PRM77 - An Open-Source Toolkit for Developing Flexible Evidence-Based Decision and Simulation Models for Value Assessment in Oncology with R

Abstract

Cost-effectiveness models have historically been built using spreadsheet software, but recently researchers have begun developing models using programming languages, such as R, to overcome its limitations. However, current software does not have the flexibility to develop cost-effectiveness models in oncology representative of appropriate evidence synthesis of the available data. Our objective was to create open-source software for developing flexible evidence synthesis based decision models to help assess value in oncology. We developed an R package that provides a general framework for decision modeling using survival analysis, partitioned survival analysis, or multi-state modeling. Cost-effectiveness analysis can either be performed for competing single or multi-line sequential treatment strategies. Relative treatment effects used in the model can either be based on a single study or an evidence synthesis of multiple studies. Parametric, spline, and fractional polynomial based analyses are supported. Parametric distributions include those available in the R package flexsurv and corresponding reparameterizations appropriate for network meta-analysis. The core code is written in C++ for speed in order to facilitate probabilistic sensitivity analysis, structural uncertainty analysis, individual patient simulation, and integration with web applications. The R package is freely available for download on GitHub. The software and methodology is illustrated with a cost-effectiveness analysis in oncology. A new toolkit for developing decision and simulation models for value assessment of medical interventions in oncology has been developed. The R package facilitates the proper integration of parameter estimates obtained with advanced meta-analysis techniques to ensure disease progression and relative treatment effects reflect the available data. Moreover, the large number of available probability distributions reduces barriers to exploring the implications of different structural assumptions.