The First Twelve Weeks
For nearly twenty years, women were told the science was settled against looking for thyroid disease early in pregnancy. It never was. A handful of physicians and scientists, working inside the guidelines, the journals and the trial networks, said so in print, again and again. This is their record, in their own words, and a thank-you.
The window, and where the trials actually started
Where the fetus depends on its mother, where the brain association is strongest, where the profession asked for a trial, and where the trials it relied on actually began.
Sources: ACOG CO 381 and PB 148; Jansen et al. 2019; ATA guidelines 2017 and 2026; Lazarus et al. 2012; Casey et al. 2017; Cooper & Pearce 2017. Every figure is quoted below.
| Row | Timing shown |
|---|---|
| Fetal dependence on maternal thyroid hormone | Before about 12 weeks — ACOG 2007 and Practice Bulletin 148. |
| Brain-structure association | Strongest near 8 weeks; undetectable after about 14 weeks — Jansen, Korevaar and colleagues, 2019. |
| Early trial requested | ATA 2017: 4–8 weeks. ATA 2026: 6–10 weeks. |
| Trials actually run | CATS: median 13 weeks 3 days. MFMU: mean 16.7 weeks. |
Start here
I am pregnant and want the short version
Start with the timeline and Questions to ask your clinician.
I am a clinician or researcher
Read the timeline, the trials section, and the 2019 debate.
I want to check a quotation
Open the quotation register and its verification legend.
Four people who made the case from inside the profession
Four profiles introduce the people whose work is quoted in the chapters that follow.
None of them called for burning anything down. They did the slower and harder thing: they wrote the model, filed the dissent, published the timing data, rewrote the viewpoint five times, and kept asking for the trial. Each of them has something no outside critic has: standing, and a seat at the table.
Antonio Bianco
Endocrinologist and physician-scientist; author of Rethinking HypothyroidismSaid what his profession would not: “a major reckoning is in order.” 2012 · 2017 · 2026Chrysoula Dosiou
Endocrinologist; author of the cost-effectiveness model; the lone dissent on the 2017 ATA task forceProved the case for looking survives losing the argument about mild disease. 2011 – 2022Alex Stagnaro-Green
First author of the 2011 ATA pregnancy guideline; co-author of the 2012 modelMoved from neutral to advocate in public, and said why. 2019 · 2026Toyah Jansen, Tim Korevaar & the Generation R team
Rotterdam; Korevaar later first author of the 2026 ATA guidelineMeasured the clock: strongest at eight weeks, gone by fourteen.“A major reckoning is in order” — Bianco, on his own profession
Bianco's account sets out how patients were dismissed and why he called for a reckoning within his own profession.
The insider who said it first, about himself.
A woman wrote to Antonio Bianco about a knee. She had lived with hypothyroidism for fifteen years and seen four doctors and endocrinologists. None of them would discuss her treatment with her. One leaned over, patted her knee, and said: “You let me worry about all that, dear.” Her summary: “So much for doctor/patient collaboration.”
Bianco reproduced that letter in his book for one reason: he had been that doctor. What makes him the right person to open this page is what he wrote next.
“Frankly, I said these things to my patients without knowing all the facts. I was just repeating what I learned—that is, to be skeptical of combination therapy, and only vaguely knowledgeable about what thyroid extract is.”
“I was just repeating what I learned.”
That sentence explains everything on this page. A doctor receives his views in training and passes them on with a confidence he has never had to defend. An institution does the same thing at scale: a standard is set, the evidence moves, and the sentence does not.
“Before I became aware of the controversies surrounding therapy for hypothyroidism, I would just dismiss these complaints and renew the prescription with what I thought was an appropriate dose of T4; I would not renew a prescription for T3 or thyroid extract, to the frustration of many.”
“My approach to symptomatic patients with hypothyroidism, which in retrospect seems insensitive, was not unique.”
He describes proposing to colleagues that they build a relationship with patients. “They looked at one another and, after a few seconds, laughed.” When the profession finally tried to hear from patients directly, the resistance was to the invitation itself:
“There were discussions and some resistance about a live debate with patients. Physicians are not used to this; I heard, 'We can only invite patients we trust.'”
The survey went ahead anyway, and its results surprised the people who ran it:
“On a scale of 0 to 10 (10 being the maximum), patients returned a median score of 5 to 6 when asked about their satisfaction level with treatment. We were surprised to see that they also returned a score of 5 to 6 when asked about their current physicians. Patients revealed that they had changed physicians multiple times, some up to ten times.”
Limit. A self-selected survey, not a representative sample of everyone with hypothyroidism. Bianco says so himself.
“One comment indicative of refusal to acknowledge fault was: 'Oh yes, we knew about this—there is nothing new in this survey. These represent the emotionally unstable patients with hypothyroidism.'”
And then his conclusion, about his own profession, from inside it:
“A major reckoning is in order, and, in my view, apologies as well.”
In pregnancy the same disagreement acquires a second party, one who cannot give evidence on her own behalf.
The question the guidance never answered
The record starts with the early period of fetal dependence on maternal thyroid hormone and distinguishes looking for disease from deciding how to treat it.
ACOG Committee Opinion 381 (2007), read alongside Casey et al. (2005).
Every guideline on this page agrees on one fact, and that fact is why the rest matters. In 2007 the American College of Obstetricians and Gynecologists wrote it down:
“Before 12 weeks of gestation, a time when the fetal thyroid begins to concentrate iodine and synthesize T4, the fetus is entirely dependent on maternal transfer of thyroid hormones. Brain development begins during this period of fetal dependency in the first trimester and continues throughout pregnancy and on into infancy.”
During those weeks the child has no working thyroid of its own. Whatever the developing brain needs, it takes from the mother’s blood, or it goes without. The same two-page document then counted the women carrying a thyroid problem into that window:
“The prevalence of subclinical hypothyroidism could be anticipated to be between 2% and 5% of women screened.”
