Counsel on teratogens and medication safety in pregnancy and lactation

In one line

Teratogenic risk is governed by timing, dose, agent and host susceptibility, never by a drug being labelled "unsafe"; the consultant task is to weigh the absolute risk a medicine adds against the often larger risk of leaving the maternal disease untreated, and to counsel with real numbers rather than a letter on a packet.

Mechanism & pathophysiology

Teratogenesis is not random toxicity. Wilson's principles still organise the whole field, and every counselling decision falls out of them.

Timing is the dominant variable. The conceptus passes through windows of differing vulnerability. In the first two weeks (pre-implantation to early implantation) exposure tends to be "all-or-nothing" — a lethal insult causes loss, a survivable one leaves an apparently intact embryo, because the cells are still totipotent and can replace damaged neighbours. Organogenesis (≈ weeks 3–8 post-conception, i.e. ≈ 5–10 weeks' menstrual age) is the critical period for structural malformation: this is when most major teratogens do their damage, and a drug taken only after it has closed (the second and third trimesters) cannot cause the classic malformation even if it is otherwise toxic. Beyond organogenesis the fetus remains vulnerable to functional and growth effects — the brain develops across the whole of pregnancy (hence valproate's neurodevelopmental signal), and the kidney and ductus arteriosus are second/third-trimester targets (ACE-inhibitors, NSAIDs). The single most useful counselling fact is therefore often when the exposure happened relative to organogenesis, not whether it happened.

Dose matters, and most teratogens have a threshold. Below a certain exposure the embryo's repair capacity is not overwhelmed; above it, risk climbs steeply. Valproate is the clean example — major malformation risk rises sharply above roughly 1000 mg/day — and the corollary is that the lowest effective dose of a necessary drug is itself a teratogen-reduction strategy, not a compromise.

Genetic susceptibility decides who is harmed. The same exposure produces a malformed child in one pregnancy and a normal child in another because maternal and fetal genotype modulate the response — variation in folate metabolism, in drug-metabolising enzymes, and in developmental pathways. This is why a teratogen produces a probability, not a certainty, and why a previously affected pregnancy raises the risk for the next.

Mechanism and access complete the picture. Each teratogen acts through a specific molecular pathway: alcohol disrupts Sonic Hedgehog and retinoic-acid signalling and is directly cytotoxic to migrating neural-crest cells, which is why the facial midline and brain are the signature targets; retinoids flood the same retinoic-acid pathway that patterns the pharyngeal arches; folate antagonists (methotrexate, and to a degree valproate and carbamazepine) impair the one-carbon metabolism that closes the neural tube; warfarin inhibits vitamin-K-dependent γ-carboxylation of bone matrix proteins, producing chondrodysplasia punctata. And the agent must reach the embryo — almost all small lipophilic drugs cross the placenta freely, whereas large molecules (heparins, most monoclonal antibodies before the late-pregnancy FcRn-mediated transfer window, insulin) largely do not, which is itself a safety lever.

Against all of this sits the background population risk: roughly 2–3% of all babies have a major congenital malformation with no identifiable cause, and the figure climbs to 4–5% if minor anomalies are counted. No counselling number is interpretable without it. A drug that "doubles the risk" of a 1-in-1000 anomaly still leaves 998 of every 1000 babies unaffected; a drug that adds 8 percentage points to a baseline of 3% is a different conversation entirely. Relative risk frightens; absolute and attributable risk inform. It also bears stating that only a small minority of malformations are drug-attributable at all — most are genetic, chromosomal, multifactorial or unexplained — so a malformation after an exposure is not proof of causation, and counselling must hold both the real risk a teratogen adds and the large background that would have produced anomalies regardless.

The corollary of the timing principle is reversibility. Structural teratogenesis, once organogenesis has miscarried a developmental step, is fixed — the cleft, the neural tube defect or the limb anomaly cannot be undone in utero. Several functional and second/third-trimester effects, by contrast, are reversible if the exposure is withdrawn: the oligohydramnios of an ACE-inhibitor or an NSAID, and the ductal constriction of an NSAID, recover when the drug is stopped. This distinction — fixed structural injury versus reversible functional effect — decides whether an exposure demands detailed counselling and a planned anomaly scan or simply prompt drug withdrawal and surveillance.

