Bootstrap-Based Confidence Intervals for Survival Analysis: Insights from Lung Cancer Data

AbstractThe study explores the bootstrap resampling to assess uncertainty in survival analysis, by using theNCCTG lung cancer dataset. To find survival probabilities and mean survival time, Kaplan-Meieranalysis was used. Cox proportional hazards models and Weibull models are used to find therelationship between age, gender, and survival. To find confidence intervals for the median survivaltime and model parameters one thousand bootstrap samples were generated. By bootstrap analysis amedian survival time was find as 310 days, with a 95% percentile interval of 284–361 days, and 95%bias-corrected and accelerated (BCa) interval of 269–353 days. By using Cox model, the coefficientfor gender was -0.5132, this corresponds to a hazard ratio of approximately 0.60. After that, thebootstrap 95% BCa interval for this coefficient ranged from -0.8084 to -0.2029. The age coeficientwas small, and its confidence interval was zero. Then by, the Weibull analysis did not provide anystrong evidence of an age-related effect. These findings shows how bootstrap resampling provideuseful measures for survival estimates and regression parameters in censored data.