Casey had already raised the timing concern in 2005
Two years before ACOG’s opinion, Brian Casey and colleagues had described the early dependency window:
“Maternal thyroxine is particularly critical early in pregnancy because the fetal thyroid gland cannot synthesize iodothyronines until after 10 weeks of gestation.”
In the Discussion, they explicitly raised the possibility that starting therapy after ten weeks would not reverse impairment already established, and cited Pop and colleagues on treatment timing:
“It may be that therapy beginning after 10 weeks of gestation would not eliminate any already established fetal neurodevelopmental impairment from hypothyroxinemia. Pop and colleagues have provided evidence that treatment may be ineffective only if given after this time.”
ACOG’s 2007 opinion nevertheless stated:
“However, data indicating fetal benefit from thyroxine supplementation in pregnant women with subclinical hypothyroidism currently are not available.”
“Without evidence that identification and treatment of pregnant women with subclinical hypothyroidism improves maternal or infant outcomes, routine screening for subclinical hypothyroidism currently is not recommended.”
Each step is honest, and the conclusion is defensible on its own premises. But a woman reading it is entitled to ask the question the document never answers: if you knew all that, why did nobody go and find out? And what should a profession do in the meantime, with two to five women in every hundred, during the weeks when the fetus depends entirely on her?
Screening and treatment are separate questions
Should we test every pregnant woman? That is a question about screening. If we find thyroid disease, what should we do about it? That is a question about treatment. They have different evidence and different owners, and for twenty years they were answered together, as though a judgment about the second settled the first. The people honoured below are the ones who pulled them apart.
Source order. The ACOG quotations are ordered to show fetal dependence, prevalence, the absence of demonstrated benefit, and the screening recommendation. Casey’s 2005 timing passages sit between prevalence and ACOG’s conclusion. The two Casey excerpts come from the Introduction and Discussion, respectively; they are not consecutive sentences in the paper. In CO 381 itself, the abstract gives the recommendation before the body gives the biology and prevalence.
What screening costs: Dosiou's 2012 model
Dosiou's model and dissent frame the economic argument for screening, including what the model assumed about treatment benefit.
Cost-effectiveness model, 2012 · lone dissent, ATA 2017 · co-author, ATA 2026.
The objection to screening always sounds decisive: we cannot recommend looking, because we have not proven that what we find is worth treating. In 2012, Chrysoula Dosiou answered it with arithmetic.
Three terms carry her work. Cost-effective means the health gained is worth the money spent, measured against a threshold the health system already uses. A QALY is one year of life in good health. Incremental means compared with the next-best option, not with doing nothing.
“A decision-analytic model compared the incremental cost per quality-adjusted life-year (QALY) gained among the following: 1) universal screening, 2) high-risk screening, and 3) no screening. Screening consisted of a first-trimester thyroid-stimulating hormone level and antithyroid peroxidase antibodies. Women with abnormal results underwent further testing and, when indicated, levothyroxine therapy.”
Read the last clause carefully: when indicated. She did not model treating every woman with a slightly raised TSH. She modelled looking, confirming, and treating when warranted. She modelled an ordinary 25-year-old in her first trimester with no known thyroid history.
“Risk-based screening and universal screening were both cost-effective relative to no screening, with incremental cost-effectiveness ratios (ICERs) of $6,753/QALY and $7,138/QALY, respectively. Universal screening was cost-effective compared with risk-based screening, with an ICER of $7,258/QALY.”
“…an ICER of $7,258 per QALY, well below the accepted threshold for cost-effectiveness of $50,000 per QALY.”
Then she did what critics demand and almost nobody does: she stripped out the benefit everyone disputes.
“Even in a scenario that assumed no benefit of treating subclinical hypothyroidism, universal screening remained cost-effective.”
The case for looking does not require the case for treating mild disease to succeed.
Here is exactly what remained once mild disease was assumed to cause no harm and to go untreated:
“In a hypothetical scenario in which subclinical hypothyroidism has no adverse outcomes and does not get treated, and the only benefits of screening consist of detection and treatment of overt hypothyroidism and closer follow-up of detected anti-TPO antibody positive women, with more prompt diagnosis of PPT and overt hypothyroidism when they develop, our results were substantially similar to the base case.”
- women whose thyroid is failing outright are found and treated;
- women carrying thyroid antibodies are followed rather than lost;
- postpartum thyroiditis is diagnosed promptly instead of months later, or never;
- hypothyroidism in the years after delivery is caught early.
And if treating mild disease did help children’s intelligence, the arithmetic turned around completely:
“Universal screening was most cost-saving, providing an additional 14.1 quality-adjusted days and saving $1,533 compared with no screening.”
“Universal screening is also highly cost-effective compared with screening of only high-risk women. … The medical community should seriously consider screening all pregnant women for AITD.”
The Endocrine Society’s own guideline admits the gap her model prices:
“…the committee acknowledges the important data confirming that such case finding will unfortunately miss 30% or more of women with overt or subclinical hypothyroidism.”
Limit. A model estimates; it does not observe. Its inputs on prevalence, outcome weights and follow-up could be challenged, and a different set could move the number. What the zero-benefit scenario shows is that the result does not depend on winning the argument about mild disease. Her base case assumed a treatment benefit; it is the sensitivity analysis that survives without one.
The 2017 dissent: one member, printed by initials
Major guidelines are negotiated for months so that clinicians receive one clean instruction. Single-member dissents printed in the body text are rare. Below Recommendation 93 of the 2017 American Thyroid Association guideline, there is one.
