{"paper_id":"469b28d0-beef-4a41-93a9-df7d93a14442","body_text":"Fever, pain, and inflammation during pregnancy are not uncommon and complex conditions to manage, where if managed poorly, these may result in maternal complications [ 1 ]. Accordingly, nonsteroidal anti-inflammatory drugs (NSAIDs) are a possible treatment option for pregnant women to treat and/or attenuate these conditions, and indications for chronic use during pregnancy are inflammatory bowel or chronic rheumatic diseases [ 2 , 3 ]. However, NSAIDs may increase the risk of embryo-fetal adverse effects, dependent on the type of NSAIDs, period of gestation, and duration of consumption, due to its mechanism of action (e.g., inhibition of prostanoid activity) [ 3 – 6 ]. Despite these concerns, exposure to NSAIDs during early pregnancy is not infrequent by either being unaware of the pregnant state or simply due to the extremely limited treatment option available for pregnant women, albeit the inconclusive data on its safety [ 7 , 8 ].\nUnlike the relatively established risks of premature closure of the ductus arteriosus and oligohydramnios with NSAID use during late pregnancy [ 9 – 14 ], existing data on early prenatal NSAID use and the risk of adverse neonatal outcomes were largely based on self-reported use, outdated, had a small sample size, or did not control for confounding by indication ( S1 \n Table ) [ 10 , 15 – 19 ]. Thus, evidence on the use of NSAIDs during early pregnancy remains unclear, with particular knowledge gaps present on cyclooxygenase (COX)-2 inhibitors [ 3 , 20 ]. The association between prenatal NSAID use and obstetric complications is even more scarce. Hence, further studies with sufficient sample size and advanced epidemiologic methods are needed to investigate into the currently indeterminate association of adverse birth outcomes, including major malformations, and obstetric complications with early prenatal exposure to NSAIDs.\nBeing aware of the underlying risk associated with prenatal NSAID use is highly important given the substantial number of women exposed to these medications and the very limited option of analgesics present for use during pregnancy [ 17 , 21 , 22 ]. There is also a strong need for a well-designed, real-world evidence to assist pregnant women or women of childbearing age and their healthcare providers to weigh the risks of using NSAIDs against its benefits during this physiologically complex period [ 23 ]. This study was therefore aimed to examine the association between early prenatal exposure to NSAIDs and the risk of adverse neonatal and maternal outcomes by using a nationwide mother–child linked cohort in South Korea.\n\nUsing South Korea’s National Health Insurance Service-National Health Insurance Database (NHIS-NHID) [ 24 ], we identified all pregnancies in women aged 18 to 44 years with live births between April 1, 2010, and December 31, 2018. The mother–offspring cohort was constructed and provided solely for research purposes by the NHIS, which used a self-developed linkage algorithm based on the unique health insurance card number shared within a family and the date of delivery; this algorithm has been internally validated by the NHIS. In brief, the NHIS-NHID includes comprehensive information on sociodemographic variables, in- and outpatient healthcare utilization including diagnoses, procedures, and prescriptions, health examination records (for both mothers and their offspring), and vital statistics linked from Statistics Korea.\nThe study protocol was approved by the Institutional Review Board of Sungkyunkwan University (2021-04-024), and all analyses were conducted according to this protocol ( S1 Protocol ); informed patient consent was not required as our study used deidentified health insurance claims data.\nAs the etiologically relevant exposure window, the outcome assessment window, and the covariate assessment window differ by study outcome, we created two separate study cohorts. First, cohort 1 was constructed for the analysis of congenital malformations by setting the window of interest between the last menstrual period (LMP) to LMP+90 (hereafter, first trimester), and we excluded the following: (1) pregnancies with exposure to known teratogenic drugs (e.g., antineoplastic agent, warfarin, lithium, systemic retinoids, misoprostol, thalidomide, androgens, antiepileptic medications [valproate, topiramate, carbamazepine, oxcarbazepine, phenobarbital, phenytoin]); (2) infants with chromosomal abnormalities, genetic