Comparative Efficacy of Treatments in Advanced/Metastatic Pancreatic Cancer: Demonstrating the Power of Multi-Level Network Meta Regression

OBJECTIVES: Multilevel network meta-regression (ML-NMR) is a flexible method for indirect treatment comparisons that lends itself well to comparing survival outcomes. The aim was to assess the efficacy of treatments for pancreatic cancer using survival data from seven studies, each comparing gemcitabine (GEM) monotherapy with GEM in combination with one of six other treatments, using an ML-NMR.

METHODS: Using the R package multinma, a Bayesian ML-NMR was performed on pseudo individual patient data (IPD) reconstructed from published overall survival (OS) curves using the Guyot algorithm. The proportion of male patients in each study was used as a covariate, and the shared effect modifier assumption was made. The best fitting model was selected according to the lowest Leave-One-Out-Information-Criterion (LOOIC) score. The restricted-mean survival time (RMST) and median OS were assessed as the main outcomes of interest.

RESULTS: The FE log-logistic model gave the lowest LOOIC score out of all 12 models. Under this model, GEM in combination with nab-paclitaxel (GEM+NAB) gave the highest estimated median OS and RMST in each study population. GEM plus capecitabine (GEM+CAP) gave similar RMST estimates to GEM+NAB, but worse median OS.

CONCLUSIONS: GEM+NAB and GEM+CAP provide increased OS compared to GEM and several other treatment regimens for advanced/metastatic pancreatic cancer. Despite not including any IPD in the network, the ML-NMR method still produced results in line with an NMA from 2014 by Gresham et. al, demonstrating the power and flexibility of the method. Future work may include more treatments in the network and more covariates.