My last post explained why the statistical term “Simpson’s Paradox” isn’t very useful, because it can refer to different phenomena — non-collapsability, classical confounding, collider confounding, and ironically sciency-sounding hand-waving to explain without explaining why, sometimes, a model change produces an effect sign reversal. One of the examples I mentioned, the abortion-suicide link, has a way deep rabbit hole out of which I’m attempting to climb.
Leading scientists and other authorities make methodological mistakes that lead to misinterpretation of a wide array of evidence showing substantial possible increased suicide risk (typically around 2x or more) associated with abortion. This post focuses on collider bias, the next on other methodological mistakes in the consensus literature (which are common mistakes throughout the medical and scientific literatures), the next at how experts talk about abortion misinformation and avoid tackling the problem of non-neutrality in science, and then finally I look at the bigger historical and cultural picture and argue that (as often) what we think of as progress may not be very progressive after all. All of this hopefully leads to better science and science communication. And that hopefully makes women’s lives better.
Thanks for reading Wilde Truth! Subscribe for free to receive new posts and support my work.
Starting where I left off: This collider bias problem permeates consensus (pro-choice) abortion science. The problem is that stratifying by confounds, without first drawing out causality to see if they are really colliders, risks introducing more bias (collider stratification bias) than it corrects for. This mistake affects a large number of well-regarded sources including consensus reports by the American Psychological Association (APA) and National Academy of Sciences (NAS), and leading experts (Gissler, Steinberg) frequently repeat the mistake as evidence that there is no link, when the evidence really shows substantial risk and does not establish whether the link is causal or not.
Since this is highly politicized terrain, a note about politics and what I’m up to here…
This Friday, the briefing deadline elapses in a landmark U.S. abortion case. In the wake of the June 2022 Dobbs v. Jackson Women’s Health Organization decision overturning the U.S. Constitutional right to abortion (Roe v. Wade, 1973), a November lawsuit brought by the Alliance for Hippocratic Medicine is asking Trump appointee U.S. district judge Matthew Kacsmaryk to ban the abortion drug mifepristone. Doctors, as well as online pharmacies and other gray or black-market sources, providing medical abortion pills would then be left with the second drug in the now-standard combination, misopristol alone — which usually works, but is harder. Against this backdrop of shifting legislative landscape and the hyperpolarized debate surrounding it, problems in the science underpinning abortion care have seemingly gone unnoticed, potentially causing numerous preventable deaths.
This points to the need for a paradigm shift in abortion research incorporating the insights and tools of the causal revolution, particularly doing DAGs for drawing causal relationships before doing statistics. But there are limits of this toolset just like any other. In order to do better science and science communication, we also need to incorporate science reform lessons on accepting uncertainty into research and informed consent in this context as in others, and confront the problem of non-neutrality in science. There is no science without scientists and their perspectives, and science has nothing to say about resolving political (value pluralistic) conflicts like whether or when societies want to prioritize the right of the embryo-fetus-baby (biological continuum) to life over the liberty of the pregnant woman to not be pregnant. Credit for these core insights without blame for my mistakes goes to Sander Greenland, Judea Pearl, Richard McElreath, Isaiah Berlin, and whoever said “it’s turtles all the way down” (perhaps William James).
Caveat: My criticism of abortion research methods and risk communication is neither in the space of normal conversations — pro-life versus pro-choice — nor typical academic versions of them, e.g., concerning relationships between abortion and mental health/fertility problems for women, and mortality/morbidity risks for future offspring (huge topics). Pro-lifers say limiting abortion access is about preventing harm, chiefly to unborn children and women; pro-choicers say advocating abortion access is about women’s autonomy and well-being. Both sides argue social harms also result. Neither wants to recognize there is substantial possible preventable harm to women on both sides of abortion access, and that measuring and comparing those sometimes lethal, sometime qualitatively different harms to support an evidence-based, aggregate instrumentalist case for a particular (liberal or restrictive) abortion access regime may be impossible. Science does not and can not tell us how to solve this problem, either.
The pro-life argument that limiting abortion access is about minimizing risks of preventable harm by applying the precautionary principle fails to recognize this limitation of what science can and can not do. Philosophically, the problem here stems from Isaiah Berlin’s distinction between negative and positive liberty, or freedom-from versus freedom-to. Society can not absolutely guarantee a woman’s freedom to carry a pregnancy to term without coercion or pressure to abort from a partner or parent, and to then care for her children as she sees fit without socio-economic or other limitations; as always, individual choice has structural limits.
Empirically, comparing abortion access harms resulting from these sorts of limits with abortion denial harms resulting from limits like physician discretion or legislative restriction compares apples and oranges. Society, not science, decides how to weigh these different sorts of harms. But this kind of comparison makes one group of women suffer harm so another can avoid it (trolley problem), when it’s not clear to what extent we have to live in such a forced-choice world. In dismissing as a product of confounding the roughly doubled suicide rate associated with abortion, consensus science may privilege the preferred (pro-choice and presumed pro-woman) narrative of powerful social networks at the cost of vulnerable women’s lives.
This post proceeds as follows: First, I review several examples of typical abortion-suicide findings. Then, I go through how consensus sources misrepresent them by modeling away findings of massive risk without accounting for causality. Next, I review some findings on pregnancy loss more generally, because they’re strikingly similar. Then, I propose some possible biological mechanisms whereby collider stratification bias could be at work in the consensus models, sketch some other findings and questions, and then using that overview, draw a DAG (possibly abortion science’s first). This post should be by far the longest post in the series, so bear with me trying to put it all on the board! The idea is that the field needs to think structurally about causation, and this is just one attempt to do that.
Dubious Science Downplays Risks of Abortion Access
Comparing women who experienced “problem” pregnancies and those who didn’t probably selects for women with problems (e.g., trauma exposure, poor social support, and mental health problems). Current literature (e.g., Gissler’s standard citation critiqued in my last post, Steinberg’s more recent contribution; and the secondary but seminal APA 2008, UK AMRC 2011, and NAS 2018 reports) reads this as an indication to correct for these confounds. But they’re colliders, so correcting for them risks introducing more bias (collider stratification bias) than it reduces. This situation is worse when colliders are also mediators, especially if there are unobserved variables (see Traag & Waltman 2022).
