Statistical methods in the study of toxic shock syndrome.

Analysis of statistical methods used in matched case-control studies of toxic shock syndrome shows that matching has implications for validity and precision of the studies and for the choice of analysis techniques. The studies considered accounted for the matching in the analysis, either by the Mantel-Haenszel or Miettinen-Pike-Morrow approach to 1-to-M matched designs, or by the use of conditional maximum likelihood fitting of logistic regression models. Methods of dealing with confounding and effect modification in the (matched) logistic regression model are presented in the context of studies of toxic shock syndrome. The varied statistical techniques used in these studies were generally appropriate to the matched design except that nonmatching variables were not thoroughly considered as effect modifiers or confounders.