Practice Point: Vitamin D may be getting another fifteen minutes of fame for its potential role in preventing diabetes, but please don’t start “treating to target” yet.

EBM Pearl: When trials have similar enough protocols, the data points can be combined in an individual patient data meta-analysis to increase the power of a study.


Any time a single treatment has been found ineffective for multiple conditions and then you see it being studied for yet another purpose, it’s hard not to think there is some kind of fishing expedition going on. That was our first instinct when we saw this systematic review and individual patient data (IPD) meta-analysis on vitamin D for primary prevention of diabetes in patients with prediabetes.

This meta-analysis took individual data points from three randomized trials studying different formulations of vitamin D versus placebo in 4,190 adults with prediabetes (and without vitamin D deficiency) and analyzed them together in an effort to increase the statistical power of the analysis. They were able to do that because the inclusion criteria for these trials were nearly identical, and all three of them had the same primary outcome: progression to new-onset diabetes.

None of the individual trials found a significant benefit for vitamin D. The authors of the meta-analysis that combined the data from these individual trials reasoned that the lack of significant findings was because they were underpowered. The individual studies were designed to detect a difference of at least 25 percent (most likely based on a pilot study that overestimated the magnitude of effect) and what they each found was a nonsignificant risk reduction of around 11 percent. When the investigators put all of the data together for a meta-analysis and combined intention-to-treat analysis, they found vitamin D to be associated with a 15 percent reduction in risk of progression to diabetes (hazard ratio 0.85, 95% CI 0.75-0.96), which translates to an NNT of 30.

So, if this is a fishing expedition, what’s the catch?

Although it is unusual that a higher quality trial would find significant benefit from an intervention when prior studies did not, we actually do buy the legitimacy of combining individual patient data from randomized trials in this case, given the similarity of study protocols and that the primary outcome studied was, in fact, progression to diabetes rather than some post-hoc secondary outcome.

The generalizability of the results from this meta-analysis gives us some pause, as the data only came from participants in Norway, the United States, and Japan, and we don’t know if a vitamin D-based intervention would work in places where people get a lot more sun exposure. What really concerns us, however, is the authors’ recommendation for a “treat to target” approach based on their finding that higher vitamin D levels were associated with a lower risk of diabetes and a higher likelihood of returning to normal glucose levels. This approach would require testing for vitamin D levels, which is a hard “no” for us based on the available evidence.

Giving patients with prediabetes 4,000 IU of vitamin D a day to prevent a future A1c ≥ 6.5 is a disease-oriented approach that is still of unproven clinical benefit*, and may not be without harm. More than that, these authors want us to treat patients with something that has an NNT of 30, when metformin, which has an NNT of 14 according to the Diabetes Prevention Program study, is not even recommended for patients with prediabetes. Furthermore, lifestyle intervention alone has an NNT of seven.

When the data show us that vitamin D helps patients with prediabetes live better or longer, we’ll be the first ones casting a line. Until then, we think it’s fair to remain skeptical and focus our energy on convincing our patients to make some lifestyle changes which we know will do more than improve their A1c.

*It has not been established that prevention (or prethreshold treatment) of diabetes delays or prevents complications compared with treatment after diabetes is diagnosed. DynaMed considers prevention of diabetes to be a surrogate outcome unless the studies also report clinical outcomes such as cardiovascular events. Specific glycemic thresholds can be an arbitrary outcome as the impact of having a fasting blood sugar of 124 mg/dL is unlikely to be significantly different than a fasting blood sugar of 126 mg/dL.

For more information, see the topic Prediabetes in DynaMed.

Reference: Ann Intern Med. 2023 Mar;176(3):355-363