Abstract
Background
Literature is divided regarding the risk of neonatal ventricular septal defect (VSD) associated with first trimester ondansetron use in pregnancy.
Methods
We evaluated the risk of VSD associated with first trimester exposure to intravenous or oral ondansetron in 33 677 deliveries at Magee–Womens Hospital in Pittsburgh, PA (2006–2014). Using log-binomial regression, we evaluated the risk: (1) in the full cohort, (2) using propensity score designs with both matching and inverse probability weighting and (3) utilizing clustered trajectory analysis evaluating the role of dose. Sensitivity analyses assessed the association between ondansetron and all recorded birth defects in aggregate.
Results
A total of 3733 (11%) pregnancies were exposed to ondansetron in the first trimester (dose range: 2.4–1008 mg). Ondansetron was associated with increased risk of VSD with risk ratios ranging from 1.7 [95% confidence interval (CI) 1.0–2.9] to 2.1 (95% CI 1.1–4.0) across methods. Risks correspond to one additional VSD for approximately every 330 pregnancies exposed in the first trimester. The association was dose-dependent with increased risk in women receiving highest cumulative doses compared with lowest doses [adjusted risk ratio: 3.2 (95% CI 1.0–9.9)]. The association between ondansetron and congenital malformations was diluted as the outcome included additional birth defects.
Conclusions
First trimester ondansetron use is associated with an increased risk of neonatal VSD potentially driven by higher doses. This risk should be viewed in the context of risks attributable to severe untreated nausea and vomiting of pregnancy.
Keywords: Pregnancy, ondansetron, ventricular septal defect, first trimester
Introduction
Nausea and vomiting occurs in up to 80% of pregnancies.1 Despite lack of United States Food and Drug Administration approval for indicated use in pregnancy, the most common treatment for nausea and vomiting of pregnancy (NVP) in the USA is ondansetron. Use of both the oral and intravenous formulations have increased, resulting in an estimated one-quarter of all pregnancies being exposed to ondansetron in 2014.2 High utilization results from the frequency of NVP,1 the detrimental effects associated with the illness3 and the efficacy of this drug.4
Despite widespread usage, the safety of ondansetron in pregnancy is still in question. Several studies have found no association between ondansetron use and congenital malformations studied in aggregate.5–9 However, grouping all major malformations is problematic as defects are heterogeneous and causes vary substantially.10 Further, only exposure during organogenesis is expected to be relevant to any structural birth defect. Indeed, when first trimester ondansetron use has been studied relative to individual birth defects, an increased risk of cardiac malformations, specifically ventricular septal defects (VSD), was reported.8 Conversely, recent evaluation of outpatient prescriptions in a large cohort of Medicaid pregnancies,9 along with an evaluation of two birth defects case-control studies,11 found no significant association between ondansetron exposure and neonatal VSD.9,11
Considering this uncertainty regarding cardiac malformations, our objective was to evaluate the association between first trimester ondansetron exposure and neonatal VSD inclusive of inpatient and outpatient treatment.
Key Messages
Ondansetron is the most commonly used treatment for nausea and vomiting of pregnancy despite lack of United States Food and Drug Administration approval for this indication.
We evaluated the risk of ondansetron use in the first trimester and associated risk of neonatal ventricular septal defect.
In a population of 33 677 deliveries, 3733 of which were exposed in the first trimester, we found a slight increase in the absolute risk of ventricular septal defect that was dose-dependent.
Because we lacked a quantitative measure of the harms resulting from untreated severe nausea and vomiting of pregnancy, we were unable to evaluate the risk–benefit ratio of medication use. We therefore suggest only that this be used as another input for clinical decision making in pregnancy.
Methods
Population
Our retrospective cohort is comprised of liveborn, singleton deliveries at Magee–Womens Hospital of the University of Pittsburgh Medical Center (UPMC) with UPMC Health Plan coverage from 2006 through 2014. UPMC Health Plan is the largest insurer in the region. Demographic and delivery information from the Magee Obstetric, Medical, and Infant (MOMI) Database, a registry of all deliveries at the hospital, was linked with claims from UPMC Health Plan and to the inpatient (hospitalization) and outpatient (clinic visit) electronic medical records. Data were linked and then deidentified by a third party approved by the Institutional Review Board. The MOMI database is regularly validated by comparing data with randomly chosen patient charts. The University of Pittsburgh Institutional Review Board approved this study.
