Birth Weight and Intima-Media Thickness

To the Editor: Recently, Tilling et al1 reported that birth weight was positively related to adult intima-media thickness by univariate analysis. However, adjustment for confounders reduced this association toward the null. Whenever adjustment eliminates an association, one has to reconsider whether the confounding hypothesis is correct. We suggest that a number of variables used as confounders do not fulfill the properties of a confounder according to any of the 3 major definitions of confounding. According to the classic definition, a confounder is a cause of disease and is associated with exposure.2,3 Table 3 of the Tilling paper includes as confounders variables that do not fulfill these conditions such as pack-years smoked and antihypertensive medication. The “collapsibility” definition declares a variable to be a confounder if the effect measure is homogenous across strata of this variable and if the crude and common stratum-specific values of the effect measure are unequal.3,4 Adjustment for sex by Tillich et al led to the most dramatic decline in effect measure. However, stratification by sex indicated that sex is rather an effect modifier than a confounder, because the effect measure was not homogenous across strata. The definition of confounding related to causal diagrams declares confounding to be present in a directed acyclic graph if there is an unblocked backdoor path leading from exposure to disease.5,6 An unblocked backdoor path has an arrowhead pointing to exposure and also has no collider.5Figure 1 shows a directed acyclic graph of the analysis of Tillich et al. Variables used as confounders belong to 1 of the 3 groups indicated by sex (S), intermediates (M), or later-life risk factors (R). Birth weight (B) is related to intima-media thickness in adulthood (I). At least in part, this effect is mediated by causal intermediates (M) such as body mass index, high-density lipoprotein and low-density lipoprotein cholesterol, and diabetes. The effect of sex (S) on (B) can be regarded to be direct, but its effect on (I) might be mediated by (M). Other risk factors acting in later life (R) such as smoking are causally linked to intermediates (M) and to disease (I), but not to birth weight (B). Note that in this causal diagram, (S) is not a confounder without considering (M), whereas neither (M) nor (R) alone are proven confounders in this study according to the confounding criterion of directed acyclic graphs.5,6 Therefore, adjustment for these variables will lead to another confounded estimate.5,6FIGURE 1.: Simplified causal diagram (directed acyclic graph) of the analysis by Tilling et al.1Adjusted estimates are superior to unadjusted estimates only if the underlying confounding hypothesis is correct,7 which might not be the case here. Drawing wrong conclusions about the absence of a role of high birth weight as a risk factor for later cardiovascular morbidity might have wide-ranging consequences given the increasing incidence of high birth weight.8 Thomas Harder Andreas Plagemann Clinic of Obstetrics, Division of Experimental Obstetrics, Campus Virchow-Klinikum, Charite-University Medicine Berlin, Berlin, Germany, [email protected]

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.