Prenatal exposure to oral glucocorticoids and risk of long-term neurodevelopmental disorders

Abstract

We investigated whether prenatal exposure to oral glucocorticoids (GCs) is associated with long-term neurodevelopmental disorders (LNDDs) in offspring. Using South Korea’s National Health Insurance Service database, we identified all live births from January 1, 2011, to December 31, 2014 (n = 1,553,505) and followed offspring through December 31, 2023. Maternal oral GC exposure during pregnancy defined the exposed group. After 1:10 propensity-score matching, 33,940 exposed and 338,921 unexposed offspring were analyzed. LNDD incidence was 11.5% in the GC group versus 9.8% in the unexposed group (odds ratio [OR] 1.20, 95% confidence interval [CI], 1.16–1.24; P < 0.001). In duration analyses from the full cohort, adjusted odds were 1.06 (95% CI, 1.05–1.08; P < 0.001) for 1–6 days and 1.13 (95% CI, 1.08–1.19; P < 0.001) for ≥ 7 days versus none. By timing, adjusted odds were 1.21 (95% CI, 1.15–1.27; P < 0.001) for first, 1.16 (95% CI, 1.07–1.26; P < 0.001) for second, and 1.21 (95% CI, 1.14–1.28; P < 0.001) for third trimester exposure. Prenatal exposure to oral GCs was associated with a modest increase in LNDD risk; given the observational design and potential residual confounding, these findings are hypothesis-generating and should inform—rather than prescribe—clinical decision-making.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-026-35516-3.

Introduction

Glucocorticoids (GCs) are commonly prescribed during pregnancy to manage various maternal conditions, including autoimmune diseases and preterm labor^1,2. While their therapeutic benefits are well-recognized, concerns have been raised regarding their potential impact on fetal neurodevelopment³. GCs can cross the placental barrier⁴, and excessive exposure may influence the developing fetal brain, potentially leading to long-term neurodevelopmental disorders (LNDDs)^4,5.

Previous studies have explored the association between prenatal GC exposure and adverse neurodevelopmental outcomes in offspring^3,6–10. However, many of these studies have focused on parenteral antenatal corticosteroids or other systemic, non-oral regimens, and data on the effects of oral GC use during pregnancy remain limited^3,6–10. Furthermore, several influential studies assessed outcomes in infancy or early childhood, making it challenging to determine long-term implications^3,6–10. A recent large-scale cohort study reported elevated risks for autism spectrum disorders, attention-deficit/hyperactivity disorder, and mood/anxiety/stress-related disorders among offspring prenatally exposed to systemic glucocorticoids; estimates for intellectual disabilities were smaller and not consistently significant⁷.

Given the widespread use of oral GCs during pregnancy and the paucity of comprehensive long-term safety data, there is a critical need for population-based studies to evaluate potential risks. In this nationwide cohort study, we aimed to investigate the association between prenatal exposure to oral GCs and the risk of LNDDs in offspring, utilizing data from South Korea’s National Health Insurance Service (NHIS) database.

Methods

Study design and ethical considerations

We conducted a nationwide, retrospective cohort study using routinely collected administrative data, following the STROBE reporting guidelines¹¹. The study complied with the principles of the Declaration of Helsinki (1975, revised 2008) and applicable national regulations. Ethical approval was obtained from the institutional review board (IRB) of Seoul National University Bundang Hospital (IRB approval No. X-2407-910-902) and the NHIS (NHIS; approval No. NHIS-2025-02-1-010), which also granted access to the database. As the study used anonymised, routinely collected data, informed consent was waived by IRB of Seoul National University Bundang Hospital.

Data source

The NHIS is South Korea’s universal, single-payer health insurance system, with mandatory enrollment for all residents, including foreigners residing for over six months. The NHIS database captures detailed information on healthcare utilisation, including prescription records, diagnostic codes, procedures, mortality, and sociodemographic characteristics. Diagnoses are recorded using the International Classification of Diseases, 10th Revision (ICD-10). Although most clinicians in South Korea practice privately, their clinical activities are subject to state regulation, ensuring standardised reporting across the healthcare system¹². Data on all relevant ICD-10 diagnostic categories were included for analysis in this study.

