Prenatal substance exposure and birth weight: Findings from the HEALthy Brain and Child Development Study

Abstract

Objective:

Estimate associations of more than minimal prenatal nicotine, alcohol, cannabis, and opioid exposures with gestational age, birthweight, and birthweight-for-gestational age.

Methods:

Data were drawn from the HEALthy Brain and Child Development (HBCD) Study, a multisite, longitudinal study in the United States. Predefined recruitment thresholds for each substance were assessed using maternal self-report, maternal toxicology, and newborn substance exposure-related diagnoses. Birth outcomes included gestational age at delivery (weeks), birthweight (grams), and birthweight-for-gestational age (centiles). Mean differences and risk ratios for the associations between substance exposure and birth outcomes were estimated using multilevel mixed-effect linear regression or multilevel mixed-effect Poisson regression.

Results:

Among 660 mother-infant dyads, 17% (n=115) of participants met recruitment thresholds for prenatal cannabis, 15% for nicotine (n=102), 13% for alcohol (n=86), and 5% for opioids (n=32). In adjusted models, prenatal cannabis and opioid exposures were each associated with lower birthweight (cannabis: −272.2g (95% CI −444.6, −99.8); opioids: −295.4g (95% CI −574.9, −15.9) and birthweight centiles (cannabis: −8.2 centiles (95% CI −15.3, −1.1); opioids: −14.4 centiles (95% CI −25.5, −3.4) although results were sensitive to model specifications. Prenatal nicotine and alcohol estimates were in similar directions but not statistically significant. No significant associations between exposures and gestational age at delivery were detected.

Conclusions:

In this initial HBCD Study data release, more than minimal exposure to cannabis and opioids was associated with smaller birth size, adding evidence to an inconsistent literature. Future studies from HBCD can more deeply interrogate timing and dose of each substance, and expand to childhood outcomes.

Introduction

In the United States, approximately 20% of pregnancies involve substance exposure,¹ most commonly nicotine (8-11%), alcohol (7-10%), and cannabis (4-8%).^1-3 Among pregnant individuals with substance use, polysubstance use is common.³ Fetal exposure to substances has been linked with a variety of physical and behavioral health outcomes.^4,5 Some of the earliest reported associations of prenatal substance exposure are shorter gestations and lower birthweight or size-for-gestational age^6-13 –factors also known to be associated with later negative neurodevelopmental functioning in children.^14-16 However, findings related to prenatal substance exposure and adverse birth outcomes are limited and inconsistent,^17-20 leading to uncertainty about risks for pregnant individuals and clinicians.^21-25 Inconsistent results are often due to small sample sizes, reliance on recall and self-report, limited adjustment for maternal factors, including co-use, of other substances, and limited data on dose, timing, and duration of substance exposure.^26-29

To characterize associations of prenatal substance exposure, large prospective studies that collect detailed information on substance exposure during gestation are needed. The HEALthy Brain and Child Development (HBCD) Study is the largest US study of brain development from birth into middle childhood.⁵ The HBCD Study protocol, described in detail elsewhere,^{4,5,38-44,30-37} establishes the type of deeply phenotyped data currently lacking to elucidate relationships between prenatal substance exposure and early childhood outcomes. The study began recruitment in 2023 with the goal to recruit a generally representative sample of over 7,000 pregnant individuals and their offspring, followed through the first decade of life.

To coincide with the first public release of HBCD data, and add well-characterized, prospectively collected evidence to the existing literature on prenatal substance exposure and birth outcomes, we examined associations of more than minimal nicotine, alcohol, cannabis, and opioid exposures with gestational age, birthweight, and birthweight-for-gestational age. By highlighting protocol elements pertaining to prenatal substance use and providing analytic code, this paper may guide future investigators analyzing HBCD data on prenatal substance exposures and childhood outcomes.

