An interactive application to support the selection of parametric hazard functions for use in decision analytic models. FORECAST

Abstract

Objectives A common challenge for construction of a multistate decision analytic model is to estimate a parametric rate of events or “hazards” that govern the transitions from one state to another over time. This article describes an application, called FORECAST (FOrward REsearch on Clinical And Survival Trends), to help the analyst select an appropriate function. Methods FORECAST interacts with the user via a portable version of R but requires no knowledge of R and the user does not need to have R installed. In practice, the user will deal with a standalone Shiny interface that opens on the user’s web browser. Results Taking as input a spreadsheet containing the digitalized coordinates of published Kaplan-Meier survival curves and number at risk, the FORECAST application provides an integrated software tool to help the user choose a parametric hazard function of an event. The tool presents statistical diagnostics (proportional hazards test, Akaike Information Criterion) and visualisations (log-cumulative hazards, predicted hazards, predicted probability of survival) to aid model selection, and output that can be downloaded to a spreadsheet (predicted hazards and coefficients of the models). Conclusion FORECAST is aimed at academic researchers, health technology evaluators, and industry analysts preparing economic evaluations to support health technology assessment dossiers. R provides powerful programs but requires continuous updates and a steep learning curve. FORECAST provides a suite of diagnostics and data visualizations to choose an appropriate function, without providing advice. Users are invited to adapt it under creative commons license.