Genetic predisposition to serum 25 hydroxyvitamin D concentrations does not influence the risk of decreasing celiac disease in European ancestry: Evidence from meta-analysis and Mendelian randomization

Abstract

While prior research has indicated a possible association between serum 25 hydroxyvitamin D 25(OH)D and celiac disease (CD), key questions remain unresolved without clear consensus. To assess association, we perform Mendelian randomization and a systematic review and meta-analysis of existing studies in European ancestry. A systematic review and meta-analysis performed by using both standardized mean difference and a Bayesian framework. Moreover, Mendelian randomization conducted via Generalized Summary-data-based Mendelian randomization (GSMR) and two-sample Mendelian randomization. Data on serum 25(OH)D levels were obtained from the Medical Research Council's Integrative Epidemiology Unit (GWAS ID: ebi-a-GCST90000618; N = 4,96,946), while data on CD included 11,812 cases and 229 controls (GWAS ID: ebi-a-GCST005523). The main method used was the random inverse variance weighted (IVW) approach, supplemented by various sensitivity analyses to strengthen the findings. Both the meta-analysis and Mendelian randomization analysis indicate that vitamin D levels have no effect on the risk of developing celiac disease. Using a random-effects model, the standardized mean difference was calculated as -0.79 (95% CI: -2.02, 0.45), which was not statistically significant (T-Statistic=-1.77, P = .152). The Bayesian meta-analysis reinforced these results, showing a point estimate of -0.6 with a wide 95% credible interval of (-13.6, 13.2), highlighting substantial uncertainty regarding the true effect size. The GSMR also found no significant association between 25(OH)D levels and CD (OR = 0.988; 95% CI: 0.853-1.145; P = .878). This null finding was corroborated by two-sample Mendelian randomization (ORIVW = 0.72, 95% CI: 0.49-1.05, P = .091). The Steiger test confirmed the absence of a directional relationship, and cluster analysis revealed no heterogeneity patterns (ORIVW = 0.374, 95% CI: 0.238-0.588). These findings were consistent across multiple MR methods, including penalized MR, robust MR, penalized robust MR, MR-Egger (both t-distribution and radial approaches), SIMEX, MR-PRESSO, Maximum likelihood, MR-Mix, RAPS, MR-Least Absolute Shrinkage and Selection Operator, debiased IVW, and contamination mixture models. By questioning the presumed direct connection between 25(OH)D and CD, it appears that 25(OH)D may not be a primary contributor to CD, which could impact future prevention approaches and alter clinical guidelines.