The common abortion science mistake of modeling away substantial raw risks has practical, moral, ideological, and psychological components. The literature making it forms the basis for current standard abortion care informed consent (e.g., from Planned Parenthood, Women On Waves, etc.), which frequently informs women that there are only possible health benefits and no risks of abortion compared to term pregnancy — except where state laws have created different requirements, which are themselves often accused of constituting misinformation. This status quo blames women with mental health problems before abortion for mental health problems after, including a substantial proportion with trauma exposure. It supports the financial and professional interests of organizations and individuals who have built their careers on abortion care as a women’s rights issue.
I focus on the APA and NAS reports, because the American Psychological Association and National Academy of Sciences are two of the primary U.S. institutional scientific bodies that produce reports that are widely seen as neutral, scientific, consensus (and indeed reality)-reflecting documents. The Academy of Medical Royal Colleges, National Collaborating Centre for Mental Health, is a UK analogue. They all issued relevant reports and got the science wrong, downplaying risks on the basis of models correcting for confounds without accounting for causality, when the raw data show substantial possible harm.
So what do the data look like? When you compare women in whole populations who had abortions with those who did not, what’s the relative suicide risk?
Abortion-Suicide Link, Example 1: Finland, 2-6x increased risk
Mika Gissler’s Finnish population register studies show consistently more than 2x increased suicide risk even after Finland implemented an (apparently somewhat effective) intervention to address the problem. Earlier research showed up to 6x increased risk comparing women who aborted with those who gave birth, and 3x increased risk compared to non-pregnant women.
Gissler argues for correcting for variables like disadvantaged background, but that’s a collider bias problem — they may be associated with unintended pregnancy, abortion, and mental health problems alike. This would mean that treating SES (for example) as a confound and stratifying on it in statistical analyses could introduce collider bias that is just as severe as (or more severe than) confounding. Pretending that outcomes like suicide and accidental death are statistically independent when they’re clearly not, just because it lets us implicitly correct for confounds like this, is no better.
In other words, we can’t make causal confusion go away by throwing statistics at it. We have to think about causality first. And if we don’t make strong causal assumptions going in, we can’t get causes coming out (channeling McElreath on Nancy Cartwright). This sucks for societies in which people would like objective, neutral science to drive policy, instead of admitting that story drives statistics, so there is no exit from belief mattering — no apolitical realm to resolve political conflicts bloodlessly.
Back in the rabbit hole du jour, some researchers, like Gissler et al 2004, frame the abortion-suicide link as a “healthy pregnant woman effect.” This supposes that we know what’s cause and effect here, and being healthier tends to cause childbirth more than abortion. It could be that better physical health causes better mental health outcomes (e.g., because inflammation also causes neuroinflammation which can cause depression). Or it could be that abortion causes worse health outcomes including suicide. Or both.
While it’s true that Finnish cuisine includes a dish called Blodplätter — blood pancakes made with whipped reindeer blood — Finland is, in this context, not strange. Subsequent to reading the Finnish results, I did a PubMed search (1970 through May 30, 2022) for the terms "case-control abortion suicide" that returned two relevant results, both of which report a strong, statistically and practically significant association between abortion and suicide. Here they are…
Example 2: Denmark, up to over 3x increased risk among young people
In a 2012 Danish register-based, nested case–control study on attempted suicide risk in children and youths after contact with somatic hospitals, Christiansen and Stenager find medical abortion increased risk of attempted suicide, with 30 such cases matched to 172 controls correcting for age and sex confounds, showing a highly statistically significant relative risk of 3.18 (95% CI 2.71-3.73), and an adjusted RR of 1.15 (0.72-1.84). The authors interpret these results as showing that “After adjustment for confounders, medical abortion remained a risk factor for attempted suicide,” stating “This association calls for the same interpretation as the association between exposure to violence and increased risk of attempted suicide in children and youths.” In other words, abortion looks bad for young women’s mental health. Gissler et al had noted the same finding, but cautioned that the sample size of teen pregnancies was too small to read into.
Example 3: Taiwan, up to 5.5x increased risk
In a study of 350 completed and 485 attempted suicide cases matched on age and year of delivery with 10 randomly selected controls in Taiwan from 2001-2011 using data from the population-based National Health Insurance Research Database, National Birth Registry, National Death Registry, Weng et al report similar findings, with substantial increases in completed suicide rates among women who experienced a fetal loss across subcategories of stillbirth, miscarriage, and abortion, with a 3.12 adjusted OR for abortion (95% CI 1.77-5.5) as compared to live birth, and substantially increased attempted suicide rates among women who experienced either miscarriage or abortion, reporting the latter's aOR as 2.5 (95% 1.63-3.82).
Other examples: U.S., Italy, meta-analyses
Some people dismiss the U.S. and Italian results as cultural artefacts due to hyperpolarization and Catholic stigma, but it’s worth noting the substantial abortion-suicide link recurs here. Numerous U.S. studies co-authored by noted pro-life researcher David Reardon show substantially higher suicide rates associated with abortion, including a 2.54x risk increase in a 1989-1997 California Medicaid record study. Similarly, Lega et al observe that, in Italy, the reported suicide rate was 1.8/100,000 postpartum, 2.77 after induced abortion, and 2.90 after miscarriage. Both U.S. and Italian examples find the same ballpark doubling of risk as in numerous other cases.
Comparing the country level globally, there only appears to be one published analysis, from Agnus M. Kim’s letter to the editor in the Asian Journal of Psychiatry (2021), showing abortion rate correlates with suicide rate. Interestingly, suicide rate correlates with alcohol consumption, but abortion rate doesn’t; and suicide rate doesn’t correlate with homicide rate, but abortion rate does. Alcohol depresses the central nervous system, worsening depression, so the alcohol-suicide link is pretty well-explained.
Geneviève Bloch-Torrico’s 2014 psychology doctoral dissertation (University of Montréal) finds abortion associated with statistically and substantially increased risk of completed or attempted suicide. This is a beautiful dissertation, and Bloch-Torrico does a great job saying that we don’t know if the link is causal.
Her findings are consistent with findings in the first PubMed hit for “abortion suicide meta-analysis,” Miranda-Mendizabal et al’s Int J Public Health 2019 study finding abortion is a risk factor for suicide attempts. The next hits on this search are Coleman’s 2011 meta-analysis, then Reardon and Thorp’s 2017 meta-analysis, and finally an irrelevant, last hit on intentional iron overdose in pregnancy. Coleman, Reardon, and Thorp are noted pro-life researchers, their findings are consistent with that perspective, and this is not a comprehensive review of the abortion-suicide meta-analytic literature. I’m just tracing the outline of the terrain here.
