Adverse pregnancy outcomes in women with celiac disease: a systematic review and meta-analysis

Search strategy

A systematic literature search of Medline, Cochrane Library and Scopus was conducted from inception till 12^th April 2022, to identify studies reporting on the risk of adverse pregnancy outcomes in women with CD. Key questions were formulated according to the PICO method: “Do pregnant women diagnosed with CD have an increased risk for adverse pregnancy outcomes compared to pregnant women without CD?” Text words and, if applicable, database subject heading fields (e.g., Medical Subject Headings), were used to perform the searches: “celiac disease,” “CD,” “coeliac disease,” “gluten enteropathy,” “pregnancy,” “premature,” “obstetric,” “complication,” “preterm delivery,” “spontaneous miscarriage,” “spontaneous abortion,” “preeclampsia,” “gestational hypertension,” “stillbirth,” “cesarean delivery,” “postpartum hemorrhage,” “gestational diabetes,” “placental abruption,” “small for gestational age”, and “fetal growth restriction”. Furthermore, we examined the references of each of the retrieved studies to identify further articles that met our criteria. We did not utilize any search software. The search was filtered for human and English language studies only. The title and abstract of studies identified in the original search were reviewed by 2 independent authors (KA, a gastroenterologist, and AS, an obstetrician/gynecologist) to eliminate studies that did not answer our research question, based on predetermined inclusion and exclusion criteria. The complete text of the included articles was then examined to see whether it provided any relevant information. The coefficient of agreement between the 2 reviewers for article selection (k ¼ 0.86, 95% confidence interval [CI] 0.79-0.95) was excellent. Conflicts in study selection were resolved by consensus, referring back to the original article and, if agreement could not be reached, a third author (IT, a biostatistician) was consulted and settled the differences.

Selection criteria

Studies in this meta-analysis had to be observational cohort or case-control studies that met the following inclusion criteria: 1) diagnosed CD according to serology testing and/or endoscopic duodenal biopsy [10]; 2) reported incident cases of at least 1 of the following obstetric complications: spontaneous abortion (SA), fetal growth restriction (FGR), preeclampsia, stillbirth, preterm delivery (PTD), cesarean delivery, postpartum hemorrhage and 5-min Apgar score <7; 3) included a non-CD population for which the aforementioned event rates were calculated (or could be inferred as expected event rates from a reference population); 4) reported relative risk (RR), rate or risk ratio, odds ratio (OR), with 95%CI or provided raw data for their calculation. Peer-reviewed observational controlled data (case-control and cohort studies) from hospitals, referral centers and population-based studies were included. Cross-sectional studies, studies without a control group, meta-analyses, review articles, short surveys, letters to the editor, notes, case reports, pilot studies and conference abstracts were excluded. In addition, studies that did not contain primary data were excluded and duplicates of studies were removed. The selection was not limited by the number of participants in each study.

Data extraction

Two investigators (KA and AS) reviewed and abstracted the data independently onto a standard pre-determined data extraction form. The following data were collected from the studies: first author and year of publication, study design, period study conducted, origin of study population, type of exposure (CD and control population), a priori outcomes of interest alongside their frequencies, total number of participants in each group (CD pregnant women vs. non-CD pregnant women), as well as any confounding factors reported in each study. FGR was defined as sonographic estimated fetal weight <10^th percentile for gestational age and PTD was defined as birth after <37 weeks of pregnancy. Authors were contacted in case of missing data in any of the eligible studies. If more than one published study came from the same population, only data from the most recent comprehensive report were included.

Outcome measures

The primary analysis focused on assessing the relative risk of predefined adverse pregnancy outcomes in pregnant women diagnosed with CD, compared with non-CD pregnant women originating from the general population, hospital or referral center. In addition, based on information available from individual cohort studies, we performed a subgroup analysis assessing the risk of predefined adverse pregnancy outcomes amongst pregnant women with diagnosed CD and those with undiagnosed CD, compared to the general population of healthy pregnant women. This analysis aimed to examine the impact of early CD diagnosis and subsequent adherence to a GFD during pregnancy on the risk of obstetric complications. The populations with sufficient information about the time of diagnosis in relation to pregnancy were subdivided into early diagnosed and undiagnosed CD groups. Whenever the same control group was utilized in a study for both the “early diagnosed” and “undiagnosed” CD study groups, the control group was equally subdivided into 2 study subgroups for the analysis [11,12].