syndromes, and malformation syndromes with known causes; (3) pregnancies with no NSAID prescription during the first trimester, but with ≥1 NSAID prescription within 3 months before the LMP (LMP-90 to LMP-1); and (4) pregnancies with only 1 NSAID prescription during the first trimester to minimize potential for misclassification of exposure. Second, cohort 2 was constructed for the analysis of nonmalformation outcomes of low birth weight, antepartum hemorrhage, and oligohydramnios, by setting a broader window to evaluate the effects of exposure prior to the outcome assessment period (LMP to 19 week of gestation [hereafter, early pregnancy]), where we excluded (1) pregnancies with no NSAID prescription during early pregnancy, but with ≥1 NSAID prescription within 3 months before the LMP, and (2) pregnancies with only 1 NSAID prescription during early pregnancy.\nWe considered pregnant women prescribed ≥2 NSAIDs during the etiologically relevant window to be defined as those exposed to NSAIDs; if a woman refilled her prescription for NSAIDs, we assumed that she probably took them. The etiologically relevant window for major congenital malformations (cohort 1) was defined as the first trimester and for low birth weight, antepartum hemorrhage, and oligohydramnios (cohort 2) as early pregnancy. Pregnancies were considered unexposed if not prescribed any NSAIDs from 90 days before the LMP throughout the end of the first trimester or end of early pregnancy for cohorts 1 and 2, respectively. The LMP date was estimated by using a previously validated algorithm to estimate gestational age in administrative healthcare databases [ 25 ].\nMajor congenital malformations were defined as infants in the first year of life with major congenital malformations according to the classification system of the European Surveillance of Congenital Anomalies (EUROCAT) subgroups of major congenital anomalies [ 26 , 27 ]. The presence of congenital malformations was defined using previously validated algorithms based on inpatient or outpatient diagnostic and procedural codes, which have shown to identify the outcomes with high specificity (i.e., ≥1 date with the respective diagnostic code or 1 diagnostic code and a code for infant death between delivery and 1 year after delivery) [ 28 , 29 ]. We excluded major congenital malformations with known causes and minor defects according to the EUROCAT exclusion list. Major congenital malformations were also categorized into 12 organ-specific malformations: (1) nervous system; (2) eye; (3) ear, face, and neck; (4) heart defects; (5) respiratory system; (6) oral clefts; (7) digestive system; (8) abdominal wall defects; (9) urinary system; (10) genital organs; (11) limb; and (12) other malformations. Nonmalformation outcomes of interest were low birth weight, antepartum hemorrhage, and oligohydramnios. We defined low birth weight using neonatal inpatient or outpatient diagnosis claims recorded within 30 days of delivery (International Classification of Disease, 10th Revision: P07) and used maternal inpatient or outpatient diagnosis claims recorded from week 20 of gestation to delivery to identify antepartum hemorrhage (O44.1, O46) and oligohydramnios (O41.0). Detailed definitions for each outcome available in  S2 Table .\nWe assessed a broad range of potential confounders: maternal sociodemographic characteristics (e.g., maternal age, income level, parity, multiple gestations), comorbidities (e.g., gastrointestinal diseases, fever), co-medication use (e.g., antiepileptics, opioid analgesics), and general markers of burden of illness (e.g., obstetric comorbidity index; [ 30 , 31 ]). Maternal characteristics were measured at delivery, and comorbidities and co-medication use were measured from 6 months before pregnancy to end of the first trimester for cohort 1 and to the end of early pregnancy for cohort 2. Measures of healthcare utilization were assessed during the 6 months before but not during pregnancy (LMP-180 to LMP-1), such that these measures were unaffected by the early detection of pregnancy complications.