OBJECTIVES: To facilitate comparative effectiveness in METex14 skipping NSCLC, five real-world datasets (RWD) were pooled (516 therapy lines from 248 patients) and used to provide data for immunotherapy, chemotherapy, and the tyrosine kinase inhibitor crizotinib. This study details the steps taken to validate the analyses.

METHODS: Following dataset pooling, the analyses were evaluated for internal and external validity. As there were relatively few patients (range: 21–91) in each study, a ‘leave one out’ analysis was performed to assess internal validity, where one study was omitted from the pooled dataset in successive runs, and survival outcomes were compared. To assess external validity, matching adjusted indirect comparisons were conducted, reweighting RWD to estimates from published studies in patients with METex14 NSCLC and comparing time-to-event outcomes for consistency.

RESULTS: The ‘leave one out’ validation resulted in similar estimates of progression-free survival (PFS) and overall survival (OS) in all cases, including point estimates of survival at key timepoints and the shape of Kaplan–Meier curves. This finding was seen across all comparators (median PFS: 2.7–3.6 months for immunotherapy, 4.1–6.4 months for chemotherapy, and 8.1–10.0 months for crizotinib). The external validation by reweighting RWD to match published immunotherapy data led to a median PFS of 3.3 months compared with the published estimate of 4.7 months, and median OS of 15.0 versus 13.1 months, respectively. This finding also held when comparing MET inhibitors.

CONCLUSIONS: Pooling of data allows for better estimation of outcomes despite small sample sizes in individual datasets, though with concerns regarding heterogeneity. Where typical measures of heterogeneity (such as I²) are unavailable, alternative approaches (such as ‘leave one out’ analysis) can be used to demonstrate the internal consistency and external validity of real-world comparator datasets, before using the pooled dataset to perform indirect comparisons with novel interventions.