Generalized Estimating Equations (GEEs), developed for the analysis of binary or categorical data in longitudinal studies, provide a flexible method for identifying associations between risk factors and repeated injury outcomes. They are highly useful for repeated measures datasets, particularly in settings where individuals sustain multiple injury events during the study period, e.g. falls in older adults, occupational injuries in high-risk industries, and injuries in athletes.
Unfortunately, GEEs provide very limited information about the longitudinal associations within the repeated injury outcomes themselves. This is troublesome, since previous injury is frequently an important predictor of future injury.
A new method, Alternating Logistic Regression (ALR), overcomes this problem by explicitly modeling the association between the injury outcomes at various time points. Logistic regression is used to estimate the dependence within the outcome measures over time, while a GEE model estimates the association between the predictor variables and the outcomes. The algorithm switches back and forth between the GEE and the logistic regression, hence the name ALR.
We illustrate the application of ALR to data on injury in rugby players. There is interest in identifying high-risk periods of the sports season. An ALR model with terms representing each bi-weekly period indicates that the first 6 weeks of the season carry the greatest risk, with, for example, weeks 3-4 (early season) having an odds ratio for injury of 1.67 (95%CI: 1.03, 2.70) relative to weeks 11-12 (mid-season). Improved pre-season conditioning may reduce the risk of early-season injury.
Learning Objectives: Participants will learn about Alternating logistic regression and its application to correlated and longitudinal injury data
Keywords: Statistics, Methodology
Presenting author's disclosure statement:
Organization/institution whose products or services will be discussed: None
Disclosure not received
Relationship: Not Received.