Using multi-level regression to determine associations and estimate causes and effects in clinical anesthesia due to patient, practitioner and hospital or health system practice variability

Abstract

In this research methods tutorial of clinical anesthesia, we will explore techniques to estimate the influence of a myriad of factors on patient outcomes. Big data that contain information on patients, treated by individual anesthesiologists and surgical teams, at different hospitals, have an inherent multi-level data structure (Fig. 1). While researchers often attempt to determine the association between patient factors and outcomes, that does not provide clinicians with the whole story. Patient care is clustered together according to clinicians and hospitals where they receive treatment. Therefore, multi-level regression models are needed to validly estimate the influence of each factor at each level. In addition, we will explore how to estimate the influence that variability—for example, one anesthesiologist deciding to do one thing, while another takes a different approach—has on outcomes for patients, using the intra-class correlation coefficient for continuous outcomes and the median odds ratio for binary outcomes. From this tutorial, you should acquire a clearer understanding of how to perform and interpret multi-level regression modeling and estimate the influence of variable clinical practices on patient outcomes in order to answer common but complex clinical questions.