A Protective Supplement with a Lopsided Funnel Plot

An I² of 79% is substantial: most of the variation across studies is not explained by chance. The panel should ask whether the studies differ by dose, baseline 25-hydroxyvitamin D level, residential status, calcium co-supplementation, or fracture ascertainment. The funnel plot asymmetry is also important. Smaller studies cluster around relative risks of 0.55-0.75, while the largest cohort is null. This pattern is compatible with publication bias, small-study effects, or differences in study quality. Supplement users are often healthier, more physically active, and more likely to have adequate calcium intake than non-users, so confounding by healthy adherer status could create a protective observational association. Reverse causation is possible if frailer people are less likely to take supplements and more likely to fracture. Inconsistent adjustment for frailty, sun exposure, and prior fracture means residual confounding is likely. The panel should not treat the pooled 0.82 as a causal estimate. It should downgrade confidence for inconsistency and suspected publication bias, and it should explore whether any subgroup, such as institutionalized adults with documented deficiency, shows a more consistent effect.

Observational evidence starts as low certainty. The panel should then consider downgrading further. Inconsistency is present because I² is 79% and the study effects range from strongly protective to null. Publication bias or small-study effects are strongly suspected from the asymmetric funnel plot and the contrast between small studies and the largest cohort. Indirectness may also apply if the studies mix community-dwelling and institutionalized populations, different doses, and different baseline vitamin D status. Imprecision is borderline: the 95% CI excludes 1 but the upper bound of 0.99 is close to no effect. The panel should not upgrade for large effect because the effect is modest and inconsistent, and there is no dose-response demonstrated. The overall certainty is therefore very low. By contrast, the randomized trial is more directly designed to answer the causal question, although its confidence interval includes both modest benefit and harm. The observational pooled estimate should be presented as hypothesis-generating, not as a basis for a strong population recommendation.

The randomized trial substantially changes the interpretation. A null trial with 5,000 participants is less vulnerable to healthy-adherer confounding than the observational studies, and its confidence interval is compatible with no meaningful fracture prevention. The observational pooled estimate is likely inflated by confounding and small-study effects. For community-dwelling adults without documented deficiency, the panel should recommend against routine vitamin D supplementation solely to prevent hip fracture, while encouraging proven measures such as fall prevention, strength and balance exercise, and osteoporosis treatment when indicated. For adults with documented deficiency, the recommendation should be to treat the deficiency according to endocrine or nutrition guidance, not to promise fracture prevention. The panel should state that the absolute benefit, if any, is small: even if the observational relative risk of 0.82 were causal, a baseline five-year hip-fracture risk of 2-3% would translate into only a small absolute reduction. The recommendation should be conditional and should explicitly note the very low certainty of the observational evidence and the null randomized trial.