\nWe described baseline characteristics for each cohort, comparing NSAID-exposed versus unexposed pregnancies, and assessed between-group covariate balance using standardized mean differences (SMDs), considering its absolute value ≥0.1 as important imbalances [ 32 ]. For all outcomes, we estimated its absolute risk per 1,000 births or pregnancies. To adjust for underlying differences between groups, we derived propensity scores (PS) from the predicted probability of being exposed to NSAIDs using a multivariable logistic regression model that included all covariates mentioned above as independent variables. We excluded observations in nonoverlapping regions of the PS distributions and identified 50 equally sized PS strata based on the distribution of NSAID-exposed women (PS fine stratification weight) [ 33 – 35 ]; weights for the NSAID-exposed group are set to 1 and the referent group are reweighted based on the number of treated patients residing within their stratum ([number of exposed in PS stratum i / number of total exposed] / [number of referent in PS stratum i / number of total referent]). We then estimated PS-adjusted relative risks (RRs) with 95% confidence intervals (CIs), or the average treatment effect among the treated, by using generalized linear models (PROC GENMOD with a weight statement, log-link function, and binomial distribution [log-binomial model]); adjustment was achieved through weighting. The primary analysis aimed to address the safety of NSAIDs versus no drug treatment by comparing prenatal NSAID exposure to unexposed women in the etiologically relevant exposure window; this would assess the safety of exposure to NSAIDs. We used SAS Enterprise version 6.1 (SAS Institute, Cary, NC, USA) for all statistical analyses, and this study is reported as per the Strengthening the Reporting of Observational Studies in Epidemiology guideline ( S1 STROBE Checklist ).\nWe conducted four secondary analyses, where PS were reestimated for each analysis. First, we compared early prenatal exposure to NSAIDs against (1) early prenatal acetaminophen use, defined as women prescribed ≥2 acetaminophen during the etiologically relevant window (to address the comparative safety of NSAIDs versus a commonly used medication with similar indications to minimize potential confounding by indication); (2) past users of NSAIDs, defined as women exposed to NSAIDs before but not during pregnancy (to address the safety of continuing NSAIDs during pregnancy). Second, given that aspirin was not included as NSAIDs in this study due to discrepancies in medical practice and indications of these drugs (nonaspirin NSAIDs versus aspirin), we also compared early prenatal exposure to aspirin against unexposed pregnancies; an analogous exposure definition to that of the main analysis for NSAIDs was used, but further excluding pregnancies exposed to ≥1 NSAIDs during early pregnancy across both aspirin-exposed and unexposed groups. Third, we repeated the primary analysis but redefined the exposure of interest by the type (nonselective NSAIDs, COX-2 selective inhibitors) and three most frequently prescribed individual NSAIDs. Last, we conducted subgroup analyses for cumulative duration of NSAID use (<5, 5 to 10, >10 days) to address duration–response relations.\nNine distinct sensitivity analyses were done to test the robustness of our main findings, where the overall findings were interpreted in light of these results [ 36 ]. First, we restricted the cohort to nulliparous pregnancies to account for correlations among women with multiple pregnancies. Second, we restricted to singleton pregnancies to eliminate the potential confounding effect of multiple gestations. Third, we restricted to pregnant women with data on health examination records to assess residual confounding effects from body mass index and smoking status. Fourth, we restricted to pregnancies with indications for NSAIDs (e.g., inflammatory disease, respiratory infection, fever) to further address confounding by indication. Moreover, we also restricted to pregnancies with severe respiratory infections (i.e., diagnosis of respiratory infections with antibiotic prescriptions or hospital admissions). Alternatively, we also adjusted for the number of (a) outpatient visits, (b) inpatient admissions, and (c) emergency department visits with respiratory infection diagnosis during early pregnancy in the outcome model as continuous variables as proxies of respiratory infection severity. Fifth, for cohort 1 only, we redefined the exposure window as the fourth to 10th week of the gestational period, a previously reported duration of organogenesis. Sixth, we conducted a negative control analysis for all outcomes by redefining the exposure window as the 5 to 8 months before the LMP to examine the possibility of residual confounding in our data. Seventh, we excluded pregnancies exposed to ≥1 NSAIDs any time during pregnancy after the first trimester (cohort 1) or early pregnancy (cohort 2); accordingly, this new definition would allow the specific assessment of exposure to NSAIDs “only” during early pregnancy. Eighth, we performed sibling-matched analyses by using conditional logistic regression models that stratified on the mother’s unique identification number to estimate odds ratios (ORs) for major congenital malformations and low birth weight to address potential confounding from within-family shared factors; only sibling pairs with discordant exposure status contributed to the estimates [ 37 ]. Last, we evaluated the potential effect of excluding pregnancies that were terminated through quantitative bias analyses as we included only pregnancies ending in live births ( S1 Appendix ).\n\nOf 3,129,715 eligible pregnancies between April 2010 and December 2018, we identified 1,898,397 and 1,895,848 pregnancies for the analyses of congenital malformations (cohort 1) and nonmalformations (cohort 2), respectively ( Fig 1 ).\nLMP, last menstrual period; NHIS, National Health Insurance Service; NSAID, nonsteroidal anti-inflammatory drug. *First trimester is defined as the period between the LMP to LMP+90 (cohort 1).  † Early pregnancy is defined as the period between LMP to 19 week of gestation (cohort 2).\nThere were 112,119 (5.9%) pregnancies prenatally exposed to NSAIDs in the first trimester (cohort 1) and 124,957 (6.6%) pregnancies prenatally exposed to NSAIDs during early pregnancy (cohort 2). Important imbalances (absolute SMDs ≥0.1) in health insurance type, nulliparity, several comorbid conditions (e.g., asthma, respiratory infection) or co-medication use (e.g., anxiolytics, antiacids), and all general markers of burden of disease were observed between NSAIDs-exposed and unexposed pregnancies before applying PS fine stratification weights. However, after weighting, balance was achieved for all covariates between groups, with absolute SMDs <0.1 ( Table 1 ); baseline characteristics for NSAID-exposed versus acetaminophen-exposed pregnancies (active comparator comparison) before and after PS fine stratification weights are shown in  S3 Table .\nCOPD, chronic obstructive pulmonary disease; DMARD, disease-modifying antirheumatic drugs, NSAID, nonsteroidal anti-inflammatory drug; OCI, obstetric comorbidity index; PS, propensity score; SD, standard deviation; SMD, standardized mean difference.\n*To account for PS, the unexposed observations were weighted using the distribution of the exposed observations among 50 PS strata. Observations from the nonoverlapping regions of the PS distributions were trimmed.\n† Value >0.1 indicates significant imbalance between the exposed and unexposed groups.\n‡ Counts less than five are suppressed due to internal regulations of the National Health Insurance Service, South Korea.\nComparing pregnancies exposed to NSAIDs to those unexposed (primary analysis), the PS-adjusted RRs were 1.14 (95% CI, 1.10 to 1.18) for overall malformations, 1.29 (1.25 to 1.33) for low birth weight, 1.05 (0.99 to 1.12) for antepartum hemorrhage, and 1.09 (1.01 to 1.19) for oligohydramnios. In the results of secondary analyses that compared with acetaminophen-exposed pregnancies or past users of NSAIDs, generally similar trends of associations were observed for the risks of study outcomes ( Fig 2 ); results of aspirin-exposed versus unexposed pregnancies were also similar to the primary analysis ( S4 Table ). The risk of heart defects remained significantly higher across all comparisons of NSAID exposure during the first trimester with unexposed pregnancies (PS-adjusted RR, 1.19 [95% CI, 1.13 to 1.24]), acetaminophen-exposed pregnancies (1.17 [1.10 to 1.25]), and past users (1.19 [1.13 to 1.25]) ( S5 ,  S6 , and  S7  Tables ).\nCI, confidential interval; NSAID, nonsteroidal anti-inflammatory drug; PS, propensity score; RD, risk difference; RR, relative risk. *RD 1,000  = risk difference per 1,000 births or pregnancies.  † Units: births for outcomes of overall congenital malformations and low birth weights; pregnancies for outcomes of antepartum hemorrhage and oligohydramnios.  ‡ Associated risk of major congenital malformations and other adverse outcomes in women who used NSAIDs in early pregnancy was compared with three distinct pregnancy groups: women not prescribed any NSAIDs (unexposed), women prescribed ≥2 acetaminophen (acetaminophen), and women who stopped taking NSAIDs during pregnancy but were prescribed them before (past users).