Design and analysis of clinical trials of nutrients: Author Reply

I thank Dr. Martineau for his comments and for the opportunity to clarify some of the points he addresses.1 We both recognize that systematic reviews (SR) and meta-analyses (MA) have two major purposes: 1) to increase statistical power and get a better fix on whether an effect occurs and, if so, on its size; and 2) to use observed heterogeneity of response to clarify why some studies find an effect and others do not. The guidelines for SR and MA, to which Dr. Martineau responds,2 were pointed solely at the first of these purposes, and for two reasons. First, for most nutrients, the controverted question is whether a nutrient produces a given effect; e.g., does vitamin D alter immune response to a clinically meaningful degree; or does calcium intake reduce menopausal weight gain? Second, with nutrients (unlike drugs), the principal reasons for heterogeneity are already known. I list four of them in the suggested guidelines for individual studies,2 i.e., basal status, dose, duration of treatment, and conutrient status. To answer the question of whether an effect really occurs, pooled studies must be homogeneous across these factors, particularly when it is known that the response is nonlinear. To pool studies with different doses, basal status values, treatment durations, and conutrient status values would certainly not clarify the underlying uncertainty that led to a MA in the first place. Moreover, the four listed factors, alone or in combination, present 15 separate ways to render a set of studies heterogeneous, which is far more than could feasibly be tested in a plausible meta-analysis. Dr. Martineau also insightfully notes the ethical dilemma presented by those randomized controlled trials (RCTs) in which the entrance criterion into a study must be low nutrient status. This problem is often ignored, but presents an unavoidable ethical challenge. Unlike with drugs, the underlying hypothesis with nutrients is not that a particular nutrient prevents or ameliorates a problem. Rather, it is that deficiency of the nutrient causes disease or dysfunction. Such an hypothesis can only be tested in a context in which specific harm is known to occur (i.e., deficiency); and for the cognate hypothesis to be tested, the control subjects must be placed at risk, if not from the effect being tested, then from some other effect, as all nutrients are essential. The relevant questions with regard to nutrients are not whether they are efficacious, but “for what are they efficacious?” and “how much is needed?” Ethics aside, RCTs are poorly suited to answer such questions. It was precisely to suggest alternative approaches that I presented in this journal two years ago a set of physiological criteria that could arguably be used to define nutritional normal for select, and perhaps most, nutrients.3