Dear Editor,

I read with interest the paper by Bizzarro et al. In their case- control study, for each case with Serratia bacteraemia they have selected two uninfected controls matched for birth weight and date of birth. The advantages of individually matching cases and controls on potential confounding factors is that the sample size required for the study to have adequate power is much smaller than if there was no matching. Unfortunately, in their multivariate analysis, they used a logistic regression model as if it was a classical unmatched case-control study. In such a study where cases are individually matched with controls, a conditional logistic regression model should have been used to preserve the matching, as the cases and controls are not truly independent but purposefully individually matched.[1-3] Their meticulous matching efforts in this important study deserve no less.

References:

1- Breslow NE, Day NE. Statistical Methods in Cancer Research. Volume I— The design and analysis of case-control studies with complex sampling. Vol 32 of IARC Scientific Publications, 1980. Lyon: International Agency for Research on Cancer.

2- Breslow NE. Statistics in epidemiology: the case-control study. Journal of the American Statistical Society 1996;91: 14–28.

3- Katz MH. Multivariable analysis. A practical guide for cinicians. 1999. Cambridge University Press.