\nResults of secondary analyses that assessed the type of or frequently used individual NSAIDs were also largely consistent with the primary analysis; results for COX-2 selective inhibitors had wide CIs likely due to small number of events and pregnancies. Of note, the risk of oligohydramnios was more than 2-fold higher with COX-2 selective inhibitors versus unexposed pregnancies (PS-adjusted RR, 2.64 [95% CI, 1.47 to 4.75]) ( Fig 3 ). Meanwhile, a clear duration–response relation was observed only for low birth weight as the PS-adjusted RRs increased proportionally with cumulative duration of NSAID use in early pregnancy (<5 days: PS-adjusted RR, 1.22 [95% CI, 1.17 to 1.28]; 5 to 10 days: 1.28 [1.23 to 1.38]; >10 days: 1.48 [1.39 to 1.58]); the risks of overall malformations (1.23 [1.14 to 1.33]) and low birth weight (1.48 [1.39 to 1.58]) were highest when using NSAIDs for >10 days during early pregnancy ( Table 2 ).\nCI, confidential interval; COX, cyclooxygenase; NSAID, nonsteroidal anti-inflammatory drug; PS, propensity score; RR, relative risk. *Units: births for outcomes of overall congenital malformations and low birth weights; pregnancies for outcomes of antepartum hemorrhage and oligohydramnios.\nCI, confidential interval; NSAID, nonsteroidal anti-inflammatory drug; PS, propensity score; RR, relative risk.\n*Units: births for outcomes of overall congenital malformations and low birth weights; pregnancies for outcomes of antepartum hemorrhage and oligohydramnios.\n† Mean (standard deviation) and median (Q1, Q3) cumulative duration for each outcome were: 8.6 (6.5) and 7.0 (5.0, 10.0) for overall congenital malformations; 8.8 (8.1) and 7.0 (5.0, 10.0) for low birth weight, antepartum hemorrhage, and oligohydramnios.\nOur main findings were generally consistent across all sensitivity analyses, with regard to the point estimates ( Fig 4  and  S8 ,  S9 ,  S10  and  S11  Tables ). Risks of overall malformations (OR, 1.13 [95% CI, 1.00 to 1.29]) and low birth weight (1.30 [1.14 to 1.49]) associated with early prenatal exposure to NSAIDs versus unexposed pregnancies remained statistically significant in sibling-matched analyses that further accounted for familial factors ( S12 Table ). Results of the quantitative bias analysis found that the corrected RR for overall malformations and congenital heart defects remained <1.41 under the most extreme scenario ( S1 Appendix  and  S2  and  S3  Figs ).\nCI, confidential interval; NSAID, nonsteroidal anti-inflammatory drug; PS, propensity score; RR, relative risk.  † Restrict the study cohort to those who underwent health screening examination for whom information on BMI and smoking was available. *Units: births for outcomes of overall congenital malformations and low birth weights; pregnancies for outcomes of antepartum hemorrhage and oligohydramnios.\n\nThis large-scale, nationwide cohort study of 1.8 million pregnancies found that, as compared with unexposed pregnancies, early prenatal exposure to NSAIDs was significantly associated with higher risks for most outcomes examined, especially for major congenital malformations, likely driven by heart defects, low birth weight, and oligohydramnios. In particular, the associations for overall malformations, low birth weight, and oligohydramnios remained significant despite comparing NSAIDs against acetaminophen or past users. Moreover, associations of adverse neonatal and maternal outcomes may be stronger with COX-2 selective inhibitors or use of NSAIDs for >10 days. Generally consistent findings, with regard to point estimates, observed across a range of sensitivity analyses provide some reassurance; however, residual confounding by indication or from unmeasured confounders may still be present due to inherent limitations of the observational nature of this study using routine care data.\nThe observed positive associations of overall malformations and low birth weight with early prenatal NSAID use were inconsistent with existing literature ( S1 Table ), which generally reported null associations [ 10 , 16 – 18 , 20 , 38 – 44 ]. However, previous studies had important limitations, making it difficult to make any meaningful comparisons. For example, most of previously reported effect estimates had very wide CIs, likely due to small sample size, resulting in uncertainty in the study’s ability to examine the research question with accuracy and precision. Moreover, former studies were unable to fully account for various sources of confounding bias (e.g., confounding by indication) by comparing against only unexposed pregnancies. Furthermore, evidence of this topic that addressed the effects of NSAIDs on Asians was absent to our knowledge. Meanwhile, this study accounted for these limitations by not only accounting for >50 covariates but also comparing NSAIDs to unexposed as well as acetaminophen and past users among 1.8 million pregnancies in South Korea. Hence, we believe that our findings provide evidence from a real-world setting for healthcare providers on the safety of NSAIDs during early pregnancy in an Asian pregnant population, which could help fill current knowledge gaps on this space of both clinical and public health importance.