Of the 84 277 singleton deliveries at Magee–Womens Hospital in the study period, 34 341 (41%) had UPMC insurance and were therefore eligible for the study. Deliveries missing gestational age at delivery were excluded (n = 664), resulting in a final sample of 33 677. Missing data for continuous variables were replaced with the median cohort value for that variable, and categorical variables maintained a status of unknown. All analyses were repeated excluding variables with ≥10% missing data (Table S1 in Supplementary material, available as Supplementary data at IJE online).
Ondansetron exposure
Ondansetron (Zofran, GlaxoSmithKline) exposure (intravenous and oral) was extracted from both the inpatient electronic medical record for treatment administered within the hospital and through insurance claims for outpatient prescriptions. Exposure was evaluated as binary (any vs none) in the first trimester, defined as 0–13 weeks and 6/7 days,12 and as a continuous variable reflecting dose prescribed on each day of gestation.
Gestational age at delivery was estimated using ultrasound or last menstrual period when ultrasound was not available. Gestational age, date of delivery and start date of therapy, were used to estimate the gestational age at each exposure to ondansetron.
For continuous dosing analyses, outpatient dose was calculated as though women followed daily dosing as prescribed. For example, a prescription written as ‘take 1 tablet every 8 hours’ for 21 tablets of 4 mg ondansetron and filled on day 100 of pregnancy, was analysed as exposure to 12 mg per day for days 100 through 106 of her pregnancy. Routes of administration were combined, as route is not expected to impact the association with VSD, after converting oral ondansetron doses to intravenous bioequivalent dose.13 Oral ondansetron exposure was converted to intravenous based on bioavailability data using the equation: intravenous dose = oral dose x 0.6. For example, if a woman received 4 mg intravenous and 8 mg orally on the same day, her exposure would be analysed as 8.8 mg for that day (4 mg + 8 mg x 0.6).
Neonatal cardiac anomalies
The primary outcome evaluated was VSD. Secondary outcomes were septal defects [ventricular or atrial septal defect (ASD)] and any heart defect (ASD, VSD, atrial-ventricular canal, hypoplastic left heart, coarctation of the aorta, mitral or aortic valve lesions, peripheral pulmonic stenosis or teratology of Fallot). All women delivering at Magee–Womens Hospital have an anatomy ultrasound at 18–22 weeks which entails careful evaluation of the fetal cardiac anatomy. When the fetal heart cannot be adequately visualized, a fetal echocardiogram is performed. At the time of delivery, heart murmurs suspected to be a VSD are further evaluated with an echocardiogram before being documented in the chart and registry. Confirmed abnormalities are captured as International Classification of Diseases (ICD)-9 codes and documented in the perinatal database at the time of delivery.
In sensitivity analyses, we assessed all birth defects captured through ICD-9 codes in the perinatal database in aggregate. In addition to cardiac anomalies, the combined birth defect outcome also included: cleft palate, arrhythmia, cardiomyopathy, gastroschisis, structural intestinal obstruction, congenital anatomical central nervous system (CNS) abnormality, or anomalies of the lower urinary tract, renal system, eye, ear, thoracic cage or musculoskeletal system.
Potential confounders
Data were available from the electronic medical record on the number of prenatal visits with a diagnosis of nausea, vomiting, hyperemesis, gastrointestinal complaint or gastrointestinal infection and the total number of hospital and office visits during pregnancy. ICD codes were reviewed by clinicians to identify those indicative of an encounter related to NVP (see Supplementary material, available as Supplementary data at IJE online). For primary analyses these counts were categorized to 0, 1, 2, 3 and ≥4 encounters with an NVP-related diagnosis. The perinatal database provided information on maternal characteristics and comorbidities detailed below.
Statistical analyses
Using log-binomial regression with robust standard error to account for women with multiple pregnancies, we evaluated the association with three distinct designs: (1) full cohort, (2a) propensity score-matched sample, (2b) propensity score-inverse probability weighted sample, (3a) clustered trajectory analysis of daily dose and (3b) clustered trajectory analysis of cumulative dose. We considered approaches 1 and 2 primary analyses, and approach 3 as exploratory to assess dose–response relationships between exposure level and VSD.
