On the trail of necrotizing fasciitis in children

I read with interest the article by Tauyee Hsieh and colleagues describing a case–control study of necrotizing fasciitis in children.¹ The authors indicated that they enrolled control subjects who were “matched to the case subjects” by date of admission and by date of birth. Although it was not specifically stated, one might assume by the wording that the control subjects were individually matched to the case subjects. The authors also noted that, for the multivariate analysis, they were unable to obtain odds ratios from conditional logistic regression (which would take the matching into account) but verified their estimates by an alternative approach that did adjust for matching.

However, it is not clear whether appropriate analyses that take matching into account were utilized in their univariate comparisons, and whether the comparisons displayed in their Table 1 represent univariate or multivariate comparisons. They indicated that they analyzed their data using Fisher's exact test for categorical variables and the Wilcoxon rank-sum test for continuous variables. If the authors did not use matching, would the comparisons in Table 1 have been different if they had? Would this have altered the “significant” risk factors included in the multivariate analysis?

Signature

Gary M. Liss
Department of Public Health Sciences University of Toronto Toronto, Ont.