UK folic acid fortification plans
The UK plans to start fortifying flour with folic acid to prevent around 200 cases of neural tube birth defects annually. This will prevent some abortions of non-viable fetuses and some disabilities, but cause some new cancer cases. It would be nice to see the cost-benefits calculation made public (assuming there is one), and even nicer to see new research into some of the many possible ways to prevent NTD without causing as many cancers (e.g., offering reproductive-aged women free folic acid or methylfolate supplements instead of dosing everyone who eats wheat).
Formula feeding amount and indication recommendation disparities and prioritizing harm prevention
After reviewing Fed Is Best: The Unintended Harms of the ‘Breast Is Best’ Message and How to Find the Right Approach for You and Your Baby, I found myself scrutinizing bottle sizes while ordering supplies for the incoming addition. This made me reconsider the book’s recommended 15 mL at a time feeding volume.
For comparison:
The Irish Health Service Executive says “During the first week, most full-term babies take no more than 1 to 2 ounces (30 to 60 mL) at feedings” — 2-4x what FIB recommends.
Nemours Children’s Health (a major U.S. pediatric health system) says “On average, a newborn drinks about 1.5-3 ounces (45-90 milliliters) every 2-3 hours” — 3-6x what FIB advises.
Why would an organization dedicated to preventing underfeeding complications recommend feeding volumes substantially below common guidance? This discrepancy might stem from FIB’s promotion of elements of the “Breast is Best” mythology they present themselves as critiquing.
An alternative approach could advocate for more liberal supplementation. Both the amount and the indications for formula use should prioritize harm prevention. As commenter Agnes L noted, the FIB chapter on exclusive breastfeeding “should come with a lot of warnings about the risk of doing so, especially in the first few days of life”; and it doesn’t make sense to prioritize blood tests over formula-feeding hungry newborns.
Historically, early supplementation was widely practiced. Well-organized wetnursing professions, co-nursing, and prelacteal feeding traditions were the norm before formula-feeding replaced breastfeeding in the 20th century. Modern Western reformers from the late 1970s on then drove a resurgence of “exclusive breastfeeding” based on historically and medically inaccurate assumptions about its safety and naturalness.
In a 2021 interview on common and preventable harm associated with breastfeeding insufficiencies, Michael Moritz — University of Pittsburgh School of Medicine professor of pediatrics, clinical director of pediatric nephrology, medical director of pediatric dialysis and the Pediatric Kidney Transplant Program at UPMC Children's Hospital of Pittsburgh, and an expert on breastfeeding-associated pediatric hypernatremia — recounted how he also asked what changed when he started investigating this. He was referred to an elderly pediatrician who described recent historical norms:
“ ‘When I practiced, we were scared to death of jaundice… didn’t wanna have to deal with exchange transfusions… We’d have them breastfeed, but as soon as we saw jaundice, we stopped the breastfeeding and had them bottle-feeding. That’s the way we were trained. With all the breastfeeding initiatives, we stopped that practice.’ ”
This shift was not based on evidence against supplementation or data establishing the new norms’ safety, but rather on possible risks like nipple confusion or breastfeeding cessation, which Moritz characterized as specious and weak compared to the immediate need to ensure infants are adequately fed.
Nor was the shift immediate or homogeneous. When “exclusive breastfeeding” was invented by modern Western reformers, it wasn’t clear yet what it meant in practice. Thus, liberal early supplementation was successfully piloted in Sweden at the start of the “Breast is Best” hospital initiative Baby-Friendly. Some participating hospitals supplemented the vast majority of newborns with formula until their mothers’ milk came in:
“[Swedish Baby-Friendly hospitals] gave healthy neonates prelacteal feedings (i.e., formula before their mother's milk came in) anywhere from under 10% to 90% of the time [214]” (Hofvander, cited in Wilde 2022).
It’s ironic that Sweden—featured in the 2014 “Breast is Best” documentary The Milky Way as a breastfeeding-friendly contrast to the U.S. — recently relied heavily on formula supplementation to ensure neonatal health before mothers’ milk came in. It was common sense at the time; babies need to eat. They still do. The pendulum swing away from such strategies has been driven more by ideology than data.
