Statistical primer: an introduction to the application of linear mixed-effects models in cardiothoracic surgery outcomes research-a case study using homograft pulmonary valve replacement data.

Abstract

Summary OBJECTIVESThe emergence of big cardio-thoracic surgery datasets that include not only short-term and long-term discrete outcomes but also repeated measurements over time offers the opportunity to apply more advanced modelling of outcomes. This article presents a detailed introduction to developing and interpreting linear mixed-effects models for repeated measurements in the setting of cardiothoracic surgery outcomes research.METHODSA retrospective dataset containing serial echocardiographic measurements in patients undergoing surgical pulmonary valve replacement from 1986 to 2017 in Erasmus MC was used to illustrate the steps of developing a linear mixed-effects model for clinician researchers.RESULTSEssential aspects of constructing the model are illustrated with the dataset including theories of linear mixed-effects models, missing values, collinearity, interaction, nonlinearity, model specification, results interpretation and assumptions evaluation. A comparison between linear regression models and linear mixed-effects models is done to elaborate on the strengths of linear mixed-effects models. An R script is provided for the implementation of the linear mixed-effects model.CONCLUSIONSLinear mixed-effects models can provide evolutional details of repeated measurements and give more valid estimates compared to linear regression models in the setting of cardio-thoracic surgery outcomes research.