The Power of a Simple Multivitamin
August 13, 2015
I founded and run the Foundation for Integrative AIDS Research (FIAR). Its mission has been to advocate for and develop studies of so-called "complementary and alternative" medicine (CAM), or integrative medicine. A few years ago, in collaboration with colleagues at Mount Sinai School of Medicine, FIAR obtained a National Institutes of Health (NIH) grant to undertake meta-analyses of questions related to CAM. The first question that we assessed was on the role of micronutrient supplementation.
What we found was truly astounding: a very simple, inexpensive multivitamin can slow the rate at which a person progresses from HIV infection to AIDS by nearly 40 percent! There was also evidence that such a multi could reduce the risk of dying for the millions of people living with HIV who cannot yet access HIV medications.
Our findings were among adults with HIV not yet on antiretrovirals (ARV). These findings can be incorporated into public health measures and become part of the standard of care for people living with HIV as some 30 million are clinically eligible but not yet on ARV.
How did we come to this conclusion? This article gets a bit technical -- but it will also explain some of the language and methods used in research like ours, which should assist you in reading journal articles going forward, I hope!
We think of HIV infecting a T cell; then it makes more of itself inside the cell, busts out, and the cell dies. That's basically true -- but where IS that T cell??
Something like 70 percent of CD4+ T-lymphocytes reside in the gut; this is true among people living with HIV as well as those who are not, though unfortunately, in the gut of a person with HIV, the T cells may be having a party, which can cause a lot of inflammation. This in turn means that the gut's ability to do its main job -- absorbing nutrients from the foods we eat and the beverages we drink -- is compromised. This results in levels of some micronutrients (vitamins and minerals) becoming depleted. We have known about this since the 1980s.
We started with the question of "is there clinical evidence for the use of a multiple micronutrient in HIV disease?" Meta-analysis involves collecting the data from different qualitative and quantitative studies and analyzing it in a statistical model. This allows us to synthesize evidence from a range of studies, while also assessing their methodological quality.
This is where we venture into the thorny thickets of statistics and its attendant modeling and mathematics. The usual way that the analysis is undertaken is known as a "frequentist" approach. This yields a p value that provides evidence as to whether the "null hypothesis" can be rejected. The cutoff for this is greater than 95%, so you'll see anything with a p value of less than 0.05 as being considered statistically significant. (Whether it is actually clinically significant is another matter!)
What the frequentist approach does NOT do is offer evidence for anything but the rejection of the null hypothesis. What's that null hypothesis then? It's the idea or hypothesis being tested that there will be no difference between two arms in a study. This is basically embedded in the primary endpoint of a study.
So if a study is looking at, say, the effect of fluconazole in treating candidiasis, the null hypothesis would be that fluconazole has no effect on the disease. The study will then proceed to assess the evidence that accepts or rejects this null hypothesis. If the study has 500 people, randomized to take either the treatment or a placebo, and the majority of treated people get better while only a few of the placebo recipients improve, the null hypothesis can be rejected that fluconazole does not affect candidiasis.
Conversely, with a frequentist analysis, it is not actually scientific to claim that the reverse is true: that fluconazole is an effective treatment for the disease! You can only claim that the data show that the null hypothesis of no benefit can be rejected.
At the end of the study, then, what is produced is a point estimate of a likelihood of rejecting the null hypothesis followed by a confidence interval. The wider the interval, the weaker the point estimate's validity is. So a risk reduction of 0.5 (0.4, 0.9) would be a pretty solid finding. The numbers in the parentheses are the confidence interval or the range of values that the "point estimate" (the first number) is embedded in. A narrow range indicates more confidence; if the upper number in this case exceeded one, that would suggest a non-statistically significant result. The p-value is calculated from this and, with this point estimate and range would probably be quite low (statistically significant).
Practically speaking, if you have a very strong signal that such a study would be likely to demonstrate, the clinical efficacy would not be ignored simply due to statistical concerns. (By contrast, use of relative risk reduction outcomes are commonly used more as a marketing tool to sell more drugs than a medically valid analytical tool to assess evidence of efficacy ... but I digress ...)
This is where the approach we used comes in handy, particularly if the datasets are somewhat limited in scope. Bayesian statistical analysis is rooted in the probability that a treatment does work, based in part on prior information that can be incorporated into the making of the statistical model. When the data are analyzed with an appropriate model, a range of possibilities is generated that can incorporate 100,000 or more calculations of outcomes, which evolves into a posterior probability. Instead of a confidence interval, a Bayesian analysis then also determines a point estimate with what is known as a credible interval. This permits us to say whether not only the null hypothesis is rejected, but also the 95% likelihood that the finding indicates what the data indicate.
These data, in frequentist as well as Bayesian approaches, are then depicted as forest plots (see figures). The forest plot shows the relative contribution of each study, the diamonds indicating the relative position on the point estimate, and the width of the diamond its confidence or credible interval. Here's one from our study:
Bayesian density strips of progression findings:
This article was provided by TheBody.
Add Your Comment:
(Please note: Your name and comment will be public, and may even show up in
Internet search results. Be careful when providing personal information! Before
adding your comment, please read TheBody.com's Comment Policy.)