The labelling problem — why a letter never answered the question

For decades the FDA letter categories (A, B, C, D, X) were treated as a safety grade, and they were never that. The category reflected the quality and reassurance of the available data, not the magnitude of risk, so two drugs in the same category could carry wildly different real-world hazard, and "no human data" defaulted a drug to category C — which prescribers then read as a mild warning rather than as ignorance. The system actively misled: it implied a hierarchy that did not exist and collapsed the timing/dose nuance that actually governs harm.

The FDA replaced it on 30 June 2015 with the Pregnancy and Lactation Labeling Rule (PLLR), which abolishes the letters and substitutes three narrative subsections — Pregnancy (8.1), Lactation (8.2), and Females and Males of Reproductive Potential (8.3) — each summarising the actual data, the background risk for context, and clinical considerations. The change is conceptual, not cosmetic: it forces the prescriber back to reading the evidence and the timing rather than reaching for a letter. The Australian TGA categories (A, B1–B3, C, D, X) persist and are still seen in SA-adjacent practice, and they carry the same flaw — they conflate evidence quality with risk and must be read with the same scepticism. The practical lesson is that "no data" is not "safe": the commonest counselling error is to reassure on a drug simply because no harm has been demonstrated, when the truth is that it has not been studied.

Assessment — quantifying and communicating risk

The exposed or about-to-be-exposed woman needs the same structured approach whoever she is.

• Establish exactly what, how much, and when. Drug name, dose, route, and the gestational window of exposure mapped against organogenesis. A first-trimester exposure to an agent whose only risk is third-trimester (an NSAID, an ACE-inhibitor) is, on the structural question, reassurable; a second-trimester exposure to a pure first-trimester teratogen has already missed its window of harm. Get the dates right before anything else.
• Use a real teratogen information source, not memory. UKTIS (and its patient-facing arm BUMPS — Best Use of Medicines in Pregnancy), Reprotox/TERIS, and MotherToBaby exist precisely because individual recall of drug-specific risk is unreliable. In SA, the SAMF and the NDoH guidelines anchor what is available and recommended locally; the teratology services anchor the risk figure. Consult them in front of the patient.
• Translate risk honestly. Give the absolute risk ("about 11 in 100, against a background of about 3 in 100") and the attributable risk (the ~8 in 100 added by the drug), not the bare relative risk. State the uncertainty where it exists. And always present the counterfactual: the risk of the untreated maternal disease, which for uncontrolled epilepsy, severe depression, a mechanical heart valve or untreated HIV may dwarf the drug's teratogenic risk.
• Document the discussion. Where a known teratogen has been taken, offer the appropriate detailed anomaly scan (typically the 18–22-week fetal anomaly scan, supplemented where the window allows) and fetal echocardiography where the agent targets the heart (lithium, valproate, retinoids), and record the shared decision — this is medicolegally load-bearing, the consent principles of which are developed in clinical-ethics-consent. Where a neural-tube-defect risk applies (valproate, carbamazepine, methotrexate exposure), a careful early scan of the cranium and spine and the option of maternal serum or imaging follow-up belong in the plan.
• Separate the two clocks: the woman who is already pregnant and exposed, and the woman planning pregnancy on a teratogen. The first needs accurate retrospective risk quantification and surveillance; the second needs pre-conception substitution and optimisation, which is where almost all the avoidable harm can be designed out. A chronic-disease drug review is most useful before a positive test, not after — most teratogenic damage from valproate, methotrexate, mycophenolate, warfarin and ACE-inhibitors is preventable only at the planning stage.

Management — the principle and the specific agents

The governing principle is constant: the lowest effective dose of the safest effective drug, and do not stop essential therapy. Abruptly withdrawing anticonvulsants, antipsychotics, antiretrovirals, immunosuppression or anticoagulation to "protect the baby" is one of the most damaging errors in obstetric medicine — maternal seizures, relapse of psychosis, organ rejection, thrombosis or vertical HIV transmission harm the fetus far more than a well-chosen drug. The work is substitution and optimisation before conception, not avoidance during it.

