125P - A parametric network meta-analysis of PD-1 inhibitors in second-line treatment for advanced esophageal squamous cell carcinoma

Background

The Network meta-analysis (NMA) of survival data is often based on the reported hazard ratios and relies on the proportional hazard (PH) assumption. To address the violation of the PH assumption, a non-proportional hazard (NPH) NMA was conducted using pseudo-patient-level data to assess the comparative efficacy of PD-1 inhibitors and chemotherapy in second-line treatment for advanced esophageal squamous cell carcinoma.

Methods

Embase and MEDLINE databases were searched for randomized controlled trials (RCTs) assessing PD-1 inhibitors for the treatment of ESCC. The Kaplan-Meier curves of overall survival (OS) from selected studies were digitized to generate pseudo-patient-level data using the Guyot algorithm. PH assumption testing was performed for all included studies and was found to be violated in multiple studies. A series of parametric models were fitted to the pseudo-patient-level data generated from the available KM curves using the method suggested by Ouwens et. al. The parametric NMA was performed using the “survivalnma” package in R. The Gelman-Rubin diagnostic, a measure of convergence for a list of Markov chain Monte Carlo (MCMC) sequences, was used to evaluate the convergence of the model.

Results

Five parametric distribution-based models (Exponential, Weibull, Gompertz, Log-normal, and Log-logistic) were considered for the NPH NMA. Based on the goodness of fit statistic, the best-fitted model was log-normal having a deviance information criterion (DIC) is 1445.7. The 3-year predicted survival of camrelizumab, nivolumab, sintilimab, pembrolizumab, tislelizumab and chemotherapy was 12%, 12%, 16%, 11%, 16% and 4%, respectively. The results should be interpreted with caution as the included trials were heterogeneous in terms of sample size, trial setting, and geography.

Conclusions

Parametric NMA is suitable for analyzing the comparative efficacy of multiple treatments in the presence of NPH as these methods leverage the use of distributional estimates (shape and scale) to conventional parametric survival models, fitted to patient-level data from the studies.

Clinical trial identification

Editorial acknowledgement

Legal entity responsible for the study

Heorlytics PVT. Ltd.

Funding

Has not received any funding.

Disclosure

All authors have declared no conflicts of interest.