3.1. Search Results
The detailed steps of the study selection are given as a PRISMA ﬂow diagram in Figure 1. A total of 2498 abstracts were retrieved from the databases, and two studies was added from the references; 1639 were excluded after reading titles and/or abstracts, and 75 articles were subjected to a full-text review. After reading the full text, a total of 24 cohorts were invited to participate, which included 13 cohort studies [27-39], seven case-control studies [40-46] and four cross-sectional studies [47-50]. Because three studies [29-31] of 24 measured serum 25(OH)D concentrations at two or three periods of pregnancies, our meta-analysis included nine studies [28-31,36,39,41,44,46] in the first trimester, 11 studies [27,30-35,37,38,40,43] in the second trimester, and nine studies [29-31,42,45,47-50] in the third trimester.
3.2. Study Characteristics
The full list of studies included [27-50] and their main characteristics are shown in Table 1. The included studies from the United States, Canada, Australia, New Zealand, Spain, Netherlands, Swedish, Poland, Brazil, Kenya, China, Singapore and Thailand. A total of nine studies [28,30,31,33,37,40,48-50] were from Asian countries, six [36,41,43,45-47] from American countries, five [29,34,38,39,42] from European countries, three [27,35,44] from Oceanian countries, and one  was from African countries. Among 24 studies, 21 [27-46] cohort studies or case-control studies were appraised by NOS, resulting with scores above 6, while four [47-50] cross-sectional studies were appraised by the ARHQ methodology checklist, which also showed good quality (see supplementary materials).
Most of the studies defined vitamin D deficiency as a serum 25(OH)D below 50 nmol/L or 20 ng/mL, but two studies [29,42] defined as 25(OH)D below 30 ng/mL were also included because there were data about serum 25(OH)D below 20 ng/mL; thus, the criteria for diagnosis and data extraction were agreed upon. Similarly, the majority of studies defined preterm birth as a gestational age <37 weeks and term birth as a gestational age ≥37 weeks, but one study  defining gestational age <35 weeks were also included because data about gestational age < 37 weeks were available.
These studies were carried on from 1999 to 2017, and published from 2012 to 2018. Of 24 studies, seven different assay methods were used to measure maternal vitamin D levels, which is in accordance with the Vitamin D standardization program (VDSP) . Importantly, liquid chromatography tandem mass spectrometry (LC-MS/MS) is considered to be the gold standard for the determination of vitamin D [51,52].
All of these studies described the association between vitamin D deficiency during pregnancy and preterm birth, whether negative or positive.
According to the lower serum 25(OH)D concentration (<50 nmol/L or <20 ng/mL) which diagnosed vitamin D deficiency, the results of the meta-analysis appear to be inconsistent in the different periods of pregnancy. In Figure 2, the association between maternal vitamin D deficiency in the first trimester and preterm birth was not statistically signiﬁcant (OR = 1.01, 95%CI: 0.88, 1.16, P = 0.876). In Figure 3, the pregnant women with vitamin D deﬁciency in the second trimester showed no statistical significance regarding the risk of developing preterm birth (OR = 1.12, 95%CI (0.92, 1.37), P = 0.249) in a random effect model. In Figure 4, the association between maternal vitamin D deficiency in the third trimester and preterm birth was not statistically signiﬁcant (OR = 1.05, 95%CI: 0.87, 1.27, P = 0.602).
3.4. Sensitivity and Subgroup Analysis
In the meta-analysis of the association between maternal vitamin D deficiency in the first and third trimesters and preterm birth, tests revealed no heterogeneity (I2 = 0, P＞0.1; I2 = 15.3%, P＞0.1); thus, a ﬁxed effect model was used for meta-analysis. In the meta-analysis of the relationship between maternal vitamin D deficiency in the second trimester and preterm birth, heterogeneity tests revealed that I2 = 60.3% (P < 0.1), indicative of moderate heterogeneity; thus, a random effect model was used. The sensitivity analyses, shown in Figure 5, indicated significant changes in the result when the study by Zhou et al.  was excluded. Subgroup analyses were performed according to the study design and the continents with relevant countries. A significant association was identified in two case-control studies [40,43] between maternal vitamin D deficiency and PTB in Figure 6 (OR = 1.33, 95%CI: 1.15, 1.54, P = 0.000). Stratifying the countries from different continents in Figure 7, five studies conducted in Asian countries [30,31,33,37,40] showed high heterogeneity (I2 = 77.4% (P < 0.1)), while the study by Bodnar et al. in the Americas  revealed a statistically signiﬁcant protective effect among pregnant women, with an OR of 1.32 (95% CI 1.13 – 1.54, P = 0.001). A random effect model was used for this meta-analysis because of the heterogeneity in all subgroup analyses.
3.5. Publication Bias
The Begg’s funnel plot of the effect of vitamin D deficiency in the second trimester on preterm birth appeared to be symmetrical, as shown in Figure 8. No signiﬁcant publication bias was detected (P = 0.69). Regarding other outcomes of vitamin D deficiency in the first or third trimester and preterm birth, due to the limited number of studies, publication biases cannot be excluded.