Systematic Review With Meta-analysis

The Risks of Proton Pump Inhibitors During Pregnancy

Cheng Mei Li; Alexandra Zhernakova; Lars Engstrand; Cisca Wijmenga; Nele Brusselaers


Aliment Pharmacol Ther. 2020;51(4):410-420. 

In This Article

Material and Methods

Eligibility Criteria, Information Sources, Search Strategy and Study Selection

We conducted a comprehensive systematic literature search in PubMed, Web of Science, EMBASE databases and Cochrane Database from inception, which was last updated June 2019 (Supplement 1), and we reported the results according to the PRISMA guidelines for systematic reviews and meta-analyses. The study protocol was registered at PROSPERO (CRD42018103320). The search was conducted independently by two reviewers (CML, NB) and any disagreements were solved by discussion. Backward and forward citation tracking on the identified eligible articles was used to identify articles potentially missed by the search. If no full-text version of eligible conference abstracts was found, we contacted the authors to complement the available data. All study designs (observational and intervention studies) were eligible if the association was assessed between PPI (or H2RA) exposure in pregnancy and the risk of neonatal adverse events compared to a comparison group, providing crude numbers or ratios. Exposure was defined as any type of PPIs (or H2RA), during pregnancy. Although teratogenic effects affecting the risk for congenital malformations only occur because of exposure before or during the first trimester, the critical time-window for exposure is not clear for the other outcomes, and may not always be clearly described. The comparison group included: (a) all or a selection of pregnant women not using PPIs (or H2RA) during pregnancy from the same cohort with or without the same indication, (b) the total population of pregnant women from the same study setting. Since H2RA are used for similar indications, we also assessed the risk of PPI use compared to H2RA use during pregnancy. The studies were eligible if at least one of the following neonatal adverse events were assessed (as defined by the authors): congenital malformations, preterm birth, low birth weight, small for gestational age, spontaneous abortion (miscarriage) and termination of pregnancy (for medical reasons or other), perinatal mortality and stillbirth. Information on mean/median gestational age and weight were collected from the exposed and comparison groups, if available. We excluded case studies, case series, cross-sectional studies, animal studies and articles in other languages than English. As H2RA were commercialised since the 1970 (PPIs since the 1980s) no eligible articles are expected to be published before the 1970, although the databases were searched from inception. When we identified two studies based on the same cohort, describing the same exposures and outcomes, only the largest and/or more recent one was included in these analyses.

Data Extraction

Data extraction included author, publication year, country, study type, population source and characteristics, control group, exposure (timing, type, duration), outcome definitions and categorisations.

Assessment of Risk of Bias

Two review authors (CML, NB) independently assessed the risk of bias for each included study by means of the Newcastle Ottawa quality assessment scales (NOS) for cohort and case-control studies.[27] Each study can be awarded up to nine stars, with a higher amount of stars reflecting a lower risk of bias, and we considered seven stars or more as a low risk of bias. In addition, we assessed in more depth the risk of selection bias (representativeness of the cohorts, cases and controls), information bias (ie recall bias of exposure, attrition bias for outcome), and confounding (including confounding by indication), and potential conflicts of interest.[28]

Data Synthesis

Data were analysed using Stata (metan package; StataCorp version 14). Random-effects modelling was used to pool the OR with 95% confidence intervals (CIs), based on the adjusted estimates. Since all outcomes are rare (<10%), risk ratios and odds ratios should be equivalent; yet if odds ratios were not reported, these were calculated based on the available crude numbers (so no measurements needed to be transformed).[29] A "treatment arm" continuity correction accounting for unbalanced groups was used to include studies with zero events in one arm, which takes into account the ratio of the group sizes,[30] yet a sensitivity analyses excluding these trials was presented where appropriate. Studies without events in both arms were excluded (since not always clear which outcomes were investigated in each study). Gestational weight and age could not be pooled, since rarely reported for both the exposed and comparison groups. Results on each outcome were only presented if reported in two or more studies. The I-square test shows the proportion of heterogeneity between studies, and was categorised as low (25%-50%), moderate (51%-75%), or high (>75%) statistical heterogeneity.[31] Subgroup analyses were used to the sources of heterogeneity, and were based on PPI type, study design and different trimester if at least two studies could be included. If different estimates could be used, adjusted results were preferred above unadjusted, and early pregnancy exposure (1st trimester) above late (2nd/3rd exposure) for congenital malformations, and late pregnancy for the other outcomes, unless otherwise specified.

From a developmental perspective, it is not expected that any exposure would increase the risk of all congenital malformations, since in reality a teratogen only affects a limited number of malformations. Yet the overall risk is what is generally reported in many studies, which may show associations if the effect sizes for the specific malformations are sufficiently high. In order to include as many studies as possible in the analyses for congenital malformations (to increase power), studies only reporting on specific malformations were also included unless otherwise specified. If these different malformations could not be combined into one risk estimate, the estimate of the outcome with the highest number of exposed cases was included. Since it is unexpected that teratogens increase the risk of all congenital malformations, specific malformations were assessed separately if feasible. Influence meta-analyses were conducted to investigate the influence of each individual study on the overall meta-analysis summary estimate. For the cumulative meta-analysis, all studies are sorted by the year of publication, and each line represents the accumulated effect of this study and all studies published earlier.[32] Symmetry of the funnel plots was assessed visually, and statistically by means of the Egger test, to assess small study effects (publication bias).[33]