This study was conducted within the framework of the Southern Israeli Pregnancy Registry (siPREG), a population-based initiative investigating maternal and perinatal outcomes in southern Israel. This study is reported as per STROBE guideline [20] (STROBE Checklist D1 in S1 File) and was approved by the Soroka University Medical Center (SUMC) ethics committee in accordance with the Declaration of Helsinki (approval number 0069–20-SOR; 7 March 2022). Informed consent was waived by the ethics committee due to the study’s retrospective design.

This retrospective observational study did not follow a prospective analysis plan. Analyses were specified after data extraction based on the study objectives.

The cohort included all pregnancies of women aged 15–45 years insured by Clalit Health Services (CHS) maintenance organization in southern Israel that resulted in delivery or elective pregnancy termination for suspected fetal malformations at SUMC between 1998 and 2018. Clalit insures approximately 70% of women of reproductive age in the region, and SUMC accounts for nearly all deliveries (~98%) in the district [21]. Health insurance in Israel is universal and legally mandated. No differences were reported between individuals insured by Clalit and other state-mandated HMOs in the southern district [22].

Pregnancies exposed to established teratogenic drugs (antimetabolites, isotretinoin, and anti-epileptic drugs), multiple gestations, or pregnancies with documented genetic or chromosomal diagnoses were excluded from the study. Analyses were restricted to pregnancies with complete data on model’s covariates and outcomes.

Exposure was defined as dispensation of any NSAID during the first trimester, from the first day of the last menstrual period through the end of the 13th gestational week. Gestational age was assessed by a trained obstetrician, primarily using last menstrual period dates recorded at first-trimester visits, and in a small minority of cases based on first-trimester ultrasound measurements.

In addition, exposure was evaluated separately for each NSAID: ibuprofen, diclofenac, etodolac, naproxen, indomethacin, piroxicam, and lornoxicam. The unexposed group comprised pregnancies unexposed to NSAID during the first trimester, regardless of exposure to other analgesic or antipyretic medications. Pregnancies with exposure to more than one NSAID during the first trimester were included in each relevant drug-specific analysis, such that exposure groups were not mutually exclusive.

Overall exposure was further quantified by the total number of defined daily doses (DDD) dispensed during the first trimester, categorized as no exposure (0 DDD), short-term (1–7 DDD), medium-term (8–21 DDD), and long-term (>21 DDD) exposure. The corresponding DDDs were 1,200 mg for ibuprofen, 100 mg for diclofenac and indomethacin, 400 mg for etodolac, 500 mg for naproxen, 20 mg for piroxicam, and 16 mg for lornoxicam [23].

MCMs were defined according to the Metropolitan Atlanta Congenital Defects Program (MACDP) of the Centers for Disease Control and Prevention and classified using the International Classification of Diseases, 9th Revision (ICD-9) by board certified neonatologists. We examined associations between first-trimester NSAID exposure and the overall prevalence of MCMs, as well as system-specific groups: cardiovascular (ICD-9: 745, 746, 747), central nervous system (740, 741, 742, 743), musculoskeletal (754, 755, 756), gastrointestinal (750, 751), and genitourinary malformations (752, 753).

Obesity was defined as any diagnosis in the 278.0–278.4 ICD-9 range, diabetes mellitus as ICD-9 250.* and gestational diabetes mellitus (GDM) as ICD-9 648.00–648.04. Folic acid use was defined as dispensation during the first trimester. Smoking was defined based on self-report or ICD-9 code 305.1.

NSAID indications were defined using ICD-9 diagnostic codes recorded during the first trimester of pregnancy, corresponding to conditions for which NSAID are commonly prescribed. These included musculoskeletal and joint disorders (e.g., arthritis, back pain, myalgia; ICD-9 714–716, 718–719, 724, 729), pain and inflammatory conditions (e.g., headache, abdominal or pelvic pain; ICD-9 346, 784, 789, 623–625), injuries and fractures (ICD-9,800–829, 840–848, 920–924, E-codes E931–E980), pregnancy-related conditions including threatened abortion (ICD-9 632), and fever or infection (ICD-9 780.6, 780.60; respiratory/ear, nose, and throat: 460–466, 381*, 462–463; urinary tract: 5,990, 590*; other infections/inflammation: 041*, 079*, 682*, 614*, 615*). A woman could have more than one NSAID indication, and all relevant first-trimester diagnoses were included in the composite variable.

