Heterogeneity of the Effect of Vitamin D Supplementation in Randomized Controlled Trials on Cancer Prevention

Vitamin D has been suspected to play a role in cancer prevention for the past 4 decades, but definitive proof of benefit has been difficult to determine. In a randomized, double-blind, placebo-controlled trial, Chandler et al¹ found that vitamin D supplementation reduced the incidence of advanced cancer in the Vitamin D and Omega-3 Trial study cohort. The authors found a difference in treatment effects: protective effect was observed among individuals who had normal body mass index (BMI) but not among overweight individuals, most likely because of the simple volumetric dilution of this fat-soluble vitamin in fat tissue. The notion that treatment effect may vary according to BMI indicates the Achilles heel of randomized controlled trials (RCTs).

Randomized controlled trials are the criterion standard in evaluating the effects of treatments and are the cornerstone of evidence-based medicine. However, they cannot deal with the heterogeneity of treatment effects.² Properly conducted randomization ensures that participants’ characteristics are equally distributed between trial arms and protects against confounding. However, it does not protect against the issue of heterogeneity of treatment effects because heterogeneity within the treated arm drives a null effect. In an RCT, only group effects are observed, and then compared: if half of the treated group improves but the other half deteriorates, the treatment will be assumed to have no effect. Similarly, if a fraction of the participants who received supplements benefited but there was no effect in others, the effect size will be diluted (and likelihood of reaching a statistically significant finding will be reduced).

The findings by Chandler et al¹ align closely with this: a 17% reduction in advanced cancer risk was reported overall (hazard ratio, 0.83; 95% CI, 0.69-0.99) but clear differences in treatment effects emerged after stratification by BMI. A significant 38% risk reduction was observed in normal-weight individuals (hazard ratio, 0.62; 95% CI, 0.45-0.86; BMI, < 25), but there was no effect in overweight (BMI 25-30: hazard ratio, 0.89; 95% CI, 0.68-1.17) and obese (BMI > 30: hazard ratio, 1.05; 95% CI, 0.74-1.49). It seems sensible to wonder whether overweight individuals would benefit from a higher dose of vitamin D.

Vitamin D research sometimes bears a resemblance to the Chinese whispers game. Although we set out hypothesizing that vitamin D sufficiency decreases cancer risk, our finest tools—RCTs—allow us to examine only whether vitamin D supplementation reduces the disease risk. The problem is that the relationship between vitamin D supplementation and vitamin D status is less closely correlated than is comfortable to acknowledge. Substantial variation in the increase of 25-hydroxyvitamin D (the best marker of vitamin D status) after administration of the same vitamin D dose has been well documented.³ It is hard to predict how much 25-hydroxyvitamin D concentration will change in a given individual after the supplementation; however, the increase is typically less in overweight individuals. In addition to this ambiguity in the vitamin D dose–25-hydroxyvitamin D response, another important issue exists and concerns the underlying relationship between vitamin D status and health.

For many nutrients, the association between the status and health is nonlinear; the sigmoid response curve tends to reflect the relationship with health outcomes more accurately⁴ and evidence indicates this is also true for vitamin D.⁵ The relationship is characterized by a steep improvement in health with increasing nutrient status between 2 inflection points; but there is little difference in health outcomes at low levels. Similarly, little extra benefit will be gained once a sufficient level has been achieved (and eventually toxic levels may be reached). The ascending part of the curve is the “response region,” in which improvement in status is most closely linked with improving health.

Two key factors disrupt the link between vitamin D supplementation and health outcomes in RCTs: baseline level and the increase in vitamin D status that is attributable to supplementation. The key is that the 2 are tightly linked: even the same absolute increase in 25-hydroxyvitamin D concentration will be associated with varying degree of improvement in health, depending on which part of the “response region” it is spanning; this might explain why baseline levels were not found to modify the effects in the study by Chandler et al.¹ Even a small improvement in status might have a large influence at the steep part of the curve. At low baseline levels, supplementation may not push vitamin D status to the ascending part of the curve, where benefit starts to manifest; at high baseline levels, there is nothing left to be gained with further increase in 25-hydroxyvitamin D levels. In other words, some patients benefit more and some less from the supplementation they were randomized to receive.

In public health terms, in a sizeable portion of the study population (ie, normal-weight individuals), risk reduction because of supplementation was particularly strong, at 38% (compared with 17% overall). In other words, among nonoverweight individuals, the risk was 1.6 times higher in the placebo group. This suggests that some population subgroups are at a considerable risk of advanced cancer. The identification of these individuals who would benefit most from supplementation would enable targeted prevention and could have major role in cancer prevention. That preventive effect was observed despite the high baseline 25-hydroxyvitamin D levels in the study population (30 ng/mL) suggests that anticancer effects of vitamin D might still persist beyond the concentration of 20 ng/mL that is currently considered to mark the sufficient level.

Overall, the findings from Chandler et al¹ support the role of vitamin D in cancer prevention and add substantially to the weight of the evidence in this field. However, advanced cancer was not prespecified as an outcome. Moreover, it is difficult to eliminate the possibility that in some cases carcinogenesis began before the trial commenced; it might be argued that vitamin D affected cancer progression and not necessarily occurrence. Therefore, further investigation is warranted.

But how can we move forward? Very large differences in vitamin D status can be found in any population. Body mass index, in addition to a range of personal, lifestyle, behavioral, environmental, dietary, and genetic factors, affects vitamin D status, and these together drive the heterogeneity of treatment effects primarily by determining baseline levels and the 25-hydroxyvitamin D response to supplementation.

The purpose of an RCT is to provide conclusive evidence of the effect of a treatment. Given the expected variability of treatment effects, how can an RCT using fixed-dose vitamin D supplementation provide definitive evidence of the benefit? Maybe it is time to consider a study design that can accommodate heterogeneity of treatment effects without compromising the strength of evidence.

Published: November 18, 2020. doi:10.1001/jamanetworkopen.2020.27176

Open Access: This is an open access article distributed under the terms of the CC-BY License. © 2020 Zgaga L. JAMA Network Open.

Corresponding Author: Lina Zgaga, PhD, Department of Public Health and Primary Care, Institute of Population Health, School of Medicine, Trinity College Dublin, the University of Dublin, Dublin, Ireland (zgagal@tcd.ie).

Conflict of Interest Disclosures: None reported.