“Hundreds of studies have investigated the association between homocysteine and cardiovascular disease over the past several decades; results from randomized controlled trials and observational studies have been discrepant, and the issue is not fully resolved. People with a common genetic variation of the methylenetetrahydrofolate reductase (MTHFR 677C→T), responsible for converting homocysteine, can have a decreased ability to metabolize folic acid and other B vitamins, showing an increased homocysteine concentration. Studies of this polymorphism show a link with an elevated risk of stroke, supporting a causal relation. However, these studies have also been inconsistent.
A recent meta-analysis summarizes results from randomized trials of homocysteine reduction with folic acid supplementation (with or without additional B vitamins) to reduce stroke incidence (2): Adequate folate abrogates the effect of the polymorphism and, as hypothesized, the investigators found that trials tended to be positive in regions with low folate and null in high folate regions (either naturally in the diet or because of folate fortification). The researchers conclude that if a randomized trial of folate supplementation for prevention of stroke were to be done in a low folate region, the predicted stroke risk reduction would be 22% based on existing genetic studies and randomized trials. Thus, many of the folate supplementation trials for stroke prevention might have been destined to fail because they were not undertaken in a low folate setting.
This result, perhaps obvious in retrospect, illustrates an important difference between drug trials, in which the placebo group has no intervention, and trials of nutritional supplements, in which everyone has variable intake that can be substantially greater than the amount of the intervention and can vary widely during the course of the trial. Some researchers (including ourselves) might have underestimated the baseline level of folate beyond which supplementation has no discernible effect.
A secondary conclusion is that meta-analyses are fraught with the potential for oversimplification. Summary estimates can mask true causal effects in susceptible subgroups. In their summary, Holmes and colleagues conclude that the benefit of high folate intake and low homocysteine concentration is probably limited to countries with low folate status. Even this conclusion understates the true relation, because the authors of the meta-analysis could not consider participant-level dietary intake of folate or alcohol, two factors known most strongly to affect the complex association between folate intake, MTHFR, homocysteine, and folate status (3).
Meta-analyses provide the perception of more accurate conclusions, especially when they include randomized trials. However, trials also have limitations. Randomized trials can produce null results if they are too small, have too short follow-up, or include individuals who are already above the sensitive range for interventional benefit. And, meta-analyses on seemingly parallel topics do not necessarily produce similar results. This discrepancy can be accounted for by differences in data collection methods or inclusion and exclusion criteria but is more likely due to the study-specific availability of risk estimates across effect modifiers.
Some researchers have dismissed observational studies of micronutrients and cardiovascular disease when the results appear to conflict with those of randomized trials, and characterize subgroup analyses in observational studies as data-dredging. Rather than automatic rejection of observational data when they appear to be discrepant from trials, as a kind of spinal reflex, the finding that trials, randomized controlled studies, and even meta-analyses can come to different conclusions should spur serious thought. Replication of findings from observational studies with wide distributions of exposures and effect modifiers can add value to findings from clinical trials, which have only a single fixed exposure and are often underpowered to examine associations across important biological modifiers of effect. Ultimately, evidence from both observational and experimental studies is necessary and the notion that there is rigid hierarchy in study design should be dispelled.”
Boston, August 2011