OBJECTIVES: At health technology assessment (HTA) overall survival trial data is often immature. With sufficient follow-up several immunotherapies have shown disease related event risks decreasing to zero. Extrapolation methods often cannot capture this based on immature trial data alone. The aim was to inform long term survival of therapies by Bayesian hierarchical network meta-analysis (BH-NMA), which borrows information of therapies in the same class across trials.

METHODS: A network of four trials in previously treated metastatic non-small cell lung cancer is used with PD-L1>1% comparing nivolumab, pembrolizumab and atezolizumab (twice) individually with docetaxel. The atezolizumab trial data were relatively immature. The BH-NMA approach is applied to a Weibull mixture cure (WMC) NMA and is compared with a standard WMC survival NMA approach. We applied class effects, i.e. parameters related to a certain class were assumed to share a common distribution. Two classes were distinguished, chemotherapy and immunotherapy. The class effects were applied on the cure rate of docetaxel and on the treatment effects on cure of immunotherapies. The fit-statistic, cure rates and mean survival were compared.

RESULTS: The standard WMC NMA predicts cure rates of 0.03[0.00;0.10], 0.03[0.00;0.06], 0.18[0.11;0.24] and 0.07[0.02;0.15] for atezolizumab, docetaxel, nivolumab, and pembrolizumab respectively. The corresponding incremental survival compared to docetaxel were 0.42[-0.59,1.72], 3.16[1.68,4.74], and 1.07[0.41,2.41] years. The BH-NMA model showed cure rates of 0.12 [0.03;0.23], 0.06 [0.03;0.09], 0.17 [0.11;0.23], and 0.12 [0.05;0.20] respectively. The corresponding predicted incremental survival compared to docetaxel was 1.33 [-0.05;3.71], 2.41 [1.05;3.99] and 1.67 [0.67;3.01] years.

CONCLUSIONS: The use of BH-NMA has important consequences in predicted incremental mean life years and therefore cost-effectiveness ratios and thus reimbursement decisions.