Study population

We identified all live births in South Korea between 1 January 2011 and 31 December 2014 using the NHIS database. Maternal health records were subsequently linked to each infant. Births with missing or incomplete delivery-related information were excluded to ensure data integrity and analytical robustness.

Perinatal glucocorticoid exposure

Newborns were categorized as GC-exposed if their mothers received any oral systemic GC during pregnancy; eligible agents were prednisolone, methylprednisolone, and dexamethasone. NHIS pharmacy claims include drug/ingredient, strength/formulation, dispense date, quantity, and days’ supply. The estimated last menstrual period (LMP) was derived from the delivery date using a validated claims-based algorithm: 35 weeks (245 days) were subtracted for births coded as preterm and 39 weeks (273 days) for term births; extremely preterm births cannot be distinguished in NHIS and were handled under the preterm rule^13,14.

To minimize misclassification of prepregnancy exposure, mothers with any oral GC prescription within 90 days before the estimated LMP were excluded from the unexposed group. For exposure characterization, we summed cumulative days’ supply during pregnancy across fills (not double-counting overlaps) and categorized duration as 1–6 days versus ≥ 7 days. Because reliable conversion of claim records to prednisolone-equivalent milligram doses is not consistently feasible (variable tapers/strengths/instructions), exposure was summarized by duration rather than dose. Adherence is unobserved in claims; we therefore applied an as-dispensed (prescription-fill–based) exposure assumption.

Study outcome (LNDD)

The primary outcome was the first recorded diagnosis of LNDD from live birth through 31 December 2023, identified using ICD-10 codes F70–F98 in NHIS claims. Consistent with prior work, we analyzed both the composite LNDD and three prespecified subtypes: intellectual disabilities (F70–F79), disorders of psychological development (F80–F89; including autism spectrum disorders, F84), and behavioural/emotional disorders with childhood onset (F90–F98; including attention-deficit/hyperactivity disorder, F90)^15,16. In South Korea, use of these codes is tied to government-subsidized services, which supports coding validity; nonetheless, outcome misclassification remains possible. Outcome ascertainment did not require repeat claims (first qualifying diagnosis used). This window corresponds to approximately 9–13 years of age across the 2011–2014 birth cohorts (mean follow-up 12.0 years, SD 1.15; see Results). Long-term follow-up was ensured via person-level linkage of NHIS eligibility and claims files using anonymized identifiers, capturing all reimbursed inpatient and outpatient encounters; offspring were right-censored at death or insurance disenrollment. To address potential early miscoding, age-threshold sensitivity analyses required diagnoses at ≥ 1, ≥2, or ≥ 4 years (see Statistical analysis).

Maternal covariates

Maternal covariates were ascertained from NHIS eligibility and claims files. Pre-pregnancy baseline covariates (12 months before the estimated LMP) included maternal age (modeled continuously), the Charlson Comorbidity Index (CCI), and four groups of potential GC indications (chronic respiratory inflammatory disease, inflammatory bowel disease, severe autoimmune dermatologic disease, and other systemic autoimmune/inflammatory disorders; ICD-10 code lists in S1 Table). Delivery-time covariates included parity (nulliparous vs. multiparous), pregnancy plurality (singleton vs. multiple), and mode of delivery (cesarean vs. vaginal) from delivery claims, and socioeconomic variables (household income quartiles with a Medical Aid category, employment including self-employment) plus residential area at delivery from the NHIS eligibility file for the year of delivery. Residential area was categorized as urban (Seoul or metropolitan cities—Busan, Daegu, Incheon, Gwangju, Daejeon, Ulsan; Sejong treated as urban from 2012 onward) and rural (all other regions).