We used theory-based causal diagrams to determine the minimal sufficient covariate set for confounder adjustment (Figure S1 in Supplementary material, available as Supplementary data at IJE online).14,15 We considered the following confounders: number of prenatal visits and those with NVP-related diagnoses, insurance type (private, public, self-pay), maternal age, race, education, marital status, pregravid body mass index (kg/m2), history of abortion/miscarriage/termination, illicit drug use, arrhythmia, chronic hypertension, collagen vascular disease, lower or upper gastrointestinal disease, diabetes, maternal structural heart disease, ultrasound with nuchal fold measurement, smoking status, patient of resident clinic. Results from the unadjusted and fully adjusted models are presented as risk ratios (RR) with corresponding 95% confidence intervals (CI). For cohort and propensity score designs, risk differences were used to calculate the number needed to harm (NNH = 1/risk difference).16
Propensity scores
Women who take ondansetron are inherently different from those who do not require medication. Propensity scores aim to mimic a randomized controlled trial by balancing these differences in the exposed and unexposed.17 Propensity scores were calculated for binary first trimester exposure to ondansetron. Scores were well balanced with similar distribution across exposure groups and only 10 deliveries were excluded from the region of common support (Figure S2 in Supplementary material, available as Supplementary data at IJE online).18,19
Propensity scores were incorporated into regression models through two approaches:20 (1) scores were matched using the nearest neighbour within a caliper of 0.00219; (2) scores were used to calculate standardized inverse probability weights.18 Success of matching and weighting was assessed by comparing standardized percent bias for potential confounders before and after matching or weighting.18,19 Matching 2 controls to every 1 case of VSD, based on propensity score, resulted in an analysed sample of 9056.
Clustered trajectory analyses
To evaluate potential dose–response relationships, we conducted clustered trajectory analyses using k-means clustering in R. The k-means model is an unsupervised learning approach that groups pregnancies with similar dosing trajectories into an optimal number of groups (k). A range of 2–6 groups was tested, each run at three different starting conditions. Groups were determined by minimizing the distance between a woman’s dose and the mean dose for the cluster on each gestational day. Calinski–Harabasz criteria were used to choose the optimal number. Analyses were limited to exposed pregnancies and infants born at ≥32 weeks because this approach requires equal exposure time for all pregnancies.21,22 Exposure was assessed only in the first trimester as this is expected to be the sensitive time-window of organogenesis.
We conducted these analyses first grouping pregnancies with similar daily dosing of ondansetron, then repeated using cumulative ondansetron doses throughout gestation. Dosing ranges and covariates were compared across groups (e.g. lowest-dose group vs highest-dose group). Using log-binomial regression, risk of VSD was modelled using the lowest-dose group as a referent. Regressions were fully adjusted, then adjusted only for the number of prenatal visits with an NVP-related diagnosis as severity of NVP likely increases concurrently with dose.
Sensitivity analyses
We re-ran analyses evaluating secondary cardiac malformations individually (Supplementary material, available as Supplementary data at IJE online). We also assessed the association between first trimester exposure to ondansetron and any birth defect using an adjusted log-binomial regression with robust standard error.
Primary analyses were performed in Stata Version 15.0 (College Station, TX) and R (K-Means for Longitudinal Data package version 2.4.1).
Results
Of the 33 677 singleton deliveries, 6038 (18%) deliveries were exposed to ondansetron at some point during pregnancy, including 3733 (11%) exposed during the first trimester (n = 3171 oral; n = 1074 intravenous; n = 18 injection). Median cumulative dose was 72 mg [interquartile range (IQR) 32.8, 144 mg] and ranged from 2.4 to 1008 mg after conversion based on bioavailability. Rates of each neonatal outcome were low with 0.4% of deliveries diagnosed with a VSD (n = 133), 0.6% with a VSD or ASD (n = 197) and 0.8% having any cardiac malformation (n = 279). A total of 5377 women had more than one delivery during the study period.
Women exposed to ondansetron in the first trimester were more likely than those unexposed to be African American, to have publicly paid insurance, a gastrointestinal disease, more encounters with an NVP-related diagnosis and to have used illicit drugs (Table 1). There were no meaningful differences in other characteristics. Infants with a VSD were more likely than those without a VSD to be born to mothers with structural heart disease, diabetes, collagen vascular disease or gastrointestinal disease (Table 2).
Table 1.
