Author response to ‘Comment on “Population-adjusted indirect treatment comparison of maintenance PARP inhibitor with or without bevacizumab versus bevacizumab alone in women with newly diagnosed advanced ovarian cancer” ( Ther Adv Med Oncol . 2021 Sep 30;13:17588359211049639)’

Dear Editor,

We thank Nakai and Matsumura for their Letter to the Editor on our paper entitled, ‘Population-adjusted indirect treatment comparison of maintenance PARP inhibitor with or without bevacizumab versus bevacizumab alone in women with newly diagnosed advanced ovarian cancer’. ¹ In their letter, the authors critique the appropriateness of the indirect treatment comparison (ITC) of PAOLA-1 and PRIMA because of differences in prior bevacizumab use across studies, which they argue, leads to a more favourable prognosis in PAOLA-1 than PRIMA. The authors draw upon data from the GOG-0218 trial of bevacizumab in ovarian cancer ² and other bevacizumab studies in cervical cancer in their evaluation. In this response, we outline a number of challenges with the comparisons made by the authors, and their resulting conclusions.

The cross-trial comparison of GOG-0218 with PAOLA-1 and PRIMA is confounded by differences in the characteristics of patients that entered the maintenance phase of GOG-0218 and those enrolled to PRIMA and PAOLA-1.^3,4 The cohort entering the maintenance phase of GOG-0218 included patients with stable disease or response after chemotherapy, while both PRIMA and PAOLA-1 enrolled only patients with response to prior chemotherapy.^3,4 Without patient data, it is infeasible to estimate outcomes for the responder-only cohort of GOG-0218. Enrolment to GOG-0218 was also restricted to patients with upfront surgery, ² while both PRIMA and PAOLA-1 enrolled patients with upfront or interval surgery.^3,4 Without adjustment, these differences confound the comparison of GOG-0218 with PRIMA and PAOLA-1 and may explain the difference in outcomes between maintenance bevacizumab in GOG-0218 and bevacizumab in PAOLA-1 (e.g., inclusion of patients with stable disease in GOG-0218).

The restriction to the post-baseline maintenance-only phase of GOG-0218 further complicates the authors’ assessment of potential imbalances between arms in the ITC. This is because the GOG-0218 maintenance data are themselves confounded from population imbalances due to post-randomisation selection. ⁵ Thus, showing similarity between placebo arms of GOG-0218 and PRIMA, but not between bevacizumab arms of GOG-0218 and PAOLA-1, does not conclusively demonstrate prognostic imbalances in the ITC. These differences may reflect imbalances in patient characteristics between the arms of the maintenance-only phase of GOG-0218, noting that only the ITC includes adjustment for population differences.

By matching PAOLA-1 to PRIMA on inclusion criteria and via propensity weighting for key prognostic characteristics, in our analyses we attempted to minimise population differences, including those arising from the different prior therapy use in the studies. We matched on a range of variables known to be prognostic of outcomes in ovarian cancer, including response status after chemotherapy. Residual disease status after surgery was not reported in the primary paper for PRIMA, ³ however, it became available later through reimbursement submissions. It was found to be balanced across arms in the ITC despite not being included in the matching analysis. ¹ Tumour size was not reported in PRIMA for matching. The location of tumour in terms of ovary versus fallopian tube was not found to be prognostic in exploratory analysis of PAOLA-1 and hence was excluded from the matching. ¹ The exclusion of tumour location or size from the matching is therefore not expected to materially impact on the ITC results. To our knowledge, there is no evidence to suggest that propensity for adverse events with bevacizumab is prognostic of outcome and therefore a source of bias in the analysis.

In summary, we agree with Nakai and Matsumura that any comparison of trials by ITC requires careful assessment and we welcome the opportunity to reply to the issues raised. We feel it important that the results should be placed in the context of the wider evidence base, including GOG-0218 and other trials. It is our view that the cross-trial comparison by Nakai and Matsumura does not provide clear evidence of prognostic imbalances in the ITC given population differences between studies and the confounding of post-randomisation data in GOG-0218. The validity of the ITC is supported by the similarity in the progression-free survival (PFS) comparing bevacizumab versus placebo [hazard ratio (HR) = 0.55, 95% confidence interval (CI) 0.43–0.70) with the randomised comparison of bevacizumab versus placebo in GOG-0218 (HR=0.62, 95% CI 0.52–0.75). ⁶ The ITC result showing an improvement in PFS for PARP inhibitor plus bevacizumab versus PARP alone is further supported by randomised studies in later lines of ovarian cancer. ⁷ As noted in our analyses, ¹ the ITC results are subject to limitations, and should be viewed as hypothesis generating in the absence of confirmatory randomised head-to-head trials.