Statistical analysis

Categorical variables were summarised as frequencies (%) and continuous variables as means standard deviation (SD). To address baseline imbalances between GC-exposed and unexposed groups, we estimated propensity scores (PS) for any prenatal oral GC exposure using logistic regression with the following pre-specified pre-LMP covariates: maternal age (continuous), parity, pregnancy plurality (singleton vs. multiple), mode of delivery (cesarean vs. vaginal), calendar year, socioeconomic variables (household income/Medical Aid, employment, urban/rural residence), CCI, and the four GC-indication groups. Pregnancy complications and neonatal characteristics were not included because they occur after exposure and may act as mediators or colliders.

Nearest-neighbour matching used a caliper of 0.2 SD of the logit PS, without replacement, at a 1:10 ratio¹⁷. To ensure common support, controls with PS values outside the exposed group’s PS range were discarded prior to matching. Covariate balance before and after matching was assessed using absolute standardised mean differences (ASD), with ASD < 0.10 indicating adequate balance. In the PS-matched cohort, we estimated odds ratios (ORs) and 95% confidence intervals (CIs) for any LNDD and, separately, for intellectual disabilities, disorders of psychological development, and behavioural/emotional disorders with childhood onset, using logistic regression with cluster-robust standard errors at the matched-set level.

We additionally conducted a 1:1 PS matching sensitivity analysis (nearest-neighbour, caliper 0.2 SD, no replacement) to evaluate robustness to the matching ratio; effect estimates were obtained with logistic regression using pair-clustered robust standard errors, and balance/common-support diagnostics were repeated.

Further sensitivity analyses were conducted in the full, unmatched cohort using multivariable logistic regression adjusting for the same baseline covariates as the PS model. These analyses had three aims: (1) assess consistency with the matched-cohort findings; (2) evaluate associations by prescription duration (unexposed [reference], 1–6 days, ≥ 7 days); and (3) evaluate associations by exposure timing (unexposed [reference], first, second, third trimester). Multicollinearity was assessed with variance inflation factors (all < 2.0).

To probe potential biases, we conducted three prespecified analyses: (i) a within-family (discordant-sibling) analysis restricted to mothers with ≥ 2 live-born offspring who were discordant for prenatal oral GC exposure; we fit conditional logistic regression stratified by mother (equivalent to mother fixed effects), thereby controlling for all time-invariant familial/genetic factors, and adjusted for time-varying pregnancy-level covariates (maternal age [continuous], calendar year, socioeconomic variables, CCI, and GC-indication groups); (ii) age-threshold sensitivity analyses redefining LNDD as the first recorded diagnosis after age ≥ 1, ≥2, or ≥ 4 years to reduce potential early misclassification; and (iii) indication-stratified analyses, repeating the adjusted models within mothers with any coded potential indication and within mothers without such codes. All analyses were conducted in R (version 4.3.2; R Foundation for Statistical Computing, Vienna, Austria). Two-sided P < 0.05 was considered statistically significant.

Results

Study population

Between 1 January 2011 and 31 December 2014, a total of 1,821,452 newborns from 1,231,523 mothers were recorded. After excluding 10,637 mothers and 267,947 newborns with incomplete delivery records, 1,553,505 newborns from 1,220,886 mothers were eligible for screening. Among these, 33,940 newborns were classified into the GC group based on maternal exposure to prescribed GCs during pregnancy. Of the remaining 1,519,565 newborns, 19,330 were excluded due to maternal GC prescriptions within 90 days prior to the estimated LMP, resulting in 1,500,110 newborns in the non-GC group.

PS matching was performed at a 1:10 ratio, yielding a final matched cohort of 372,757 newborns (33,940 in the GC group and 338,921 in the non-GC group). A flow diagram illustrating the selection process is shown in Fig. 1. Baseline characteristics of the GC and non-GC groups before and after matching are presented in Table 1. After matching, all ASDs for covariates were less than 0.1, confirming adequate balance between groups. The mean follow-up period for LNDD assessment was 12.0 years (SD, 1.15 years). Among baseline covariates, household income was missing for 60,927 individuals (4.0%) and was modeled using an ‘unknown’ category. Post-matching balance for income, including the unknown category, satisfied ASD < 0.10. As a robustness check, we also performed 1:1 PS matching; baseline balance was again satisfactory (all ASDs < 0.10) and results were directionally consistent with the primary analysis (S2 Table).