Alcohol and fetal alcohol spectrum disorder

There is no established safe threshold and no safe trimester; alcohol is a direct neurotoxin and the leading preventable cause of intellectual disability. The full picture is fetal alcohol spectrum disorder (FASD) — from the complete fetal alcohol syndrome (the characteristic facial triad of short palpebral fissures, smooth philtrum and thin upper lip; growth restriction; CNS dysfunction) through to alcohol-related neurodevelopmental disorder with no dysmorphology at all. This is the defining teratology issue of South African obstetrics: the Western Cape carries the highest documented FASD prevalence in the world (community studies of 135–209 per 1000 children, with one recent estimate near 310 per 1000), driven by the historical dop system and entrenched binge patterns. The intervention is population-level and pre-conceptional — there is no antenatal treatment that reverses the injury, so brief alcohol screening (a validated tool, asked without judgement) and referral at every antenatal contact, and abstinence counselling before pregnancy, are core obstetric work, not social-work overflow. The diagnostic difficulty compounds the burden: the facial phenotype is subtle and fades with age, much of the spectrum has no dysmorphology at all, and maternal alcohol history is under-reported, so FASD is consistently under-recognised and the true prevalence is almost certainly higher than even the SA figures suggest. The consultant message is unambiguous — there is no "safe glass", no safe trimester, and the only effective treatment is prevention delivered before and around conception.

Retinoids (isotretinoin)

The most reliably teratogenic prescription drug in routine use. Continued exposure past about day 15 post-conception carries a structural anomaly risk of roughly 25–35%, targeting the structures patterned by retinoic acid: CNS, craniofacial (microtia/anotia, micrognathia, cleft palate), conotruncal cardiac and thymic defects, plus later neurocognitive impairment. It is contraindicated in pregnancy; effective contraception must continue and, because of tissue sequestration, it should be stopped at least one month before conception. The pregnancy-prevention frameworks (iPLEDGE internationally) exist because a single course taken unknowingly in early pregnancy is catastrophic — every prescription to a woman of childbearing potential demands a documented contraception plan, and topical retinoids are best avoided too even though systemic absorption is low.

Sodium valproate — the agent that most needs a programme

Valproate is the highest-risk anticonvulsant and the clearest example of dose-dependent, dual (structural and neurodevelopmental) teratogenicity. First-trimester monotherapy carries a major malformation risk of up to ~11% (against a ~3% background), with neural tube defects (spina bifida), cardiac, orofacial cleft and hypospadias the signature anomalies, and risk rising sharply above 1000 mg/day. Beyond structure, in-utero valproate reduces childhood IQ and roughly triples the rate of autism spectrum disorder, with 30–40% of exposed children needing additional educational support — a harm that organogenesis-timed thinking misses, because the brain is vulnerable across the whole pregnancy.

The consequence is a formal pregnancy-prevention programme. The MHRA's PREVENT scheme holds that valproate must not be started in girls or women of childbearing potential unless there is no effective alternative and she is enrolled — annual specialist review, a signed Risk Acknowledgement Form, and highly effective contraception (UK guidance has since tightened to dual specialist sign-off and now extends caution to men). The South African position mirrors this: a SAHPRA/manufacturer Dear-Healthcare-Professional letter advises against valproate in women of reproductive age, yet SA studies show it remains over-prescribed to exactly this group — a live local safety gap, not an academic one. The pre-conception move is to switch to levetiracetam or lamotrigine (see below) and continue high-dose folic acid 5 mg daily, never to stop an essential anticonvulsant in someone with genuine valproate-responsive epilepsy without a neurologist's plan, where the seizure risk is itself dangerous. Generalised epilepsy management overlaps the wider picture at neurological-disorders-in-pregnancy.

Other anticonvulsants

The EURAP registry's comparative data drive the modern hierarchy. Major malformation rates are lowest for levetiracetam (~2.5%) and lamotrigine (~3.1%) — at or near background — making them the preferred agents in pregnancy. Topiramate (~4.9%) adds orofacial cleft and reduced birthweight (and, separately, undermines hormonal contraception). Carbamazepine (~5.4%) carries a neural-tube-defect signal, lower than valproate but real. The unifying message is that anticonvulsant choice should be optimised before conception toward levetiracetam/lamotrigine, that monotherapy at the lowest controlling dose beats polytherapy, and that 5 mg folic acid is standard for all of them.

