Greg-package: Regression Helper Functions
In gforge/Greg: Regression Helper Functions

Greg-package

R Documentation

Regression Helper Functions

Description

This R-package provides functions that primarily aimed at helping you work with regression models. While much of the data presented by the standard regression output is useful and important - there is often a need for further simplification prior to publication. The methods implemented in this package are inspired by some of the top journals such as NEJM, BMJ, and other medical journals as this is my research field.

Output functions

The package has function that automatically prints the crude unadjusted estimates of a function next to the adjusted estimates, a common practice for medical publications.

The forestplot wrappers allows for easily displaying regression estimates, often convenient for models with a large number of variables. There is also functionality that can help you comparing different models, e.g. subsets of patients or compare different regression types.

Time-splitter

When working with Cox regressions the proportional hazards can sometimes be violated. As the tt() approach doesn't lend itself that well to big datasets I often rely on time-splitting the dataset and then using the start time as an interaction term. See the function timeSplitter() and the associated vignette("timeSplitter").

Other regression functions

In addition to these funciton the package has some extentions to linear regression where it extends the functionality by allowing for robust covariance matrices. by integrating the 'sandwich'-package for rms::ols().

Important notice

This package has an extensive test-set for ensuring that everything behaves as expected. Despite this I strongly urge you to check that the values make sense. I commonly use the regression methods available in the 'rms'-package and in the 'stats'-package. In addition I use the coxph() in many of my analyses and should also be safe. Please send me a notice if you are using the package with some other regression models, especially if you have some tests verifying the functionality.