Background: Development of medicines in rare oncologic patient populations are growing, but well-powered randomized controlled trials are typically extremely challenging or unethical to conduct in such settings. Quantitative bias analysis (QBA) applied to external control arm (ECA) analyses represent an opportunity for evaluating the validity of the corresponding comparative efficacy estimates.
Objectives: We provide a brief overview of QBA approaches and specifically how QBA can be applied to ECAs through a comparative effectiveness study in lung cancer. We illustrate how QBA is used to ascertain robustness of results despite missing data and potential of unknown confounding. The focus is on the methods application, not the clinical results.
Methods: The case study assesses the comparative effectiveness of pralsetnib versus pembrolizumab ("pembro") and combined with chemotherapy ("pembro+chemo") for overall survival (OS). The single-arm data is from ARROW (NCT03037385), and RWD from the Flatiron Health (FH). Patients were filtered for availability of smoking status, disease stage, ECOG performance status (PS), and non-squamous histology.
The ECA was constructed using FH, and confounders balanced using inverse probability treatment weighting. Hazard ratios (HRs) were estimated using the Cox-proportional hazards model.
The QBA analyses for missing data used tipping point analysis and multiple imputation using chained equations. The E-value was used to assess the robustness of results against unmeasured confounding.
Results: The HRs estimated post-adjustment were for the comparison with pembro was 0.33 (95% CI, 0.18–0.61), and for pembro+chemo 0.36 (95% CI, 0.21–0.64).
The results for QBA for missing data were nearly identical for both comparisons. Across all scenarios, the resulting HRs and 95% CIs showed minimal sensitivity, with HRs ranging from 0.36 to 0.41 and the upper 95% CI limits ranging from 0.65 to 0.74.
The E-value was 3.31 for the comparison with pembro, and 3.37 with pembro+chemo. We expected our results are robust to plausible unmeasured confounding since QBA suggested it would be implausible for sufficiently large systematic differences in unmeasured prognostic variables to reverse our findings.
Conclusions: QBA represents an important tool for expanding the acceptability and confidence of external control arm studies, particularly when real-world data are used to constitute part, or all, of the control arm. Familiarity with QBA methods such as tipping point analysis and E-values are increasing, and influential regulatory and HTA bodies are calling for their explicit use in the submission of external control arm evidence. Our case study demonstrates how QBA can be used to assess study robustness to data missingness and residual confounding.
.jpg "Grace Hsu, MSc photo")