What response rate actually predicts in first-line melanoma immunotherapy

Design

individual patient data meta-analysis of four randomised controlled trials using a two-level meta-analytic surrogacy framework

Population

1,865 patients with treatment-naive metastatic melanoma on nivolumab or nivolumab plus ipilimumab versus dacarbazine or immunotherapy

Primary outcome

individual- and trial-level association of objective response rate, progression-free survival and time to next treatment or death with overall survival

Effect

ORR individual-level OR 12.29 (9.78-14.80), trial-level R-squared 0.62 (0-1.00); PFS rho 0.72, R-squared 0.73 (0.27-1.00); TNTD rho 0.77, R-squared 0.77 (0.37-1.00)

Trials increasingly report response rate or progression-free survival and let regulators infer survival. This analysis used individual patient data from 1,865 people in four randomised first-line trials of nivolumab, alone or with ipilimumab, against dacarbazine or another immunotherapy, and tested three candidates as surrogates for overall survival within the standard two-level meta-analytic framework.

At the individual level, objective response rate was strongly associated with survival (odds ratio 12.29, 95% CI 9.78-14.80). Progression-free survival and time to next treatment or death correlated moderately, with Spearman coefficients of 0.72 and 0.77. At the trial level - the level that actually matters for validating a surrogate - R-squared values were 0.62 for response rate, 0.73 for progression-free survival and 0.77 for time to next treatment or death, all with intervals so wide that they reached 1.00 and, for response rate, 0.

The practical reading is a caution rather than a finding. A patient who responds does better, which is a statement about prognosis. Whether a trial that improves response rate will improve survival is a different question, and these data say only that all three candidates predict moderately and none is established. Time to next treatment or death performed best, which is interesting because it captures something clinicians already use informally - how long a line of therapy holds. When the next melanoma trial reports response rate at an interim analysis, this is the paper that says how much weight it carries.

• Distinguish individual prognosis from trial-level validity when reading a surrogate endpoint
• Treat a response-rate-only readout as promising rather than conclusive
• Time to next treatment or death tracked survival best and is easy to record in practice
• The findings apply to anti-PD-1-based first-line therapy, not to targeted therapy or later lines
• Very wide trial-level intervals mean none of these surrogates is formally validated

The statistics, in plain English

Trial-level R-squared asks how much of the variation in survival benefit across trials is explained by variation in the surrogate's benefit. A value of 0.62 with an interval from 0 to 1.00, as for response rate, means the data are compatible with the surrogate explaining everything or nothing - four trials cannot pin this down. The individual-level odds ratio of 12.29 is precise but answers a different question: responders live longer, which was never in doubt. Cross-validation showing prediction intervals that always contained the observed hazard ratios indicates the models are stable, not that they are accurate.

Read the rest in the app

You have read your two free briefings this month. The app carries all 27 specialties, every morning, free — and this finding is waiting in it.

Get Daily Doctor — free

Scan to keep reading on your phone. No account needed to start.