The reluctance to swing the pendulum back, embracing formula supplementation, risks compromising infant health — particularly in jaundiced newborns. Research suggests that formula may expedite bilirubin excretion more effectively than breastmilk. Ignoring this evidence and wrongly equating human milk with formula in treating jaundice, as FIB does, may undermine harm prevention efforts. As I wrote in the review:
There are also places where formula’s praises could be sung more based on the evidence, but instead that evidence is omitted. For instance, the authors say more milk speeds bilirubin excretion to help treat jaundice (p. 96) — without acknowledging some evidence suggests that formula speeds bilirubin excretion better than breastmilk. (Probably because it has more calories per gulp, and that means more food in —> more poop out, bowel movements being segmental.)
Relatedly, my own analysis suggested that phototherapy risks iatrogenesis, but the discussion of phototherapy treatment for jaundice (p. 98) ignores this possibility along with the corollary that more aggressive formula supplementation has broad harm prevention potential. Instead, the authors set up banked human milk and formula supplementation as presumptively equivalent in reducing jaundice (p. 99), something the evidence does not establish. Some evidence rather suggests formula is better in this context, and it seems possible that this is just not something the authors felt it was socially or politically wise to come out and say.
Such posturing reinforces the status quo FIB was created to challenge. If the goal is to prevent harm, the messaging should emphasize feeding strategies that prevent underfeeding from birth and mitigate harm immediately when complications from it arise — rather than perpetuating myths and practices that risk compromising infant health. Hungry infants need more “fearless formula-feeding,” not more singing with the “Breast is Best” choir.
Bayes’ Rule lesson plan in the NYT on prenatal diagnostics
A few years ago, the NYT ran a beautiful example of how to apply Bayes’ Rule in the form of a lesson plan using the case study of prenatal diagnostics (h/t Franziska Hoerbst). For clarity, this is up there with Stephen Fienberg’s Table S-1 in the NAS polygraph report. The author, Dashiell Young-Saver, also founded Skew the Script for teaching high school students math is relevant (h/t FH). What a sci com wonder!
Meanwhile, NYT also links to other examples of mass screenings for low-prevalence problems in their coverage — rapid Covid testing among asymptomatic people, and facial recognition software in policing— but without incorporating Young-Saver’s (and Fienberg’s) application of Gigerenzer & Hoffrage-style frequency formats to illustrate conditional probabilities. Hopefully his readers grokked Young-Saver better than the editors apparently did.
Bias and misinformation in abortion discourse
Two recent maternal deaths in Georgia following medical abortion (abortion with pills) raise concerns about risk communication in medical, public, and scientific discourse. Both cases involved complications caused by abortion pills, yet news coverage has tended to attribute the deaths to the state’s abortion restrictions — which include a ban after six weeks — without entertaining alternative causal explanations. More neutral and accurate coverage could further harm prevention.
Case 1: Drug overdose
In one case, Candi Miller died from a drug overdose involving fentanyl, acetaminophen, and diphenhydramine after experiencing an incomplete abortion. ProPublica reported that the abortion pills she had taken were not responsible for her death, citing a statement from the Dutch doctor who heads the Austrian non-profit that sold Miller the pills online, Rebecca Gomperts of Aid Access.
However, medical abortion runs a possible double-digit risk of incomplete abortion, according to research Aid Access cites. Miller’s overdose was likely linked to physical pain and/or psychological distress caused by the medical abortion and/or this common complication of it. Particularly in settings where follow-up care is known to be limited by legal restrictions, Miller’s death underscores the need for informed consent and safer medical practices.
Aid Access’s website shows the organization misinforms women about the possible risks of medical abortion from home. It misinterprets the cited WHO study as showing “there was no evidence to say home-based abortion is any less effective or safe than medical abortions in a clinic setting.” This reflects statistical significance testing misuse, a common methodological mistake across medical and scientific literatures. The fact that the evidence is ambiguous and uncertain does not mean that there is no evidence suggesting that home-based abortions may be substantially riskier than clinic-based abortions.
In fact, the cited study’s findings show an incomplete abortion rate ranging from 3-14% for home-based and 1-20% for clinic-based abortions. In other words, we don’t know if one setting is safer than the other, or there is no difference, when it comes to this complication. These findings could reflect that clinic-based abortions are much safer, if their true incomplete abortion rate is toward the estimated lower bound (1%) and the true incomplete abortion rate of home-based abortions is toward the estimated upper bound (14%). Or the reverse (20% vs 3%). Or there could be no effect.
There are plausible causal mechanisms for either effect. For instance, maybe women are more relaxed at home, and that helps lower complication rates. Or maybe clinic staff help more women follow the protocol better, and that lowers them.
