Background
Proton pump inhibitors (PPIs) are the most effective treatment of heartburn [
1], but contra-indicated during pregnancy. Yet PPIs are still prescribed in approximately 1% of all pregnancies according to our recent meta-analysis [
2], and also available over-the-counter in several countries including Sweden [
3,
4]. Maternal PPI use might affect the child via different mechanisms. PPIs have been shown to cross the placenta [
1] and prenatal exposure to PPIs is seemingly associated with an increased risk of developing childhood asthma [
5]. In addition, the initial gut colonization is highly influenced by the maternal microbiome (vaginal and fecal) [
6,
7], and the maternal microbiome seems to play an important role in the onset of pregnancy complications [
8‐
10]. In turn, PPI use has been associated with important changes in the gut microbiome that appear to be more prominent than those related to antibiotic use [
11], also in infants as shown in our small pilot study [
12]. The Food and Drug Administration (FDA) used to classify (up till 2015) most PPIs as category B drugs (“No risk in animal studies”), except for omeprazole which was categorized as type C (“Risk cannot be ruled out”) [
13,
14]. Current recommendations are that omeprazole is not recommended during breastfeeding and that it “should be used during pregnancy only if the benefit outweighs the risk to the fetus” [
15]. There were no reports of teratogenicity, and PPI use was not associated with major adverse pregnancy outcomes or birth defects [
16‐
19], but was related to a lower birth weight [
16,
17] and an increased risk of pre-eclampsia [
20]. However, our recent meta-analysis reported an increased risk of congenital malformations associated with PPI use during pregnancy [
2]. Little research is done concerning the effect of PPIs on less severe health risks that might have long-term implications for the mother and her offspring including maternal complications (gestational diabetes, pre-eclampsia), preterm birth, and small or large for gestational age. In our previous meta-analysis, we did find a handful of studies addressing neonatal adverse events, yet none of the pooled analyses (beyond congenital malformations) reached statistical significance, which may be due to power issues and low prevalences of exposure to PPIs [
2]. This highlights the need to determine the effect of PPIs on the pregnant woman and her developing child.
This large nationwide population-based Swedish cohort study aimed to investigate the relation between the use of PPIs shortly before and during pregnancy on the risk of maternal and neonatal health complications.
Methods
The study was performed using a Swedish cohort including all live singleton births delivered between July 2006 and December 2016, and the terminology “women” and “mother” were defined based on their biological sex and pregnancy status, not their gender identity. The cohort was created by linking information from four high-quality nationwide Swedish health data registries maintained by the National Board of Health and Welfare (Socialstyrelsen), as described earlier [
21,
22]: the Medical Birth Registry [
23‐
25] (established in 1973), the Prescribed Drug Registry [
26] (established in July 2005), the Patient Registry (in- and outpatient care) [
27,
28], and the Causes of Death Registry (since 1952). Information was linked through the unique Swedish personal identification number [
29]. The study was approved by the Regional Ethics Committee of Stockholm (2017/2423–31), without the need for informed consent because of the registry-based nature of the data.
Outcomes
The maternal outcomes were pre-eclampsia, characterized by hypertension (systolic blood pressure > 140 mmHg and/or diastolic blood pressure > 90 mmHg) combined with proteinuria (24-h urine protein level > 300 mg), and gestational diabetes mellitus (GDM), defined as any degree of glucose intolerance with the onset during pregnancy (Additional file 1: Table A
1) [
30]. The neonatal outcomes included preterm birth (birth < 37 weeks of gestation), an Apgar score 5 min after birth (AS
5min) < 7, and small (SGA) and large for gestational age (LGA) based on birth weight in the 10th or 90th percentile based on gestational age, defined as birthweight below the 10th and above the 90th population percentile, respectively. [
30,
31]
Study exposure
The drugs in the Prescribed Drug Registry are classified according to the Anatomical Therapeutic Chemical (ATC) Classification System and the duration of use is expressed in defined daily doses (DDDs) per package. The exposure was prescribed PPI (ATC-code: A02BC) in the period ranging from 3 months before the last menstrual period (LMP) up to the delivery date. Women filling at least two prescriptions during the study period were considered users since lower compliance is expected for those with only a single prescription. [
32]
Covariates
Potential confounders included maternal characteristics (age at delivery, body mass index (BMI), tobacco consumption (smoking or moist snuff use), other prescribed drug use, comorbidities), pregnancy, and obstetric characteristics (Additional file 1: Table A
1). Missingness in BMI was adjusted for by creating an additional dummy variable. The use of prescription drugs was split into histamine-2 receptor antagonist (H
2RA) use and other drugs (NSAIDs, low-dose aspirin, and antibiotics). The use of H
2RA was separated because it is prescribed for similar indications as PPIs. Maternal comorbidities were identified according to their ICD-10 codes or by the prescription of associated drugs (Additional file 1: Table A
1). The comorbidities included hypertension, GDM, diabetes mellitus (type 1 and type 2), and hypo- and hyperthyroidism.
