Making the Most of Your Data: Using an Alternative Statistical Methodology to Multi-level Modeling to Investigate Hospital Effects on Acute Hospital Length of Stay Following Stroke when Number of Hospitals is Small

Recent research in stroke has shown that differences in hospital characteristics such as staffing levels and facility provision partly explain variations in mortality rates seen between hospitals. It is important to determine whether this also applies to other important stroke-related outcomes. If it can be shown that hospital characteristics play a role in determining disparities in patient outcomes after stroke, policy makers and service commissioners can target modifiable hospital characteristics to improve outcomes and increase equity in care. The influence of hospital characteristics on patient outcomes is regarded as contextual effects. Medical researchers interested in investigating contextual effects usually employ multi-level modeling techniques as it overcomes the limitations of traditional single-level regression models by modeling relationships independently at different levels. In this case study we demonstrate that the use of multi-level modeling when the number of hospitals sampled is small leads to vastly misleading results. We show how employing a much more conservative, robust approach which considers hospital as a fixed-effect should be favored. Such an approach explores the relationship between the outcome of interest (which in this example is length of stay) and hospital characteristics in a descriptive manner. By studying this case study, researchers should have an understanding as to why they should avoid the popularized multi-level modeling method when the number of hospitals is small, and consider employing this alternative approach that makes the best use of the data in the most robust way to overcome such limitations.