\nUnlike congenital malformations and low birth weight, there were little to no previous studies for comparison on antepartum hemorrhage or oligohydramnios with nonselective NSAIDs or COX-2 inhibitor use in early pregnancy. With future studies, preferably randomized clinical trials, warranted to corroborate our findings, the observed risk of antepartum hemorrhage and oligohydramnios associated with early prenatal NSAID use versus acetaminophen-exposed pregnancies suggest cautious use of NSAIDs in this period, as well as the possibility that acetaminophen could be the safer option. While few mechanisms of action of acetaminophen different to that of NSAIDs, such as indirect inhibition of COX, could possibly explain this observation [ 45 – 49 ], a more in-depth investigation is needed. Moreover, given that oligohydramnios can be an end point of multiple pathways, including fetal kidney impairment, intrauterine growth restriction, and rupture of the fetal membranes, and the observational nature of this study being able to only examine an association and not causation, cautious interpretation is warranted. Thus, based on these findings and the fact that acetaminophen has relatively minimal anti-inflammatory effects versus NSAIDs [ 49 ], the decision to prescribe or take NSAIDs over acetaminophen during pregnancy should be carefully considered only for the treatment of inflammatory conditions.\nGiven the large size of this study, we assessed with higher precision the association of neonatal and maternal adverse outcomes following prenatal exposure to COX-2 inhibitors than former studies [ 20 , 42 ]. While one study found a similar estimate for major congenital malformations with COX-2 inhibitor use in the first trimester (RR, 1.40 [95% CI, 0.70 to 2.78]) [ 42 ] to our study (PS-adjusted RR, 1.42 [95% CI, 1.02 to 1.97]), we found statistically significant associations by having greater power; another prior study also had very wide CIs (0.96 [0.28 to 3.26]) [ 20 ]. As studies on this topic were also lacking for comparisons, further investigations using multiple data sources to overcome sample size issues are likely needed to accurately examine the safety of COX-2 inhibitors in early pregnancy.\nTo our knowledge, this is the largest observational study that examined the association between early prenatal NSAID use and the risk of neonatal and maternal adverse outcomes by using nationwide data of more than 3 million mother–offspring pairs augmented with advanced epidemiological and statistical methods. Notably, this is the first study, to our knowledge, that assessed the risk of several outcomes, including antepartum hemorrhage and oligohydramnios, with NSAID use in an Asian population. The large sample allowed for analyses of rare outcomes, COX-2 inhibitors, and duration–response relations with relatively high precision. Moreover, exposure to NSAIDs is likely to be unaffected by recall bias given the use of routinely collected electronic prescribing records to define exposure status. Furthermore, largely consistent findings with regard to point estimates across all sensitivity analyses support the robustness of our primary findings. Finally, the deterministic mother–offspring link based on algorithms developed, validated, and provided for research purposes by the NHIS further improved the study’s accuracy and reliability.