Pregnancy and obstetric variables, of which some are also outcomes, expected to be associated with at least one outcome were pre-eclampsia, mode of delivery (cesarean section or vaginal delivery), preterm birth, neonatal birthweight (SGA, average for gestational age (AGA) or LGA), parity, time in months since previous delivery, and whether the outcome was present in a previous pregnancy.
Statistical analysis
The effect of PPIs was assessed by comparing PPI users with non-users. Multiple logistic regression models were used to evaluate the association between the exposure and the odds that the outcome occurred, corrected for covariates, and presented as odds ratios (OR) with 95% confidence intervals (CI).
Models were built independently for each outcome, based on the purposeful selection method described by Hosmer et al. [
33]. For more detail, see Additional file
2: Additional Methods. [
33‐
37]
All final models included PPI use, irrespective of whether it was significant, because it was the exposure of interest. The model was concluded as the final model after assessing the adequacy and fit of the model. The cohort contained women with one or more pregnancies resulting in live birth. Generalized Estimating Equations (GEE) were used to take the correlation between siblings into account [
38]
.
All analyses were performed in R version 3.6.1 [
39‐
42]. Observations with missing information on one of the outcomes (
n = 6105) were removed from all analyses. All covariates were categorical and, if necessary, included a separate category for missing values.
Sensitivity analysis (only firstborns)
Multiple linear logistic regression was performed for each outcome, including only the firstborn children. Covariates included all before-mentioned maternal characteristics. Pregnancy and obstetric characteristics included were pre-eclampsia, mode of delivery, preterm birth, and neonatal birthweight, excluding covariates considering multiple pregnancies.
Use of PPIs at different timepoints (trimesters)
To determine if the use of PPIs had a different effect depending on the timing of the prescription, the exposure was split into four time periods. The first period ranged from 3 months before up to the last menstrual period (LMP). The time between the LMP and the delivery was divided into trimesters. The first trimester lasted until 97 days after LMP, the second starting at 98 days until 202 days of gestation, and the third trimester ranged from 203 days of gestation to delivery.
Dose–response association
The dose–response association between PPI use and the outcomes was assessed in two models. First, PPI use was replaced by a categorical variable describing the number of prescriptions (≤ 1, 2, 3, ≥ 4). Second, PPI use was described by a categorical variable based on the quantiles of the DDD presented as the number of weeks.
Discussion
In this large Swedish population-based study, PPI use shortly before and during pregnancy was associated with a higher probability of pre-eclampsia, GDM, preterm birth, and being born SGA. Analysis of only the firstborn child of a mother yielded similar results. Differences in the outcomes were seen by the different exposure periods based on all pregnancies. PPI use in the period ranging from three months before LMP until the end of the first trimester was associated with increased odds of preterm birth and SGA. Similarly, the odds of preterm birth were also higher when PPIs were used during the second trimester. PPI use in the second and third trimesters of pregnancy was associated with higher odds of pre-eclampsia. No evidence for a dose–response relation between PPI use and any of the outcomes was found.
Our results are consistent with previous studies relating PPI use to a higher (or not reduced) risk of pre-eclampsia [
20,
43,
44] and low birth weight [
16,
17]. Contradicting to our results, other observational studies have reported no significant relation between PPI use and low birthweight and/or preterm birth [
17,
19]. However, compared to this large nationwide study, both studies had a much lower number of observations available and mainly focused on major anomalies. To our knowledge, there are no randomized clinical trials investigating the safety of maternal PPI use, regarding maternal and neonatal adverse events [
2,
45,
46]. We also question if it is still ethically defendable to conduct these on PPI use during pregnancy with the accumulating safety concerns based on association studies, and our increasing understanding of drug interactions and the microbiome [
2,
5,
12,
47,
48].
