Fractional Polynomial Modelling In Network Meta-Analyses Of Cancer Immunotherapies In Advanced Non-Small Cell Lung Cancer (NSCLC)

OBJECTIVES: Cancer immunotherapies may have a delayed onset of treatment effect and may lead to long-term survival in a proportion of patients, something not necessarily seen with conventional therapies. Therefore, a fractional polynomial (FP) approach to network meta-analysis (NMA)a, which does not assume proportional hazards, may be more suitable than standard hazard ratio (HR) based NMA for modeling overall survival (OS) and progression-free survival (PFS). METHODS: For the analysis of OS and PFS, we fit first order and second order fractional polynomials with powers pi and p2 from the set (01), and fixed effects and random effects with heterogeneity for the intercept. Comparisons of OS and PFS between Roche's cancer immunotherapy drug atezolizumab (anti-PDLl antibody) used as monotherapy with other treatments in previously treated advanced NSCLC were made via a Bayesian FP NMA. Model fit was assessed based on Deviance Information Criterion (DIC) and visual inspection. RESULTS: Fixed effects models often provided a better fit than random effects with the lowest DICs. Second order models often provided lower DICs than first order for OS and PFS; however, visual inspection of the curves showed a survival 'plateau', meaning a proportion of patients did not experience the event during the time horizon. First order models were thus a better fit by visual examination. The modeled HRs were not proportional over time, supporting the use of FP NMA over standard HR based NMA in this case. CONCLUSIONS: Fractional polynomials NMA can be a suitable approach when modeling non-proportional hazards, as seen with cancer immunotherapies in advanced NSCLC. a Jansen JP. Network meta-analysis of survival data with fractional polynomials. BMC Med Res Methodol 2011;11:61.