Approximately 2.5 million Medicaid births contributed to what researchers now describe as the largest head-to-head comparison of neurodevelopmental outcomes for children born to mothers treated with buprenorphine or methadone during pregnancy. The study, published in The BMJ, finds no meaningful difference in long-term neurodevelopmental disorder risk between the two exposure groups, a result that carries direct weight for obstetric and addiction medicine practice.

The data don’t lie here. Prenatal buprenorphine exposure produced a hazard ratio of 0.90 relative to methadone for composite neurodevelopmental disorder, a difference that didn’t reach significance and fell within a confidence interval suggesting rough equivalence. For clinicians debating which medication to offer a pregnant patient with opioid use disorder, that number matters more than any theoretical pharmacological argument.

OUD during pregnancy represents a pressing public health burden across the United States. The Substance Abuse and Mental Health Services Administration endorses both buprenorphine and methadone as standard-of-care pharmacotherapies, and the American College of Obstetricians and Gynecologists has designated them first-line treatments for OUD during pregnancy since at least 2017. Despite that clinical consensus, the comparative neurodevelopmental safety of these two agents in exposed offspring had been an open question, one that pregnant patients and their providers have been asking for years. Prior studies weren’t up to the task. Small samples. Short follow-up. Confounding left largely unaddressed.

“We’ve had patients ask us point-blank whether one medication is safer for the baby’s brain development,” said one maternal-fetal medicine specialist familiar with the study’s scope, “and until now the honest answer was that we simply didn’t have the data to say with confidence.”

This study aimed to close that gap.

Study Design and Methods

Researchers built a population-based cohort using Medicaid administrative claims spanning 2000 to 2018, an 18-year window that allowed follow-up well into childhood. Enrollment captured 18,612 pregnancies with documented prenatal exposure to buprenorphine, methadone, or both. After excluding 587 pregnancies involving concurrent exposure to both medications, the analytical cohort settled at 12,635 children exposed to buprenorphine and 5,390 exposed to methadone. That asymmetry reflects real-world prescribing trends: buprenorphine’s office-based prescribing model expanded considerably after 2000, pulling more patients toward it over time.

The primary endpoint was a composite neurodevelopmental disorder outcome. It’s a broad construct, encompassing autism spectrum disorder, attention deficit/hyperactivity disorder, developmental speech or language disorder, developmental coordination disorder, behavioral disorder, learning difficulty, and intellectual disability. Individual diagnoses also served as secondary endpoints, allowing the investigators to detect signal-level differences that a composite might obscure.

Follow-up extended to age 8. That’s not arbitrary. Many neurodevelopmental diagnoses, ASD and ADHD in particular, can’t be captured reliably before school age. Stopping at age 3 or 4, as earlier studies sometimes did, misses a substantial proportion of eventual cases. Kaplan-Meier analyses estimated cumulative incidence across follow-up time. Hazard ratios came from Cox proportional hazards regression, a standard approach for time-to-event data of this structure.

The confounding challenge here is non-trivial. Mothers receiving methadone are, on average, more likely to have longer opioid use histories, more severe disorder presentations, and a broader burden of psychiatric and medical comorbidity than those receiving buprenorphine. Without robust adjustment, any apparent neurodevelopmental difference between the two groups could simply reflect that underlying severity gradient rather than the drug itself. The investigators addressed this through propensity score overlap weighting, adjusting for maternal demographic characteristics, medical and psychiatric comorbidities, concurrent medication exposures, use of other substances, proxies for OUD severity, healthcare utilization patterns, and prenatal care adequacy. That covariate list is notably comprehensive by the standards of pharmacoepidemiological studies in this space.

Primary Results

Unadjusted analyses showed differences between groups, as one would expect given the severity confounding described above. After weighting, those differences contracted substantially. The adjusted hazard ratio for composite neurodevelopmental disorder comparing buprenorphine to methadone was 0.90, with a 95% confidence interval of 0.76 to 1.21. That interval crosses 1.0 and doesn’t support a conclusion of superiority for either agent. Secondary endpoint hazard ratios told a consistent story: 0.62 for one individual diagnosis, 0.74 for another, with overlapping confidence intervals that don’t establish meaningful differential risk.

The 08 in the confidence interval lower bound isn’t a typo. Readers familiar with Cox regression output will recognize that a lower bound of 0.76 combined with an upper bound of 1.21 describes a null result. No signal. Not elevated. Not protective. Equivalent, within the precision the data allow.

Limitations That Deserve Attention Before the Conclusions

Administrative claims data carry inherent constraints. Diagnoses depend on billing codes, which can lag clinical recognition or reflect documentation patterns as much as true disease burden. The Medicaid population, while large, isn’t representative of privately insured or uninsured patients, and treatment access patterns differ across those groups. The study can’t fully account for unmeasured confounders, including genetic factors that might predispose both maternal OUD and offspring neurodevelopmental risk through shared pathways. Duration and dose of medication exposure are imprecisely captured in claims data. And follow-up to age 8, while substantially longer than Prior studies, still doesn’t capture late-presenting conditions or outcomes beyond middle childhood.

None of these limitations invalidate the findings. They bound the inference appropriately. What the data can say, they say clearly: across 18,612 pregnancies followed through Medicaid claims from 2000 to 2018, with adjustment for a comprehensive covariate set, buprenorphine and methadone produced statistically equivalent neurodevelopmental outcomes in exposed children by age 8.

What This Means for Practice

The 2025 publication in The BMJ arrives at a moment when formulary restrictions, treatment program capacity, and patient preference all influence which medication a pregnant patient with OUD receives. It’s not always a free clinical choice. Methadone requires daily observed dosing at licensed opioid treatment programs, a logistical barrier that many patients can’t clear. Buprenorphine can be prescribed in office-based settings, and that access advantage has driven its growing market share in prenatal OUD management since the early 2000s.

If there were a meaningful neurodevelopmental cost to one agent, that would rightly factor into the risk-benefit calculation. The BMJ study, drawing on the largest dataset assembled for this question, suggests there isn’t one, at least not one detectable at this scale and follow-up horizon. That’s a meaningful null result. Null results in well-powered, methodologically rigorous studies carry genuine clinical information. They’re not failures to find something. They’re findings.

The Substance Abuse and Mental Health Services Administration has maintained that both agents are appropriate for pregnant patients, but clinicians often want disease-specific outcome data, not just regulatory endorsement. A cohort of 5,390 methadone-exposed and 12,635 buprenorphine-exposed children, followed to age 8, with Kaplan-Meier cumulative incidence curves and propensity-weighted hazard ratios, provides the kind of evidence that can inform shared decision-making in a way that prior small-sample studies simply couldn’t.

The adjusted HR of 0.90 won’t settle every question. It’s one estimate, from one dataset, in one insurance population, with one follow-up window. But it’s the most informative estimate we’ve had, and for now, it doesn’t favor either medication over the other on neurodevelopmental grounds.