Methods

Study design:

Each of 27 recruitment sites enrolled a pregnant sample from their site’s catchment area,⁵ relying on diverse methods including social media platforms, obstetric clinics, and substance use treatment facilities.³¹ Per protocol,⁵ participants were enrolled in the second or third trimester of pregnancy. Inclusion criteria were English- or Spanish-speaking pregnant individuals 18 years or older; exclusion criteria were any plan to terminate the pregnancy or not to retain custody of the child. To ensure sufficient exposure to study opioids, alcohol, cannabis and nicotine/tobacco, the study was designed for at least 25% of the sample to have more than minimal prenatal substance use (12% each exposure accounting for co-use between groups). Other substances were queried but were not specified in recruitment targets.

The HBCD study releases cumulative data annually. Data for this analysis were drawn from the HBCD 1.0 data release, which included all completed visits by July 1, 2024. The resulting sample contained 1,426 participants with a prenatal visit 1 (V1: maternal health and substance use, biospecimens) and 660 with a visit 2 (V2: 0-1 month post-delivery: pregnancy and birth outcomes). Additional details on study measures are available in prior papers, ^{4,5,38-42,30-37} the study website (www.HBCDstudy.org), and the data documentation website (https://docs.hbcdstudy.org).

All adult participants consented for themselves and their children, and the study was approved by a single institutional review board (IRB) in addition to IRBs of all involved institutions. Fully de-identified data were obtained through the National Institute of Health Brain Development Cohorts Data Hub (https://www.nbdc-datahub.org/) under an institutional data use certification.

Exposures, outcomes and covariates:

Prenatal substance exposure classification was based off self-report, biospecimens, and infant substance-related diagnoses as follows. At V1, individuals self-reported substance use using the NIDA-modified ASSIST,⁴⁵ expanded for the HBCD Study from 9 to 18 substances.⁴ Participants with reported substance use in the 3 months prior to or during pregnancy were subsequently interviewed by trained research staff about timing and frequency of use using a Timeline Followback (TLFB) method.⁴⁶ The TLFB asked about substance use each day for each endorsed substance for 5 weeks during pregnancy: 4 early weeks of pregnancy (last menstrual period (LMP) + 2 weeks through LMP + 6 weeks) and the week before enrollment. At V2, participants reported whether they used substances between their prenatal visit (V1) and delivery; an endorsement was followed by a TLFB interview to collect daily substance use for each endorsed substance during the last 2 full weeks of gestation. In addition, all participants (regardless of substance use report) were asked to provide a urine sample (V1) and nail clippings (V1 and V2) for toxicology, and at V2 reported any infant diagnoses of fetal alcohol syndrome (FAS; alcohol) or neonatal opioid withdrawal syndrome (NOWS; opioids). From the 7 weeks of self-reported substance use on the TLFB, toxicology and maternal-reported FAS and NOWS diagnoses, participants were categorized into each exposure group when they met or exceeded threshold definitions (as defined in Table 1⁴). Individuals not meeting substance exposure thresholds and with negative toxicology were included in the general population sample.

Table 1.

Thresholds of prenatal exposure used to assess enrollment goals for the HEALthy Brain and Child Development Study

Opioid Use

Any self-reported use of prescribed (including medications for opioid use disorder) or illicit opioids for 2 or more weeks during pregnancy in the Timeline Follow Back, OR

Diagnosis with Neonatal Opioid Withdrawal Syndrome, OR

Positive toxicology for an opioid for research-collected biospecimen.

Alcohol Use

Any self-reported use equal to 7 or more standard drinks per week for 2 or more weeks during pregnancy in the Timeline Follow Back, OR

Any self-reported use equal to 3 or more standard drinks per occasion on 2 or more occasions during pregnancy in the Timeline Follow Back, OR

Diagnosis with Fetal Alcohol Syndrome, OR

Positive toxicology for alcohol for research-collected biospecimen.

Cannabis Use

Any self-reported use of cannabis for 4 or more weeks during pregnancy in the Timeline Follow Back, OR

Positive toxicology for cannabis for research-collected biospecimen.

Nicotine / Nicotine Use

Any self-reported use of nicotine or nicotine products for 4 weeks or more during pregnancy in the Timeline Follow Back, OR

Positive toxicology for nicotine for research-collected biospecimen.