Fig. 1.

A flow diagram illustrating the selection process. PS, propensity score; LMP, Last Menstrual Period; GC, glucocorticoid.

Table 1.

Comparison of baseline characteristics between the GC and non-GC groups before and after 1:10 PS matching.

Variable	Entire cohort (n = 1,534,050)		ASD	PS-matched cohort (n = 372,861)		ASD
	GC group (n = 33,940)	Non- GC group (n = 1,500,110)		GC group (n = 33,940)	Non- GC group (n = 338,921)
Mother age, year	32.5 (4.3)	32.7 (4.1)	0.029	32.5 (4.3)	32.5 (4.1)	0.009
Household income level
Medical aid program	371 (1.1)	8,948 (0.6)		371 (1.1)	3,520 (1.0)
Q1 (lowest)	5,747 (16.9)	222,436 (14.8)	0.025	5,747 (16.9)	57,266 (16.9)	0.007
Q2	8,654 (25.5)	371,351 (24.8)	0.048	8,654 (25.5)	86,748 (25.6)	0.003
Q3	11,166 (32.9)	522,844 (34.9)	0.020	11,166 (32.9)	111,330 (32.8)	0.011
Q4 (highest)	6,714 (19.8)	314,892 (21.0)	0.045	6,714 (19.8)	67,193 (19.8)	0.003
Unknown	1,288 (3.8)	59,639 (4.0)	0.011	1,288 (3.8)	12,864 (3.8)	0.004
Having a job	23,279 (68.6)	1,041,043 (69.4)	0.033	23,279 (68.6)	232,665 (68.6)	0.031
Residence
Urban area	14,032 (41.3)	665,406 (44.4)		14,032 (41.3)	139,911 (41.3)
Rural area	19,908 (58.7)	834,704 (55.6)	0.072	19,908 (58.7)	199,010 (58.7)	0.031
Maternal condition
CCI, point	1.7 (1.4)	1.3 (1.3)	0.261	1.7 (1.4)	1.7 (1.4)	0.011
Myocardial infarction	113 (0.3)	3,431 (0.2)	0.022	113 (0.3)	1,103 (0.3)	0.007
Congestive heart failure	230 (0.7)	6,862 (0.5)	0.027	230 (0.7)	2,228 (0.7)	0.016
Peripheral vascular disease	1,471 (4.3)	44,782 (3.0)	0.071	1,471 (4.3)	14,468 (4.3)	0.013
Cerebrovascular disease	650 (1.9)	21,425 (1.4)	0.121	650 (1.9)	6,314 (1.9)	0.005
Dementia	25 (0.1)	666 (0.0)	0.038	25 (0.1)	223 (0.1)	< 0.001
Chronic pulmonary disease	21,854 (64.4)	770,757 (51.4)	0.013	21,854 (64.4)	217,400 (64.1)	0.013
Rheumatic disease	2,328 (6.9)	55,568 (3.7)	0.117	2,328 (6.9)	22,343 (6.6)	0.012
Peptic ulcer disease	13,855 (40.8)	478,842 (31.9)	0.160	13,855 (40.8)	137,244 (40.5)	0.006
Mild liver disease	9,974 (29.4)	361,666 (24.1)	0.128	9,974 (29.4)	98,898 (29.2)	0.003
DM without chronic complication	2,696 (7.9)	93,340 (6.2)	0.056	2,696 (7.9)	26,502 (7.8)	0.007
DM with chronic complication	398 (1.2)	12,642 (0.8)	0.330	398 (1.2)	3,888 (1.1)	0.002
Hemiplegia or paraplegia	50 (0.1)	1,605 (0.1)	0.007	50 (0.1)	513 (0.2)	0.009
Renal disease	201 (0.6)	5,175 (0.3)	0.036	201 (0.6)	1,922 (0.6)	0.005
Cancer	1,092 (3.2)	41,326 (2.8)	0.025	1,092 (3.2)	10,883 (3.2)	0.009
Moderate or severe liver disease	202 (0.6)	6,181 (0.4)	0.016	202 (0.6)	1,979 (0.6)	0.003
Metastatic solid tumor	100 (0.3)	3,723 (0.2)	< 0.001	100 (0.3)	1,035 (0.3)	0.011
AIDS/HIV	13 (0.0)	378 (0.0)	0.008	13 (0.0)	112 (0.0)	0.007
Maternal potential indication of GC
Chronic respiratory inflammatory disease	4,565 (13.4)	83,974 (5.6)	0.228	4,565 (13.5)	44,111 (13.0)	0.005
Inflammatory bowel disease	65 (0.2)	1,175 (0.1)	0.035	65 (0.2)	564 (0.2)	0.008
Severe autoimmune dermatologic disease	316 (0.9)	5,832 (0.4)	0.066	316 (0.9)	2,948 (0.9)	0.021
Other systemic autoimmune/ inflammatory disorders	78 (0.2)	1,544 (0.1)	0.028	78 (0.2)	709 (0.2)	0.013
Delivery information
Primiparous	24,832 (73.2)	1,181,819 (78.8)	0.134	24,832 (73.2)	248,529 (73.3)	0.005
Single baby pregnancy	33,158 (97.7)	1,476,327 (98.4)	0.049	33,158 (97.7)	331,242 (97.7)	0.027
Caesarean section	13,918 (41.0)	566,445 (37.8)	0.084	13,918 (41.0)	138,409 (40.8)	< 0.001
Year of baby birth
2011	8,160 (24.0)	391,671 (26.1)	0.048	8,160 (24.0)	81,767 (24.1)	< 0.001
2012	9,231 (27.2)	397,901 (26.5)	0.015	9,231 (27.2)	91,841 (27.1)	0.002
2013	8,426 (24.8)	355,924 (23.7)	0.019	8,426 (24.8)	84,154 (24.8)	0.014
2014	8,123 (23.9)	354,614 (23.6)	0.013	8,123 (23.9)	81,159 (23.9)	0.018