Characteristics of singleton deliveries at Magee–Womens Hospital from 2006 to 2014 with UPMC Health Plan Insurance coverage according to any ondansetron exposure in the first trimester (n = 33 677)
| Characteristic | Ondansetron use | No ondansetron |
|---|---|---|
| n (%) | n (%) | |
| n = 3733 | n = 29 944 | |
| Race | ||
| Caucasian | 2515 (67) | 21 340 (71) |
| African American | 961 (26) | 5890 (20) |
| Other | 144 (4) | 1556 (5) |
| Missing | 113 (3) | 1158 (4) |
| Mother’s age [mean (SD)] | 28 (6) | 29 (6) |
| Private insurance | 2153 (58) | 19 478 (65) |
| Patient of resident clinic | 450 (12) | 3602 (12) |
| Mother’s education | ||
| Less than high school | 264 (7) | 1969 (7) |
| High school graduate or GED completed | 731 (20) | 5816 (19) |
| Some college credit | 1591 (43) | 13 167 (44) |
| College graduate | 597 (16) | 5378 (18) |
| Missing | 550 (15) | 3614 (12) |
| Prepregnancy BMI (kg/m2) [mean (SD)] | 26 (6) | 25 (5) |
| Married | 1850 (50) | 16 217 (54) |
| Number of prenatal visits [mean (SD)] | 15 (9) | 13 (8) |
| Smoked during pregnancy | 462 (12) | 3910 (13) |
| Primiparous | 1550 (42) | 14 579 (49) |
| Maternal structural heart disease | 51 (1) | 214 (1) |
| Maternal arrhythmia | 29 (0·8) | 87 (0·3) |
| Maternal chronic hypertension | 83 (2) | 609 (2) |
| Maternal diabetes (any) | 226 (6) | 1753 (6) |
| Maternal malignancy | 2 (0·05) | 22 (0·07) |
| Maternal collagen vascular disease | 35 (0·9) | 148 (0·5) |
| Maternal lower or upper GI diagnosis | 272 (7) | 1203 (4) |
| Maternal illicit drug use | 215 (6) | 1001 (3) |
| History of miscarriage or termination | 1316 (35) | 9263 (31) |
| Ultrasound with nuchal fold measurement | 27 (1) | 171 (1) |
| Gestational age at delivery, days [mean (SD)] | 268 (17) | 270 (16) |
| Count of prenatal visits with NVP-related diagnosis | ||
| None | 2113 (57) | 26 560 (89) |
| 1 Encounter | 796 (21) | 2477 (8) |
| 2 Encounters | 377 (10) | 548 (2) |
| 3 Encounters | 176 (5) | 164 (1) |
| ≥4 Encounters | 271 (7) | 195 (1) |
UPMC, University of Pittsburgh Medical Center; NVP, nausea and vomiting of pregnancy; SD, standard deviation; GED, General Educational Development tests; BMI, body mass index; GI, gastrointestinal.
Table 2.
Characteristics of singleton deliveries at Magee-Womens Hospital from 2006 to 2014 with UPMC Health Plan Insurance coverage according to neonatal ventricular septal defect (n = 33 677)
| Characteristic | Ventricular septal defect | No ventricular septal defect |
|---|---|---|
| n (%) | n (%) | |
| n = 133 | n = 33 544 | |
| Race | ||
| Caucasian | 99 (74) | 23 756 (71) |
| African American | 23 (17) | 6828 (20) |
| Other | 8 (6) | 1692 (5) |
| Missing | 3 (2) | 1268 (4) |
| Mother’s age [mean (SD)] | 29 (6) | 29 (6) |
| Private insurance | 81 (61) | 21 550 (64) |
| Patient of resident clinic | 13 (10) | 4039 (12) |
| Mother’s education | ||
| Less than high school | 9 (7) | 2224 (7) |
| High school graduate or GED completed | 23 (17) | 6524 (19) |
| Some college credit | 66 (50) | 14 692 (44) |
| College graduate | 19 (14) | 5956 (18) |
| Missing | 16 (12) | 4148 (12) |
| Prepregnancy BMI (kg/m2) [mean (SD)] | 25 (5) | 25 (5) |
| Married | 67 (50) | 18 000 (54) |
| Number of prenatal visits [mean (SD)] | 14 (8) | 13 (8) |
| Smoked during pregnancy | 16 (12) | 4356 (13) |
| Primiparous | 62 (47) | 16 067 (48) |
| Maternal structural heart disease | 4 (3) | 261 (1) |
| Maternal arrhythmia | 0 (0) | 116 (0·4) |
| Maternal chronic hypertension | 4 (3) | 688 (2) |
| Maternal diabetes (any) | 16 (12) | 1963 (6) |
| Maternal malignancy | 0 (0) | 24 (0·07) |
| Maternal collagen vascular disease | 2 (2) | 181 (0·5) |
| Maternal lower or upper GI diagnosis | 10 (8) | 1465 (4) |
| Maternal drug use | 8 (6) | 1208 (4) |
| History of miscarriage or termination | 49 (37) | 10 530 (31) |
| Ultrasound with nuchal fold measurement | 3 (2) | 195 (1) |
| Gestational age at delivery, days [mean (SD)] | 262 (24) | 270 (16) |
| Count of prenatal visits with NVP-related diagnosis | ||
| None | 107 (80) | 28 566 (85) |
| 1 Encounter | 16 (12) | 3257 (10) |
| 2 Encounters | 4 (3) | 921 (3) |
| 3 Encounters | 5 (4) | 335 (1) |
| ≥4 Encounters | 1 (1) | 1 (1) |
UPMC, University of Pittsburgh Medical Center; NVP, nausea and vomiting of pregnancy; SD, standard deviation; GED, General Educational Development tests; BMI, body mass index; GI, gastrointestinal.