© Copyright 2025 by Elsevier. All rights reserved. Used/adapted with permission from Gurka et al, Dev Cogn Neurosci. 2025.⁴

Birth outcomes were assessed as both continuous and categorical variables. Gestational age from maternal report was calculated from the birth date (in floored gestational weeks) minus the study calculated LMP (EDD-40 weeks). Preterm birth (dichotomous) was defined as <37 completed gestational weeks. Low birth weight (LBW; dichotomous) from maternal report was defined as birthweight <2500 grams. Sex-specific birthweight-for-gestational age centiles were calculated using Intergrowth curves.⁴⁷ Small-for-gestational age (SGA; dichotomous) was defined as birthweight <10^th percentile for gestational age and sex.

Covariates for adjustment were selected based on hypothesized associations with prenatal substance exposure and birth outcomes. Covariates included self-reported maternal race-ethnicity (non-Hispanic White, non-Hispanic Black, Hispanic, Other race (Asian, American Indian/Alaska Native, two or more races)), age at delivery (years), gravidity (number of pregnancies, including current pregnancy), self-reported maternal education (dichotomized at high school diploma or equivalent versus above), health insurance payer (private, public, other source of insurance, no insurance), Supplemental Nutrition Program for Women, Infants and Children (WIC) benefits receipt (Yes/No), and maternal depression (measured with the total score of the Edinburgh Postnatal Depression Scale (EPDS)⁴⁸, all collected at V1. Although race and ethnicity are social constructs, they were included in analytic models, along with age and proxies for income, due to the documented disparities among minority, young and lower-income individuals in the receipt of substance use treatment^49,50 and in adverse birth outcomes,⁵¹ both likely due to structural factors.

Statistical analysis

Exposure was dichotomized at or above enrollment thresholds compared to below thresholds or no substance use for analysis. Analytic code to classify exposure thresholds is referenced in Supplemental Material. Mean differences for associations between prenatal substance exposure and continuous birth outcomes were estimated using multilevel mixed-effect linear regression with a random effect for the study site. Adjusted models were specified in two ways; adjusted mean differences were first estimated using the same linear regression model with the addition of inverse probability of treatment weighting (IPTW) to account for the effect of covariates. Models were then repeated with multivariable adjustment with the same covariates. The four substances were included in the same model to estimate associations independent of each other. Risk ratios for associations between prenatal substance exposure and binary birth outcomes were estimated using a multilevel mixed-effect Poisson regression with a robust variance estimator and study site as a random effect. Adjusted risk ratios were estimated using the same Poisson model with the addition of IPTW and were also repeated with multivariable adjustment. The estimation of inverse probability weights has been described previously⁵² and is detailed in the supplement. In a sensitivity analysis, the reference group for each substance was disaggregated into no exposure and exposure below the thresholds, with no exposure serving as the reference group in multivariable adjusted models. Complete case analysis was performed for all models (5% missing overall in sample). Cell sizes under 11 were suppressed. All analyses were performed in Stata SE 18.5.

Results

There were 660 participants included in the analytic sample from HBCD 1.0 (Figure 1). The sample was predominantly non-Hispanic White (51%) followed by non-Hispanic Black or African American (21%) (Table 2). Most participants had post-secondary education (81%) and private insurance (60%); 24% of participants used WIC benefits, and 10% had EPDS scores indicating probable depression (≥13).⁴⁸ Individuals who met prenatal substance exposure criteria were more likely to identify as non-Hispanic Black or African American, be insured through Medicaid, receive WIC benefits, screen positive for probable or possible depression on the EPDS, and were less likely to have education beyond a high school diploma. Characteristics by each substance are listed in Supplemental Table 2.

Figure 1.

Cohort selection of completed visits eligible for inclusion in the HBCD data release 1.0

Table 2.