GC, glucocorticoid; PS, propensity score; ASD, absolute standardized mean difference; CCI, Charlson comorbidity index; DM, diabetes mellitus; AIDS, acquired immune deficiency syndrome; HIV, human immunodeficiency virus.

Analyses in the PS-matched cohort

Table 2 summarizes the results before and after PS matching. In the matched cohort, the incidence of any LNDD was 11.5% (3,906/33,940) in the GC group and 9.8% (33,195/338,921) in the non-GC group. Logistic regression showed higher odds of LNDD with GC exposure versus no exposure (OR, 1.20; 95% CI, 1.16–1.24; P < 0.001). This corresponds to an absolute risk difference (ARD) of approximately 1.7% points (about 17 additional cases per 1,000 GC-exposed births) and an approximate number needed to harm (NNH) of 59, if interpreted causally. By subtype, odds were higher for intellectual disabilities (OR, 1.20; 95% CI, 1.09–1.31; P < 0.001), disorders of psychological development (OR, 1.17; 95% CI, 1.11–1.24; P < 0.001), and behavioural/emotional disorders with childhood onset (OR, 1.20; 95% CI, 1.15–1.25; P < 0.001). Results from the 1:1 PS-matched cohort were directionally consistent with the 1:10 analysis (S3 Table). Given the cohort exposure prevalence (~ 2.2%), the population-attributable risk (PAR) is approximately 0.4%, i.e., ~ 37 additional LNDD cases per 100,000 live births, assuming a causal relationship and treating the odds ratio as a proxy for the risk ratio.

Table 2.

Analyses before and after 1:10 PS matching.

