OBJECTIVES : To develop a novel network meta-analysis approach based on Royston Parmar model to compare the overall survival (OS) of nivolumab and pembrolizumab for 2^nd or later line of NSCLC treatment.

METHODS : Following a recently published network meta-analysis, OS of PD-L1 >=1% patients was compared for nivolumab 3mg/kg per 2 weeks and pembrolizumab 2mg/kg and 10mg/kg per 3 weeks based on the nivolumab studies Checkmate 057 and Checkmate 017 and pembrolizumab study Keynote 010. A Royston Parmar model-based network meta-analysis approach was developed using internal knots at 6 and 12 months and outer knots at 0 and 30 months for all treatments and all studies using the Weibull transformation log(-log(S)). The resulting coefficients were pooled. The docetaxel arm in Keynote 010 was used as the reference treatment. This approach is compared with applying pooled Hazard Ratios of nivolumab versus docetaxel to the docetaxel spline and with using a standard parametric extrapolation distribution based NMA.

RESULTS : Using the spline based NMA approach the nivolumab and pembrolizumab curves crossed twice. At six months, predicted OS percentages for nivolumab, pembrolizumab 2mg and pembrolizumab 10mg were 69%, 72% and 73%; at 12 months 58%, 46%, 53%; at 24 months 26%, 33%, 37% and 0.2%, 24% and 23% at 60months.

Both the pooled Hazard Ratio NMA approach and the standard parametric extrapolation methods were not sensitive to capture the survival differences at 12 months: Using the pooled Hazard Ratios, the nivolumab survival percentages were estimated to be 74% at 6 months, 49% at 12 months, 24% at 24 months and 3% at 60 months. In the standard parametric survival distribution based NMAs the curves only crossed once.

CONCLUSIONS : For the comparison of nivolumab with pembrolizumab, the new spline-based approach better captured and utilized the non-proportional changing hazard rates of OS over time.