“One task force member (CD) dissented from this recommendation, feeling that universal testing for maternal TSH and anti-TPO antibodies soon after pregnancy confirmation is warranted given the existing support for the obstetrical benefits of treatment, with minimal risk of harm with appropriate monitoring. There might be an additional benefit of treatment on fetal neurodevelopment, as shown in animal studies. Unfortunately, the timing of initiation of LT4 treatment after completion of the first trimester in the RCTs that examine child IQ as the outcome is a very serious design flaw that biases these studies to producing negative results, since they miss the relevant window of opportunity to influence neurodevelopment. Nonetheless, universal screening would allow identification and treatment of all women with overt hypothyroidism, which is prevalent in 0.3%–0.5% of pregnant women (259,260,583) and asymptomatic in about 70% of patients (606). It would also allow identification and selective treatment of women with autoimmune subclinical hypothyroidism, and appropriate close monitoring of euthyroid women with isolated thyroid autoimmunity, some of whom will develop hypothyroidism during gestation or PPT.”
Four members of that 2017 task force had sat on the Endocrine Society’s pregnancy committee five years earlier, when five named members recommended screening every pregnant woman by the ninth week. One of the five was John Lazarus, lead investigator of CATS, the negative trial whose timing was by then the central question. The document records one dissent. It does not record how anyone else voted. The only disagreement it printed came from the member who had done the arithmetic.
The dissent also sets out a whole program in a few lines: find and treat overt disease, treat autoimmune subclinical disease selectively, and closely monitor antibody-positive women. Nine years later, the association’s own figures supplied the number that makes risk-based screening hard to defend:
“Various studies have shown that risk factors recommended in the previous version of these guidelines would lead to screening of 55–78% of all pregnant women, while detection rates varied between 75% and 85% for overt hypothyroidism and between 54% and 60% for subclinical hypothyroidism.”
The chosen strategy already tests most pregnant women, and it still leaves up to a quarter of the overt disease undetected. Everyone agrees that disease must be treated.
How the ATA cited Dosiou's model: behind a condition
“Importantly, if a screening paradigm is conclusively proved effective, universal screening of women in the first trimester has been shown to be cost effective.”
“In sensitivity analyses, even when the benefits of screening were limited to detection and treatment of overt hypothyroidism, screening was shown to be highly cost-effective at <$8000/quality-adjusted life year (QALY).”
If. The finding was real, it was cited, and it was placed behind a condition the same document said had not been met. Her work was not suppressed. It was subordinated: it appeared in the discussion while the recommendation above it went the other way.
In 2026 Chrysoula Dosiou was an author of the association’s guideline again. She was in the room every time. The disagreement is real, reasonable people looked at the same arithmetic, and her case has to be won on its merits. It is still the clearest case on this page.
Stagnaro-Green: find the disease, then argue about treatment
Stagnaro-Green argues for finding disease while keeping the decision to treat separate.
First author, ATA 2011 · co-author of the 2012 model · advocate 2015–2022.
Dosiou made the economic case. Alex Stagnaro-Green made the clinical one, and he was in the room when the economic case was built: he is one of the six authors of the 2012 model. One group of people built both halves of an argument, and the profession answered only one of them.
He is no outsider. He was first author of the American Thyroid Association’s 2011 pregnancy guideline. In 2020 he published a paper whose title is a position:
“Universal screening for thyroid disease during pregnancy should be performed.”
“Limited evidence concerning the impact of untreated and treated sub-clinical disease and thyroid autoimmunity has distracted from the core rationale for universal screening – the beneficial impact of detecting and treating overt thyroid disease. … universal screening for overt disease stands independently from that of subclinical and autoimmune disease. The time to initiate universal screening is now.”
He moved the mild cases out of the way and argued from the disease nobody disputes. How badly does the risk-factor approach miss? His review gathered the studies:
“The values range from 18% to 89%.”
Limit. That range counts overt and subclinical disease together. Overt disease is rarer, and how much of it case-finding misses has not been cleanly measured, which is why the 2026 ATA figure in the Dosiou section matters.
“Targeted screening, which is recommended by most professional associations, fails to detect a large proportion of pregnant women with thyroid disease. In fact, an increasing number of providers are performing universal screening for thyroid disease in pregnancy, contrary to society guidelines.”
Clinicians, in other words, had already begun screening without waiting for permission. Then the sentence that meets Dosiou’s, eight years after hers:
“While there is still debate regarding whether or not treating subclinical hypothyroidism decreases maternal/fetal adverse outcomes, the cost-effective studies demonstrate that universal screening is superior to targeted screening even if one assumes no benefit for identifying and treating subclinical hypothyroidism.”
From "insufficient evidence" to "the time has come"
His own 2011 guideline had been neutral. It was not negative:
“There is insufficient evidence to recommend for or against universal TSH screening at the first trimester visit.”
Six years later, commenting on the association’s revision, he wrote:
“Although agreeing with most of the recommendations, I present an alternative perspective on the normal range of TSH in pregnancy, management of women on levothyroxine preconception and universal screening for thyroid disorders during pregnancy.”
“…I believe that the time for universal screening for thyroid disorders during pregnancy has come.”
| Year | Position |
|---|---|
| 2011 | First author of the ATA guideline: insufficient evidence for or against universal TSH screening. |
| Feb 2015 | JAMA Viewpoint: screen for overt disease, even granting the subclinical argument. |
| 2017 | Nature Reviews Endocrinology: “the time for universal screening … has come.” |
| 2019 | Argues the motion at ENDO 2019 in New Orleans. Full debate → |
| 2020 | “Universal screening for thyroid disease during pregnancy should be performed.” |
February 2015: the JAMA viewpoint ACOG never answered
Two months before ACOG graded its recommendation against universal screening Level A, Stagnaro-Green made the case in a journal that obstetricians and family physicians read. He conceded the contested questions first:
“This Viewpoint discusses these criteria and suggests that sufficient evidence exists for the institution of routine screening to detect and treat overt thyroid disease during pregnancy. However, there are insufficient data to support universal screening to detect subclinical hypothyroidism or the presence of thyroid antibodies.”
