To the Editor.
—Dr Fagon and colleagues1 have convincingly demonstrated that nosocomial pneumonia contributes to the death of patients in intensive care units (ICUs) independent of the confounding effect of other predictors. However, in their article they did not address whether there was any interaction between nosocomial pneumonia and the other variables that were kept in their multiple logistic regression model.It is possible that nosocomial pneumonia is completely independent of the other variables as a predictor of death in ICUs, but it is equally (or more) likely that the risk of dying from nosocomial pneumonia varies in magnitude according to the severity of the patients' baseline condition or the presence of other predictors of death. In epidemiology this phenomenon is known as "effect modification."2 Taking, for example, nosocomial bacteremia, if this variable were shown to be an effect modifier, it would imply that there was a synergism
Nishioka SDA. Nosocomial Pneumonia and Mortality. JAMA. 1996;276(4):284. doi:10.1001/jama.1996.03540040028027