Meta-analysis: Vitamin D and Non-alcoholic Fatty Liver Disease

M. Eliades; E. Spyrou; N. Agrawal; M. Lazo; F. L. Brancati; J. J. Potter; A. A. Koteish; J. M. Clark; E. Guallar; R. Hernaez


Aliment Pharmacol Ther. 2013;38(3):246-254. 

In This Article


Search Strategy and Study Selection

We identified all studies that assessed the association between vitamin D and NAFLD in humans. We developed a search engine adapted for PUBMED and EMBASE up until 22 April 2013 (Table S1), with no language restrictions. In addition, we examined the reference lists of relevant original papers and review articles.

We included all studies that reported data on vitamin D levels and NAFLD or examined the association between vitamin D levels and NAFLD in multivariate models. We excluded papers without original data; animal or in vitro studies; studies that examined other types of liver disease (e.g. alcoholic liver disease, viral hepatitis, hepatocellular carcinoma, toxin-induced liver injury); and studies whose main focus were conditions primarily affected by vitamin D metabolism (end-stage renal disease, primary hyperparathyroidism) or medications that affect vitamin D metabolism like antiepileptic medications.

Data Abstraction

Three investigators (ME, ES, NA) independently reviewed identified abstracts and selected papers for full review. Discrepancies were resolved by a fourth reviewer (RH). For each selected publication, we abstracted key study characteristics, including publication year, country, study design, participant characteristics (age, gender, ethnicity and body mass index), season, method of diagnosis of NAFLD and controls, serum vitamin D assay and the cut-off level for defining vitamin D deficiency. We assessed the quality of the reporting by adapting the Strengthening the Reporting of Observational studies in Epidemiology (STROBE) checklist[15] (Table S2), and followed the Meta-analysis of Observational Studies in Epidemiology (MOOSE) guidelines for the current meta-analysis.[16] All data were double checked by one investigator (ME).

Statistical Methods

For studies that reported mean and standard deviation of vitamin D levels for NAFLD participants and controls, we combined the standardised mean differences (SMD) using Hedge's adjusted g to correct for small sample bias in a random effects model.[17] For studies that reported dichotomous outcomes (vitamin D deficiency), we pooled the odds ratios (OR) using the inverse variance method with a random effects model. For studies that reported ORs for quartiles/tertiles of vitamin D levels, we obtained a single OR for each study by combining all odds ratios using a fixed effects model and the single OR as the measure for the association between low vitamin D and NAFLD in that study. For two studies that did not report measures of association between low vitamin D and NAFLD,[18,19] we estimated the unadjusted odds ratio and 95% confidence interval (CI) by creating a 2 × 2 table of NAFLD and controls by vitamin D status (in absolute frequencies).

Recently published reports from Asia[20–23] suggest an ethnic difference between Asian and non-Asian populations in the association between vitamin D and NAFLD. This might be explained by allele frequency of vitamin D receptor (VDR) polymorphism between Chinese and Western populations;[24] therefore, we opted to stratify our analyses in Western (participants from USA, Italy, Canada and Israel), and Eastern (China and Korea) participants.

We assessed statistical heterogeneity with Cochran's Q-test and with the I 2 statistic. Publication bias was assessed using Harbord's regression test for funnel-plot asymmetry and Egger's test. All analyses were carried out using the commands metan and metabias, available in STATA 10 (College Station, TX, USA).