“Treatment of overt thyroid disease is universally recommended, regardless of pregnancy status, and thus a placebo-controlled trial of the efficacy of treatment for overt thyroid disease during pregnancy is not required.”
“In a sensitivity analysis, screening solely for overt hypothyroidism had a cost-effectiveness ratio of $6776/QALY … when compared with no screening and $7335/QALY when compared with targeted screening … For comparative purposes, the cost-effectiveness ratio for amniocentesis is $15,000/QALY and screening for gestational diabetes mellitus is $12,078/QALY.”
“However, the lack of data regarding the treatment of subclinical hypothyroidism should not affect the decision to screen for overt thyroid disease.”
He also named, plainly, the reasons that seldom get written down:
“Other concerns regarding routine screening may include that the prevalences of diseases already screened for in pregnancy are extremely low; that once screening for thyroid disease has begun it will be impossible to discontinue …; or that possible medicolegal risks may arise for failure to screen.”
Two months later, ACOG issued Practice Bulletin 148 and graded its recommendation against universal screening Level A. The bulletin did not cite the JAMA piece.
Why Stagnaro-Green changed position
No new trial moved him. CATS was published in 2012, between his neutral guideline and his advocacy, and it was negative. His stated ground was the evidence underneath the threshold itself:
“…only 26% (5 of 19) of the papers cited reported an upper limit of normal ≥ 4.0 mU/l, and in four of the five studies the population evaluated comprised less than 750 women. On the other hand, in the five largest studies cited (each including more than 5,000 pregnant women), only one reported an upper limit of normal exceeding 3.5 mU/l.”
“In essence, a universal TSH demarcation distinguishing the upper range of normal from the lower range of abnormal does not exist.”
If there is no universal line, every threshold is a choice, and that choice decides how many women are labelled. He has also been careful about the limits of his own case:
“However, these studies have some limitations, including that levothyroxine therapy is not initiated until the second trimester, thereby missing the critical time of neurodevelopment.”
Limit. “Second trimester” is his shorthand. CATS began treatment at a median 13 weeks 3 days, right at the boundary. The MFMU trial began at a mean 16.7 weeks, clearly in the second trimester.
The timing evidence: strongest at 8 weeks, gone by 14
This chapter sets out the Generation R timing association and the limits of what an observational finding can show.
Generation R, Rotterdam · Lancet Diabetes & Endocrinology, 2019.
Everything above is argument. This is a measurement. A Rotterdam team followed nearly two thousand mother–child pairs in the Generation R cohort. They had the mothers’ thyroid function measured in pregnancy and the children’s brains scanned by MRI at about ten. And because the blood had been drawn across a range of gestational ages, they could ask what no trial had asked: does it matter when you look?
“Maternal TSH had an inverted U-shaped association with offspring total grey matter volume (p=0.007) and with cortical grey matter volume (p=0.022). … Analyses stratified for gestational age at blood sampling showed an inverted U-shaped association of maternal TSH with child total grey matter volume and cortical grey matter volume, which was most evident at 8 weeks gestation. After about 14 weeks of gestation, TSH was no longer associated with child brain morphology.”
Eight weeks. And gone by fourteen.
“These novel findings suggest that embryonic brain development is particularly vulnerable to altered maternal thyroid function.”
“The absence of association from the 14th week onwards is an important novel finding that should be considered during clinical risk assessment and when timing interventions during clinical practice, as well as in the design of future studies of the effects of levothyroxine treatment for mild thyroid disease in pregnancy.”
In the design of future studies. They told the field, in a journal, that the studies it had been running were timed wrong. Then they said so outright:
“…some randomised controlled trials examining the effect of levothyroxine treatment of maternal subclinical hypothyroidism or hypothyroxinaemia in pregnancy on offspring IQ showed no beneficial effect. The initiation of treatment in these trials, with thyroid hormone replacement therapy starting between 13 and 18 weeks of gestation, was later than our estimated optimal timing.”
| When | What |
|---|---|
| ≈ 8 wk | Association with child grey-matter volume strongest (Jansen 2019) |
| ≈ 14 wk | Association no longer detectable |
| 13 wk 3 d | CATS: median start of treatment, at the edge of the window, with most participants past its peak |
| 16.7 wk | MFMU: mean randomization, beyond it |
Seven years later, Tim Korevaar, the second author of this study, was first author of the 2026 American Thyroid Association guideline. That guideline asked for a trial beginning at six to ten weeks.
Limit. Observational, so confounding is possible. It measures brain structure, not IQ. Women with known or treated thyroid disease were excluded. The median blood draw was at 13.1 weeks, so the eight-week estimate rests on the women sampled earlier. The formal test of whether timing changed the association gave p = 0.053 and p = 0.086, which is suggestive but short of the usual line. It locates where to look for an effect; it does not show that treating there produces one.
The trial the ATA asked for in 2011, 2017 and 2026
The ATA's requests for an early trial are presented alongside the limits recorded in its guidelines.
Task forces of 2011, 2017 and 2026.
One body asked for the decisive trial three times: the American Thyroid Association, in each pregnancy guideline it published. In 2011, before the large trials reported:
“Of concern to the task force is that the double-blind placebo control studies either recently completed, or presently underway, began screening and intervention after the first trimester. As such these studies will not be able to address the impact of LT4 treatment in the first trimester in women with SCH, isolated hypothyroxinemia, or thyroid antibody positivity on the mother and developing fetus. A trial that screens women preconception and then randomizes women with SCH or isolated hypothyroxinemia and TAb+ euthyroid women to either a treatment or no treatment arm is needed.”