In any event, by contrast, surgical abortions (D&Cs) may run much lower risks of complications including incomplete abortion. NHS puts that risk at 35/1,000, or 3.5%. This could be comparable to medical abortions’ risks — or it could be more than five times less risky, depending on where the true risk falls in the estimated probable range (if it does fall in that range). We don’t know.
That said, it’s generally accepted that medical abortion risks heavier, longer bleeding, more pain, and higher rates of incomplete abortion than surgical abortion. But we shouldn’t expect people who make a living selling abortion pills (like Gomperts) to volunteer that — because of these risks — best medical practices involving abortion care for women where follow-up care is restricted involve surgical abortion whenever possible. That would threaten their business along with their psychosocial well-being and professional standing.
This is not the first time an organization headed by Gomperts has systematically misinformed women about the possible risks of the abortion services she makes a living selling. Gomperts’ Dutch NGO Women on Waves has long denied abortion carries possible acute psychological, fertility, or long-term health risks. This denial is inconsistent with the available evidence. A wide array of findings suggest possible risks of adverse mental health effects including at least 2x increased suicide risk. The evidence shows substantial possible risk, and does not establish whether the link is causal or not. There are no statistical pyrotechnics that can change that.
Similarly, Aid Access doesn’t mention possible psychological risks of abortion, instead reiterating the contested pro-choice claim that abortion is safer than childbirth. In fact, their relative risks depends on who you are, what might kill you, and what kind of medical care you have access to.
For example: If, like Miller probably did from lupus and diabetes, you have a compromised immune system and chronic systemic inflammation, then the risks of incomplete abortion from medical abortion in a setting without follow-up care access might be greater than the risks of pregnancy with access to medical care. We don’t know. Supporting informed consent by comparing abortion and pregnancy/birth risks should include at least noting the absence of such risk stratification and the depth of uncertainty involved in the comparison.
Case 2: Sepsis
In the other Georgia maternal death case, Amber Thurman died from sepsis caused by retained fetal tissue. Again, an incomplete abortion. Thurman had scheduled a D&C but hit traffic. Her best friend, Ricaria Baker, told ProPublica the clinic said it couldn’t hold “spot longer than 15 minutes — it was inundated with women from other states where bans had taken effect.” So it offered her the pills instead — knowing she would take them and return to a region where follow-up care could be inaccessible or delayed. The hospital that subsequently treated Thurman for sepsis then delayed her D&C so long, it killed her.
This all occurred despite the fact that Thurman was pregnant with twins. There are reasons to suspect multiple pregnancies may require a different medical abortion protocol than singletons. We don’t have enough information to know for sure. But, as in Miller’s case, the possibility of heightened risk should have invoked a higher bar for caution among healthcare practitioners.
Refocusing on harm prevention
These cases and their coverage illustrate a broader issue: political narratives about abortion often overshadow evidence-based discussion of possible risks and harm prevention strategies. Pro-choice reporting like ProPublica’s tends to emphasize restrictive laws as the root cause of harm, while pro-life perspectives focus on the inherent risks of abortion itself. Both frames risk oversimplifying complex realities.
But the pro-choice narrative is the dominant one in mainstream media and scientific discourses. In this case, the BMJ exhibited less neutrality than ProPublica, with its September news coverage entitled “Two US women died because of a lack of abortion care, report says” and November opinion essay “Banning abortion prevents us from providing safe care to all pregnant women.”
The former headline appears to be inaccurate. The “report” in question isn’t public and the BMJ doesn’t claim they obtained access to it. They’re probably referring to internal Maternal Mortality Review Committee reports leaked to ProPublica detailing two Georgia women’s post-abortion deaths in September.
ProPublica’s reporting didn’t say the report says that two “women died because of a lack of abortion care,” either. Rather, it cited a committee member (presumably the leaker, or one of them) who thought so — presumably the motivation for the leak. What ProPublica actually specified the committee found was that Thurman’s death was likely preventable, had the Atlanta hospital treating her for sepsis not delayed her D&C 19+ hours.
The BMJ opinion piece, for its part, argues for a liberal first and second trimester abortion regime in the interests of women’s health. But it’s not clear that such unrestricted abortion access would net benefit instead of net harming women’s health. We would need data on possible harms and benefits of both restrictive and unrestrictive abortion regimes to evaluate that claim.
More broadly, both pieces’ titles and content assume their interpretation of the causal story here is correct or even the only possible interpretation of the facts. The latter is false. For instance, similar coverage could have been entitled “Two U.S. mothers died following a common medical abortion complication, absent follow-up care.”
