465Simplifying the Differences between Causal and Prediction Analyses

Abstract

Abstract Focus of Presentation Multivariable regression models can be used to answer a variety of clinical questions. The two main objectives of regression models are to either 1) understand an association between one or more exposures and an outcome; or 2) predict future outcomes based on certain exposures or variables. To simplify this, we will consider the former causal analysis, and the latter prediction analysis. This presentation will explain the steps in model development and assessment using a clinical case study, highlighting the similarities and differences. This presentation is aimed at trainees. Findings The key differences between causal and prediction models include: the purpose and research questions, power calculations, variable selection, model specification, testing model fit, and the desired outcome of each model. The case study demonstrates these differences, while working through a causal and prediction model with similar clinical questions. Conclusions/Implications It is important for researchers to consider the purpose of their research question and to tailor the model accordingly. This will guide the model development and interpretation, which are different for causal and prediction analyses. Key messages A thorough understanding of the types of models available, their assumptions, and the process of model development and assessment is essential to conducting research that is valid and applicable to the clinical environment, enabling knowledge translation.