OBJECTIVES: Due to limited follow-up in randomized trials (RCTs), survival extrapolation beyond the follow-up period is essential to inform reimbursement decisions. Depending on the hazard rates, more advanced extrapolation survival models might be justified. NMAs are used to compare studies in the absence of direct evidence. This study aims to compare survival extrapolation outcomes using different NMA methods.

METHODS: A recently published NMA (Bosma et al. 2022) in first-line systemic therapies for mRCC was used as an example. The NMA included six different RCTs and seven different IO treatments (i.e., sunitinib, nivolumab+ipilimumab, avelumab+axitinib, nivolumab+cabozantinib, pembrolizumab+axitinib, atezolizumab+bevacizumab, pembrolizumab+lenvatinib). The Kaplan-Meier graphs from the RCTs were used to reconstruct pseudo individual patient-level data. The proportional hazard assumption (PHA) was assessed with a Schoenfeld test. The statistical fit of the models was assessed using the leave-one-out information criterion (LOOIC). Outcomes with a p-value<0.05 were considered statistically significant.

RESULTS: PHA violation and the delayed treatment effect in the included studies justified the use of PNMA and piecewise (datacut 12-months) NMA methods, respectively. Based on the LOOIC, lognormal and log-logistic were the best-fitted distributions for the PNMA and the piecewise NMA, respectively. Survival outcomes between the different NMA methods varied significantly. A statistically significant increase (i.e., p-value<0.05) in the IMS was estimated for the following treatments compared to sunitinib: nivolumab+ipilimumab [IMS: 3.38 (95%CI:1.40-5.33)], nivolumab+cabozantinib [IMS: 4.86 (95%CI:2.81-6.88)], pembrolizumab+axitinib [IMS: 3.53 (95% CI:1.54-5.49)], pembrolizumab+lenvatinib [IMS: 3.84 (95%CI:1.83-5.82)] in HRNMA; nivolumab+ipilimumab [IMS: 2.40 (95%CI:1.16-3.63)] in PNMA and nivolumab+ipilimumab [IMS: 1.90 (95%CI:0.51-3.27)] in piecewise NMA.

CONCLUSIONS: Different NMA methods can lead to significantly different survival outcomes. The appropriate NMA method should be selected carefully to avoid biased estimates in health economic models used for reimbursement decision-making.