\nThis study also has a few limitations, which should be carefully taken into account when interpreting our findings. First, exposure misclassification is possible as NSAIDs may be purchased over-the-counter, which are unavailable for assessment in the claims data used. Hence, differential misclassification in unexposed pregnancies cannot be ruled out as women obtaining their NSAIDs over-the-counter are unlikely to have been prescribed NSAIDs as well; the true proportion of NSAID-exposed pregnancies could be larger among unexposed pregnancies. Moreover, although we were unable to determine whether patients actually took the NSAIDs prescribed, we applied a strict exposure definition of ≥2 prescriptions of NSAIDs to address for this uncertainty; under reasonable assumptions, if a woman refilled the respective drug’s prescription, she probably took them. Meanwhile, exposure misclassification due to inaccurate timing of gestational age is another possibility, which, however, would be expected to bias the estimates towards the null. Second, we could have underestimated the risk of antepartum hemorrhage or oligohydramnios as local clinicians and specialists do not always report them in routine clinical practice. Yet, this misclassification is likely to be nondifferential between groups as, although failure to record certain conditions are related with demographic factors; these were all well balanced in the weighted cohort. Hence, any bias arising from this, under reasonable assumptions, would have led the estimates towards the null. Third, residual confounding by indication is still possible. However, the results of the analysis that compared NSAIDs with acetaminophen and the point estimates of the sensitivity analysis that restricted to pregnancies with indications of NSAIDs (e.g., inflammatory disease) or severe respiratory infections, and exclusive use in early pregnancy suggest that were consistent with the main findings. Fourth, residual confounding from unmeasured confounders cannot be ruled out despite rigorous adjustment of 52 potential confounders and the sibling-matched analysis that further accounted for familial factors. Fifth, assessment of first trimester or early pregnancy exposure starting from the LMP adds 2 weeks of exposure prior to conception, which, with the short NSAID half-life, would again bias toward the null. Sixth, we restricted our study cohorts to live births only, which could have led to selection bias given that severe malformations resulting in pregnancy terminations would have been missed. However, the results of quantitative bias analyses showed that the corrected RR for overall congenital malformations and congenital heart defects remained <1.41 under the most extreme scenario. It indicates that selection bias, if present, is likely to have been minimal ( S1 Appendix  and  S2  and  S3  Figs ). Last, given the nature of this observational study and the use of routinely collected insurance claims data to examine the safety of NSAIDs in early pregnancy, our findings are limited to suggesting only an association and not causation or causal relationships. Hence, caution is warranted in interpreting our findings.\n\nIn this comprehensive, nationwide cohort study of 1.8 million pregnancies, exposure to NSAIDs during early pregnancy, as compared with no exposure, was associated with slightly higher risks of major congenital malformations, largely driven by heart defects, low birth weight, and oligohydramnios. Notably, neonatal and maternal adverse outcomes were more strongly associated with the use of COX-2 inhibitors or NSAIDs for >10 days during early pregnancy. These findings therefore suggest that clinicians should carefully weigh the benefits of prescribing NSAIDs in early pregnancy against its modest, but possible, risk of neonatal and maternal outcomes. Further, if possible, both clinicians and pregnant women should consider prescribing or receiving nonselective NSAIDs for <10 days, respectively, with continued careful monitoring for any safety signals. Future studies are warranted to corroborate our findings and to examine the causation on this topic.\n\n(DOCX)\nClick here for additional data file.\n(DOCX)\nClick here for additional data file.\n(DOCX)\nClick here for additional data file.\n(DOCX)\nClick here for additional data file.\n(DOCX)\nClick here for additional data file.\n(DOCX)\nClick here for additional data file.\n(DOCX)\nClick here for additional data file.\n(DOCX)\nClick here for additional data file.\n(DOCX)\nClick here for additional data file.\n(DOCX)\nClick here for additional data file.\n(DOCX)\nClick here for additional data file.\n(DOCX)\nClick here for additional data file.\n(DOCX)\nClick here for additional data file.\n(DOCX)\nClick here for additional data file.\n(DOCX)\nClick here for additional data file.\n(DOCX)\nClick here for additional data file.\n(DOCX)\nClick here for additional data file.\n(DOCX)\nClick here for additional data file.","source_license":"CC-BY-4.0","license_restricted":false}