OBJECTIVES: With a connected network of randomized controlled trials (RCTs), NMA is often utilized to simultaneously compare the included interventions. This study explored the application of available methods allowing for the integration of a disconnected study into an NMA, using a case study in first-line treatments for R/M-HNSCC.

METHODS: The network featured six RCTs, evaluating pembrolizumab+platinum+5-FU, platinum+5-FU+cetuximab, platinum+5-FU, cisplatin+docetaxel+cetuximab, cisplatin+paclitaxel, cisplatin, 5-FU, and methotrexate. The disconnected study was KEYNOTE-B10, a single-arm trial of pembrolizumab+carboplatin+paclitaxel (B10 regimen). Aggregate-level matching (ALM), additive component NMA (CNMA), random effects on baseline (REB), and reference prediction (RP) were used to integrate aggregate-level data from KEYNOTE-B10 into the network. Individual patient-level data (IPD) from KEYNOTE-B10 were also reweighted (via a matching-adjusted indirect comparison) and included in KEYNOTE-048 (MAIC+NMA), which was determined as the most similar trial in the network to KEYNOTE-B10. Fixed-effect NMA models in Bayesian framework were used to estimate odds ratios (ORs) with 95% credible intervals (CrIs) for comparisons of objective response rate (ORR) for B10 versus the above-mentioned interventions.

RESULTS: ALM, CNMA, and MAIC+NMA showed B10 significantly improved ORR in five comparisons (ranges of ORs: 2.96-9.76, 2.14-14.15, and 3.24-10.27, respectively). Results were not significant in the three remaining comparisons; ORs favored B10 in ALM and MAIC+NMA, while the direction of effects differed in CNMA. REB and RP had more variable estimates with wider CrIs and no significant results.

CONCLUSIONS: ALM was shown to estimate ORs with reasonable accuracy when only aggregate-level data were available. Additive CNMA was not as reliable given some interactions between treatments were not captured. REB and RP provided large uncertainty, likely due to their strong assumption of exchangeable referent treatment. Leveraging IPD in MAIC+NMA yielded more precise and reliable estimates due to adjustments for population differences, demonstrating superior/comparable efficacy of B10 versus other first-line treatments in R/M-HNSCC.