The bottom line is that the abortion-suicide link consistently looks massive. We don’t know if it’s causal. But leading researchers and institutions consistently dismiss these findings as artefacts of confounding to be modeled away with statistical tricks, when that is bad science. Here are just a few of the most prominent examples.
Expert Misrepresentation, Example 1: The 2008 Report of the APA Task Force “Mental Health and Abortion”
This report cites Gissler et al’s 2004 study, “The largest and most methodologically rigorous Finland study,” finding that abortion is associated with violent pregnancy-associated death. Then it states “Abortion is a marker of risk for violence, not a cause of violence. Thus it is important to control for violence exposure in studies of pregnancy outcome.”
This is a tautology. If you want to assume that abortion is purely a consequence and not also a cause of violence, fine. Assume that and say so. But don’t say you’ve proven it, and that’s what proves it, when that’s the open empirical question you’re supposed to be addressing. In other words, Gissler’s assumption that suicide, accidental death, and homicide are independent is tenuous.
The tautology here masks the confound problem, as discussed previously. It could be that violence exposure, mental health problems, and poor social support are (any or all) associated with unintended/unwanted pregnancy, abortion, and mental health problems — in which case the first set of covariates would be not just risk factors but also colliders. So controlling for them would risk introducing more bias (collider stratification bias) than it corrects for.
This is just one example of APA misreading the evidence overall. The methodological problem is the causal salad (in McElreath’s terms), and not knowing what to do with confounds. The psychological problem is then misrepresenting this uncertainty by downplaying risks, saying there is no causal link between abortion and mental health harms when there are substantial associations and we do not know if they are causal. (Greenland calls this misrepresenting uncertainty as certainty “dichotomania,” and the next post gets into that more.)
APA could have said: a bunch of studies show substantial raw associations between abortion and bad mental health outcomes like suicide, but we don’t know what to make of that causally because the real world is so messy, especially in complex, heterogeneous situations characterized by private information and undesired outcomes.
Instead, the report concluded with more certain determinations, buying that modeling adjustments to raw data explain away evidence of substantial possible harm including suicide. They say (p. 92) that one first-trimester abortion probably doesn’t increase mental health risks, while abortion under other circumstances (multiple abortions, late abortions, and aborting wanted pregnancies due to coercion or fetal malformations) probably does — and individual variation, data limitations, and uncertainties abound. The evidence that APA reviewed does not support the first part of that conclusion.
The APA Monitor more recently repeated the same myth, in more black-and-white (and thus even less accurate) terms, reporting in a 2022 post that we know there is no abortion-mental health problem link. This is the liberal, ostensibly neutral party line that powerful social networks expect scientists to repeat. It’s not established by evidence.
Example 2: The 2018 NAS Consensus Study Report “The Safety and Quality of Abortion Care in the United States” (highlights, full book) similarly buries the substantial links reported in the studies it cites (including Gissler and Reardon) under (1) dubious analysis that makes them go away (Gissler), or (2) the criticism that they should have done such dubious correction for confounding (Reardon). Both dismissals make the same mistake: They assume that if you see a raw link, and there’s a confound story you believe in, you should just correct for the confound and accept the answer that results as the right one. But this risks introducing more bias than it corrects for. We need to care about the raw data and do better science when it comes to accounting for causality.
As a sidenote, another reason NAS dismisses findings linking abortion with mental health harm is that some evidence comes from women’s self-reports. We should be extremely wary of dismissing data like this. As I’ve written previously, women reported depression from birth control pills for a long time before doctors laundered their reports into quantitative, official medical data, published it, and got credit for announcing the “discovery” of what women had been saying for decades. Screw that.
By the same token, there’s a longstanding move in several prominent pro-life abortion researchers’ work to bolster causal claims with data from women who had abortions, experienced subsequent mental health problems including substance abuse, and perceive these problems to stem directly from abortion. But generally, we have to worry about sample selection bias, attrition, recall bias, social desirability bias, and other things when people give us causal stories. We have bad (causal, statistical, scientific) intuitions as a species.
Still, when a bunch of women say that something hurt like hell, the raw data says a bunch of them killed themselves, and a bunch of (often male) doctors, researchers, and policymakers dismiss these women’s experiences as a product of things like mental health problems — we should take another look.
Example 3: Steinberg’s 2019 Lancet Psychiatry article
Steinberg et al use Danish population registry data to focus on the relationship between first-time abortion and first-time suicide attempts, adjusting for age, calendar year, socioeconomic status, history of childbirth, mental health, parental mental health, and physical health — most of which are colliders. To take just one example, there’s a causal ambiguity in the common contention that researchers should “correct for” “risk factors” like mental health problems in modeling whether abortion increases risk of mental health adverse outcomes like suicide. Since depression increases your risk for future, worse depression, it’s entirely possible that pregnancy loss causes depression which makes future depressions worse in women who were already prone to mental health problems. But then abortion is still causing a mental health harm; it’s just harming a particularly vulnerable group. Researchers and physicians are supposed to care about that harm more, not dismiss it.
Going forward trying to learn more about causality in the abortion-suicide story, one of the interesting sets of related findings is on pregnancy loss more generally, as Gissler et al 2004 and many others have noted.
Pregnancy Loss Hurts
Whether the cause of pregnancy loss is intentional (abortion) or accidental (miscarriage, stillbirth), mental anguish often follows. As you would expect, stillbirth is the worst. But abortion and miscarriage produce strikingly similar patterns of clinically significant distress in a minority of women.
To quote Robert Sapolsky, women’s heightened vulnerability to depression, and its association with menstrual periods and the two weeks after childbirth, screams biology. And so, too, I think, do strong associations between pregnancy loss and suicide — a particularly unusual depression outcome for women, who are generally far less likely to use violent methods and so to succeed in killing themselves.
Let’s look at some findings: In a nested case-control study we visited before, Weng et al use data from the Taiwanese national population-based registries (relatively complete, linked registers of national health insurance, births, and deaths) to show suicide risk increased substantially in women who lost a fetus through stillbirth, miscarriage, or induced abortion. The effects are massive: up to 15x increased risk of completed suicide following stillbirth, 5x risk for miscarriage, and 5.5 for abortion; plus more than 2x increased risk of attempted suicide following miscarriage and over 3x increased risk following abortion.