Warfarin

A first-trimester teratogen with a precise window: exposure between roughly the 6th and 9th (up to 12th) post-menstrual week produces warfarin embryopathy — nasal hypoplasia and stippled epiphyses (chondrodysplasia punctata), from inhibition of vitamin-K-dependent bone-matrix carboxylation — with CNS and haemorrhagic risks later in pregnancy. Risk is dose-dependent and substantially lower (though not absent) at ≤5 mg/day. The hardest case is the woman with a mechanical heart valve, where warfarin is the most effective anticoagulant but the most teratogenic, and the low-molecular-weight-heparin alternatives carry a higher maternal valve-thrombosis risk. The trade-off is brutal and quantitative: warfarin throughout gives the best protection against valve thrombosis but exposes the first-trimester fetus to embryopathy, while switching to LMWH for the first trimester removes the embryopathy risk but, even with anti-Xa monitoring, carries a higher maternal thrombosis and mortality risk. Practice individualises by the warfarin dose — many units continue warfarin where the maintaining dose is ≤5 mg/day (low embryopathy risk) and switch to dose-adjusted LMWH where it is higher, converting back to warfarin in the second trimester and to heparin again near delivery. There is no single correct answer; the decision is shared, anti-Xa-monitored, and documented, and the broader anticoagulation reasoning lives with the antiphospholipid and thrombophilia work at sle-and-antiphospholipid-syndrome.

ACE-inhibitors and ARBs

The signature second- and third-trimester fetopathy: fetal renal hypoperfusion causes oligo- or anhydramnios, neonatal acute kidney injury or renal dysplasia, skull hypocalvaria, pulmonary hypoplasia and limb contractures — a recognisable, sometimes lethal pattern. First-trimester exposure does not cause the classic fetopathy (fetal urine only drives liquor volume from the second trimester), and although an older signal suggested a first-trimester cardiac/CNS malformation excess, that is confounded by the underlying maternal hypertension and diabetes and is no longer held to be a strong independent effect. The practical rule is to switch to a pregnancy-compatible antihypertensive (labetalol, nifedipine, methyldopa) pre-conception or at the first positive test, and not to panic-terminate over a first-trimester exposure.

Methotrexate and mycophenolate

Both are absolute contraindications that demand a pre-conception switch. Methotrexate (a folate antagonist, also used as an abortifacient) produces aminopterin-type embryopathy — craniosynostosis, skull hypoplasia and wide fontanelles, limb defects, growth restriction — above roughly 10 mg/week in a window around 6–8 weeks; it must be stopped, with a washout, before conception and replaced where ongoing immunosuppression is needed. Mycophenolate mofetil is among the most teratogenic immunosuppressants, with malformation in roughly a quarter of exposed pregnancies: microtia/anotia with external-auditory-canal atresia, cleft lip and palate, ocular coloboma, and cardiac and limb defects, plus increased early loss. The pre-pregnancy plan for the transplant or lupus patient is to switch to azathioprine (compatible) well before conception — a decision that belongs to pre-pregnancy counselling, not to the first booking visit.

Lithium

The teaching point here is corrected fear. Older case-register data claimed a 400-fold increase in Ebstein's anomaly, and a generation withheld lithium on that basis. The large modern cohort (Patorno, NEJM 2017) found cardiac malformation in 2.4 vs 1.2 per 100 — an adjusted risk ratio of about 1.65, dose-dependent and concentrated in right-ventricular-outflow defects — and the pooled meta-analysis found no significant cardiac excess and no Ebstein cases at all. Lithium is therefore a modest, dose-dependent cardiac teratogen, not a forbidden drug: in a woman whose bipolar disorder genuinely needs it, the answer is the lowest effective dose, level monitoring (with the volume-of-distribution shifts of pregnancy and the peripartum period in mind), a fetal echocardiogram, and continued treatment — because relapse of severe bipolar disorder in pregnancy and the puerperium is itself dangerous.

