Predictive Models and Visualizations according to Outcome Variables Using R – Focusing on Regression Analyses

Abstract

Predictive models have recently become increasingly important across various fields. In particular, in clinical research, the main purpose is to build a model that can find risk factors and predict a specific disease. Predictive models can help clinicians make fast and accurate decisions by capturing relationships between multiple factors related dependent variables. Accordingly, this paper describes a predictive model construction method and visualization that can be useful in clinical research. As dependent variables can be divided into continuous, categorical, and survival variables, the concepts and principles of linear, logistic, and cox regression analyses for building predictive models are explained in this paper. In addition, we investigated how to select variables to create an optimal model and how to evaluate the discrimination and calibration of the model. A visualization method that can help interpret according to each regression analysis model is also described. This paper will provide basic knowledge for clinical researchers to more easily build predictive models and evaluate them for practical use.