Model-Based Meta-Analysis on the Efficacy of Pharmacological Treatments for Idiopathic Pulmonary Fibrosis

Aims: Recently the FDA approved the first two drugs (pirfenidone and nintedanib) indicated for the treatment of idiopathic pulmonary fibrosis (IPF), a therapeutic area that has highly variable placebo effects across clinical trials but more consistent treatment effects within trials1. The purpose of this analysis was to leverage publically available data to understand and quantify comparative efficacy in order to support future trial designs and decision making as part of model-informed drug development in IPF.

Methods: An augmented database with efficacy and safety endpoints was developed using published data from PubMed literature, FDA summaries, and reports that includes 40 citations from 32 trials. A model-based meta-analysis (MBMA) with a longitudinal profile of placebo-corrected change from baseline (mean difference) of percent predicted forced vital capacity (%predicted FVC) was characterized by a regression model using maximum likelihood estimation as implemented in R. Multiple levels of heterogeneity were described as mixed-effect variability terms. Covariates of population characteristics were imputed for trial arms with missing values, and standard deviations were modeled and predicted. Treatment effects, time courses, and potential dose-response relationships were estimated, and covariate effects on treatment effects were investigated.

Results: The analysis dataset is composed of %predicted FVC summary-level data of 5090 subjects from 45 arms in 20 trials, 17 of which were double blinded and placebo controlled. The trial durations ranged from 8 to 470 weeks. A non-parametric placebo model was used to fit the longitudinal %predicted FVC data. Treatment effects were estimated for all 14 active treatments in the analysis dataset, and the dose-response of pirfenidone was characterized using a step-function (low vs. high dose). The final model includes a time course of the drug effect that is described by an increasing form of the exponential decay function: 1-exp(-lambda*time), where lambda determines the steepness of the curve. Covariates investigated for their potential to influence treatment effects included baseline %predicted FVC, disease duration, age, gender, race, and smoking status. The covariate distribution ranges were narrow, and ultimately baseline %predicted FVC was the only covariate included in the final model. Model diagnostic plots showed that the model predictions adequately fit the observed data. Pirfenidone and nintedanib were the only drugs identified to have significant estimated positive treatment effects, with 95% confidence intervals not overlapping zero. Model simulations were performed to further evaluate the treatment and covariate effects on longitudinal change from baseline %predicted FVC.

Conclusion: A mixed-effect model was developed using publically available data to describe the treatment and covariate effects on longitudinal profiles of change from baseline %predicted FVC. The results from this MBMA approach allow quantification of treatment profiles and estimation of relative efficacy between treatments that have not been directly evaluated in the same trial, which is essential for drug development in a therapeutic area such as IPF. Additionally, clinical trial simulations using the final model developed from the MBMA will be used to inform future trial designs.

1. FDA Clinical Pharmacology and Biopharmaceutics Review(s) for Pirfenidone, 2014.