Outcome	Event (n, %)	OR (95% CI)	P-value
Total LNDD
Before PS matching
Non-GC group	141,862/1,500,110 (9.5)	1
GC group	3,906/33,940 (11.5)	1.25 (1.20, 1.29)	< 0.001
After PS matching
Non-GC group	33,195/338,921 (9.8)	1
GC group	3,906/33,940 (11.5)	1.20 (1.16, 1.24)	< 0.001
Intellectual disabilities
Before PS matching
Non-GC group	17,086/1,500,110 (1.1)	1
GC group	488/33,940 (1.4)	1.27 (1.16, 1.39)	< 0.001
After PS matching
Non-GC group	4,085/338,921 (1.2)	1
GC group	488/33,940 (1.4)	1.20 (1.09, 1.31)	< 0.001
Disorder of psychological development
Before PS matching
Non-GC group	47,778/1,500,110 (3.2)	1
GC group	1,295/33,940 (3.8)	1.21 (1.14, 1.28)	< 0.001
After PS matching
Non-GC group	11,101/338,921 (3.3)	1
GC group	1,295/33,940 (3.8)	1.17 (1.11, 1.24)	< 0.001
Behavioural and emotional disorders with onset typically during childhood or adolescence
Before PS matching
Non-GC group	114,020/1,500,110 (7.6)	1
GC group	3,157/33,940 (9.3)	1.25 (1.20, 1.29)	< 0.001
After PS matching
Non-GC group	26,717/338,921 (7.9)	1
GC group	3,157/33,940 (9.3)	1.20 (1.15, 1.25)	< 0.001

PS, propensity score; OR, odds ratio; CI, confidence interval; LNDD, longterm neurodevelopmental disorder; GC, glucocorticoid.

Analyses in the entire cohort

Multivariable logistic regression results from the entire cohort are presented in Table 3. In Model 1, prenatal oral GC exposure was associated with higher odds of any LNDD compared with no exposure (OR 1.20, 95% CI 1.16–1.25, P < 0.001). In Model 2 (by prescription duration), odds were 1.06 (95% CI 1.05–1.08, P < 0.001) for 1–6 days and 1.13 (95% CI 1.08–1.19, P < 0.001) for ≥ 7 days, versus no exposure. In Model 3 (by exposure timing), odds were 1.21 (95% CI 1.15–1.27, P < 0.001) for first trimester, 1.16 (95% CI 1.07–1.26, P < 0.001) for second trimester, and 1.21 (95% CI 1.14–1.28, P < 0.001) for third trimester exposure, each versus no exposure. Adjusted ORs with 95% CIs for all covariates in Model 1 are provided in S4 Table.

Table 3.

Multivariable logistic regression analyses in the entire cohort.

Outcome	OR (95% CI)	P-value
Multivariable model 1
Non-GC group	1
GC group	1.20 (1.16, 1.25)	< 0.001
Multivariable model 2
Non-GC group	1
GC prescription: 1–6 days (n = 29,746)	1.06 (1.05, 1.08)	< 0.001
GC prescription: ≥ 7 days (n = 4,197)	1.13 (1.08, 1.19)	< 0.001
Multivariable model 2
Non-GC group	1
1st trimester exposed GC group (n = 15,704)	1.21 (1.15, 1.27)	< 0.001
Non-GC group	1
2nd trimester exposed GC group (n = 6,127)	1.16 (1.07, 1.26)	< 0.001
Non-GC group	1
3rd trimester exposed GC group (n = 12,112)	1.21 (1.14, 1.28)	< 0.001

OR, odds ratio; CI, confidence interval; GC, glucocorticoid.

Discordant sibling analysis

In a discordant sibling analysis (Table 4), prenatal oral GC exposure remained associated with any LNDD (OR 1.16, 95% CI 1.10–1.24, P < 0.001); subtype estimates were similar: intellectual disabilities 1.12 (95% CI 1.08–1.18, P < 0.001), psychological development disorders 1.18 (95% CI 1.03–1.27, P = 0.023), and behavioural/emotional disorders 1.19 (95% CI 1.11–1.28, P < 0.001).

Table 4.

Discordant sibling analysis.