The cohort was constructed by linking four databases. Pregnancy and delivery data were obtained from the SUMC Obstetrics and Gynecology Division, and MCMs diagnosed within the first year of life from the SUMC hospitalization database. Malformations identified before elective terminations were manually retrieved from the SUMC Committee for Termination of Pregnancies registry. All diagnoses were confirmed by board-certified specialists and coded using ICD-9. Medication dispensation data, including prescription and over-the-counter NSAID with Anatomical Therapeutic Chemical classification system (ATC) codes and DDD, were extracted from the CHS database. Databases were merged via national identification number and hospitalization number to link maternal, neonatal, and termination records (Appendix A1 in S1 File).

Statistical analyses were performed using R version 4.5.1 [24]. Categorical variables are reported as n (%) and compared using χ² tests or Fisher’s exact tests. Continuous variables are summarized as mean ± standard deviation or median [Interquartile Range (IQR)], according to distribution, and compared with two-sample t-tests or Wilcoxon rank-sum tests.

Unmatched-adjusted-Relative Risks (aRRs) were estimated using Poisson regression with a logarithmic link function. Marginal (average) effects were derived with G-computation, and robust variance estimates were obtained with cluster-robust heteroskedasticity-consistent (“sandwich”) estimator clustered by patient identifier [25–28]. Analyses were conducted using the marginaleffects package in R [29].

For matching-adjusted models, generalized full matching [30] was applied on the propensity score to further account for confounding, targeting the average treatment effect (ATE) [31]. The propensity score was estimated using a probit regression of the treatment on the covariates and achieved adequate balance. Matching weights were derived from this procedure and incorporated into the outcome model (no units were discarded by the matching). Matched-aRRs were then estimated using G-computation with weighted Poisson-regression and two-way cluster-robust standard errors, clustered on maternal identifier and matching subclass.

The conceptual model connecting consistent first-trimester NSAID exposure with MCMs is depicted in the directed acyclic graph (DAG), Fig 1. Covariates accounted for in the final models included maternal age, ethnicity, lack of perinatal care, diabetes, obesity, folic acid supplementation, gravidity, calendar year, smoking, NSAID indication, and exposure to other analgesics or antipyretics. These were selected a priori by two experts (S.D. and A.A.H.) to block potential confounding via “backdoor” paths. Maternal diabetes and obesity are associated with cardiac malformations [32–34] and higher NSAID use [35,36]. Lack of perinatal care and folic acid supplementation influence both exposure and malformation risk [37,38]. Ethnicity accounts for higher malformation prevalence and greater use of over-the-counter medications among Bedouins [39]. Calendar year adjusts for temporal changes in NSAID dispensations and detection of MCMs due to increasing prenatal ultrasonography. NSAID indication (e.g., fever or inflammatory conditions) was included to control for confounding by underlying conditions [40,41]. The models were additionally adjusted for first-trimester exposure to other analgesics or anti-pyretics (acetaminophen and dipyrone) to account for other underlying indications for NSAID use.

To evaluate the potential impact of exposure misclassification due to over-the-counter ibuprofen purchases not captured in the dispensation database, we conducted a probabilistic sensitivity (“tipping-point”) analysis. Assuming dispensation-based exposure ascertainment is highly specific but may be imperfectly sensitive, we treated recorded ibuprofen exposure as a reliable indicator and simulated increasing degrees of missed exposure by randomly reclassifying individuals from the unexposed to the exposed group. To represent a conservative scenario in which misclassification could bias the association away from the null, we constrained the reclassified subset to have the same prevalence of MCMs observed among the recorded exposed group (8.3%). The proportion reclassified ranged from 0% to 3% of the total cohort in 0.05% increments; at each increment, we performed 100 random reallocations and refit the primary matching-adjusted model to obtain a distribution of adjusted risk ratio estimates (Full details are provided in Appendix A2 in S1 File).

The matching procedure and outcome model included NSAID indication to account for the underlying clinical indication. In response to peer review comments, to further characterize the relationship between NSAID exposure and underlying conditions, we compared the distribution of NSAID indication categories between exposed and unexposed pregnancies. In addition, exposure to other analgesic/antipyretic medications (acetaminophen and dipyrone) was incorporated into the matching preprocessing and outcome model as an indirect proxy for NSAID indications. However, because these medications may be used concurrently with NSAID, their inclusion may introduce temporal ambiguity, overadjustment, or collider bias. We therefore performed four additional sensitivity analyses: (1) removing the exposure to other analgesic/antipyretic medications variable from both matching and model adjustment; (2) limiting matching and adjustment to exposures that occurred before NSAID use; (3) limiting matching to exposures that occurred before NSAID use and additionally adjusting in the outcome model for exposures occurring before and after NSAID use as separate covariates; and (4) excluding pregnancies with exposure to other analgesic/antipyretic medications from both the exposed and unexposed groups.