Similarly, Shigemi et al’s J. Psychiatr Res. 2021 Japanese retrospective cohort study using the Diagnosis Procedure Combination database, a national database for acute-care inpatients in Japan, looks at all pregnant women who were hospitalized for attempting suicide between January 2016 and March 2018. They found “critical perinatal outcomes” — abortion, miscarriage, intrauterine fetal death, and successful maternal suicide — were significantly associated with violent suicide attempt methods (OR = 3.57 [95% CI = 1.15–11.1]). Violent methods are much more likely to succeed, so this means pregnancy loss predicts suicide.
In a Danish population-based record linkage study, Coleman et al 2013 find massive increased death risks associated with abortions as well as miscarriages, with more losses having much worse effects: “Increased risks of death were 45%, 114% and 191% for 1, 2 and 3 abortions, respectively, compared with no abortions after controlling for other reproductive outcomes and last pregnancy age. Increased risks of death were equal to 44%, 86% and 150% for 1, 2 and 3 natural losses, respectively…” A lot of pro-choice researchers dismiss studies by Coleman and Reardon because they’re pro-life, but their results are not sui generis.
Bellieni and Buonocore’s “Abortion and subsequent mental health” 2013 Psychiatry and Clinical Neurosciences review finds some evidence for worse short and long-term mental health risks associated with abortion, but notes miscarriage may increase short-term anxiety and depression risks more.
Overall, pregnancy loss looks bad for mental health, and abortion looks more like a type of pregnancy loss than a type of normal elective medical procedure in terms of possible mental health risks. We don’t know that associated risks like suicide are causal; we don’t know that they’re not.
So what’s going on here in the pregnancy loss data, and what’s going on in the denial of it in scientific and medical discourse? Harkening back again to Sapolsky, I think biology, its interplay with the psychological and the social, and our removed attitude toward our biopsychosocial natures help answer both questions.
Possible biological explanations for the abortion-suicide link are under-recognized. We are mammals who think of ourselves as rational actors, still often on the Cartesian model (body and mind — never the twain shall meet). This is a model that Peirce, James, Damasio, and others have long questioned or even radically rewritten (i.e., thoughts come from feelings, which come from bodily sensations). On this model, I sketch out several possible biopsychosocial mechanisms for the abortion-suicide link, in service of sketching a causal diagram (DAG). In particular, the next section suggests previously unrecognized causal pathways whereby biology could causally contribute to the abortion-suicide link in ways that may introduce or worsen the collider stratification bias that afflicts current consensus science.
Better Science Is Possible — Imperfectly
Abortion-suicide risk models often include some version of socially disadvantaged status, mental health, and physical health. I hone in here first on mental health as a collider, and look at some ways in which biological mechanisms could help explain how “correcting for” mental health to make effects attenuate gives wrong answers. Then I review some other findings that suggest variables that might need to go in the DAG before drawing it.
Collider bias: Is psychiatric vulnerability cause and consequence of abortion?
Shonewille et al BMC Pregnancy Childbirth 2022 review evidence from eleven studies reporting on psychiatric vulnerability and unintended pregnancy. They find that having a history of psychiatric disorder(s) may substantially increase women’s risk of unintended pregnancy (OR 1.34, 95% CI 1.08-1.67). In the U.S., over half of unintended pregnancies result in abortion. This suggests that mental health problems may contribute causally to unintended pregnancy and abortion, making mental health a collider in models dealing with abortion and mental health outcomes. Some suggest the link is diminished advanced planning capacities in general, with risky sexual behavior associated with mania in particular. Others note that exposure to violence, especially sexual violence, and the presence of PTSD symptoms are common in women seeking abortions. So in one model, violence exposure contributes to mental health problems, which contribute to unintended pregnancies, which contribute to abortions, which contribute to mental health problems.
Here are some examples of how this might play out mechanistically. The first three — oxytocin sensitivity and deficit; progesterone withdrawal and deficit; and anemias— are biology-focused, but biology does dialogue with psychology and social structures in them all. One could also imagine more psychological and social mechanisms on a triplanar biopsychosocial spectrum. For example, abortion may contribute to recurrences of prior mental health problems like depression that tend to get worse when they recur, mental grooves in the default mode network wearing deeper (among other things). It may strengthen the self-death association that may predict suicide, which may already be stronger in some women who abort. And in the simplest counter-factual, abortion prevents the appearance of a new human being some months later, entirely dependent on its caretaker(s); Durkheim famously observed that people with children committed suicide at far lower rates than others, and we might note that his finding generalizes to women who give birth as opposed to aborting. Being needed may keep people from killing themselves. So correcting for vulnerabilities that relate to lack of social power or worth (real or perceived, by self or others) — a grouping which we might think of as including mental health problems — may introduce collider bias, too.
Example 1: Oxytocin Sensitivity and Deficit (Jonesing for Love)
What if some women’s brains respond more strongly to possible psychotropic effects of sex, orgasm, and semen? What if some women’s brains respond more strongly to the possible psychotropic effects of pregnancy? What if those groups overlap? Might a heightened affective response to semen help explain why women with mental health problems appear to be more at-risk for unintended pregnancy? Might the same sort of heightened affective response help explain why women with previous mental health problems may be more likely to experience post-abortion mental health problems?
These are taboo questions, and there’s always someone saying that talking about this hurts women because abusers will cite research as an excuse (bad argument). But there are empirical reasons to ask them. Oxytocin sensitivity could mechanistically link these effects, and may be affected including through a threshold mechanism by traumatic experiences, which reduce endogenous oxytocin, and happen to also be a strong predictor of mental health problems including in women in general and women who have abortions in particular. In other words, traumatized people may have less oxytocin, need it more, and react differently when they get it; and you can get oxytocin from unprotected sex and pregnancy. Maybe that’s why some evidence suggests traumatized women seem to engage in more high-risk sexual behavior. We tend to think of this behavior as maladaptive, but heightened oxytocin sensitivity (i.e., endogenous deficit plus heightened affective response when the deficit is corrected) may be seen as adaptive in contexts where experiencing connectedness can be a matter of life or death. If resources are scarce and threats abound, and you’re a subordinate primate female (like all primate females), you probably want a mate for safety.
