REPEATED-MEASURES ANALYSIS OF BINARY OUTCOMES - APPLICATIONS TO INJURY RESEARCH

Repeated measures are reasonably common in injury research and thus to ols are required for appropriate analysis in order to account for the correlated nature of this type of data. Three methods for analyzing re peated measures binary outcome data are presented and contrasted: gene ralized estimating equations (GEE), a survey sample methodology, and l ogistic regression. These methods are applied to data collected from a cohort study of rugby players, designed to examine the risk and prote ctive factors for rugby injury. It is not, however, the purpose of thi s paper to present causal models of rugby injuries. The GEE approach i s attractive because it is able to account for the correlation among a subject's outcomes and several covariates can be included in a model. The survey sample method approach, which also accounts for the correl ation but is restrictive in terms of the number of covariates it can h andle, is another approach which is described. These two methods are c ontrasted to logistic regression, which assumes independence among a s ubject's outcomes. Under certain circumstances, the three methods do n ot differ substantially from one another. Under other circumstances, s ince logistic regression ignores the correlated nature of the data, st andard errors may be incorrectly estimated and thus certain covariates may be incorrectly identified as significant predictors in a model. C opyright (C) 1996 Elsevier Science Ltd