Description of HEALthy Brain and Child Development (HBCD) study cohort (release 1.0), by any substance use

	Total	None or below threshold substance use	At or above threshold substance use
	n=660	n=432	n=228
Substance use parameters
Alcohol (any source)	86 (13%)	0 (0.0%)	86 (38%)
Cannabis (any source)	115 (17%)	0 (0.0%)	115 (50%)
Nicotine (any source)	102 (15%)	0 (0.0%)	102 (45%)
Opioids (any source)	32 (5%)	0 (0.0%)	32 (14%)
Maternal and pregnancy characteristics
Maternal race
Non-Hispanic White	338 (51%)	244 (56%)	94 (41%)
Non-Hispanic Black or African American	136 (21%)	60 (14%)	76 (33%)
Hispanic	82 (12%)	58 (13%)	24 (11%)
Other race (includes >1 race)	86 (13%)	60 (14%)	26 (11%)
Maternal age at delivery; years (median, IQR)	33 (29-36)	33 (30-36)	31 (26-34)
Number of pregnancies (median, IQR)	2 (1-4)	2 (1-3)	2 (1-4)
Maternal education (> high school diploma)	537 (81%)	389 (90%)	148 (65%)
Insurance payer
Private	395 (60%)	317 (73%)	78 (34%)
Medicaid	159 (24%)	61 (14%)	98 (43%)
Other (Tricare, Healthcare.gov, Indian Health Service, etc.)	71 (11%)	40 (9%)	31 (14%)
No Insurance	29 (4%)	12 (3%)	17 (7%)
Supplemental Nutrition Program for Women, Infants, and Children	159 (24%)	67 (16%)	92 (40%)
Possible depression (EPDS ≥9)	158 (24%)	81 (19%)	77 (34%)
Probable depression (EPDS ≥=13)	68 (10%)	33 (8%)	35 (15%)
Edinburgh Postnatal Depression Scale score (median, IQR)	5 (2-8)	4 (2-7)	5 (3-10)
Infant sex (male)	335 (51%)	224 (52%)	111 (49%)

Substance use

Prenatal substance exposure criteria were met by 17% of participants for cannabis (n=115), followed by nicotine (15%; n=102), alcohol (13%; n=86), and opioids (5%; n=32) (Table 2); 432 (65%) did not meet criteria for any of the four exposures. The specific criteria (self-report, toxicology, or infant diagnoses) that determined prenatal substance exposure status varied by substance (Supplemental Figures 1-4). For alcohol, the most common source was self-report alone (88%). For cannabis, the most common sources were self-report, nail and urine toxicology (30%), followed by nail toxicology alone (19%). For nicotine, the most common sources were self-report, nail and urine toxicology (34%), followed by urine toxicology alone (18%). For opioids, the predominant sources were self-report, nail and urine toxicology (41%).

We also examined single versus polysubstance prenatal exposure (Supplemental Figure 5). The most common single prenatal substance exposure was alcohol alone (10%; n=60), followed by cannabis alone (7%; n=45). The most common prenatal polysubstance exposure was a combination of nicotine and cannabis (6%; n=40). Additional combinations of substance exposure by substance are described in supplemental files. Generally, alcohol was the substance most likely to be used alone, while nicotine was the least likely substance to be used alone.

Infant outcomes

Gestational age at delivery

The median gestational age at delivery was 39 weeks (interquartile range (IQR): 38-40). In unadjusted and adjusted models, point estimates for the four exposures signaled a reduction in completed gestational weeks, but were not statistically significant (Figure 2, Supplemental Table 3). Eight percent (n=53) of the sample were born prematurely. In models for preterm birth (Supplemental Table 4), there were no statistically significant associations.

Figure 2.

Beta estimates for substance exposure and gestational weeks at delivery. Adjustment via inverse probability of treatment (IPTW) (blue) and multivariable adjustment (MVA) (green).

Birthweight

The median weight at birth was 3260 grams (IQR: 2920-3600). In unadjusted models, each substance exposure was associated with lower birthweight, although alcohol exposure was not statistically significant (Figure 3, Supplemental Table 3). In adjusted models, opioid and cannabis exposure remained significantly associated with a reduced birthweight, although cannabis estimates were sensitive to model specification. Eight percent (n=52) of the sample was LBW. In models for LBW (Supplemental Table 4), none of the exposures were statistically significant.

Figure 3.

Beta estimates for substance exposure and birthweight. Adjustment via inverse probability of treatment (IPTW) and random effect for site (blue) and multivariable adjustment (MVA) and random effect for site (green).