Trauma exposure correlates with structural vulnerabilities. We can imagine many causal reasons for this (e.g., abusers seeking out vulnerable victims, decreasing socio-economic status increasing crime exposure, intergenerational network effects). Structurally vulnerable women may choose abortion at higher rates, because having children and taking care of them takes a lot of resources, and resource access is about power. So here is a causal story in which structural vulnerability contributes to trauma exposure, changing and perhaps magnifying existing oxytocin sensitivity along with creating or magnifying overlapping mental health problems in some subgroups — a combination that contributes to some relatively vulnerable women having unsafe sex, getting pregnant, and having abortions.
But with the pregnancy loss comes the loss of its oxytocin high — pushing particularly oxytocin-sensitive, structurally powerless, traumatized women with mental health problems off a hormonal cliff with considerable apparent risk compared to giving birth, while experts assure them that it’s risk-free or even risk-mitigating. What you might expect to see if this causal story is true, is a substantial increase in suicide risk post-abortion that attenuates when researchers stratify by the confounds of structural disadvantage, violence exposure, and previous mental health problems (which is what current consensus science shows) — but the attentuation would not, in this case, indicate lack of causality. Wrongly thinking that it does may generate victim-blaming among healthcare professionals such as mental health support people, who may inform suicidal trauma victims post-abortion that their problems pre-exist the abortion and are not causally related according to the current research literature.
As a sidenote, mainstream biochemical theories of mental health problems, like the serotonin-depression and dopamine-ADHD myths, have rightly fallen out of fashion with scientists familar with work by experts including Bob Whitaker and Peter Gøtzsche. In my opinion, this is an area where women’s health really does look different from the bigger picture of which it is a part. In the mainstream, biochemistry has had its (lucrative) heyday. But in women’s health, the biochemistry in play in long-studied relationships like abortion-suicide and breastfeeding-postpartum depression remains largely unexplored in ways that tend to dismiss both huge chunks of scientific evidence and huge numbers of women’s experiences. I think this is because it is not potentially hugely profitable for drug companies to explore relevant treatment options (more later). And because it has been too easy for experts to dismiss relatively powerless women’s experiences.
Example 2: Progesterone Deficiency and Anxiety (Blocking Nature’s Dope)
Very stressed women may have heightened DHEA, which can cause or correlate with progesterone deficiency, which would be remedied by continuing pregnancy (because pregnancy jacks up progesterone) — but potentially exacerbated by medical abortion using pills that block progesterone (mifepristeone’s mechanism). Higher progesterone increases GABA function; GABA is a neurotransmitter that inhibits neuronal activity associated with anxiety, thus producing a calming effect. Progesterone is Nature’s dope.
So pregnancy loss in general and medical abortion with mifepristone in particular may produce an anxiety spike that is worse in very stressed women, and that some are unable to organically recover from due to stress-related progesterone deficiency. This may help explain why sleep disorders are more common in women with recent abortions. Sleep problems tend to compound mental heatlh problems and are a particular risk factor for suicide. Progesterone helps with sleep; progesterone deficiency is commonly associated with insomnia, for instance, in women with PCOS (polycystic ovarian syndrome), or who are in perimenopause or menopause.
Example 3: Anemias and Depression (Bleeding Sanity)
Abortion causes blood loss, sometimes quite substantial; this is particularly true of medical abortion — which often entails considerably heavier and longer bleeding than surgical abortion. This can cause or compound common deficiencies including iron and folate-deficiency anemias. Both can cause or contribute to depression. This form of depression would be harder to prevent or treat if you were unable to eat. Misopristol (the second drug in standard medical abortions) causes stomach problems that can impair eating. Researchers in Camilleri et al’s 2019 Front. Neurosci. lab study of medical abortion observed apparent anhedonia including lack of normal preference for sucrose solution (sugar water) among rats given medical abortions.
Some subgroups of women are probably at increased risk for anemia-related depression before and after abortion. For example, those with a common genetic mutation (set of possible mutations) that impairs folate metabolism, have a greater risk of folate deficiency; and supplementing with the more bioavailable form, methylfolate, seems to help in treatment-resistant depression. That’s a relatively purely biological risk group vis-à-vis depression-causing anemias and abortion, but there are also biopsychosocial ones. For example, one could imagine subgroup effects from socio-economic status, social support quality, eating behaviors, and other mental health problems entering in here, too. Less money means less money to spend on food, which may impair nutritional diet quality. Less social support means less help cooking and feeding yourself when you don’t feel well. Vegetarians and vegans are already prone to nutritional deficiencies including these, as are women with disordered eating due to a range of mental health problems (e.g., anhedonic depression, anorexia, anxiety, bulemia, OCD).
When “Correcting for,” Isn’t
This is by no means an exhaustive list of possible causal mechanisms whereby something in the biology of pregnancy loss in general and/or abortion in particular contributes to the experience of mental distress, sometimes rising to suicide. Inflammatory and immune processes could also contribute, with progesterone tending to downregulate both; taking away the downregulation may tend to spike inflammation. It’s also not impossible that the body “knows” it has lost an offspring, and responds accordingly with a cascade of something resembling a biochemical basis of grief.
My point here is that we need to see why correcting for colliders in abortion models is dumb, and these possible biological mechanisms show that. In all these examples, “correcting for” confounds may inadvertently hide the causal mechanisms whereby social factors influence psychology and biology, and biology influences the risk of mental health harm including suicide from pregnancy loss including abortion. This problem isn’t specific to biopsychosocial phenomena; but it doesn’t help that this terrain is so fluid and multicausal, either.
So how do we know when “correcting for” confounds, isn’t? Drawing out causal relationships using DAGs is the way. This is not a DAG tutorial; there are lots of excellent resources out there that do that, and I linked many in the last post. But before you can draw DAGs, you need to know the terrain you’re drawing on. All of it. Which is impossible. So you need to know at least enough of it that you can start thinking comprehensively like a subject-area expert. This is probably ideally done as a group brainstorming exercise, with experts from different camps. Who’s organizing the picnic?
Here’s a quick pass through some things I observe in the abortion literature that might be relevant to drawing the causal story in DAG form.
People have often assumed that the youngest women are most at risk of distress, grief, and other possible adverse effects, probably because minors are generally considered at-risk in medical research, and teen pregnancies socially problematic in the first place. But results like those in Luo et al’s BMC Public Health 2018 study raise an implicit critique of this assumption, at least in its linear incarnation. They find up to over 2x adjusted odds of suicidal ideation among women who had an abortion in the past year (95% CI 1.46-2.44). They don’t report unadjusted results, which is a shame because they adjust for colliders, like most of the rest of the literature. The association was considerably weaker among those < age 20 (95% CI .27-2.11), stronger among those aged 20-25 (95% CI 1.14-2.19), and considerably stronger among those >25 (95% CI 2.16-5.28).