The late-pregnancy and miscellaneous group

• NSAIDs: the FDA's 2020 labelling change advises avoidance from 20 weeks (fetal renal prostaglandin inhibition → oligohydramnios and neonatal renal impairment) and from about 30 weeks because they constrict the ductus arteriosus, risking persistent pulmonary hypertension. Short early-pregnancy use is broadly tolerated; the danger is the third trimester.
• Tetracyclines discolour teeth and deposit in bone after about 14–16 weeks and are avoided from then; first-trimester exposure is not a structural teratogen but alternatives exist.
• Live vaccines (MMR, varicella, yellow fever) are avoided in pregnancy on theoretical grounds, and conception is deferred ~4 weeks after them — though inadvertent exposure has not been shown to cause the predicted harm and is not an indication for termination. The inactivated vaccines that matter (influenza, Tdap, and COVID-19) are actively recommended.
• Antithyroid drugs illustrate the trimester-switch principle: carbimazole/methimazole carry an embryopathy (aplasia cutis, choanal and oesophageal atresia) concentrated in the first trimester, so propylthiouracil (with its own small hepatotoxicity risk) is preferred in the first trimester, often switching back after organogenesis — a clean example of choosing the agent by window. Thyroid management in pregnancy is developed at thyroid-disorders-in-pregnancy.
• Ondansetron is the live controversy in routine antenatal prescribing: a small absolute increase in orofacial cleft has been reported in some datasets and not others, so it is reasonable for refractory hyperemesis when first-line antiemetics fail, counselling the woman that the absolute risk (if real) is very small against a background cleft rate — neither banning it reflexively nor reaching for it first.
• Phenytoin and phenobarbital are older anticonvulsants with their own fetal hydantoin/barbiturate syndromes and are not first choices where levetiracetam or lamotrigine are available; they are mentioned because they still appear on constrained formularies and the same lowest-effective-dose, 5 mg folate logic applies.

Psychiatric, autoimmune, anticoagulation and HIV — the "do not stop" disciplines

Perinatal mental health is the area where teratogen anxiety most often produces harm. Untreated severe depression and bipolar disorder carry real maternal and fetal risk, and abrupt cessation of psychotropics drives relapse; the right move is informed continuation of the safest effective agent (most SSRIs are broadly compatible; the late-pregnancy considerations are neonatal adaptation and a small persistent-pulmonary-hypertension signal, not structural teratogenesis) rather than reflexive withdrawal. The same "do not stop" logic governs autoimmune disease (substitute mycophenolate/methotrexate for azathioprine/hydroxychloroquine, do not simply stop), mechanical-valve and antiphospholipid anticoagulation, and — critically in SA — HIV: antiretroviral therapy must be continued throughout for prevention of vertical transmission (PVT), where stopping ART to "avoid drugs in pregnancy" would be a grave error that the modern SA regimens (tenofovir/lamivudine/dolutegravir-based) are specifically designed to avoid. Dolutegravir is illustrative of how teratogen science self-corrects: an early SA-derived neural-tube-defect signal (the Tsepamo surveillance) prompted caution, but with accumulating data the absolute risk attenuated to close to background, and current SA guidance recommends dolutegravir-based ART in pregnancy with periconceptional folate — the over-reaction would have been to withhold the most effective, best-tolerated regimen on an early, later-attenuated signal.

Medication in lactation

Most maternal drugs reach milk at a low relative infant dose (RID) — the proportion of the weight-adjusted maternal dose the infant receives — and an RID under ~10% is generally considered compatible. The variables that raise infant exposure are a long maternal half-life, high oral bioavailability in the infant, a high milk-to-plasma ratio, and a premature or very young infant with immature clearance. The practical approach is to check a curated resource (LactMed, UKDILAS/SPS) rather than reach for a blanket "avoid", to time doses where a short-half-life drug allows, and to recognise that the agents genuinely incompatible with breastfeeding are few. The reflex of stopping breastfeeding "to be safe" usually trades a real benefit for a theoretical risk; agents requiring genuine caution include lithium, amiodarone, retinoids, antineoplastics and radioisotopes, while most analgesics, most antibiotics, most antihypertensives and most antidepressants are compatible. The decision is never the drug alone — a thriving term infant with a compatible drug needs no change; a sick or preterm neonate, or a drug with a long half-life and high infant bioavailability, shifts the calculus toward closer monitoring or, occasionally, a held feed around peak concentration.

Guidelines compared

The substantive disagreement is small and mostly about labelling philosophy (narrative vs categorical) and regulatory strictness (valproate). On the clinical facts — the agents, the windows, the doses — the bodies broadly agree; the SA-specific divergence is one of access and local burden (EML-constrained substitution, the FASD epidemic, persistent valproate prescribing) rather than of risk science.