The adjusted RRs ranged from 1.7 (95% CI 1.0–2.9) to 2.1 (95% CI 1.1–4.0) across methods (Table 3). In the primary analyses, RRs correspond to ∼1 additional VSD for every 263–385 pregnancies exposed to ondansetron in the first trimester. Estimates of risk were consistent regardless of method for propensity score implementation. RRs were not influenced by varying which variables were included in adjustment.
Table 3.
Summary of all results assessing association between ondansetron exposure in the first trimester and neonatal ventricular septal defect
| Method | Population at risk | Events | Unadjusted risk per 1000 livebirths | Unadjusted odds ratio (95% CI) | Adjusteda risk ratio (95% CI) | Number needed to harm16,a |
|---|---|---|---|---|---|---|
| Regression | ||||||
| Log-binomial regression | 385 | |||||
| No ondansetron | 29 944 | 109 | 3.6 | Referent | Referent | |
| Ondansetron | 3733 | 24 | 6.4 | 1.8 (1.1–2.8) | 1.7 (1.0–2.9) | |
| Propensity scores | ||||||
| Propensity score-matched | 263 | |||||
| No exposure | 5779 | 22 | 3.8 | Referent | Referent | |
| Ondansetron | 3727 | 24 | 6.4 | 1.8 (1.0–3.2) | 2.1 (1.1–4.0) | |
| Propensity score-IPTW | 344 | |||||
| No exposure | 29 934 | 109 | 3.6 | Referent | Referent | |
| Ondansetron | 3733 | 24 | 6.4 | 1.6 (1.0–2.6) | 1.9 (1.1–3.1) |
Adjusted for count of encounters for nausea/vomiting/hyperemesis/gastrointestinal infection, total encounters, insurance type, maternal age, race, years of education, marital status, pre-gravid BMI, parity, history of termination/miscarriage/abortion, drug use, arrhythmia, chronic hypertension, collagen vascular disease, lower gastrointestinal disease, upper gastrointestinal disease, diabetes (any), maternal structural heart disease, ultrasound with nuchal fold measurement in pregnancy, smoking status and patient of Magee-resident clinic.
IPTW, inverse probability treatment weighting; CI, confidence interval.
There were 5096 pregnancies exposed to ondansetron before 32 weeks gestation, and therefore eligible for the exploratory clustered trajectory analyses. Individual pregnancies are graphed as black lines in Figure 1. The k-means model clustered pregnancies into 2 daily and 3 cumulative dosing trajectories over the first trimester. For cumulative dosing, Group A (68% of pregnancies) was comprised of pregnancies with lowest doses compared with Groups B and C (median total dose: 4 mg vs 144 mg vs 360 mg, respectively). Adjusted risk for VSD increased concurrently with cumulative ondansetron dose over pregnancy and was 3 times higher in the highest-dose (Group C) compared with the lowest-dose Group A (Table 4). Results from the daily dose analysis demonstrated similar relationships (Figure 1, Table 4), with a peak in dose from days 50 to 100 when assessing the entire pregnancy (data not shown).
Figure 1.
Ondansetron cumulative and daily dosing trajectory groups in the first trimester of pregnancy. Solid black lines represent the dose on each gestational day for individual pregnancies. The lines labelled by letter are the mean cumulative dose at each gestational day for each group.
Table 4.
