Epidemiology, health policy and outcomes
This session is intended to provide an overview of available approaches to analyze paired/correlated outcomes. Paired data are common in rheumatology studies where multiple joints within the body, or different sub-regions within the same joint may be involved. The measurements are more similar between the joints of the same individual than between different individuals due to similar genetic and/or environmental exposures, and the possible effect of one joint on another. Paired-unit design affects cross-sectional, longitudinal and clinical trial studies with continuous, discrete, ordinal and time-to-event outcomes. A first question to consider at study initiation is whether the person or the joint is the unit of analysis. The choice depends on the scientific question and should be done with an understanding of the strengths and weaknesses of both approaches. If the best strategy involves joint-level, rather than person-level analysis, the failure to incorporate the correlation between joints will lead to incorrect inferences. We will cover scenarios when analyzing the joint, rather than the person, would be advantageous and describe statistical methods that take into account the correlated components of paired data. At this bootcamp session, faculty will review examples from literature and demonstrate how to apply these methods to a real data set.