival in patients with NSCLC. A model of combining independent patient and tumor factors provides only modest accuracy. Further study to integrate treatment modality and biologic factors is warranted to generate a more accurate survival model for precision medicine in NSCLC.