Vitamin D and the limits of randomized controlled trials

tion Examination Survey (NHANES III), levels below 50 nmol/l were associated with a substantially increased risk of total and cardiovascular mortality. Approximately one-third of participants fell into this deficiency category and this proportion is a large underestimate because in northern states samples were collected only in the summer months. Levels in the winter months in northern latitudes are of most concern for vitamin D deficiency. For example, in the winter months in northern states in the USA, almost all African Americans have levels below 50 nmol/l. In contrast to downplaying any potential beneficial non-skeletal role for vitamin D, any evidence of potential harm was magnified. The IOM report cited a few studies that demonstrated an apparent U-shape or inverse J-shape for some health outcomes, and the potential for harm was prominently stated, even reflected in its two sentence summary statement. Some of the selected examples require a stretch of the imagination to see the evidence for a U-shaped pattern. For example, the results of one study of 25(OH)D and CVD were described as: y‘there was no additional reduction in risk at levels greater than 75 nmol/L and that the dose–response relationship may be U-shaped above 75 nmol/L’. Upon inspection of the figure for this ‘U-shaped’ relationship, there was no credible statistical evidence for an increased risk – because only six cases had levels .75 nmol/l, the confidence intervals were extremely wide and essentially uninformative. A better description of the results from this study is that risk of CVD decreases with increasing 25(OH)D but the benefit levels off at about 50–60 nmol/l. Why the report placed large emphasis on the potential harm of high levels rather than the potential benefits may reflect largely the general approach for reviewing data, which, while appropriate for evaluating drug efficacy on specific diseases, may not be ideal to evaluate a ‘lifestyle’ biological factor, which I consider vitamin D to be. If we think of vitamin D as a pharmaceutical agent, we implicitly assume zero as the ‘default’ level and would look carefully for any evidence of risk, even from observational studies, and require evidence of benefit largely if not exclusively from randomized trials. However, ‘natural’ levels of 25(OH)D from healthy individuals with relatively high sun exposure (reflecting most of human history) are typically in the 125 to 175 nmol/l range. Because sun exposure tends to be low in current society, the highest levels are typically in the range of 75–100 nmol/l. Benefits and risks associated with vitamin D can be considered bidirectionally – for example, using 75 nmol/l as a starting point, the question of whether levels lower than this are deleterious is as relevant as the question of whether higher levels are harmful. The vast majority of studies for various endpoints including some cancers, total mortality, CVD, hypertension, skeletal health and some autoimmune and infectious diseases find higher risk at levels below 75 nmol/l; sometimes the trend appears inversely linear up to this point and sometimes the threshold for no further benefit may be lower (for example, at 50 nmol/l), but the group in the range of 75–100 nmol/l is typically the lowest-risk group. A much greater body of observational evidence supports that levels of 25(OH)D below 75 nmol/l are deleterious v. levels of 75 nmol/l or higher, than supports deleterious effects at higher levels. Excessive concern for potential adverse effects may also extend from the example of b-carotene, where randomized trials did not support hypothesized benefits and even indicated harm. However, b-carotene trials tested intakes about tenfold higher than would be consumed by those on a diet naturally high in b-carotene, so any supraphysiological effects could not be predicted by experience in human subjects. I would not consider a seemingly high vitamin D dose of 25 mg (1000 IU/d), for example, as supraphysiological, as this amount could be made though several minutes of sun exposure. Ongoing and future randomized trials will undoubtedly generate important information, but are unlikely to address all relevant issues. For example, for some diseases, the timing of the relevant exposure could be decades before the diagnosis of the disease. Given our current state of knowledge, it is a larger concern that many people are not getting enough vitamin D rather than many are getting too much.