Why might it make sense if older women were at greater risk for post-abortion distress? Biology, in dialogue with the psychological and social realms, again offers several possible explanations. Progesterone levels decline with age, so taking away Nature’s dope could cause a particularly low and durable drop in older women. This could be compounded by stress as older women may have to be lower-status or more highly stressed to choose abortion, even if they really want a baby, the older they get. That may be, and the distress may also be greater directly (no biochemistry required), because older women’s opportunity costs of ending a pregnancy are higher, as they have fewer future possible pregnancies.
The other side of this picture may be consistent with the opportunity cost story: If fewer future possible pregnancies for older women should mean more mental health risk from abortion, then more future possible pregnancies for younger women should mean less mental health risk for them. That is what we seem to see, for example, in Suvi Leppälahti’s doctoral dissertation on the effects of teenage pregnancy in Finland. But again, it could also or instead be a biochemical story — younger women tend to have higher sex hormones including progesterone, and maybe this gives them a greater resilience to going off the cliff.
Whatever the mechanism, these findings start to paint a picture of possible differential abortion benefits and risks across patient categories like age groups, which might be something you would want to tell patients about, as Reardon and Thorp have argued. And this sort of risk communication would really matter clinically, since abortion rates seem to be bimodal — with very young and very old women aborting much more frequently on average than their mid-reproductive age counterparts (e.g., Figure 4, Mazuy et al, 2015, Population & Societies). So a relatively small number of abortion patients (younger and older ones) may carry a disproportionate amount of the risk of preventable harm, and clinicians should know that.
In passing, as a matter of filling in some information one might need to be critical about the source, Lu et al’s article “was supported by the grants from the National Natural Science Foundation of China (71273174), the National Natural Science Foundation of China (71673187) and the Shanghai Key Discipline Construction Project in Public Health (15GWZK1002).” The Chinese government used to coerce women to have only one child, and still coerces ethnic minorities to have fewer kids. Now it’s facing a demographic time bomb, so it’s changed its calculation about the state’s interest in (some) people having babies or not: it wants (some) women to have more babies and fewer abortions. Accordingly, China has also been increasing public health messaging recently about the fertility risks of what it calls medically non-necessary abortions, as part of a larger policy shift to encourage population growth. Yet, these warnings are evidence-based, and there is no neutrality in science anyway (more on this in another post).
Bottom line, Lu et al’s suicidality findings are generally consistent with typical findings (e.g., those reviewed above from Taiwan, Japan, Denmark, the U.S., and Italy). They’re also consistent with other age-specific findings. Look at the big age spike in Gissler et al 1996’s Figure 1.
Pain intensity in medical abortion tends to be high. In one Finnish hospital sample, 57.7% of girls and women undergoing it reported experiencing pain they rated at least 70 on the visual analogue scale 0-100mm.
Medical abortion seems worse than surgical abortion in terms of pain and immediate possible complications like prolonged bleeding, heavy bleeding, hemorrhage, and incomplete abortion requiring (ostensibly) a surgical follow-up. Around 30 years ago, 195 Scottish women who didn’t care how they got an abortion were randomized to receive either surgical or medical abortions. Henshaw et al reported that women who got medical abortions were much more likely to say they would choose the other method in the future (22% versus 2%, p.001). Most of the difference (95%) came from women who got medical abortions who were over 50 days gestational age, suggesting that the physical experience wasn’t something they would choose to repeat.
There are some good reasons to suspect that more physical pain or complications means more mental pain or complications, and the two interact in more ways than one. We are meaning-making monkeys. What if, following Peirce, Damsio, etc., sometimes we make mental pain out of physical pain, even if that makes us more pain along with more sense?
In a 1978 study re-purposing data from an RCT on aspirin and vacuum aspiration abortion pain (N = 214) to see what increased “psychosomatic reactions,” Bracken reported that being married correlated with a difficult abortion decision, but we don’t know why. Did husbands choose abortions for resistant wives? Did willful wives choose and hide wanted abortions from controlling husbands? Did poor relationship quality in both sorts of cases net make everything more difficult for the subgroup of married women in shitty marriages? We don’t know.
But we do know that higher pre-abortion anxiety plus less skillful operator increased procedure pain, and those things together with a difficult decision and not having had kids before made for more postabortion anxiety, which was the chosen endpoint in part because it correlates with postabortion depression.
Becker makes it complicated, concluding: “Future research into the psychological reactions to abortion, therefore, should seek to specify the source of particular reactions (reaction of partner, reaction to surgery, reaction to possible complications, etc) rather than operationalizing a large number of highly interelated concepts (such as fear, anxiety and regret) and relating them to a single independent variable called ‘abortion.’ ” But the evidence he presents could also be interpreted as suggesting that feeling psychologically bad and then experiencing pain makes you feel psychologically worse, maybe because the initial psychological badness primes you to make more meaning from the biological badness of pain. Maybe we’re just story-telling monkeys.
These sorts of findings about the severity and relevance of pain to post-abortion mental health are concerning in the context of current global trends. Countries with already liberal abortion regimes (e.g., Finland, New Zealand) have been moving to all-medical abortion. Countries with increasingly conservative abortion regimes (e.g., the U.S.) have seen surgical abortion restricted, which may tend to increase medical abortion. In both sorts of countries, more medical abortions could mean more physical and psychological pain for women.
Data bias: What abortions do we not get data on?
It looks in a lot of samples as though being poor is a risk factor for abortion and/or repeat abortion (e.g., Jones et al J women’s Health 2018). This could be because wealthier women get better medical care and so have fewer unintended or unwanted pregnanies. Or it could because more privileged women have abortion access we don’t know about, or lie more about it. Do some physicians perform abortions but code them differently for different sorts of patients? Do wealthier women lie more about having had abortions?
How would you get data to answer these questions? It’s not clear a doctor survey sample should be limited to gynecologists, or even doctors, given the increasing availability of medical abortion pills online. The first question might already have answers split into two eras: surgical and surgical + medical abortion eras. The surgical answer might have involved gynecologists servicing wealthier clients more discreetly, while the surgical + medical answer might depend less on physicians and more on Internet access. What data would exist on medical abortions over the Internet? Could we get it? Maybe some evil surveillance technology or infrastructure has left a useful trail…
The second question could be addressed using weird methods survey researchers have worked out to get better data when people are inclined to lie about sensitive issues like sex and drug use (bogus pipeline, dice, Internet anonymity). Nothing is perfect.
