**3. Results**

### *3.1. Study Selection*

Initially, 290 potential studies were identified from the search. Figure 1 shows a flowchart of the study selection process based on the PRISMA statement [23]. After the initial pass, 58 were excluded as duplicates. 212 were excluded after reading the title and abstract because of evident irrelevance. In the second pass, the full text of the 20 studies selected in the first pass were read and 10 studies were excluded for not meeting the search criteria. Two articles were excluded because they did not provide sufficient data for the calculation of Ors with 95% CI [32,33]. Three papers were excluded because they were family-based [34–36]. Two papers were excluded as they assessed associations between polymorphisms not in linkage disequilibrium with the selected variants [37,38]. Two papers did not investigate the association between the selected variants and T1D, investigating a different outcome [39,40]. Only one study was excluded due to using the same sample population [24]. Therefore, 10 studies were included in this systematic review.

**Figure 1.** Flowchart illustrating the literature search and study selection.

### *3.2. Characteristics of Included Studies*

The summary characteristics of included studies are shown in Table 1. The studies were published between 1999 to 2021, conducted in different geographical locations. Of the 10 included studies, seven were case-control studies [25,40–44], and three had a cohort design [45–47]. Most studies focused on T1D in childhood, as indicated by the mean age of onset in cases. Appropriate genotyping methods and diagnostic criteria were used in all included studies. Of the studies selected, six studies [40–43,47] fulfilled the WHO diagnostic criteria for T1D, while the majority of the remaining studies [25,44–46,48] indirectly captured criteria by description from multiple case sample populations. Polymerase chain reaction-restriction fragment length polymorphism (PCR-RFLP) was used by half the included studies as the genotyping method.

Similarly, none of the eligible studies endeavoured to control for vitamin D dietary intake through infancy and/or childhood, a known risk factor of T1D. However, when study quality was assessed, all included studies presented with a low risk of bias using the CASP tools, with no deviation from the Hardy-Weinberg equilibrium in controls reported in all case-control studies, and only some studies presenting with one item partially unmet (Supplementary Table S5).

Statistical methods to control confounding varied between studies. Most studies adjusted for different potential confounding factors, such as age, sex, genotype batch, geographical origin and BMI (see Table 1). Two remaining papers were matched casecontrol studies to control for known potential confounding variables. Hussein et al. [41], matched by age and ethnic origin, while Mahmoud et al. [42] matched by gender. Six studies [40,43–47] did not report OR results directly, and some, but not all, of the studies, generated adjusted ORs.


**Table 1.** Characteristics of observational studies evaluating the association between vitamin D genetic variants and type 1 diabetes included in the meta-analysis.


**Table 1.** *Cont.*

OR, odds ratio; VDBP, vitamin D binding protein. a Each effect allele represents the 25(OH)D concentration increasing allele, as defined by Sunlight Consortium [21]. b Identified using LDproxy, coded by 25(OH)D concentration decreasing alleles (see methods) c Effect allele direction reversed based on 25(OH)D concentration increasing, as defined by Sunlight Consortium (see methods) [21]. d Low/undetectable serum C-peptide and presence of 1+ pancreatic autoantibodies.

### *3.3. Findings from the Meta-Analysis*

All specified polymorphisms (namely rs10741657 G/A (*CYP2R1*), rs117913124 A/G (*CYP2R1* low frequency), rs12785878 G/T (*DHCR7/NADSYN1*), rs3755967 T/C (*GC*), rs17216707 C/T (*CYP24A1*), rs10745742 C/T (*AMDHD1*), rs8018720 C/G (*SEC23A*) were reported in three or more studies and taken forward to the meta-analyses. Associations between the SNPs and T1D, using individual and pooled OR estimates, are displayed in Figure 2 and Supplementary Figure S1.


**Figure 2.** Meta-analysis for the association between selected genetic variants affecting serum 25-hydroxyvitamin concentrations and type 1 diabetes with the random effects model (variants coded by 25-hydroxyvitamin D increasing alleles). Squares represent the individual odds ratio estimate. Diamonds show the pooled effect. Horizontal bars represent the 95% confidence intervals.

For rs10741657 G/A (*CYP2R1*), the reported ORs ranged from 0.46 to 1.11 (Figure 2). The random-effects pooled OR was 0.97 (95% CI 0.93, 1.02; *p* = 0.01) with little heterogeneity among the studies (I2 = 25.1%). For rs117913124 A/G (*CYP2R1* low frequency), the ORs ranged from 1.00 to 1.07 (Figure 2) with a pooled OR of 1.02 (95% CI 0.94, 1.11; *p* = 0.78; I = 0.0%). For rs12785878 G/T (*DHCR7*/*NADSYN1*), the ORs ranged from 0.78 to 1.06 (Figure 2), with a pooled OR of 0.99 (95% CI 0.92, 1.07; *p* = 0.02). There was evidence of moderate between-study heterogeneity (I2 = 64.8%). For rs3755967 T/C (GC), the OR ranged from 0.99 to 1.53 (Figure 2), with a pooled OR of 1.02 and no sign of heterogeneity (95% CI 0.99, 1.06; *p* = 0.97; I = 0.0%). In the evaluation for publication bias, asymmetry in Begg's funnel plot was observed for *GC* rs3755967 (Supplementary Figure S2). For rs17216707 C/T (*CYP24A1*), the OR ranged from 0.96 to 1.03 (Figure 2). The randomeffects model pooled OR was 1.00 (95% CI 0.95, 1.04, *p* = 0.37), with little indication of heterogeneity (I2 = 18.0%). For rs10745742 C/T (*AMDHD1*), the OR ranged from 1.00 to 1.02 (Figure 2) with a pooled OR of 1.00 (95% CI 0.97, 1.04; *p* = 0.90). Again, there was no sign of heterogeneity (I2 = 0.0%). For rs8018720 C/G (*SEC23A*), the OR ranged from 0.97 to 1.05 (Figure 2). The REM yielded a pooled OR of 1.01 (95% CI 0.95, 1.07, *p* = 0.19) with little heterogeneity among the studies (I2 = 42.8%). In view of these individual estimates, under the studied models no statistically significant associations between any of the seven SNPs alone (or their proxies) and T1D were found. Other than in rs3755967 (*GC*), no other asymmetry in Begg's funnel plot was observed. No outcome reporting bias was detected in any of the studies.

Furthermore, a sensitivity analysis was also performed to assess the influence of each study using the leave-one-out method. The pooled ORs were not changed materially and remained not significant, indicating good stability of results (range of pooled OR: 0.97–1.02). A subgroup analysis performed on the Caucasian population found no manifestations of association, with no major changes in primary outcomes (Supplementary Figure S1). Analyses showed all seven selected polymorphisms (or their proxies) were not associated with T1D risk under the studied models (range of pooled OR: 0.98–1.02).
