Overall Survival In Patients With Non-Small Cell Lung Cancer: A Comparison Of Clinical Trial Versus Real-World Outcomes Using A Propensity Score Reweighting Approach

Abstract

Overall survival (OS) is an established outcome in randomized controlled trials (RCTs) addressing new cancer treatments. However, because characteristics of patients included in RCTs typically deviate from real-world patient characteristics, it is unclear whether OS outcomes as shown in RCTs can be used to estimate the real-world OS associated with observed treatments. We aimed to compare the OS reported in a recent RCT for advanced non-small cell lung cancer (aNSCLC) patients receiving first-line chemotherapy with the real-world OS of that chemotherapy. Using claims data from a German sickness fund, we identified patients diagnosed with aNSCLC initiating a first-line chemotherapy (minimum follow-up 18 months). OS data from the aNSCLC-RCT were extracted by digitizing the reported Kaplan-Meier curve (KMC) of the chemotherapy group. The real-world KMC was adjusted using propensity score reweights with regard to the aggregated aNSCLC-RCT patient characteristics. Survival outcomes were finally compared using Cox proportional hazards regression. 95 aNSCLC-patients initiating first-line chemotherapy were identified in the real-world dataset, median OS was 12.2 months (95%CI:7.2-14.8). The RCT included 151 patients in the chemotherapy group, median OS was not reached after 18 months. Once an adjusted patient sample was used (adjustment for differences in age, gender, brain metastases, smoking status), median OS of observed 95 patients decreased from 12.2 to 8.1 months (95%CI:4.3-12.0). If this KMC is compared to the RCT-KMC, patients seemed to have a better survival prognosis in the RCT in comparison to the real-world (Hazard ratio:1.89; 95%CI:1.35-2.65). Our analysis confirms that the real-world OS of a treatment is likely to be worse than OS of that same treatment in RCTs, even if an adjustment for differences in patient characteristics is done. Whether this is related to any unobserved differences in patient characteristics or to differences in the diagnosis/treatment framework needs to be further investigated.