Colossus: Bridging the gap between big data and radiation epidemiology.

Abstract

Software to fit complex models using big data sets is needed to answer persistent and emerging questions in radiation epidemiology. The open-source R package Colossus has been developed to meet this need. Colossus was designed to take advantage of the input and graphing flexibility of R scripts, employ multi-core systems to run analyses faster, and permit the straightforward addition of future capabilities. The Million Person Study has used Colossus for analyses that estimate exponential hazard ratios and linear excess relative risks, both with Wald confidence boundaries and likelihood-based boundaries. Incorporating methods to propagate covariate uncertainty into model parameter uncertainty is the next major focus area. In addition, piecewise dose-response models, gradient descent algorithm options, and other statistical tests are being implemented.