Birthweight-for-gestational age

The median birthweight-for-gestational age was 65th percentile (IQR 38-86). Opioid, cannabis, and nicotine exposures were significantly associated with reductions in birthweight-for-gestational age, while estimates for alcohol exposure were not (Figure 4, Supplemental Table 3). After adjustment, cannabis and opioid exposure remained negatively associated with birthweight-for-gestational age centile, while estimates for nicotine exposure were attenuated. Five percent (n=32) of the sample were small for gestational age at birth. In models for small for gestational age (Supplemental Table 4), there were no significant associations.

Figure 4.

Beta estimates for substance exposure and birthweight centile. Adjustment via inverse probability of treatment (IPTW) and random effect for site (blue) and multivariable adjustment (MVA) and random effect for site (green).

Sensitivity analysis

Most infants in each reference group had no exposure to that substance (negative toxicology, no report on TLFB, no NOWS/FAS: alcohol: n=402 (70% of reference group); cannabis: n=519 (95%); nicotine: n=544 (97%), opioids: n=620 (99%)). When the reference group was disaggregated into ‘no exposure’ and ‘exposure below threshold,’ results were unchanged when comparing exposure at or above thresholds to no exposure (Supplemental Table 5, Supplemental Figures 6-8). Estimates of exposure falling below thresholds for each substance were null, and typically attenuated relative to estimates of use above the thresholds, although all confidence intervals for the two exposure groups overlapped.

Posthoc analyses

To test the robustness of the findings, we repeated multivariable linear models for cannabis with exposure limited to positivity from biospecimens (with or without self-report; n=105 exposed) and separately from self-report (with or without biospecimens (n=75 exposed) for birthweight and birthweight for gestational age. There was little difference between source of exposure (Supplemental Table 6, Supplemental Figure 9) for either outcome. There were too few opioid-exposed infants to stratify results by positivity source.

Discussion

In data from the HBCD Study, more than minimal levels of prenatal cannabis and opioid exposure were associated with lower newborn birthweight and birthweight-for-gestational age. Cannabis findings were robust when examined by source of exposure (toxicology vs. self-report). Estimates from nicotine and alcohol models were not statistically significant. These findings are relevant given the inconclusive literature and uncertainty among pregnant individuals and clinicians regarding select prenatal substance exposures, particularly cannabis.^21-25

In our US birth cohort study, participants met exposure thresholds for common substances, including cannabis (17%), nicotine (15%), alcohol (13%), and opioids (5%). Because of oversampling for opioids^53,54 and use of thresholds to define ‘more than minimal’ substance exposure, HBCD prenatal substance exposure estimates are not directly comparable with substance use rates in population-based samples like the National Survey on Drug Use and Health¹ or PRAMS.³ However, HBCD prevalence estimates by information source are informative to other studies, as most data collection relies on self-report. Although over 90% of alcohol use in HBCD was captured with self-report, only 66% of cannabis, 62% of nicotine, and 72% of opioids were captured by self-report (data not fully shown due to cell suppression). The prevalence discordance between sources was expected; toxicology classified any positive result as expose, while self-report required meeting use thresholds, and urine, fingernail and toenail specimens vary in toxicology look back periods.⁵⁵ Nevertheless, this approach likely provides more inclusive exposure ascertainment, and future work should continue to analyze whether findings differ by source.

In the HBCD sample, co-use of the four substances was common. Although alcohol was primarily used alone, most who used nicotine, cannabis, and opioids also met criteria for at least one other substance. These findings underscore the importance of considering co-exposure in clinical care and research.

Our findings with respect to birth outcomes were nuanced. All four substances of interest have been associated with lower gestational age, birthweight, and birthweight-for-gestational age in prior studies,^9-11,13,56 although the literature is inconsistent and many studies are noted as having low methodological quality.^17-20 In our models, many findings attenuated upon adjustment, reflecting confounding common in studies of prenatal substance exposure. Nicotine models appeared most affected by confounding, which in post hoc analyses were largely driven by race-ethnicity, education, and depression. In addition, approximately 40% of participants with self-reported nicotine had discontinued use by enrollment (data not shown), potentially minimizing the strength of associations with birth outcomes.⁵⁷ For all substances, our imprecise estimates are also a function of exposure definitions, as pre-defined exposure levels reflect more than minimal use, which may include lower levels of use relative to other prenatal exposure studies. Further, substances were aggregated across types (e.g. opioid exposure consists of medications for opioid use disorder and prescription and nonprescription products), which increases heterogeneity in estimates.