Log-binomial regressions of risk of ventricular septal defect by daily and cumulative dosing trajectory group over first trimester of pregnancy among women exposed to ondansetron
| Dosing approach over entire pregnancy | Cumulative dose (mg) median (IQR) | Population at risk | Events (n) | Unadjusted risk per 1000 live born infants | Adjusteda odds ratio (95% CI) | Adjustedb odds ratio (95% CI) |
|---|---|---|---|---|---|---|
| Daily Dose | ||||||
| Cluster A | 12 (0, 72) | 4193 | 21 | 5.0 | Reference | Reference |
| Cluster B | 240 (176, 318) | 903 | 6 | 6.6 | 1.4 (0.56, 3.5) | 1.3 (0.53, 3.2) |
| Cumulative dose | ||||||
| Cluster A | 4 (0, 48) | 3452 | 14 | 4.1 | Reference | Reference |
| Cluster B | 144 (97, 192) | 1328 | 9 | 6.8 | 1.8 (0.73, 4.4) | 1.6 (0.67, 3.7) |
| Cluster C | 360 (288, 440) | 316 | 4 | 13 | 3.9 (1.2, 13) | 3.2 (1.0, 9.9) |
Adjusted for count of encounters for nausea/vomiting/hyperemesis/gastrointestinal infection, total encounters, insurance type, maternal age, race, years of education, marital status, pre-gravid BMI, parity, history of abortion, drug use, arrhythmia, chronic hypertension, collagen vascular disease, lower gastrointestinal disease, upper gastrointestinal disease, diabetes (any), maternal structural heart disease, ultrasound with nuchal fold measurement in pregnancy, smoking status and patient of Magee-resident clinic.
Adjusted for count of encounters for nausea/vomiting/hyperemesis/gastrointestinal infection.
IQR, interquartile range; CI, confidence interval.
Associations between first trimester ondansetron exposure and cardiac malformations persisted but were weakened as the outcome was expanded to include additional cardiac defects (Table S2 in Supplementary material, available as Supplementary data at IJE online). Sensitivity analyses demonstrated further dilution of effect with the broadening of outcome to any birth defect (Table S3 in Supplementary material, available as Supplementary data at IJE online). There was no association between ondansetron and any birth defect.
Discussion
We found that ondansetron was associated with an increased risk of VSD on the order of one additional VSD for approximately every 330 exposed neonates exposed in the first trimester. This relationship was dose-dependent, with doses peaking in the first trimester. Nevertheless, ondansetron exposure was not related to risk of birth defects when all major malformations were combined.
Our findings are consistent with the studies that combined all congenital malformations and found no association.6–9,23 Three previous studies8,9,11 analysed the association of first trimester ondansetron and cardiac septal defects separately, producing conflicting results. Using a cohort of Swedish deliveries from 1998 to 2012 (n = 1 501 434) and a combination of self-report and prescription registry data, Danielsson et al.1,8,11 demonstrated an increase in the risk of septal defects for 1349 pregnancies exposed to ondansetron compared with the unexposed (RR 2.1, 95% CI 1.2–3.3). Though this conclusion was derived from only 17 (1.3%) septal defects in the exposed (0.7% in unexposed), these results are similar in magnitude to our study. In contrast to our and Danielsson’s findings, Parker et al. found no association between VSD and ondansetron using the National Birth Defect Prevention Study (1997–2011) and Slone Birth Defect Study (1997–2014).11 Unlike our study, these results are subject to recall bias as use was self–reported up to 2 years after delivery. In a recent impressive cohort of pregnant, Medicaid recipients in the USA from 2000 to 2013 (n = 1 816 414), authors demonstrated no association between first trimester ondansetron and cardiac malformations.9 The authors found a slight increased risk of cardiac malformation driven by VSD [RR 1.14, (95% CI 1.04–1·27)] and ASD (1.37, 95% CI 1.19–1.57) that was eliminated after adjustment using propensity score stratification. Notably, Medicaid data provided exposures to other medications which may play a confounding role but would not explain a dose–response relationship with ondansetron. Exposure was limited to outpatient prescriptions in this study. This approach fails to capture inpatient use, which comprises 32% of our exposures. Moreover, of the 1074 (29%) who received ondansetron intravenously, 58% (n = 622) would be misclassified as unexposed as they never filled a prescription in the first trimester. This misclassification of exposure in women with severe nausea may dilute the association and contribute to null findings.
