Got milk?

The article by Mirzaei and colleagues 1 is another thoughtful and clever use of the Nurse’s Health Study. The evaluation of the mother’s data from this study is carefully conducted with an innovative approach to taking self-reported outcomes and validating their approach to estimating actual serum 25(OH)D levels. Further, by focusing on the gestational period, the problems of telescoping and sequencing effects in retrospective studies are mitigated to a degree. There are careful considerations in the handling of design issues and the authors acknowledge the obvious limitations of such studies, such as differential recall bias by cases vs controls. Acknowledging it, of course, does not eliminate it, but there is equally no evidence to demonstrate the findings result from recall bias. Thus, the authors should be congratulated on a very thoughtful and well done work. Nevertheless, we want to express some cautions, limitations, and overinterpretations of the data. It is easy to see the potential headlines that could result from this work or the public health perceptions that all pregnant mothers should be consuming greater amounts of milk to prevent multiple sclerosis (MS). These concerns can be directly traced to language in the paper such as ‘‘33% of the MS cases among mothers who drank less than 2 glasses of milk per day could have been hypothetically prevented.’’ While this is carefully worded and analytically correct in its statement, it is too easy to take the current fervor for vitamin D and a cross-sectional retrospective analysis and demand pregnant women not drink or smoke, and now drink more milk. The concern over an overly simplistic message is opposite the article’s extremely complex nature and analyses of the evidence. It incorporates not only direct variables, but also covariates that may be highly correlated. It is difficult to assess the interrelated variables and isolate their impact on the results. For example, milk consumption has north to south gradients. The Northeast on average consumes 335ml per day; North Central 291ml/ day; West 276ml/day, and the South 214ml/day. While the authors adjusted for location at birth, they did so as North, Middle, South, which may have left unadjusted some of the milk consumption and latitudinal effects in a combination that with highly correlated environmental, genetic, and vitamin D effects give rise to an association with the milk consumption data. The evidence for the vitamin D effect is mostly through a gradient as individual levels of milk consumption are not significantly different except at the highest levels of consumption. Furthermore, the gradient effect may not be the appropriate paradigm for the impact on disease. With rickets, as with many other nutritional diseases, it is a threshold that is key and while there is some evidence for this, the basic association rests with the gradient. In addition, what is the biological basis by which the estimated level of vitamin D associated with the selfreported intake could work to prevent, in utero, subsequent MS? It is not totally clear what level of vitamin D intake is required to sustain an adequate serum level.