Then it answered, in advance, anyone who says such a trial cannot be done:
“The task force is aware of the difficulties inherent in performing such a trial, and the ethical challenges to be faced. Nevertheless, we believe that such a trial is feasible, can be ethically performed with appropriate study design and safeguards, and will yield invaluable information related to the optimal care of the pregnant women and the developing fetus.”
Feasible. Ethically performed. Invaluable. In 2017 the warning returned, nearly word for word:
“Of concern to the task force is that most of the double-blind placebo-controlled studies either recently completed, or presently underway, began screening and intervention after the first trimester. As such, these studies will not be able to address the impact of LT4 treatment in the first trimester …”
The 2017 paragraph is the 2011 paragraph lightly edited. The association carried its own warning forward, intact, into the edition that rested on the trials it warned about. It listed the trial again in its research agenda:
“A RCT of early LT4 intervention (at 4–8 weeks) in women with either subclinical hypothyroidism or isolated hypothyroxinemia to determine effects on child IQ.”
And in 2026, with a wider window and a firmer specification:
“A randomized trial of early LT4 intervention (preconception or early pregnancy, i.e., 6–10 weeks' gestation) in women with either subclinical hypothyroidism or isolated hypothyroxinemia to determine effects on adverse pregnancy outcomes and child IQ.”
“There are no data from randomized trials on early pregnancy LT4 treatment for isolated hypothyroxinemia.”
What the 2026 guideline decided, and what it admitted
“In women without known thyroid disease, it is suggested that TSH testing be offered upon a positive pregnancy test to those at increased risk of thyroid dysfunction during pregnancy.”
That is still risk-based testing, the same policy obstetrics has held since 2007. But a few pages later, in its own rationale, the guideline says:
“The risk factors that are considered an indication for thyroid function testing during pregnancy are not exhaustive. Furthermore, the use of this set of risk factors will not enable the identification of all women with thyroid dysfunction, specifically overt hypothyroidism.”
Set the three statements side by side. The association asks for a trial that must find women at six weeks. It recommends a testing strategy that, by its own account, will not find all of them, overt disease included. And the trials that were run found their patients by screening them when routine antenatal care began. The policy does not forbid the evidence. It leaves every early trial to build its own early contact with pregnant women, from nothing.
“In developing the Guidelines, the task force frequently struggled with the paucity of high-quality double-blinded placebo-controlled trials in the field of thyroid and pregnancy. In fact, only a minority of the 97 recommendations (24%) in the present Guidelines were graded at the highest American College of Physicians Grading System level of evidence.”
That honesty is to the association’s credit. A profession that grades three-quarters of its advice below the top level has already accepted that it must act under uncertainty. The real question is which uncertainties it chose to act on.
Across the same years, ACOG’s three pregnancy-thyroid documents (2007, 2015 and 2020) contain no research recommendation at all.
The trials started too late to test the question
The timing, results, and qualifications of CATS and the MFMU trial are placed side by side.
CATS (Lazarus et al., 2012) · MFMU (Casey et al., 2017) · Cooper & Pearce (2017).
The two trials later treated as settling the question are large, careful and honest. Their authors wrote down what they had done, and that candor deserves thanks too.
“We conducted a randomized trial in which pregnant women at a gestation of 15 weeks 6 days or less provided blood samples for measurement of thyrotropin and free thyroxine (T4).”
Blood was drawn at a median of 12 weeks 3 days. Treatment began at 13 weeks 3 days. In their own discussion, the CATS investigators said:
“A possible explanation for the negative results of our study is that screening was performed and levothyroxine therapy initiated too late in gestation to have a major influence on brain development (up to 20 weeks' gestation; median, 13 weeks 3 days). The timing of screening in our study was chosen because it is at this time that routine antenatal care in hospitals is initiated in the majority of women.”
The reason was not scientific. It was the timetable of antenatal clinics. The American trial’s authors were equally direct:
“A limitation of the two trials is the relatively late time during gestation at which women were randomly assigned.”
“Although we did not enroll women earlier than at 8 weeks of gestation in order to avoid enrolling women who might have an early miscarriage, we believe that these trials, which were performed at centers that were actively screening women on presentation for prenatal care, probably reflect what could be accomplished by means of routine thyroid screening during pregnancy in the United States.”
That same week in the New England Journal of Medicine, David Cooper and Elizabeth Pearce replied:
“However, both the present trials and the CATS trial are limited by the late initiation of treatment. In the present trials, randomization did not occur until an average of 17 weeks of gestation, which is even later than in the CATS trial. … The findings of Casey et al. indicate that screening and treatment for subclinical hypothyroidism, if performed well into the second trimester of pregnancy, are unlikely to be beneficial.”
“However, because more than 75% of women in the United States have their first prenatal visit before 12 weeks of gestation, earlier treatment appears to be feasible.”
The evidence against screening
Both trials also contain findings against the timing explanation, and they belong on this page.
“However, in post hoc analyses, we likewise did not find a benefit of screening and treatment in the subgroup of women who started treatment earlier, although it should be acknowledged that our trial was not powered for this assessment.”
“We also found no significant interaction according to the time of initiation of therapy (<17 weeks of gestation vs. ≥17 weeks of gestation).”
The timing explanation is not established. It is a hypothesis with indirect support: a plausible mechanism, an observational window, and the trial authors’ own belief that they started too late. Neither internal test reached the period the hypothesis is about. CATS’s early subgroup was not powered, and the MFMU split falls at 17 weeks.
The trials were not wrong. The use made of them was.
They showed, durably, that treating subclinical hypothyroidism from the late first or second trimester does not improve children’s cognitive outcomes. They were then cited as though they had shown that looking early does not help, a question they never tested.
To screen, or not to screen?