Confounding by indication of NVP is frequently encountered in the literature. For this bias to be present, NVP must be associated with both exposure to ondansetron, which it undoubtedly is, and with VSD. However, NVP has never been associated with congenital anomalies.12 In fact, the 2018 ACOG guidelines specifically state that even hyperemesis gravidarum has never been shown to increase risk.3,12 Although confounding by NVP severity is a critical consideration when evaluating perinatal outcomes resulting from NVP-related malnutrition, such as fetal growth restriction, it does not appear to be a major concern when evaluating birth defects.
The slight increased risk of VSD with first trimester ondansetron use may be explained by its effect on the embryonic heart during organogenesis given that ondansetron is known to cross the placenta.24 Danielsson et al. showed a concentration-dependent association between ondansetron and embryonic heart rhythms and ultimately cardiovascular malformations in pregnant rats.25 Our study extends this literature by demonstrating a similar dose–response relationship in an epidemiologic study.
Our findings must be interpreted with an understanding of its limitations. A limitation we share with most pharmacoepidemiologic studies is reliance on claims and prescription data, which may be subject to misclassification bias. We had no measure of adherence and thus assumed that oral medication was taken as prescribed. This resulted in unusually high doses, despite conversion to bioequivalence, as prescriptions could be written for frequent daily use (e.g. ‘every 4 hours’) for up to 90 days. For this reason, we caution readers not to extrapolate a safety dose threshold from the dosing trajectories. We did however, have access to both inpatient and outpatient claims, providing us a full representation of a woman’s exposure. Despite this broad approach to capturing exposure, findings of increased risk are based on only 24 VSDs occurring in the 3733 infants exposed in the first trimester. We also lacked information on other medication use in pregnancy. This results in a potential unmeasured confounding drug and eliminates the possibility for evaluation against an active comparator. Finally, though protocols at our institution require an echocardiogram to confirm VSDs before documentation into the chart (excluding middle muscular VSD which can often be diagnosed based on unique characteristics alone), we were unable to validate the outcome of VSD in this data set.
It is crucial to emphasize that VSD is often a small defect that closes spontaneously26 and has limited haemodynamic consequence. Moreover, ondansetron is an effective treatment for severe NVP,27 and the lack of a quantitative measure of NVP’s impact prevents us from calculating the risk–benefit ratio. Such a measure would include quality of life indicators (e.g. missed work) and perinatal outcomes related to malnutrition (e.g. preterm birth). Each of these factors must be considered in clinical decision making.
Our findings demonstrate a slight increase in the absolute risk of VSD with first trimester exposure to ondansetron. Increased risk was dose-dependent in the first trimester of pregnancy. More research is needed to define the critical dose at which risk increases substantially. Clinically, NVP treatment decisions must consider the increased VSD risk associated with ondansetron while bearing in mind the dangers associated with withholding this effective medication.
Funding
This work was support by the Obstetric and Fetal Pharmacology Research Centers from the National Institute of Child Health and Human Development [grant HD047905].
Conflict of interest: None declared.
Supplementary Material
References
- 1. Matthews A, Haas DM, O’Mathuna DP, Dowswell T, Doyle M.. Interventions for nausea and vomiting in early pregnancy. Cochrane Database Syst Rev 2014;3:CD007575. [DOI] [PubMed] [Google Scholar]
- 2. Taylor LG, Bird ST, Sahin L . Antiemetic use among pregnant women in the United States: the escalating use of ondansetron. Pharmacoepidemiol Drug Saf 2017;26:592–96. [DOI] [PubMed] [Google Scholar]
- 3. Veenendaal MV, van Abeelen AF, Painter RC, van der Post JA, Roseboom TJ.. Consequences of hyperemesis gravidarum for offspring: a systematic review and meta-analysis. BJOG 2011;118:1302–13. [DOI] [PubMed] [Google Scholar]
- 4. Griddine A, Bush JS.. Ondansetron In: Abai B, Abu-Ghosh A, Acharya AB et al (eds). StatPearls .Treasure Island, FL: StatPearls Publishing, 2018. [Google Scholar]