Although the analytic sample for the HBCD Study will continue to grow over the next several years, these early findings may contribute to prenatal counseling. Misperceptions of cannabis as ‘natural’ or ‘safe’ persist, and studies note a lack of clear information or communication about cannabis-related risks.^58,59 Our findings support public health messaging and clinical recommendations to discontinue cannabis use in pregnancy.⁶⁰ Moreover, our findings on opioids and lower birthweight may assist clinicians and pregnant people when preparing for opioid-exposed births. Notably, prenatal agonist-medication treatment for opioid use disorder (OUD) is an evidence-based approach, with documented reductions in maternal morbidity and mortality and improved birth outcomes. This analysis was unable to separate medications for OUD (n=17) from other opioid exposures (n=16, with overlap between the two) due to the small sample size, and also could not consider timing of treatment initiation. As the HBCD sample grows, those analyses may be possible.

These findings should be viewed considering study limitations. Although each site attempted to enroll a generally representative sample from their catchment area, recruitment denominators were not captured and representativeness cannot be determined. These initial findings from the first HBCD data release are based on <10% of the anticipated final sample size of the HBCD Study, resulting in imprecise estimates and models that were sensitive to model specifications. In addition, we did not assess the mode of use (e.g., vaping versus edibles), type of use (e.g., opioid-agonist medication treatment versus nonprescription opioids), dose, timing, duration, or joint effects of exposures, although this information will be available in future data releases. Additionally, although the HBCD Study recruitment design specified these four relatively common prenatal exposures, the study is collecting detailed information on 18 substances, including stimulants, which have increased in use.^61-63 The prevalence of remaining exposures remains low, and we did not include other exposures due to small cell sizes. This may have resulted in biased estimates. It is also possible that some prenatal substance exposure was misclassified. In order to minimize participant burden and potential recall bias, only 7 weeks of pregnancy were queried on TLFB interviews,⁴ and toxicology has variable look-back periods.⁵⁵ Both could lead to misclassification of exposure, potentially attenuating results. It should be noted, however, that the combination of the TLFB interviews and biomarkers assessed in both maternal urine and nail clippings is one of the most comprehensive in the field. Additionally, although beyond the scope of this analysis, future analyses can look at maternal nutrition as a mediator and modifier of the relationship between prenatal substance exposure and birth outcomes, which has previously been documented.⁶⁴ Finally, we relied on maternal report of birthweight and length of gestation, which may have resulted in misclassification and should be validated in future data releases when medical record data are available.

In conclusion, we found that prenatal cannabis and opioid exposures were each associated with smaller birth size. These findings align with existing recommendations to abstain from cannabis in pregnancy and support the need for disaggregated analyses of opioid exposure.

Article summary:

In early data from the HEALthy Brain and Child Development Study, prenatal cannabis and opioid exposures were associated with lower birthweight and birthweight-for-gestational age.

What’s known:

Prenatal substance use during pregnancy is highly prevalent but its potential impact on offspring outcomes remains poorly understood. Large, well characterized datasets with both self-report and toxicology are necessary to fully characterize and analyze prenatal substance exposure.

What this study adds:

Our findings of an association between both cannabis and opioid exposures and birth size add to an inconsistent literature. Future studies from this growing cohort can confirm and expand findings into childhood outcomes.

Abbreviations

EPDS

Edinburgh postnatal depression scale

EDD

expected due date

HBCD

HEALthy Brain and Child Development

IPTW

inverse probability of treatment weighting

LBW

Low birthweight

MVA

multivariable analysis

SGA

Small for gestational age

TLFB

Timeline Followback

WIC

Women, Infants and Children