The 2019 debate brings the competing screening arguments into one session, with the complete transcript available to check.
Moderator David Cooper · for: Alex Stagnaro-Green · against: Kristien Boelaert.
On 26 March 2019, the Endocrine Society staged a formal debate in New Orleans: Thyroid Disease in Pregnancy: To Screen, or Not to Screen? David Cooper moderated. Stagnaro-Green argued for. Kristien Boelaert, an investigator on the TABLET trial and the most credible opponent the motion had, argued against.
“Based on prevalence and impact of overt thyroid disease in pregnancy, which is 1.15%, it is incumbent upon us to do universal screening whether or not the identification and treatment of subclinical hypothyroidism or thyroid autoimmunity has any beneficial effect.”
“I am not making the argument that we should screen all women to diagnose and treat subclinical hypothyroidism. And I am not making the argument that we should screen all women to diagnose and treat women with autoimmune antithyroid disease. My argument is very simple. We have missed the forest for the trees. … So what we are really saying is we are trying to avoid uncertainty at all costs. We are putting our head in the sand. And we are saying we only want to diagnose that what we know how to handle.”
During the questions, Cooper asked whether screening identifies women early enough for treatment to matter, and whether it should be done before conception. Boelaert answered first:
“I agree. Screening should be done preconception. As I've alluded to.”
“The intervention trials that have been done, you can see that treatment was actually started late. And, you know, I would agree that both the CATS trial and the Casey study treatment was started too late.”
“These studies are exceedingly difficult, and it's probably a bit of wishful thinking that we will get decent data from intervention studies where this has actually started early enough.”
She went on to argue for targeted case-finding and closed with “we should not screen all women during pregnancy.” But the final slide of her case read: Let’s conduct RCTs early in pregnancy. On the timing, the two sides agreed. They disagreed about what to do for patients while the right window went untested.
She also made the strongest point anyone made against this page, and it deserves its place here:
“So I know it is small numbers of patients but even in these 50 women, nearly 50 women that were treated, there was no difference. So again I put it to you, are we overegging things a little bit?”
A subgroup of about fifty was never powered to detect a difference in child cognition. Every guideline cited here, including her own position, still recommends treating overt hypothyroidism. The randomized evidence for treating overt disease in pregnancy is thin for the same reason the early window is untested: nobody has run the trial.
Special sectionThe complete 2019 debate, minute by minute
The full 60-minute transcript with every slide in place, the before-and-after audience polls, and the key moments marked.
READ THE DEBATE →The ACOG recommendation that changed by two words
This chapter compares ACOG's recommendation language and how it describes the positions of other societies.
Practice Bulletins 148 (2015) and 223 (2020).
Against all of that stood a sentence. In 2015, ACOG placed this among its Level A recommendations, the tier it reserves for “good and consistent scientific evidence”:
“Universal screening for thyroid disease in pregnancy is not recommended because identification and treatment of maternal subclinical hypothyroidism has not been shown to result in improved neurocognitive function in offspring.”
In 2015 that claim was narrow, and it was defensible: the available trials measured cognition, and the cognition result was null. The same bulletin’s discussion also said:
“Currently, there is no evidence that identification and treatment of subclinical hypothyroidism during pregnancy improves these outcomes.”
In 2020 the revised bulletin kept the grade, kept the recommendation, and widened the sentence:
“Universal screening for thyroid disease in pregnancy is not recommended because identification and treatment of maternal subclinical hypothyroidism has not been shown to result in improved pregnancy outcomes and neurocognitive function in offspring.”
The 2020 bulletin added “pregnancy outcomes and”. The grade stayed the same, and so did the underlying cognition trials. A reader of the Level A list now meets a broader negative claim, written after randomized reports of obstetric benefit in subgroups had appeared (Ma 2015; Nazarpour 2018), neither of which the bulletin discusses. The discussion goes further still:
“The results of the CATS study and the 2017 Maternal–Fetal Medicine Units Network's Randomized Trial of Thyroxine Therapy for Subclinical Hypothyroidism or Hypothyroxinemia Diagnosed During Pregnancy trial demonstrate that screening and treatment of women with subclinical hypothyroidism during pregnancy does not improve the cognitive function of their children at age 3 years and 5 years, respectively.”
“Does not improve” is a statement about the world. “Has not been shown” is a statement about the literature. And the paragraph names both trials without the weeks at which they began.
What ACOG attributed to three medical societies
“Therefore, the American College of Obstetricians and Gynecologists, the Endocrine Society, and the American Association of Clinical Endocrinologists recommend against universal screening for thyroid disease in pregnancy and recommend testing during pregnancy only for women who are at increased risk of overt hypothyroidism.”
The footnote for the second body is the Endocrine Society’s 2012 guideline. Here is what that guideline says:
“The committee could not reach agreement with regard to screening recommendations for all newly pregnant women. Two versions are therefore presented.”
“Some members recommended screening of all pregnant women for serum TSH abnormalities by the ninth week or at the time of their first visit.”
“Some members recommended neither for nor against universal screening of all pregnant women for TSH abnormalities at the time of their first visit. These members strongly support aggressive case finding to identify and test high-risk women… and they recognize that in some situations ascertainment of the individual's risk status may not be feasible. In such cases, and where the local practice environment is appropriate, testing of all women by wk 9 of pregnancy or at the first prenatal visit is reasonable.”
The Endocrine Society did not recommend against universal screening. Its committee could not agree. Five named members recommended screening every pregnant woman by nine weeks, and the rest went neither for nor against while calling universal testing by nine weeks “reasonable.” Fairness requires the rest of the record, though:
“Universal screening for the presence of anti-TPO antibodies either before or during pregnancy is not recommended.”
