Nivolumab + Relatlimab vs BRAF/MEK Inhibitors
2026-02-13
How much can reweighting really buy us?
A Journal Club on nivolumab plus relatlimab versus BRAF/MEK inhibitors in BRAF-mutant advanced melanoma — and, more fundamentally, on how matching-adjusted indirect comparisons try to answer a treatment question when the randomized comparison does not exist.
Thirty people joined — and the methods kept us there
People represented 30 Unique Teams connections in the attendance export
Median time together 82 min Half the room stayed at least this long
Stayed ≥ 60 minutes 57% The meeting ran just over two hours
NIVO + RELA versus BRAF/MEK inhibitors
Efficacy of nivolumab plus relatlimab versus BRAF/MEK inhibitors for first-line treatment of BRAF-mutant advanced melanoma: a matching-adjusted indirect comparison
Miller DM, et al. BMJ Oncology. 2025;4:e000912.
doi: 10.1136/bmjonc-2025-000912
The clinical question was straightforward. The evidentiary problem was not: what can we infer about NIVO + RELA versus BRAF/MEK therapy when no head-to-head randomized trial exists?
The meeting spent real time opening the black box
Much of the session was devoted not simply to the paper’s result, but to the mechanics of the method itself. A matching-adjusted indirect comparison starts with individual patient data from one trial and reweights those patients until their weighted baseline characteristics resemble the published aggregate population from another trial.
01 · Start with IPD RELATIVITY-047 provided patient-level data for the BRAF-mutant NIVO + RELA cohort.
02 · Match what is observable Weights are chosen so the weighted NIVO + RELA cohort matches baseline characteristics reported in each comparator trial.
03 · Compare outcomes Weighted outcome models then compare OS, PFS, ORR, and safety across the reweighted populations.
The overlap constraint You can only adjust for characteristics measured in the IPD trial and actually reported by the comparator publication.
The weights matter Patients who resemble the target comparator population contribute more; patients who resemble it less contribute less.
ESS is the price Reweighting can reduce effective sample size, revealing how much information remains after forcing the populations to look alike.
What MAIC cannot do: an unanchored MAIC can only adjust for measured and reported cross-trial differences. Unmeasured prognostic factors, treatment-effect modifiers, eligibility differences, assessment schedules, subsequent therapies, and censoring conventions can still bias the comparison.
A clinically familiar pattern emerged
Longer-term OS favored NIVO + RELA
After matching, NIVO + RELA was associated with better longer-term OS versus the BRAF/MEK doublets; the comparison with the atezolizumab-containing triplet also favored NIVO + RELA overall.
Early and late effects were different
The proportional-hazards assumption did not hold cleanly. Early PFS was similar or sometimes favored targeted therapy, while later PFS tended to favor NIVO + RELA. That motivated interval Cox models rather than one global hazard ratio.
BRAF/MEK therapy still won on ORR
Objective response rate was lower with NIVO + RELA than with each comparator — an important reminder that early tumor shrinkage and long-term survival are not interchangeable treatment attributes.
Severe adverse events favored NIVO + RELA
After matching, grade 3/4 adverse events were less frequent with NIVO + RELA than with the BRAF/MEK doublets, with a similarly favorable comparison against the atezolizumab-containing triplet.
The methodological lesson was at least as important as the treatment comparison
Show the analytic machinery
The presentation deliberately walked through synthetic data, Kaplan–Meier reconstruction, variable matching, patient weights, rescaling, and weighted outcome models. The goal was to make MAIC feel traceable rather than magical.
The comparison is limited by what both studies tell you
A clinically important variable cannot be adjusted for if it was not measured in the IPD source or not reported in the comparator publication. The apparent precision of a weighted estimate does not remove that structural limitation.
Balance can cost information
The original BRAF-mutant NIVO + RELA cohort contained 136 patients, but effective sample sizes after matching varied substantially across comparisons. ESS is therefore not a decorative diagnostic; it helps show how aggressive the reweighting had to become.
Association after adjustment is not randomization
The final comparison may be more credible than a naïve cross-trial contrast, but it remains an unanchored observational comparison across trials. Residual confounding should stay visible in the conclusion.
Discussion standouts
A special thank-you to five colleagues whose questions, explanations, and perspectives helped make an unusually methods-heavy Journal Club accessible and useful.
Adewunmi O. Adelaja Sameer Gupta Frank Worden David M. Miller
The Teams record is retained with the meeting materials. There was no archived community-tenure survey available for this recap, so the table below sticks to the attendance record itself.
| Name | Minutes | Camera | Unmuted | Raised hand |
|---|---|---|---|---|
| Sonia Cohen | 122 | — | — | — |
| David M. Miller | 98 | ● | ● | — |
| Frank Worden | 97 | ● | ● | — |
| Isaac Brownell | 97 | ● | ● | ● |
| Juliane Andrade Czapla | 97 | — | ● | — |
| Shinya U Amano | 97 | ● | ● | — |
| Rhoda Myra Alani | 95 | ● | ● | — |
| Truelian Yu | 95 | — | — | — |
| Vern Sondak | 95 | ● | ● | ● |
| Shailender Bhatia | 93 | ● | ● | — |
| Christine C. Cimoch | 92 | ● | ● | — |
| Sameer Gupta | 92 | ● | ● | — |
| Elizabeth I. Buchbinder | 90 | ● | ● | — |
| Ross D. Merkin | 89 | ● | ● | — |
| Vatche Tchekmedyian | 88 | ● | ● | — |
| Manisha Thakuria | 75 | — | ● | — |
| Natasha Hill | 69 | — | — | — |
| Kamaneh Montazeri | 55 | ● | ● | — |
| Krista M. Rubin | 51 | ● | — | — |
| Alex Sorrentino | 45 | ● | — | — |
| Riley M. Fadden | 45 | — | ● | — |
| Adewunmi O. Adelaja | 44 | — | ● | — |
| NIkhil Khushalani | 36 | ● | ● | ● |
| Sunandana Chandra | 32 | ● | ● | ● |
| Ade Adamson | 29 | ● | — | — |
| Vishal Patel | 24 | — | ● | — |
| Taylor Harp | 15 | — | — | — |
| Ajay N. Sharma | 5 | — | — | — |
| Sg | 4 | — | — | — |
| Song Park | 1 | — | — | — |