In the comment section of a recent Vitamin D Council review of a study that showed no link between vitamin D and pancreatic cancer, Dr. William Grant pointed out that a major limitation of prospective vitamin D studies is the lag time between the vitamin D measurement and the event of interest. Dr. Grant recommends that such studies control for follow-up time and, if possible, measure vitamin D levels every two-to-four years.
Grant WB. Effect of interval between serum draw and follow-up period on relative risk of cancer incidence with respect to 25-hydroxyvitamin D level: Implications for meta-analyses and setting vitamin D guidelines. Dermato-endocrinology. Jul 2011;3(3):199-204.
The basic problem is that each individual’s vitamin D level changes over time. In his articles, Dr. Grant demonstrates the problem using multiple statistical techniques.
There’s yet another way to talk about this limitation of prospective vitamin D studies using the concept of regression toward the mean. In this concept, the word regression just means movement – it has nothing to do with statistical regression. The idea is that if you repeatedly measure a group of individuals, the individuals who score in the tails of the distribution in one measurement (i.e., those whose scores are unusually high or unusually low), are highly likely to have a score that is closer to the mean in other measurements.
Daniel Kahneman has a particularly good discussion of regression toward the mean in his book Thinking, Fast and Slow (see chapter 17). Kahneman says,
“Whether undetected or wrongly explained, the phenomenon of regression is strange to the human mind. So strange, indeed, that it was first identified and understood two hundred years after the theory of gravitation and differential calculus.”
Let’s look at how the problem of regression toward the mean impacts the results of another recently reported prospective study of vitamin D.
Robinson-Cohen C, Hoofnagle AN, Ix JH, et al. Racial differences in the association of serum 25-hydroxyvitamin D concentration with coronary heart disease events. JAMA : the Journal of the American Medical Association. Jul 10 2013;310(2):179-188.
The study is based on data collected in the Multi-Ethnic Study of Atherosclerosis (MESA), which collected baseline data on individuals in the U.S. who were free of heart disease from July 2000 through September 2002. The baseline data included blood samples, which were frozen. Follow-up calls were made to individuals in the study every 9 to 12 months to determine if they had had any cardiovascular events – if so, these events were verified by medical records or death certificates.
In 2012 the researchers obtained a stored blood sample for each individual in the study and tested the samples for 25(OH)D, the marker of vitamin D status. The researchers then used the statistical technique of Cox proportional hazards regression to develop their findings. They found that low vitamin D status was associated with more heart disease in whites (and in a very small group of ethnically Chinese subjects), but not in Hispanics or blacks.
Besides all the problems Dr. Grant has pointed out, this study has the additional problem that hardly any blacks had what the study considered high levels of vitamin D. Those who did were well within the upper tail of the vitamin D distribution for blacks. Because of regression toward the mean, we’d expect that many of these individuals — had their vitamin D been measured a second time — would have had levels that wouldn’t be considered high in the context of this study. This problem didn’t have as big an impact on the white subjects because their overall vitamin D levels where higher. Those in the upper tail of the distribution could have still had what the study considered high levels after regressing toward the mean.
There is one additional problem with this study. The researchers suggest that the mechanism for the effects they found is genetic differences between blacks and whites. But race is a social construct – there is no biological test for race.
For example, it’s well known that President Obama had a white mother from Kansas and a black father from Africa. Using the social construct of race, President Obama is black, but genetically, he is as white as he is black. If the researchers had used actual genetic differences as their grouping variable, that would carry some weight. But to suggest that genetic differences can be reliably identified based on the social construct of race is sloppy.
The take-away: Prospective observational studies examining the associated effects of vitamin D status have to be carefully watched. Ideally, prospective observational studies will have multiple vitamin D measurements during the course of the study. If not, it’s important to pay heed when the vitamin D level was drawn and how long the follow-up period was.