Outcome	OR (95% CI)	P-value
Total LNDD
Non-GC group (n = 5,207)	1
GC group (n = 5,463)	1.16 (1.10, 1.24)	< 0.001
Intellectual disabilities
Non-GC group (n = 5,207)	1
GC group (n = 5,463)	1.12 (1.08, 1.18)	< 0.001
Disorder of psychological development
Non-GC group (n = 5,207)	1
GC group (n = 5,463)	1.18 (1.03, 1.27)	0.023
Behavioural and emotional disorders with onset typically during childhood or adolescence
Non-GC group (n = 5,207)	1
GC group (n = 5,463)	1.19 (1.11, 1.28)	< 0.001

OR, odds ratio; CI, confidence interval; LNDD, long-term neurodevelopmental disorder; GC, glucocorticoid.

Sensitivity and subgroup analyses

In sensitivity analyses restricting LNDD to diagnoses recorded after age ≥ 1, ≥2, and ≥ 4 years (Table 5), prenatal oral GC exposure remained associated with higher odds of LNDD—OR 1.17 (95% CI 1.12–1.22, P < 0.001), 1.20 (95% CI 1.13–1.24, P < 0.001), and 1.16 (95% CI 1.12–1.20, P < 0.001), respectively. In subgroup analyses stratified by maternal potential indications for GC therapy (Table 6), prenatal oral GC exposure remained associated with higher odds of LNDD, with similar magnitudes in both strata: maternal-indication stratum OR 1.22 (95% CI 1.12–1.34, P < 0.001) and no-indication stratum OR 1.21 (95% CI 1.16–1.25, P < 0.001).

Table 5.

Sensitivity analyses by minimum age at LNDD diagnosis.

Outcome	OR (95% CI)	P-value
Diagnosis of LNDD 1 years after birth (n = 145,525)
Non-GC group	1
GC group	1.17 (1.12, 1.22)	< 0.001
Diagnosis of LNDD 2 years after birth (n = 127,352)
Non-GC group	1
GC group	1.20 (1.13, 1.24)	< 0.001
Diagnosis of LNDD 4 years after birth (n = 101,974)
Non-GC group	1
GC group	1.16 (1.12, 1.20)	< 0.001

OR, odds ratio; CI, confidence interval; LNDD, long-term neurodevelopmental disorder; GC, glucocorticoid.

Table 6.

Sensitivity analyses by indication of GC.

Outcome	OR (95% CI)	P-value
LNDD
Maternal potential indication of GC group (n = 96,843)	1
Non-GC group	1.22 (1.12, 1.34)	< 0.001
GC group
Non-maternal potential indication of GC group (n = 1,437,207)
Non-GC group	1
GC group	1.21 (1.16, 1.25)	< 0.001

OR, odds ratio; CI, confidence interval; LNDD, long-term neurodevelopmental disorder; GC, glucocorticoid.

Discussion

In this large, population-based cohort with up to 13 years of follow-up, prenatal exposure to oral glucocorticoids was associated with a modest increase in the odds of LNDD (≈ 18% in the PS-matched cohort). The association was consistent across analyses by exposure duration and trimester and remained after multiple robustness checks (including 1:1 matching, age-threshold outcome definitions, and a discordant-sibling design). Given the observational design and potential residual confounding, these results should be interpreted as evidence of association rather than causation.

GC readily crosses the placenta and binds to glucocorticoid receptors expressed in fetal tissues, including the developing brain^18,19. The fetal brain is particularly sensitive to exogenous GC because of the immature enzymatic barrier in the placenta, which normally inactivates maternal cortisol¹⁸. High levels of GC during gestation have been shown to alter hippocampal development, suppress neurogenesis, and induce long-term changes in the hypothalamic-pituitary-adrenal axis, potentially predisposing offspring to cognitive, emotional, and behavioural dysregulation²⁰. Animal studies have demonstrated that antenatal GC exposure leads to altered brain architecture, increased neuronal apoptosis, and changes in synaptic plasticity²¹.