Maybe this isn’t even a form of data bias, since it doesn’t look like other researchers normally consider it; whereas they have considered the bias innate in getting data on legal but not illegal abortions… But I don’t see why this wouldn’t be an additional bias to worry about.
Another possible set we don’t get data on are women who were considering or planning to abort, but miscarried instead. It could be that some of the same stressors that contribute to abortion decisions also contribute to miscarriages (including physical and psychological health problems), so the occurrences are not independent. But it could also be that women who were going to abort but miscarried first instead are the best comparison group we could get, even though the outcomes aren’t independent. I don’t see any studies like this, possibly because it would be a subset of a subset of women, self-reporting biases might be substantial, and asking sensitive questions like this is not going to be standard in any ordinary survey.
Outcome variables: what makes sense?
It would be really difficult to do a comprehensive abortion risk review, and it’s not clear what outcomes to look at. Short-term suicide is a worst-case, rare outcome. It’s really remarkable that there’s evidence of such substantial suicide risk increase, and that it’s been so widely and inappropriately dismissed. We have to care about bodies, and, because effects of most treatments attenuate, it makes sense to look at bodies in the short-term. But quality of life, disability, and long-term health matter, too, and there are more questions here.
Medium-term: There’s a wide range of associated mental health problems. For instance, Coleman et al 2009 find substantial increases in panic attacks, agoraphobia, PTSD, bipolar, major depression, and substance abuse disorders associated with abortion. There’s a wide range of associated physical health problems, including possible increased breast cancer risk (future rabbit hole), as well as fertility problems that could have permanent affects on future children (e.g., prematurity and attendant neurodevelopmental risks). And then there’s the long-term…
Frailty: What’s the connection?
What about the grief/frailty connection in this context? Mourtzi et al reported in 2018 that Greek women with more abortions have higher frailty risks, with more abortions associated with higher risks. But we don’t know which came first: Maybe physical weakness causes more abortions because women rightly don't feel strong enough to carry pregnancies and raise children, and then these weaker women at time 0 also become weaker women at time 4. Or maybe abortion causes adverse mental health effects (depression, suicidality, traumatic stress, grief) that in turn causally contribute to behaviors like under-eating from anhedonia and not exercising from psychomotor retardation, that in turn cause frailty. And we should be cautious about only entertaining these two possibilities, because Mayerl et al suggest that frailty and depression might share common causes in the short and long-term.
This goes back to it looking in the data like abortion codes as pregnancy loss, which can incur grief, which can punch women in the guts so hard that some of them die. Some of those who don’t may then live with ripple effects of that grief that physically mark them for life, like tree trunks with rungs showing the seasons they starved. Frailty is not just a marking, either.
Frailty matters more broadly because it predicts negative health outcomes including death; the frailty index is a sensitive survival predictor according to Mitnitski et al. We would want to know if abortion decades previously causally contributed to early deaths. But maybe it’s just a confound for social status or social health, since effective frailty interventions (strength exercises and protein supplementation) look like things you would already be getting in the course of life if you were active and cared-for, and/or feeling up to taking good care of yourself. We would like to know what’s going on here, and we don’t, but we could at least know more about what we do know.
The abortion-frailty link, together with the frailty-protein supplementation and exercise fixes, evokes again Camilleri et al’s study of medical abortion in rats. Their post-abortion rats “displayed a significant reduction in sucrose consumption/preference during Treatment Week relative to Pre-Treatment Week,” which looks like anhedonia. “Regression analysis indicated that pregnancy termination was a predictor variable for body weight, food intake and all locomotor activity parameters measured,” consistent with the idea that depression can cause behavioral changes (less eating and movement) — that can cause frailty over time. So I really want to see data from a public health agency like the U.S. FDA on weight loss and mood following medical abortion.
But I can’t find that data. It seems like it should have been part of the drug’s pre-approval safety portfolio, since both drugs are well-known for being hard on the stomach, and you would want to know how that affects weight to know if you can give them safely to everyone. This gap in the safety data seems even more concerning in light of Camilleri et al’s findings. It may seem odd that medical abortion has become standard care worldwide without this basic safety data. But it’s not unusual, in the sense that a lot of standard medical practices are based on inadequate or faulty evidence (e.g., lack of experimental evidence for most surgical interventions, historical misunderstandings and bad science for “exclusive breastfeeding”).
Anyway, frailty in the abortion context presents a great causal diagramming problem, because it has lots of possible arrows and loops. So we need to see more extensive causal chains drawn out and think about confounding of various kinds here.
Another one of the ideas I’d like to see tested here involves association between self and death…
Mechanism: Does abortion increase suicides by strengthening the self-death association?
Normally I’m very wary of implicit association test (IAT) research for various reasons, but there’s a cool study showing that the strength of the self-death association predicts suicide attempts. (Here’s a failure to replicate; grain of salt.) This is a relatively hard thing to do that has bothered doctors for a long time. So one possible mechanism whereby abortion increases suicide risk is through increasing this self-death association. But… Is that really the mechanism even if it turns up on pre- and post-testing?
This association could be driven by underlying biology, the association could in turn influence biology, or both. But it’s hard to measure “biology” to test the biological abortion-suicide hypothesis (a hypothesis that seems to be way under-hypothesized and studied), in part because there are so many things that could be going on here. Pregnancy makes progesterone skyrocket, and ending it pushes women off a hormonal cliff; allopregnanolone (a progesterone metabolite) is a promising treatment for postpartum depression, which seems to involve going off the same cliff, albeit at a different height, and with a different outcome, in different average social circumstances. So could giving depressed aborted women progesterone prevent suicides? Maybe, but then why not screen and treat pregnant women seeking abortions for mental health and social problems first, instead?
Once you get into treating troubled, vulnerable, and pregnant women, no one wants to do any good experimental research anymore (tiny violin) — because it’s unethical as hell. But you could give women a short self-death IAT measure before and after abortion, and see if it correlates with suicidality. Or could you? A good IRB would tell you to do everything you could for these women to help them avoid preventable harm and reduce suicidality. But we don’t know enough about the relative risks of abortion or how to reduce deaths in at-risk pregnant women, to know how to do that.
