Multivariable fractional polynomial method for regression model

One assumption in creating generalized linear model (GLM) is linearity in its link function. For example, in logistic regression model, covariates are assumed to be linearly associated with response variable in logit scale. However, it is not always the case and the assumption may be wrong. For example, lactate is associated with mortality outcome, but the relationship is not linear (1). Quadratic or cubic terms can be added to an explanatory variable to account for the non-linearity relationship. However, this requires subject-matter knowledge to determine the form of a variable. In exploratory study, such knowledge is always lacking and investigators have to rely on data to determine the functional form. Multivariable fractional polynomial (MFP) method is such a method that it allows software to determine whether an explanatory variable is important for the model, and its functional form (2,3). MFP can be used when investigators want to preserve continuous nature of covariates and suspect that the relationship is non-linear. The article aims to describe how to perform MFP methods by using R package. Fundamentals on MFP are also provided to make the article more readable.