A further plea for rigorous science and explicit disclosure of potential conflicts of interest

Abstract

Dr. Morfeld recently published in this journal an editoriallisting threats to validity in epidemiologic studies (Morfeld2009). The threats to validity enumerated included publica-tion bias, exposure misclassiWcation and biased analysesand distorted presentations of epidemiologic studies. Toillustrate this last type of bias, Dr. Morfeld took the exam-ple of our study of air pollutants and term birth weight in aMunich birth cohort published in 2007 in EnvironmentalHealth Perspectives (Slama et al. 2007). These criticismsare unfounded, as we brie Xy discuss below.The main criticisms of our study concerned the analysisof birth weight as a binary outcome rather than as a contin-uous outcome, the numerical stability of the log-binomialmodels and Wnally the interpretation of the model adjustedfor all atmospheric pollutants simultaneously.Dr. Morfeld’s assertion that, because birth weight isrecorded as a continuous outcome, it is most appropriatelyanalysed as a continuous outcome, ignores much of thepublished work on perinatal outcomes. Indeed, as indicatedby Wilcox (2001), “For most of the previous century, birth-weight has been treated as a dichotomy.” Consistent withthe clinical notions of low birth weight and small for gesta-tional age newborn (van Wassenaer 2005), epidemiologistsoften treat birth weight as a binary outcome, using an arbi-trary cutoV (low birth weight) or a cuto V corresponding to agiven percentile of the sex- and gestational age-speciWcbirth weight distribution in a reference population (Savitzet al. 2002). Treating birth weight as a continuous outcomeis also an option, allowing shifts in entire birth weight dis-tribution associated with exposure to be studied. However,such shifts in the entire birth weight distribution are mostlydriven by shifts within the dominant distribution. They maynot occur if an exposure only causes an increase in theproportion of births in the lower tail of the birth weightdistribution (Savitz et al. 2002), a situation that may beidentiWed by dichotomizing birth weight. For this reason,both approaches should be seen as complementary and areoften used in parallel.We analysed birth weight as a continuous outcome, anddid not Wnd an association with maternal exposure to atmo-spheric pollutants; however, we did observe an associationwith atmospheric pollutants when we treated birth weightas a binary outcome with a cutoV at 3,000 g (Slama et al.2007). Due to the limitations on word count and the numberof tables imposed by the journal, we could not describe indetail our analysis of birth weight as a continuous outcome,nor state these results in the abstract. However, we clearlystated that we had undertaken this analysis and indicated its