The maternal and neonatal adverse events investigated can affect short- and long-term health of both the mother and the child. Pre-eclampsia is a cause of worldwide maternal and perinatal morbidity and mortality [
49]. GDM has previously been associated with an increased frequency of maternal hypertensive disorders and an increased risk of type 2 diabetes after pregnancy. GDM has been associated with the child having a higher odds of developing obesity, glucose intolerance, and diabetes in late adolescence and young adulthood [
50]. Preterm birth is a major cause of neonatal and infant morbidity. Children that are too small or too light at birth, have a higher risk of hypertension, obesity, and diabetes mellitus type 2 later in life [
51] and a lower quality of life when young adults [
52]. We do acknowledge that causality cannot be established and that unknown confounders may still affect the results. Although we did adjust for BMI, we did see that obesity was more prevalent among PPI users (21.6% vs. 11.7%) and was associated with higher odds of pre-eclampsia and gestational diabetes, preterm birth and large for gestational age, as previously described in the literature. [
53,
54]
Despite being contra-indicated, 1.4% of pregnant women in this cohort were PPI users (excluding over-the-counter use). Many studies reporting PPI utilization during pregnancy report a prevalence below 2% [
2,
5,
55,
56] with a few studies reporting up to 6% [
43,
57,
58]. Although this is lower than reported utilization in non-pregnant adults [
59‐
63], over-the-counter use is usually not included [
64]. With 4 million pregnancies born in the European Union in 2020 alone, a 1% PPI prevalence equals 40,000 pregnancies per year [
65]. As our results do support previous safety concerns [
16‐
19], more awareness to potential consequences of PPI use during pregnancy seems warranted. To note, PPI use has been considered inappropriate in up to 70% of (non-pregnant) long-term users [
66,
67].
The underlying mechanisms on how PPIs affect our health need further exploration; as well as safer (non-pharmaceutical) alternatives for treating of gastro-intestinal symptoms and discomfort during pregnancy. PPIs may still have a place for restricted indications during pregnancy, yet widespread and unsupervised over-the-counter use should be discouraged.
This study has several strengths including the large registry-based nature of the data and its high completeness, resulting in a large nationwide and population-based study with highly valid data on outcomes, exposure, and covariates.
Despite the high completeness of the registries and adjustments for confounders, confounding by indication could not be entirely ruled out. Nausea, vomiting (hyperemesis gravidarum), gastro-esophageal reflux, and/or peptic ulcers may be more common and/or severe among PPI users than non-users. Although these indications may increase the risk for adverse events, it remains unclear if PPI use can reduce the risk [
68]. Reverse causation could be affecting the associations, particularly for third-trimester exposure, for which the effect could also be underestimated since not all deliveries reach the end of the third trimester.
In addition, information on the outcomes and some covariates was incomplete (< 1% overall). Observations with any of the outcome variables missing (0.006%) were removed. Information on the exposure was limited by the availability of PPIs over the counter and lack of confirmation whether the women actually used the drugs, although it is expected that most women will use prescribed drugs only (after advice from their midwife/clinician). We included only women with at least two dispensed prescriptions, indicating that they were utilizing the drugs. This, however, could lead to misclassification of women filling a single prescription taking the drug, or women who only took PPI over the counter. Another potential concern is that a woman with previously diagnosed diabetes might have been missed and/or misclassified as having GDM. Due to a lack of power (only 0.2% of this cohort used H2RA), it was not possible to assess the effect of H
2RA use on the odds of an outcome. H
2RA is prescribed for similar indications as PPIs and it is recommended only to prescribe PPIs if antiacids and H
2RA do not sufficiently relieve symptoms [
1,
13]. The registries did not provide data on potential confounders such as whether the mothers took tocolytics to suppress preterm delivery, chronic hypertension, and hyperemesis. We also lacked information on ethnicity and socio-economic status. Nonetheless, our results were adjusted for important factors including diabetes mellitus, gestational hypertension, and hypo- and hyperthyroidism. The current analysis only included whether the outcome of interest was present in a previous pregnancy, but not if any of the other outcomes was. Others reported on a previous LGA child increasing the risk of GDM [
69] and previous SGA the risk of preterm birth [
70]. Furthermore, we would not expect socio-economic differences to have influenced our results significantly, because pregnancy-related health care in Sweden is highly standardized, equally accessible for the entire country, and free for the expecting mother. We chose to categorize exposure by trimester above time-varying exposures, since this is the most applicable to clinical antenatal practice (in particular since any antenatal PPI use is contra-indicated). We also do not have the exact period of exposure, since the duration of use is estimated based on the average use per package.
Publisher's note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.