It’s worth pointing out bias and misinformation in the abortion discourse, because lack of neutrality is arguably hindering harm prevention. For instance, while Georgia’s abortion restrictions limited these women’s options, their deaths also underscore the importance of informed consent and access to follow-up care. A more neutral investigation might explore whether women in regions with abortion restrictions would be better served by doctors in other regions performing surgical abortions (D&Cs), which may carry substantially lower risks of incomplete abortion, rather than providing abortion pill access that may more frequently require follow-up care likely to be unavailable locally.
Perfect neutrality isn’t possible, but we should be wary of taking at face value the statements of sources like Gomperts and the abortion clinic that provided Thurman’s pills. They have perverse incentives to misrepresent medical abortion as safer than it is (e.g., they profit from it), to misrepresent their conduct in these cases as blameless (e.g., they could be liable for these and other deaths), and to tell the preferred story of powerful sociopolitical networks (i.e., minimizing or denying possible risks of abortion itself while assuming relevant deaths were caused by abortion restrictions). (Someone like Gomperts might rightly respond that abortion providers who profit from D&Cs similarly have perverse incentives to question the safety of medical abortions; the reality is that there’s a lot we don’t know about how their possible risks compare.)
In terms of cognitive bias, this is a confirmation bias story. Believe abortion mitigates risk, and you likely see these deaths as caused by abortion restrictions. Believe abortion carries risks, and you likely see these deaths as caused by abortion. The point here is not that one narrative is right or wrong. It’s that this is interpretive.
Moreover, focusing on an (unestablished) black-and-white story in which abortion restrictions caused these deaths distracts from possible harm prevention actions like improving the accuracy of information abortion providers give women, prioritizing surgical over medical abortion for women without follow-up care access (perhaps in a research setting), holding providers accountable for ensuring women have access to appropriate follow-up care (perhaps through criminal proceedings), improving research on possible abortion risks (though it may not be possible to answer key questions due to confounding and ethical constraints), and enhancing neutrality in the public discourse. By addressing these gaps in reporting and practice, both publishers and healthcare systems can better serve women, regardless of their political or geographic context.
(See also ProPublica’s most recent report on how the state’s public health commissioner couldn’t identify the leaker, and so dismissed the whole committee.)
Freedom of information request results on UK Covid vaccine damages
Recent freedom of information (FOI) request results on UK Covid vaccine damages, combined with recent Parliamentary questioning, suggest that the government has spent more on reviewing than on compensating claims, and has imposed a budgetary ceiling on the latter, Tom Jefferson and Carl Heneghan report at Trust the Evidence.
The NHS Business Services Authority responded on Oct. 21 to Sheila Ward’s FOI request on WhatDoTheyKnow about how many claims have been made to the Vaccine Damage Payment Scheme (15,804), how many claimants have been notified of an outcome (7,936), how many claims have been successful (188, overwhelmingly following AstraZeneca vaccination), how many claims have been rejected (7,748), and how many did not meet the criteria for medical assessment (716). Of the uncompensated claims, 391 “met the criteria for causation,” but “the independent medical assessor recommended that the vaccine has not caused severe disablement.”
This suggests Covid vaccines may be relatively risky. For comparison, Jefferson and Heneghan report “653 claims not involving a COVID-19 vaccine have been received by the VDPS. Fewer than five of these claims have been awarded.” That’s a 4/653 (~.006) hit rate for non-Covid vaccine damages, versus a 188/15,804 (~.012), 188/7,936 (~.024), 579/15,804 (~.037), or 579/7,936 (~.073) hit rate for Covid vaccine damages (with the denominator being either claims or claims notified of an outcome, and the numerator being either successful claims — or claims that were successful or met the criteria for causation).
Possible immortal time bias in meta-analysis on Covid vax in pregnancy
This week, Ghent University Postdoctoral Research Assistant Johan Steen gave a Karolinska Institutet CAUSALab Methods Series talk on immortal time bias, an under-recognized bias in which researchers accidentally select on survival or something that correlates with it. In a DAG, this is a form of collider-stratification bias. The solution is getting time zero aligned correctly. The key citation here seems to be “Specifying a target trial prevents immortal time bias and other self-inflicted injuries in observational analyses,” Hernán et al 2016, J Clin Epidemiol, 79:70-75. (The video’s not online yet, but should eventually appear here.)