Quality and risk of bias assessment of the studies

Two investigators (KA and AS) independently assessed the included studies. The quality was assessed using the Newcastle-Ottawa scale [13]. This scale is based on a “star system” that ranges from 0-9, with 0 being the lowest possible quality, and judges study quality according to 3 perspectives: selection of the study groups (4 questions), comparability of the groups (1 question) and ascertainment of the outcome of interest (3 questions). Each question was rated with a maximum of one star except for “comparability of the groups”, for which separate stars were awarded for controlling maternal age and/or any other additional covariate (maximum 2 stars). The risk of bias was assessed with the Quality In Prognosis Studies (QUIPS) tool. The Cochrane Prognosis Methods Group recommends it for assessing risk of bias in prognostic factor studies [14]. Six important domains should be critically appraised when evaluating validity and bias in studies of prognostic factors: (1) study participation [15]; study attrition; (3) prognostic factor measurement; (4) outcome measurement; (5) study confounding; and (6) statistical analysis and reporting [16]. The different domains contain between 3 and 7 prompting items to be rated on a 4-grade scale (yes, partial, no, unsure). Finally, based on their assessments of the included items, the rater gives an overall, decisive judgment of the risk of bias within each domain. This risk is classified into 3 levels (serious, moderate, and low risk of bias) [14]. Any disagreements between the 2 investigators were resolved by re-evaluating the original study.

Data synthesis and statistical analysis

The Cochrane Collaboration’s Review Manager Software (version 5.4) was used to perform data analysis. In the analysis of case-control studies, the ORs (95%CI) were calculated for each of the endpoints, whereas in the analysis of cohort studies, RRs (95%CI) were calculated for every single endpoint for CD pregnant women in comparison with controls. We used 2 approaches to analyze heterogeneity between study-specific estimates. First, the Cochran Q statistical test for heterogeneity, which tests the null hypothesis that all studies in a meta-analysis have the same underlying magnitude of effect. Because this test is underpowered to detect moderate degrees of heterogeneity, the presence of statistically significant heterogeneity across the studies was evaluated by utilizing a P-value <0.10. Second, to estimate which proportion of the total variation across studies was caused by study-related factors (clinical setting, methodological or statistical differences) rather than chance, the I² statistic was calculated, where I²=100%×(Q−df)/Q representing the magnitude of the heterogeneity—moderate: 30-60%, substantial: 50-90%, considerable: 75-100% [17,18]. In all analyses, dichotomous outcomes were pooled using the Der Simonian and Laird random-effects model, as it suited our analysis given the heterogeneity generally observed between observational studies. A probability level of 0.05 was considered statistically significant for all tests (excluding heterogeneity). Funnel plots and Egger’s test were used through the R studios to test for publication bias.

Eligible studies

The search strategy identified 1427 articles (Supplementary Fig. 1) After the removal of duplicates and screening of titles, abstracts and keywords, 28 papers underwent full-text review. During this process, 9 articles were excluded because of irrelevant outcomes [19-27]. One study was excluded because it neither provided risk estimates for the outcomes of interest nor offered data for calculations [28]. The remaining 18 studies [29-46], published between 1990 and 2021, fulfilled the selection criteria. Our meta-analysis included 14 cohort studies and 4 case-control studies, the general characteristics and results of which are described in Tables 1 and 2.