To account for NSAID dispensations that may have started shortly before conception and extended into the first trimester, we conducted another sensitivity analysis redefining first-trimester exposure to include NSAID dispensations initiated within the two weeks preceding the estimated date of conception.

Among 20,202 NSAID-exposed pregnancies, 1,651 (8.2%) were diagnosed with MCMs, compared with 16,998 (7.0%) among the 244,656 unexposed pregnancies. NSAID exposure was associated with an increased risk of MCMs in crude analyses (RR = 1.18 (95% CI[1.12,1.24])); however, no association was found after matching (matched-aRR = 0.99 (95% CI [0.90,1.10])) (Fig 4). Cardiovascular malformations were identified in 885 (4.4%) of exposed and 7,873 (3.2%) of unexposed pregnancies (p < 0.001), with a higher crude risk (RR = 1.36 (95% CI [1.27,1.46])) that did not persist in the matched analysis (matched-aRR = 1.05 (95% CI [0.92,1.20])). No evidence of increased risk for musculoskeletal malformations (matched-aRR = 1.03 (95% CI [0.77,1.39])), central nervous system malformations (matched-aRR = 0.77 (95% CI [0.53,1.11])), cleft palate (matched-aRR = 0.95 (95% CI [0.47–1.91])), gastrointestinal malformations (matched-aRR = 1.03 (95% CI [0.64–1.63])), or genitourinary malformations (matched-aRR = 0.99 (95% CI [0.72,1.35])) was observed.

No associations were observed between exposure to any of the NSAID and the risk of overall MCMs (ibuprofen: matched-aRR = 0.97 (95% CI [0.86–1.09]); diclofenac: matched-aRR = 1.02 (95% CI [0.80,1.30]); naproxen: matched-aRR = 0.97 (95% CI [0.73,1.30]); indomethacin: matched-aRR = 1.76 (95% CI [0.59, 5.21]); etodolac: matched-aRR = 0.88 (95% CI [0.64, 1.21])). No significant associations were observed between individual NSAID and congenital malformations across different organ systems (Fig 4).

The prevalence of MCMs was 8.2% among short-term NSAID users (1–7 DDD), 8.6% among medium-term users (8–21 DDD), and 9.3% among long-term users (>21 DDD), compared with 7.0% in the unexposed group. No association was found with MCMs across the different dosage categories (Table B1 in S1 File).

In simulations, the adjusted risk ratio increased as a larger proportion of the cohort was assumed to be misclassified and reassigned from unexposed to exposed. Using the 1.3% missed-exposure estimate from prior validation work at SUMC as a plausible degree of misclassification, reclassifying 1.3% of the cohort [42] (with an 8.3% MCMs prevalence among reclassified individuals) yielded a nonsignificant association in the matching-adjusted analysis (matched-aRR = 1.06 (95% CI [0.97,1.15])). Overall, plausible levels of unrecorded over-the-counter ibuprofen exposure did not materially change the direction or statistical inference of the primary results (Appendix A2 in S1 File).

Concerning the distribution of indications among exposed and unexposed pregnancies, musculoskeletal and pain or inflammatory indications were rare in both groups (<0.1%), while fever or infections and pregnancy-related indications were more common among exposed pregnancies (3.6% versus 2.0% and 9.9% versus 7.0%, respectively) (Table A1 in S1 File). Additional sensitivity analyses, performed to address potential temporal ambiguity and collider bias from concurrent use of other analgesics, also showed no association with overall or organ-specific malformations (Tables A2 and A3 in S1 File). Extending NSAID exposure to two weeks before conception yielded similar results (matched-aRR = 1.02 (95% CI [0.90,1.12])).

The research was approved by the Soroka University Medical Center ethics committee in accordance with the Declaration of Helsinki (study number: 0069–20-SOR, Approval date: 07/03/2022). Due to the retrospective nature of the cohort, the ethic committee waived the need of informed consent to participate.

Due to the retrospective nature of the cohort, the ethic committee waived the need of informed consent to participate.

Due to the retrospective nature of the cohort, the ethic committee waived the need of informed consent to publish.