Analyses of injury count data: some do's and don'ts

The analysis of injury data requires different considerations from the analysis of other types of outcomes because an individual can experience the outcome many times. When describing injury patterns using numerator-only data (e.g., proportion of upper-extremity injuries vs. lower-extremity injuries), simple comparisons of proportions are inappropriate because 1) individuals are compared with themselves and 2) multiple testing increases the potential for incorrect inference. Bootstrapping (resampling) techniques can be used to determine confidence intervals and whether the frequencies significantly differ across categories. When describing injury rates, the authors suggest plotting the observed injury rate against the number of exposures to obtain a visual representation of the heterogeneity of risk across individuals. Because the distribution of injury rates is often skewed, some research questions may be best addressed by comparing the weighted median injury rates instead of the weighted mean injury rates (which are given by standard formulae). Again, resampling techniques can be used to obtain a null distribution for injury rates in order to determine whether there are subjects who have unexpectedly high injury rates. More advanced analyses are required to account for multiplicity.