Table 1 Characteristics of the studies included in the meta - analysis regarding the risk for adverse pregnancy outcomes of women with celiac disease

Table 2 Results of the meta - analysis regarding obstetric complications and pregnancies with celiac disease compared to healthy pregnancies, alongside results of the subgroup analyses regarding the impact of early diagnosis of celiac disease on the course of the pregnancy

Quality and risk of bias assessment of the included studies

Based on the Newcastle-Ottawa scale [13], 2 studies were rated as 9-star, 6 studies as 8-star, 9 studies as 7-star, and 1 study as 6-star (Table 1). All studies provided a clear definition of the diagnosis of CD, including the details of confirmation based on serology testing and/or endoscopic duodenal biopsy. The risk of bias according to the QUIPS tool for every individual study will be depicted next to the forest plots. A letter was assigned to every domain of the tool: A for Study Participation, B for Study Attrition, C for Prognostic Factor Measurement, D for Outcome Measurement, E for Study Confounding, and F for Statistical Analysis and Reporting.

Analysis of dichotomous outcomes

Spontaneous abortion

Seven studies (5 cohorts) [34,35,37,40-42,44] were analyzed, including 7978 pregnant women with CD and 78,241 non-CD pregnant women; SA was detected in 11.5% (n=920) vs. 9.2% (n=7202), respectively. A statistically significant positive correlation was found between the risk for SA and CD (RR 1.35, 95%CI 1.10-1.65). This association was also observed in the analysis of the cohort studies alone. The heterogeneity was high in the overall analysis, as well as among the cohort studies (Fig. 1).

Figure 1 Forest plot demonstrating the relative risk for spontaneous abortion of pregnant women with CD compared to a healthy control group

CI, confidence interval; CD, celiac disease

FGR

We evaluated data from 5 cohort studies [33,34,38,43,45], with 5105 pregnant women with CD and 17,466,900 controls; FGR was detected in 3.9% (n=201) vs. 2.1% (n=364,192), respectively. A significant correlation between CD in pregnancy and FGR was observed (RR 1.68, 95%CI 1.34-2.10). The heterogeneity was moderate (Fig. 2).

Figure 2 Forest plot demonstrating the relative risk for fetal growth restriction of pregnant women with CD compared to a healthy control group

CI, confidence interval; CD, celiac disease

Analysis of continuous outcomes

Subgroup analyses

In total, 11 studies offered relevant data and were included in the subgroup analysis (Table 1). No statistically significant risk was observed among the early diagnosed and the undiagnosed CD groups with spontaneous abortion, compared to the control group (Fig. 3). A statistically significant positive correlation with FGR was identified solely among the undiagnosed CD (RR 1.94, 95%CI 1.37-2.73) (Fig. 4), while only women with undiagnosed CD had an higher risk for stillbirth compared to healthy controls (RR 1.73, 95%CI 1.16-2.56) (Supplementary Fig. 9). The undiagnosed CD subgroup had an elevated risk of PTD compared to controls (RR 1.41, 95%CI 1.17-1.69), whereas no indication of elevated risk was observed for the early diagnosis subgroup with regard to PTD; a significant difference among the subgroups of undiagnosed vs. early diagnosed CD was identified (P=0.02) and the initial heterogeneity was also minimized (Supplementary Fig. 10). No greater risk of cesarean delivery was detected in either the early diagnosed or the undiagnosed CD group compared to healthy controls (Supplementary Fig. 11). Interestingly, the undiagnosed subgroup had a lower mean BW compared to healthy controls (MD -280.53, 95%CI -456.13 to -104.93), while no difference was found among the early diagnosed and the healthy control group; a significant difference was noted in the subgroups of undiagnosed vs. early diagnosed CD (P<0.001) (Supplementary Fig. 12). All subgroup analyses regarding the impact of early diagnosis of CD on the course of the pregnancy are summarized in Table 2.

Figure 3 Subgroup analysis on spontaneous abortion based on the time of CD diagnosis

CI, confidence interval; CD, celiac disease

Figure 4 Subgroup analysis on fetal growth restriction based on the time of CD diagnosis

CI, confidence interval; CD, celiac disease

Publication bias

PTD was the outcome included in the majority of the included studies, so it was tested for publication bias (Supplementary Fig. 13,14). The funnel plot in which the individual studies were scattered symmetrically on the vertical axis showed no indication of publication bias, while Egger’s test confirmed this (P=0.12). It should be noted, however, that the PTD meta-analysis contained only 9 studies. Thus, both the funnel plot and Egger’s test may have lacked the statistical power to detect publication bias (Supplementary Fig. 15).