The evidence & the controversy

Three threads are worth holding at consultant level. First, the first-trimester ACE-inhibitor question: the early reports of a cardiac/CNS malformation excess have not survived adjustment for the underlying maternal cardiometabolic disease, so the modern position treats inadvertent first-trimester exposure as a reason to switch agents and scan, not as a reason to terminate — over-reacting here is its own harm. Second, the lithium recalibration is the cleanest example in teratology of a "known" fact being overturned by better data: the 400-fold Ebstein figure was an artefact of biased early registries, and withholding lithium from a woman who needs it now causes more harm than the modest, dose-dependent cardiac risk the cohort data actually show. Third — and increasingly examined — is the wider environmental-exposure debate: endocrine-disrupting chemicals and phyto-oestrogens (bisphenols, phthalates, the much-discussed compounds in some sanitary and cosmetic products) are biologically plausible developmental disruptors, but the human teratogenicity evidence is contested and largely observational, and the honest position is to acknowledge the signal, avoid clearly avoidable exposures, and resist both alarmism and dismissal until the data mature. The thread that runs through all three is the same discipline: weigh attributable risk against the counterfactual, distrust both the reassuring letter and the frightening anecdote, and let the timing and the numbers — not the label — drive the plan.

Landmark trials & key evidence

Exam traps & red flags

• Stopping essential therapy "to protect the baby." Withdrawing anticonvulsants, antiretrovirals, immunosuppression, anticoagulation or psychotropics typically harms mother and fetus more than the drug — the answer is substitution and optimisation before conception, not avoidance during it.
• Reading "no data" as "safe." A drug that has never been studied in pregnancy is unknown, not benign; reassuring on absence of evidence is the commonest counselling error.
• Trusting the letter category. A/B/C/D/X (and TGA letters) grade evidence quality, not risk magnitude; the PLLR abolished them for exactly this reason — counsel from the data and the timing.
• Quoting relative risk to a frightened patient. Without the ~2–3% background and the attributable risk, "doubles the risk" is meaningless and needlessly terrifying.
• Panicking over a post-organogenesis or post-window exposure. A first-trimester NSAID or a single inadvertent live vaccine is not an indication for termination; map the exposure against its actual window before counselling.
• Missing the SA FASD burden. Alcohol is the leading preventable cause of intellectual disability and SA's defining teratology problem; not screening for it at antenatal contacts is a systemic miss.
• Forgetting valproate in a young woman. Continuing valproate in a woman of childbearing potential without a pregnancy-prevention programme or a pre-conception switch is both a clinical and a regulatory failure.
• Over-restricting breastfeeding. Most maternal drugs are compatible (RID <10%); stopping breastfeeding reflexively trades real benefit for theoretical risk.
• Terminating over a first-trimester ACE-inhibitor. The classic fetopathy is a second/third-trimester effect; switch the drug and scan rather than over-react.

Evidence anchors

• Patorno et al. — Lithium Use in Pregnancy and the Risk of Cardiac Malformations, N Engl J Med 2017
• Munk-Olsen et al. — Lithium in pregnancy: international collaborative meta-analysis of six cohorts, Lancet Psychiatry 2018
• Tomson et al. — Comparative malformation risk of eight anticonvulsants (EURAP), Lancet Neurol 2018
• Tomson et al. — Dose-dependent malformation risk of anticonvulsants (EURAP), Lancet Neurol 2011
• Meador et al. — Cognitive function at 3 years after fetal antiepileptic-drug exposure (NEAD), N Engl J Med 2009
• Meador et al. — Fetal antiepileptic-drug exposure and cognitive outcomes at age 6 (NEAD), Lancet Neurol 2013
• UKTIS — Use of sodium valproate in pregnancy
• UKTIS — Use of angiotensin II receptor antagonists in pregnancy
• UKTIS — Use of mycophenolate mofetil in pregnancy
• FDA — Pregnancy and Lactation Labeling Rule resources
• AAFP — FDA implements new pregnancy/lactation labelling (PLLR overview), 2016
• South African Medicines Formulary (SAMF) and NDoH Maternity Care / HIV (PVT) guidelines — locally available agents, valproate-in-women-of-childbearing-potential cautions, and ART continuation in pregnancy (no stable public URL).
• Teratogen information services for risk quantification at the point of counselling: UKTIS/BUMPS, Reprotox/TERIS, MotherToBaby; LactMed for lactation.