The timing and duration of GC exposure appear to be critical factors influencing neurodevelopmental outcomes²². In our study, elevated LNDD risk was observed across all trimesters, but with slightly higher effect sizes in early and late gestation, suggesting potential windows of vulnerability. This aligns with previous findings that the fetal brain undergoes rapid region-specific maturation at different gestational stages, and that even short-term GC exposure during sensitive periods may disrupt neuronal differentiation and circuit formation²³. Furthermore, while synthetic GCs are clinically beneficial for fetal lung maturation, they are more potent and longer-acting than endogenous cortisol, raising concern about prolonged receptor activation in the fetal brain¹⁰.

Our findings are consistent with prior studies reporting an increased risk of neurodevelopmental or psychiatric outcomes following antenatal GC exposure. A Finnish registry-based cohort study found that antenatal corticosteroid treatment was associated with a higher incidence of childhood mental and behavioral disorders, including anxiety and attention-deficit/hyperactivity disorder⁹. More recently, a population-based cohort study demonstrated that prenatal exposure to systemic GCs was associated with an elevated risk of mental disorders in offspring, reinforcing concerns about long-term neurodevelopmental impacts⁷. A separate investigation conducted in China observed that antenatal GC exposure—particularly to dexamethasone—was linked to impaired cognitive outcomes in infants at one year of age⁸. However, most existing studies have focused on preterm populations, high-dose systemic administration, or short-term neurodevelopmental endpoints. In contrast, our study uniquely examined oral GC exposure within a general obstetric population using nationwide data and a 13-year follow-up. These distinctions enhance the generalizability of our findings and suggest that even low-to-moderate prenatal GC exposure may have lasting neurodevelopmental consequences.

This study has several limitations that should be acknowledged. First, as with all observational studies, residual confounding cannot be fully excluded despite the use of PS matching and multivariable adjustment. Second, data on indications for GC prescription were not available, limiting our ability to separate the effects of GC itself from those of the underlying maternal conditions. Third, although the outcome definition was based on validated diagnostic codes within the Korean NHIS database, the use of administrative data may still lead to underdiagnosis or misclassification. Fourth, we were unable to account for exposure to known neurodevelopmental teratogens—such as antiepileptic drugs, alcohol, tobacco, or environmental toxins—that may independently influence LNDD risk. Fifth, We could not evaluate a mg-based dose–response, as claims do not consistently permit accurate conversion to prednisolone-equivalents; moreover, adherence is unobserved, so nondifferential exposure misclassification is possible and would likely bias estimates toward the null. Lastly, parental mental health, and parenting behavior—all important contributors to child neurodevelopment—were not available in the dataset.

Despite these limitations, the study’s strengths include a large, nationwide cohort, long-term follow-up of up to 13 years, and rigorous analytic methods. Clinically, these results indicate an association between prenatal exposure to oral glucocorticoids and long-term neurodevelopmental risk. Because causality cannot be inferred, the findings should inform shared risk–benefit discussions rather than constitute treatment recommendations or guideline changes.

In conclusion, prenatal exposure to oral GC was associated with a small but consistent increase in the long-term risk of LNDD in offspring. While oral GCs are commonly used during pregnancy to manage maternal autoimmune or inflammatory conditions, our findings highlight the importance of considering potential neurodevelopmental consequences for the child. Future research should disentangle medication effects from underlying maternal conditions and identify potentially vulnerable subgroups or exposure windows. These findings indicate an association between prenatal oral glucocorticoid exposure and a small absolute increase in long-term LNDD risk. They should inform shared risk–benefit discussions while confirmatory studies clarify causality and identify vulnerable subgroups.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Conceptualisation: In-Ae Song, Tak Kyu OhData curation and formal analysis: Tak Kyu OhMethodology and supervision: In-Ae SongWriting – original draft: Tak Kyu OhWriting – review & editing: In-Ae SongGuarantor: In-Ae Song.

Funding

This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.

Data availability

Individual-level data are held by NHIS and available under license per NHIS policies; aggregated results and analysis code are available from the corresponding author on reasonable request and with NHIS permission.

Declarations

Competing interests

The authors declare no competing interests.