- 5. Einarson A, Maltepe C, Navioz Y, Kennedy D, Tan MP, Koren G.. The safety of ondansetron for nausea and vomiting of pregnancy: a prospective comparative study. BJOG 2004;111:940–43. [DOI] [PubMed] [Google Scholar]
- 6. Colvin L, Gill AW, Slack-Smith L, Stanley FJ, Bower C.. Off-label use of ondansetron in pregnancy in Western Australia. Biomed Res Int 2013;2013:1. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7. Pasternak B, Svanstrom H, Hviid A.. Ondansetron in pregnancy and risk of adverse fetal outcomes. N Engl J Med 2013;368:814–23. [DOI] [PubMed] [Google Scholar]
- 8. Danielsson B, Wikner BN, Kallen B.. Use of ondansetron during pregnancy and congenital malformations in the infant. Reprod Toxicol 2014;50:134–37. [DOI] [PubMed] [Google Scholar]
- 9. Huybrechts KF, Hernández-Díaz S, Straub L . Association of maternal first-trimester ondansetron use with cardiac malformation and oral clefts in offspring. JAMA 2018;320:2429–37. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10. Mitchell AA. Studies of drug-induced birth defects In: Strom BL. (ed). Pharmacoepidemiology. 5th edn.Chichester, UK: John Wiley and Sons, 2012, pp. 595–608. [Google Scholar]
- 11. Parker SE, Van Bennekom C, Anderka M, Mitchell AA.. Ondansetron for treatment of nausea and vomiting of pregnancy and the risk of specific birth defects. Obstet Gynecol 2018;132:385–94. [DOI] [PubMed] [Google Scholar]
- 12.Nausea and vomiting of pregnancy. Practice Bulletin No. 189.American College of Obstetricians and Gynecologists. Obstet Gynecol 2018;131:e15–30. [DOI] [PubMed] [Google Scholar]
- 13. Zofran [Package Insert]. Research Triangle Park, NC: Glaxo-SmithKline, 2014. [Google Scholar]
- 14.Single World Intervention Graphs (SWIGs): A Unification of the Counterfactual and Graphical Approaches to Causality. Working Paper Number 128. Center for Statistics; the Social Sciences, University of Washington. http://www.csss.washington.edu/Papers/wp128.pdf (26 August 2018, date last accessed).
- 15. Shrier I, Platt RW.. Reducing bias through directed acyclic graphs. BMC Med Res Methodol 2008;8:70. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16. Citrome L. Compelling or irrelevant? Using the number needed to treat can help decide. Acta Psychiatr Scand 2008;117:412–29. [DOI] [PubMed] [Google Scholar]
- 17. Rubin DB. On principles for modeling propensity scores in medical research. Pharmacoepidem Drug Safe 2004;13:855–57. [DOI] [PubMed] [Google Scholar]
- 18. Austin PC, Stuart EA.. Moving towards best practice when using inverse probability of treatment weighting (IPTW) using the propensity score to estimate the causal treatment effects in observational studies. Stat Med 2015;34:3661–79. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19. Austin PC. Balance diagnostics for comparing the distribution of baseline covariates between treatment groups in propensity-score matched samples. Stat Med 2009;28:3083–107. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20. Austin PC, Mamdani MM.. A comparison of propensity score methods: a case-study estimating the effectiveness of post-AMI statin use. Stat Med 2006;25:2084–106. [DOI] [PubMed] [Google Scholar]
- 21. Genolini C, Falissard B.. Kml: a package to cluster longitudinal data. Comput Methods Prog Biomed 2011;104:e112–e121. [DOI] [PubMed] [Google Scholar]
- 22. Palmsten K, Rolland M, Hebert MF . Patterns of prednisone use during pregnancy in women with rheumatoid arthritis: daily and cumulative dose. Pharmacoepidemiol Drug Saf 2018;27:430–8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23. Fejzo MS, MacGibbon KW, Mullin PM.. Ondansetron in pregnancy and risk of adverse fetal outcomes in the United States. Reprod Toxicol 2016;62:87–91. [DOI] [PubMed] [Google Scholar]
- 24. Siu SN, Chan MT, Lau T.. Placental transfer of ondansetron during early human pregnancy. Clin Pharmacokinet 2006;45:419–23. [DOI] [PubMed] [Google Scholar]
- 25. Danielsson B, Webster WS, Ritchie HE.. Ondansetron and the teratogenicity in rats: evidence for a mechanism mediated via embryonic hERG blockade. Reprod Toxicol 2018;81:237–45. [DOI] [PubMed] [Google Scholar]
- 26. Dees E, Baldwin HS.. Developmental biology of the heart In: Gleason A, Juul SE (eds). Avery’s Diseases of the Newborn. 10th edn.Philadepphia, PA: Elsevier, 2018, pp. 724–740.e3. [Google Scholar]
- 27. O’Donnell A, McParlin C, Robson SC . Treatments for hyperemesis gravidarum and nausea and vomiting in pregnancy: a systematic review and economic assessment. Health Technol Assess 2016;20:1–268. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.