Abstract
OBJECTIVES: Adding ramucirumab to docetaxel after failure of chemotherapy improves response rates and survival in aNSCLC. TREAT-LUNG aimed to characterize real-world use of ramucirumab+docetaxel or docetaxel, following ICIs and chemo for aNSCLC.
METHODS: Retrospective study of real-world data abstracted/aggregated from 3 EHR databases (Flatiron/COTA/IQVIA). Eligible patients had received second-line (primary-2L-cohort) or third-line (3L-cohort) ramucirumab+docetaxel or docetaxel for aNSCLC, following ICIs and chemo, between 03/2015-12/2019. Primary endpoint was real-world objective response rate (rwORR) among patients with evaluable tumor response assessment data. Minimum required sample size for primary analysis of rwORR was 286, based on a priori power analysis. Treatment-group comparisons were adjusted using propensity-score stratification.
RESULTS: After adjustment, real-world tumor response assessment was unknown/missing for 121 of 243 patients (50%) in the primary-2L-cohort, 206 of 361 patients (57%) in the 3L-cohort, and 311 of 557 patients (56%) in the combined-(2L+3L)-cohort. Among the subset of the 2L-cohort with evaluable real-world tumor response data (n=121), rwORR was 43% in ramucirumab+docetaxel-group (n=78) and 39% in docetaxel-group (n=43), odds ratio, 1.20; 95% confidence interval, 0.54-2.65. This lack of significant difference between groups was consistent for disease control rates and across 3L (n=155) and combined-(n=246)-cohorts. Secondary effectiveness outcomes such as real-world PFS and OS among all eligible patients were similar between groups (data to be presented). Patterns of adverse events were consistent with established safety profiles for ramucirumab and docetaxel, with some (≤5%) occurrence of immune-related toxicities in both groups.
CONCLUSION: rwORR differences between groups did not reach statistical significance. Findings must be interpreted in context of study limitations including insufficient sample size for primary and secondary analyses, potential selection bias inherent in passively collected EHR datasets, and residual confounding despite propensity adjustment. Perhaps, more importantly, the high rate of missing tumor response data (>50% overall) may have implications for the design of real-world data studies.
