For the most part, the disorders that have shaped nutrition as a science have had 2 characteristics: each nutrient was the major factor in its cognate disease, and disease expression had short latency, which facilitated recognition of the connection between cause and effect. More recently, an extensive body of epidemiologic associations between nutrient intake and various outcomes has raised the possibility that insufficient nutrition may contribute to the burden of chronic disease. But true causal connections between specific nutrients and these other disorders have been hard to establish, partly because most chronic diseases are highly multifactorial and partly because they have long latency periods. Both factors make the evaluation of causality inherently difficult.
Vitamin D presents a good case in point. There has been an explosion of reports relating vitamin status to various disorders in the past 10 y (1). This issue of the Journal contains another of these reports. Visser et al (2), from the Amsterdam Longitudinal Aging Study, report that the likelihood of nursing home admission was inversely related to baseline vitamin status.
This finding has biological plausibility, inasmuch as low vitamin D status impairs lower-extremity function and contributes significantly to the risk of falling (3, 4).
The high prevalence of vitamin D deficiency mandates that physicians assess and correct vitamin D status in their elderly patients (5).
However, the same high prevalence must cause us to ask why a global intervention should not be given serious consideration. Most of the studies concerned (including the study by Visser et al) are observational in character. Do such studies constitute sufficient evidence for a population-level intervention?
The randomized controlled trial (RCT), which has become the gold standard for establishing the efficacy of pharmacologic agents, is poorly suited to the evaluation of nutritional effects, a fact that I believe many have been reluctant to acknowledge. Several important differences between nutrients and drugs lead to this conclusion. In addition to long latency and multifactorial causation for the diseases concerned, nutrients and drugs differ in 3 crucial respects. First, whereas a drug-free state exists that can be contrasted with a drug-added state, with respect to nutrients, the only contrast can be between different intakes, both usually well above zero. Second, most nutrients have what is known as threshold behavior, ie, some physiologic measure improves as intake rises up to a level of sufficiency, above which higher intakes produce no additional benefit. Third, most nutrients have beneficial effects on multiple tissues and organ systems, and thus a focus on a single or “primary” outcome measure, which is favored by RCTs, is often procrustean.
As a consequence of the second point, investigators using the RCT design must contrast 2 groups of subjects, at least one of which has a distinctly inadequate intake of the nutrient concerned. Failure to do that, as occurred in the calcium arm of the Women’s Health Initiative (WHI) (7, 8), constitutes an invalid test of the corresponding hypothesis. However, the assignment of subjects to an intake that is inadequate by current standards, for the span of time required to produce the necessary difference in serious outcomes, raises significant and probably insurmountable ethical problems (9). Whatever the explanation, there was no low-calcium contrast group in WHI.
Unfortunately, the reports from WHI primarily stressed the outcome of the controlled intervention, as if calcium and vitamin D had been drugs, and tended to deemphasize the observational data that the study had also generated. Examples are the fact that, in addition to a baseline calcium intake in the participants nearly twice that predicted from the third National Health and Nutrition Examination Survey, the hip fracture rate was approximately half of that predicted from Medicare. No connection seems to have been made between these 2 departures from expectation. There was a similar lapse with respect to the connection between vitamin D intake and colon cancer. Whereas the designed low dose of vitamin D in WHI did not significantly alter colon cancer incidence, baseline vitamin D status was, in fact, significantly inversely related to cancer risk. The lowest 25(OH)D quartile had a risk 2.5 times that of the highest quartile.
Such circumstances raise 2 related questions: 1) to what extent should current standards of proof be relaxed for nutrient benefits? and 2) what alternative investigational design might be used to produce results that could be used as a basis for nutritional policy?
Public policy has long been comfortable in using a more relaxed standard of proof for potentially harmful effects. The studies leading to a food label for trans fats would not likely have been considered adequate to support the promotion of the use of trans fats (had the relation been the other way around). Nor do we require RCTs to set the cutoffs for various environmental toxins, from lead to polychlorinated biphenyl. Should the standard of proof be relaxed for benefit, as in the instance of a fortification program that elevates nutrient status (in this case vitamin D) to demonstrably safe intakes? If the balance of the potential for harm tilts in favor of fortification, then the answer should be “yes,” irrespective of the level of evidence.
With respect to study design, there is one approach that offers a reasonably practicable alternative to the RCT, ie, the nonconcurrent cohort study. As are all cohort studies, it is prospective. It is “nonconcurrent” in the sense that the exposure to the various amounts of nutrient precedes the investigation; ie, the entering of subjects into study is done after the exposure but before any analysis.
The foregoing approach could be readily evaluated in existing databases, but, whatever the study design, the problem of proof remains, and, as a consequence, a substantial potential for reduction in the burden of chronic disease hangs in the balance.
Referência :
(1) Am J Clin Nutr. 2006 Sep;84(3):471-2.