Pregnancy medication safety still rests on incomplete evidence
Pregnancy remains one of the least studied periods in medicine, even though medication use is common and often unavoidable. The evidence gap is not theoretical. It affects bedside decisions every day, from antiemetics and antibiotics to analgesics and long-term treatment for epilepsy, asthma, and other chronic disease. Safety data are still thin because pregnant people have historically been excluded from clinical trials, leaving clinicians to make decisions from observational evidence that is often incomplete, delayed, and difficult to interpret.
A reminder of the stakes sits behind every discussion of fetal drug exposure. The thalidomide disaster, with its fetal abnormalities and miscarriages, remains the classic warning of what can happen when medication risk is misunderstood. More recent concerns about antiepileptics and neurocognitive delay, along with public debate about acetaminophen and neurodevelopmental differences, show that the problem has not gone away. In routine care, the question is often not whether a pregnant person should take a medication, but how much uncertainty can reasonably be tolerated when the mother’s health and the fetus’s development are both in play.
Observational studies are doing the work trials have not
Because pregnant people have been excluded from most trials, observational research has become the main source of medication safety information in pregnancy. Recent examples in PLOS Medicine illustrate how much of the current evidence base depends on retrospective data, registry linkage, and careful confounding control. One cohort study found no association between first-trimester nonsteroidal anti-inflammatory drugs exposure and major congenital malformations. Other recent work found no association between first-trimester macrolide use and major malformations, or between antibiotic exposure and childhood autoimmune disease. A meta-analysis of prenatal exposure to asthma medications did report an association with autism spectrum disorder, but the evidence remains inconclusive because confounding by indication cannot be ruled out.
This is the central problem in pregnancy pharmacoepidemiology. The mother who needs treatment is not interchangeable with the mother who does not. Underlying illness, disease severity, and pregnancy-related complications all shape outcomes, and they can also shape who receives treatment in the first place. Severe maternal illness increases the risk of miscarriage, prematurity, and fetal growth restriction, so an apparent drug signal may actually reflect the illness that prompted treatment. For some conditions, such as cystic fibrosis or epilepsy, withholding treatment may be more dangerous than continuing it. For milder conditions, the risk-benefit balance looks different.
One sentence captures the clinical reality: maternal illness and fetal outcome travel together.
Trial inclusion is improving, but it will not close the gap alone
Policy changes from the US Food and Drug Administration and the European Medicines Agency are moving in the right direction by encouraging inclusion of pregnant people in certain clinical trials. Even so, broad inclusion on paper will not quickly solve the evidence deficit. Pregnancy complicates trial design in ways that standard adult studies do not face. Few participants will become pregnant during a trial, and some will be pregnant only part of the way through follow-up. Physiologic changes during pregnancy can alter pharmacokinetics, risks, and response, which means larger samples may be needed to detect clinically meaningful effects. Efficacy questions may also require subgroup analyses specific to pregnant participants, making underpowered studies a serious concern.
For many medicines, the more realistic path remains observational. Administrative datasets can support pharmacovigilance and target trial emulation, but only if investigators handle time carefully. Exposure status changes during pregnancy. Concomitant treatments change. Gestational age changes. The pregnancy itself changes. Those moving parts make temporal ordering unusually fragile, and a simplistic model can distort causality. Rare outcomes and long-latency outcomes are particularly hard to capture, especially when fetal or child effects emerge years after exposure rather than during pregnancy itself.
Registries need better detail, not just bigger numbers
The article argues for a more systematic reporting infrastructure that gathers all available pregnancy exposure data in one place. That includes pregnancies that begin during trials, exposures that occur before pregnancy is recognized, individual case reports, and reports from drug safety information services such as teratology counseling networks. Registries remain essential, but they need more precise fields if they are to support reliable dose-response analysis. Gestational age at exposure and dose are not optional details; they are the variables that determine whether a signal is biologically meaningful or just statistical noise.
Linkage between maternal and child records is equally important. Without that connection, fetal exposure can be documented but long-term child outcomes can be missed. Neurodevelopmental effects, in particular, require extended follow-up. A pregnancy registry that stops at delivery leaves the most difficult questions unanswered. Harmonized data collection across registries would also make pooled analyses more feasible, which matters because many individual registries are too small to answer rare safety questions on their own.
A more forceful idea appears in the piece: pharmaceutical companies could be required to complete registry studies for every drug they bring to market. That would shift some of the burden away from scattered academic groups and charity funders and make safety surveillance a built-in part of drug development rather than an afterthought.
Placental pharmacology can separate exposure from speculation
Better epidemiology needs a biological partner. Placental perfusion models, cell systems, and physiologically based pharmacokinetic modeling can help estimate whether and when a medicine crosses the placenta. Biomarkers measured in cord blood, neonatal blood spots, or hair can further quantify fetal exposure. These approaches matter because observed associations are easier to interpret when fetal exposure itself is known. If a drug does not meaningfully cross the placenta, a later child outcome is less likely to be explained by direct fetal exposure and more likely to reflect maternal illness, shared genetics, or other confounding factors.
In practice, placental pharmacology helps prioritize which associations deserve the closest scrutiny. It also helps determine whether a signal has biologic plausibility. Without that context, epidemiology can overcall risk or miss it entirely.
Method standards will determine whether the evidence is usable
Data alone will not solve the problem. Agreement on analytic methods is just as important. Perinatal stages are hard to define cleanly, and outcomes are not captured with equal completeness. Early pregnancy loss is especially difficult to record, and different medication classes may require different causal frameworks. A uniform analytic approach that ignores these realities will produce inconsistent answers even when large datasets are available.
The authors call for collaborative, systematic efforts to generate timely analyses for all appropriate medications in pregnancy. They also argue for methodological standards that improve consistency and reporting. That is the part most likely to determine whether future studies can be compared, pooled, and trusted.
Pregnancy will never be a simple exposure window. The evidentiary challenge is to treat it as a high-stakes biologic state with changing physiology, changing risk, and changing treatment needs. The remaining barriers are concrete: inadequate registry linkage, incomplete long-term follow-up, inconsistent methods, and too little integration between epidemiology and placental pharmacology. Until those problems are addressed, medication safety in pregnancy will remain one of the most consequential blind spots in clinical medicine.