Overall Survival Analysis-Lung Cancer patients treated with Pembrolizumab

Due to restricted access to individual patient level data (IPD) from pembrolizumab clinical trials, we employed a validated, evidence-based data synthesis approach to generate a structured dataset suitable for risk modeling. We conducted a systematic literature review across three major databases: Pubmed, Embase , and the Cochrane Library , yielding 2,770 records. After title, abstract, and full-text screening, 36 studies met the inclusion criteria.For each study, we extracted time-to-event data from published Kaplan--Meier (KM) survival curves using image digitization and reconstructed IPD using the algorithm described by \cite{guyot2012ipd}, which preserves original sample sizes and censoring distributions. Demographic and clinical covariates were simulated to match published summary statistics: categorical variables were sampled from multinomial distributions using reported proportions, and continuous variables were generated from parametric distributions fitted to published means and standard deviations. When only medians and interquartile ranges were available, standard moment-matching techniques were applied to approximate the underlying distributions. To reflect uncertainty inherent in the imputed covariates and event times, we generated ten multiple imputed datasets and combined estimates across imputations using Rubin’s rules. In each dataset, we encoded 12-month mortality as a binary outcome, where \( y_i = 1 \) if death occurred within 12 months, and \( y_i = 0 \) otherwise. This short-term binary endpoint was used solely as a validation step to confirm the fidelity of the reconstructed IPD and was not the primary focus of our analysis. As an additional validation step, we evaluated the consistency between the reconstructed and published survival curves by overlaying Kaplan--Meier plots and computing the root mean square error (RMSE) between survival probabilities at reported time points, following best practices for IPD reconstruction.