“Universal screening is not recommended for patients who are pregnant or are planning pregnancy, including assisted reproduction. Grade B, BEL 1”
So AACE did recommend against universal screening, on grounds it described as “insufficient evidence for lack of benefit”. The Endocrine Society recommended against universal antibody screening. On the question that mattered, TSH screening, ACOG’s sentence turned a split committee into a consensus. Several explanations fit the record: condensation, a case-finding reading, or composite drafting. None of them requires a villain. The effect is the same either way: a woman told “all the major societies agree” was told something those societies’ own documents do not bear out.
What the drug label says about pregnancy
The quoted regulatory reviews and prescribing information set out what the product record says about pregnancy.
FDA review, Synthroid NDA 21-402 (2002) · Tirosint-SOL NDA 206977 (2016) · current prescribing information.
The regulators knew the stakes early. In April 2002, reviewing Synthroid’s application, an FDA medical officer wrote:
“Hypothyroidism during pregnancy is associated with a higher rate of complications, including spontaneous abortion, pre-eclampsia, stillbirth and premature delivery. Maternal hypothyroidism may have an adverse effect on fetal and childhood growth and development.”
Those words are essentially the ones in levothyroxine labels today. The knowledge was there in 2002, the urgency was there, and so were FDA’s own reviewers years later, proposing this text in a review they drafted themselves:
“Levothyroxine should not be discontinued during pregnancy and hypothyroidism diagnosed during pregnancy should be promptly treated.”
The current Synthroid label carries a strong instruction, in the right place for a clinician adjusting a dose:
“For pregnant patients with pre-existing hypothyroidism, measure serum TSH and free-T4 as soon as pregnancy is confirmed and, at minimum, during each trimester of pregnancy.”
One gap remains, and it is easy to fix. Levothyroxine binds iron and calcium, and prenatal vitamins contain both. Here is the label’s warning:
“Inform patients that agents such as iron and calcium supplements and antacids can decrease the absorption of levothyroxine. Instruct patients not to take SYNTHROID tablets within 4 hours of these agents.”
Across the eight levothyroxine labels examined, the word “prenatal” does not appear, and neither does “multivitamin.” The warning sits in the pharmacology and patient-counselling sections, not in section 8.1 on pregnancy. A clinician can read the entire pregnancy section without meeting it.
Limit. Drug compendia, pharmacy systems and electronic alerts often carry the interaction. The point is narrower: it is absent from the label’s pregnancy section, and it never names the product a pregnant woman is actually taking.
Women already on treatment
The record turns to women already receiving levothyroxine and the practical monitoring issues described in that group.
Taylor et al. 2014 · Endocrine Society 2012 · ATA 2017 · Lemieux et al. 2021.
Nobody argues about this group. These women are already diagnosed, already prescribed and already in care, and the record shows that care still fails them in ordinary ways.
“Among pregnant women who had their TSH measured in the first trimester, 62.8% had a TSH level greater than 2.5 mU/L, with 7.4% greater than 10 mU/L.”
Limit. A proportion of the women who were tested, not of all treated women, and not a count of anyone harmed. In that study the miscarriage association was not significant between 2.5 and 4.5 mU/L and became clear above it: a gradient, not a cliff.
The guidance for these women has been clear for years:
“The T4 dose usually needs to be incremented by 4 to 6 wk gestation and may require a 30% or more increase in dosage.”
“For women receiving LT4 preconception, a prospective, randomized study has provided evidence that supports a single dose-adjustment strategy… Dosage augmentation should occur as soon as possible when a missed menstruation or suspected pregnancy occurs, and this should be discussed with every patient in the prepregnancy setting.”
In Alberta, a universal health system, every woman in this cohort delivered after the 2012 recommendation:
“Of the 10,680 pregnancies, 8774 (82.2%) had at least one TSH measurement … and 6707 (62.8%) had at least two TSH tests during pregnancy. … it is concerning that 17.8% did not.”
The same study also found harm running the other way:
“TSH in pregnancy below 0.10 mIU/L increased the odds of preterm delivery when compared with control pregnancies (adjusted odds ratio, 2.14 [95% confidence interval 1.51–2.78]).”
“The data suggested a potential harm of overtreating pregnant women on thyroid replacement, as a TSH level below 0.10 mIU/L at any time during pregnancy was associated with a doubling of the odds of preterm delivery.”
Being untested is a failure of delivery. Being over-treated is a failure of titration. Monitoring as it is actually practised did not prevent over-treatment in a well-resourced system. Any screening program therefore has to build careful titration in as a condition, not leave it as a hope.
The objections, and the answers
The objections are stated before the page sets out the screening program it supports.
The case against, stated by those who make it.
The strongest objection deserves its best statement, and the American Thyroid Association gave it:
“…the majority of patients identified through any universal screening process have TSH concentrations between 2.5 and 5.0 mU/L—a population in whom a treatment benefit is not well established. Furthermore, such a strategy could have detrimental effects, labeling many patients with a biochemical abnormality, and in many cases leading to initiation of possibly inappropriate long-term treatment.”
Everything in that sentence is true. It is an argument about what happens after the test, and it has an answer that does not require abandoning the test: confirm before treating. A mild elevation in a woman with no antibodies and no symptoms is a reason to repeat the test within one to two weeks, not to prescribe.
The screening program this record supports
Offer every woman the test
A TSH and antibody test at first contact, ideally before ten weeks, from the same blood draw. She may decline, and her decision is recorded without argument.
Treat overt disease at once
A raised TSH with a low free T4, or a TSH above 10, is treated without waiting.
Confirm mild results quickly
Repeat within one to two weeks rather than four, so confirmation does not recreate the trials’ timing. Treatment in the mild range is a shared decision.
Follow antibody-positive women
A written result, a TSH six to eight weeks after delivery, and a TSH before the next pregnancy. This is surveillance, not treatment.