This is generally interesting because the causal revolution dealing with time properly struck me (still new here) as its farthest-reaching innovation; most analyses just don’t do this, and diagramming causality out (using DAGs) forces it. And specifically it interests me because I recently noticed the sign of the possible stillbirth effect associated with Covid vaccination in pregnancy reversing, like it can due to collider-stratification bias.
Denoble et al’s recent U.S. Vaccine Safety Datalink analysis misinterpreted their results but found a 26% decrease to 83% increase in stillbirth risk within 6 weeks of vaccination or index date compared with no vaccination. By contrast, (Smriti) Prasad et al’s 2022 meta-analysis of “23 studies including 117,552 COVID-19 vaccinated pregnant people, almost exclusively with mRNA vaccines” found “The risk of stillbirth was significantly lower in the vaccinated cohort by 15% (pooled OR 0·85; 95% CI 0·73–0·99, 66,067 vaccinated vs. 424,624 unvaccinated, I2 = 93·9%).” This suggests a 1-27% decrease in stillbirth risk associated with the vaccine.
However, most included studies did not report outcomes according to trimester at vaccination (much less gestational age) or number of doses. This suggests a widespread failure to account for time. If the vaccine caused adverse pregnancy outcomes, we might expect to see gestational age/trimester, time from vaccination, and dose (along with vaccine type) matter. These factors should be accounted for in analyses.
Oddly, Prasad et al grappled with the implications of this failure for miscarriage and preterm birth risks, but not for stillbirth. And they fleshed out a possible causal pathway of concern that applies to all three adverse outcomes, without noting that it does:
The mRNA vaccine causes both antibody and cellular immune responses; given the importance of T-cell suppression in placental development and fetal wellbeing42, concern has been expressed that the vaccine may increase miscarriage risk.
A breakdown of maternal immune tolerance for the fetus has similarly been proposed as a cause of preterm birth and stillbirth.
For miscarriage, Prasad et al concluded “there was no increased risk of miscarriage following COVID-19 vaccination, based on data from large national registries and reports of inadvertent exposures in early pregnancy during RCTs.” But the underlying evidence for this claim was scant and the results ambiguous, they acknowledged elsewhere:
Although miscarriage rate was reported in multiple studies20,24,27,29, only two compared vaccinated and unvaccinated populations and accounted for time-varying confounding27,29. The pooled odds ratio meta-analysis showed no significant effect of vaccination on miscarriage (pooled OR 1.00; 95% CI 0.92–1.09, 15,684 vaccinated vs. 108,249 unvaccinated population, P = 0.988, I2 = 19.8%, Fig. 6). These findings were consistent with data from five randomised trials1,2,35,36,37 that reported miscarriage rates after inadvertent exposure to COVID-19 vaccination in early pregnancy. As the number of reported exposures was small (N = 4–43) and the vaccine types varied (i.e., two mRNA and three viral vector), quantitative synthesis was not undertaken.
Similarly for preterm birth rates, they noted “the data do not account for time-varying confounding” and expressed concern that “the [32%] reduced rate in the CDC study is likely explained by the fact that most (~60%) vaccinated people were in their third trimester at the time of vaccination, so many were too late to have a preterm birth.” This explanation does not make sense: the third trimester is when you can have a preterm birth, and the baby is most likely to survive it.
They also noted the UKHSA data conflicted, suggesting a 9% increase in preterm birth rates associated with vaccination in pregnancy. Again, a case of switching signs. The sign switched again when they combined “two estimates of hazard ratio (Lipkind et al. 28 and Blakeway et al. 8) which adjust for time-varying and other possible confounders,” producing “a non-statistically significant reduction in preterm birth by 10%.”
They interpreted this finding as “consistent with multiple studies linking COVID-19 infection to preterm birth.” But the possibility that Covid infection contributes to adverse pregnancy outcomes including preterm birth does not exclude the possibility that Covid vaccination contributes to them, too. And these sign-switching findings could also be consistent with collider-stratification bias.
It’s causally plausible that there are adverse causal effects of both Covid infection and vaccination on pregnancy outcomes; both trigger inflammatory processes when the body is already under enough strain. The point is that we don’t know what the net cost-benefit analysis says about them. And, in the real world, people’s choices around both infection exposure and vaccination can be constrained in ways that confound the data we get on associated risks. So it’s really hard to know what’s going on.
But the failure to diagram causality in these analyses — including setting time zero at a point where researchers are not selecting on the outcome of interest in the treated group due to immortal time bias — undermines the credibility of statements about Covid vaccination’s effects on pregnancy outcomes.