Monitor to a range
TSH every four weeks to mid-pregnancy and again near 30 weeks, titrated to a pregnancy-specific range so women are not over-replaced.
Run the trial inside it
A program that finds women at six weeks is the recruitment platform the ATA’s requested trial has never had.
Limit. This program is closest to the ATA’s summary of Dosiou’s overt-disease scenario (under $8,000 per QALY). It cannot borrow her $7,258 headline, and the honest next step is to re-run her model with today’s inputs. If an early trial found no benefit, if screening at scale caused measured net harm, or if confirmation could not be delivered, the case would need to be argued again.
Questions to ask your clinician
These questions provide a way to discuss the record with the clinician caring for your pregnancy.
Questions the evidence supports. Not medical advice.
This is not medical advice, and it is no substitute for the person caring for your pregnancy. But the evidence on this page supports asking these questions.
Is my thyroid being checked, and when?
With known thyroid disease: as soon as pregnancy is confirmed, then every four weeks to mid-pregnancy. Without it: you can ask whether you have risk factors, or ask for the test.
What was my TSH, and against what range?
Pregnancy has its own reference range. A non-pregnancy range can call a value normal when it is not.
If I start treatment, when is the next test?
Four weeks is the interval the guidelines use. If no one has told you, ask.
I already take levothyroxine. What happens now?
The ATA advises that your dose go up promptly at a positive pregnancy test, and that this be discussed before you conceive.
What should I not take it with?
Not with your prenatal vitamin, and not within four hours of iron or calcium. If mornings are impossible with nausea, ask about taking it at bedtime.
When did treatment start in the studies?
At 13 weeks 3 days in CATS and 16.7 weeks in the MFMU trial. Treatment at six or eight weeks has never been tested.
Credits and thanks
The closing acknowledgments credit the people whose work made this record possible.
For saying, about yourself and your profession, what most physicians never say: that you were repeating what you learned, and that a reckoning, and an apology, were owed.
For the arithmetic that survives its critics’ strongest assumption, for the dissent printed under your initials, and for staying in the room.
For changing your mind in public and giving your reasons, for the JAMA viewpoint you told the field you rewrote five times, and for making the case at a microphone in front of your peers.
For measuring the clock and telling the field plainly that its trials had started later than the window you found.
And to the trialists, task-force members and editors who wrote down the limits of their own work: you made it possible for a patient to read the record and see what was, and was not, known.
Back to top ↑About this site
I’m Joshua Davis, an independent patient researcher. I compiled this record to show the case physicians and researchers have made for looking for thyroid disease early in pregnancy, in their own words.
The 89 quotations were matched against text conversions of primary documents. They have not all been checked against the original page images. Please check those images before reusing a quotation in print.
The quotation register explains the checks behind each label. The labels are unchanged; this revision does not add a new verification of the quotations.
In the debate transcript, [V] marks reconstructed or uncertain wording that needs checking against the audio before quotation. A dagger (†) marks a short passage reconstructed where the audio is partly unintelligible. Appendix D identifies the passages and the checks they need.
This is an evidence record, not medical advice. Screening decisions should be discussed with your own clinician.
Back to top ↑Corrections and updates
1 November 2026: tables made visible on mobile; Canada figure corrected to 47.1%; THESIS figure standardised to 39.7%; Mitchell 2021 wording corrected; Beltrão 2026 and Jensen 2022 footnotes added.
— Personal laboratory record corrected
Corrected my personal account to state that switching to bedtime dosing solved three years and nine months of hypothyroidism. Added my dated TSH and free-T4 results since March 2020 and removed unsupported personal narration.
— Information Gap added
Added The Levothyroxine Information Gap. Its publication corrections record the corrected citation, survey figures, preference wording and formulation exception. The existing 89 quotations, Casey sequence and complete debate record are unchanged.
— Casey’s timing passages restored
Previously, the tribute omitted Casey et al.’s 2005 statements about early maternal thyroxine and treatment begun after ten weeks. They are now quoted in the timing chapter, between ACOG’s prevalence estimate and its statements about benefit evidence and screening, in the sequence of my original blog post. The register includes both passages as entries 88 and 89.
These words belong to Casey et al. (2005), not ACOG Committee Opinion 381. The two excerpts appear in different parts of the paper; the treatment passage now includes Casey’s preceding sentence and his attribution to Pop and colleagues.
. I corrected the poll summary and clarified the page’s headings and links. The original 87 quotations, transcript speech, speaker attributions, displayed timestamps and recorded poll values are unchanged.
Accuracy corrections
- Thyroid dysfunction poll heading
Previously: “The poll: 71% before the debate, 55% after.”
Corrected: “The poll: thyroid dysfunction, 71% before, 54% after.” The live poll board shows 54%; the comparison slide shows 55%. Both source figures remain visible on the debate page.
- Appendix C’s change calculation
Previously: “−17 points yes (71 → 54–55%).”
Corrected: “−17 points using the live board (71% → 54%); −16 points using the comparison slide (71% → 55%).” The earlier summary applied one calculation to two different displayed results.
- Number of key moments
Previously: “Six minutes that matter.” Corrected: “Eight moments that matter.” There are eight selected moments.
Links and wording
Links for the repeated timestamps at 22:00, 54:11, 59:05 and 59:14 now distinguish the two turns at each time. The displayed times and spoken text are unchanged.
The authored card heading “The 150 micrograms dose in CATS” now reads “The 150-microgram dose in CATS.” The spoken dose descriptions in the transcript are unchanged.
The withdrawn audit
I withdrew the earlier levothyroxine-label audit. That audit and its data are not published on this site, and its old address now returns “not found.” The quotation register and complete debate are the source records for the argument on this site; they are not a republication of the withdrawn audit.
Back to top ↑