So maybe someone could study the IAT mechanism and see if it pans out. It might also make sense to measure the extent to which women view the embryo/fetus as a life, to see if that belief correlates with strengthened self-death association post-abortion, and that correlates with depression/suicidality/self-harm. If it does, then you might theorize that aborting a fetus codes as murder, cementing the self-death association, causing heightened suicide risk, much as suicide rehearsal behaviors (like starting to take pills) correlate with increased completed suicide risk. Something about deepening those default mode network grooves. Abortion could be like a (or be a) suicide rehearsal behavior where the self is rehearsing death, and it is a good idea to discourage suicide rehearsal behaviors if you don’t want people to commit suicide.
How do you square the circle of respecting women’s autonomy as a first principle while limiting it when it comes to preventing harm as a first principle? You have to accept multiple values exist, they often conflict, and society is in the business of promoting one over another in cases like this. Noted anthropologist Sarah Hrdy writes (on p. 472 of Mother Nature): “So far as genes and tissue are concerned, embryo-fetus-baby represents a biological continuum.” Maybe women who recognize this ambiguity shouldn’t have abortions if they’re already in a risk group for mental health problems and would rather not abort anyway, but don’t feel they have the resources to raise a child. Maybe practicing being disempowered and creating a situation where they’re not needed causally increases suicide risk by increasing anomie, in Durkheim’s terms.
But is it then a doctor’s role to say so? How would you feel if a doctor questioned your life choice? On the other side, how would you feel if a patient reported considerable distress from a medical procedure you performed? Defensiveness on both sides would tend to hinder dialogue.
A lot of treatment questions are moot because of widespread healthcare provider denial that post-abortion trauma exists, even though there’s suggestive evidence that it may. This denial means that there’s little research into preventing and treating it. Otherwise one might ask if it looks more like PTSD-classic or PTSD-moral injury, and what treatment modalities work best for what phenotype of PTSD predominates in this context…
Abortion trauma syndrome: Is prevention/treatment research feasible?
There’s a whole literature on abortion trauma syndrome, aka post abortion stress syndrome, in which pro-lifers argue it exists, pro-choicers argue it doesn’t, and no one convinces anybody else. It seems to be a microcosm of the rest of the abortion discourse, with one side saying their read of the science is right and the other side is myth, the other side saying the same thing, and neither side seeming to produce evidence that would speak to the prevention and treatment of traumatic stress, suicidality, and other problems commonly seen in women who have had abortions. So maybe the right question to ask isn’t necessarily “does it exist?” But rather, “how might things we know might work for preventing and treating trauma fit in here?”
Potential traumatic stress prevention includes methylphenidate, beta blockers, and ketamine. MDMA is the single most effective traumatic stress treatment; no one has tried it in a novel context like abortion trauma, where it might just as easily be applied before, during, and after the procedure. Why? Pro-lifers want to prevent abortion trauma by preventing abortion; pro-choicers want to deny abortion trauma exists.
A sizeable minority of women report experiencing severe and lasting harm resulting from abortion, like those here, here, here, and here. In Table 1, Biggs and Antonia 2016 found 39% of women who had abortions reported post-traumatic stress symptoms, with 14% citing the index pregnancy as the reason. We can worry about selection bias, question the influence of culture and stigma, and mistrust sources with ideological interests. But it is inappropriate for experts to dismiss women’s experiences.
There’s a methodological case to be made for triangulating observational evidence that abortion as modern societies currently practice it substantially increases risks of adverse mental health outcomes including suicide, with qualitative evidence that it traumatizes a substantial minority of women, and experimental evidence that medical abortion causes biological and behavioral markers of depression and anxiety in lab rats, to conclude that we don’t know, but abortion may cause common and preventable harm to women in the form of mental distress, and the form of it that is swiftly becoming dominant worldwide may be riskier than the surgical form it replaces.
Everyone is interested in proving or denying this harm, but no one is interesting in preventing or treating it. Screw that. There is too much evidence from too many different sources (e.g., many diverse OECD countries) and methods (e.g., population register studies and self-reports) that abortion increases mental health risks, to deny that this is a possibility and try to help women accordingly. Prevention/treatment research on abortion harm may not be politically feasible, but it’s scientifically simple (experiments with drugs are fun like that). And often, when you don’t know what’s really going on, experiments can help clarify this by producing causal effects.
The biggest harm prevention potential here is in preventing instead of treating abortion trauma, and the biological mechanisms proposed above (oxytocin, progesterone, iron and folate) can be manipulated before just as easily as after abortion. There’s a case for heightening oxytocin before and during abortion care: It’s used in tandem with counseling for traumatic stress treatment, and abortion in many places comes with counseling. Why not see if MDMA-assisted abortion counseling changes abortion decisions by helping women feel safe and open up, which might decrease unwanted abortions? The same treatment, a week later, might help lessen the pain of abortion, which is often reported as severe; MDMA is an analgesic and induces oxytocin release, which is part of labor and might help complete abortion by strengthening uterine contractions. Such a procedure might help prevent abortion trauma. But it’s probably too taboo for psychedelic science to touch.
There’s really no reason that progesterone is not already standard care for sleep problems in women who have experienced completed pregnancy loss across the board. Okay, there’s one reason: it’s a common clinical belief that you want the loss to complete, so you want to wait until bleeding is finished — which can take many weeks or even months. But you can screen for signs for complications (sepsis is rare). The trade-off here between unknown physical health risk if “remains of conception” remain in the uterus longer versus unknown mental health risk if women don’t sleep following pregnancy loss needs to be weighed on a case-by-case basis. It’s possible that a good trial on women who have experienced pregnancy loss, randomizing to progesterone treatment for sleep problems or “standard care” (none) could change practice by highlighting harm prevention. (Update 04/19/23: Actually, we’re not sure if progesterone might contribute to some cancers. So that would be a good reason to exercise caution in its use.)
Similarly, there’s really no reason that testing for anemia is not standard in women pre- and post-abortion. It’s not that expensive to test, and you don’t want to perform abortions on women who are depressed for a fixable biochemical reason that affects their judgment and will get worse when they lose blood, particularly when substantially elevated suicide risk is associated with the procedure and we don’t know why. An alternative would be to treat before and after without testing, measuring depression and fatigue (anemia symptoms). This would be another simple randomized experiment.
I grouped these experimental answers to the earlier hypothesized biochemical mechanisms down here, because an experimental science answer to the abortion trauma question seems like the most neutral, productive response to this hyperpolarized research body. At least in theory, ethical experiments on mitigating possible mental health harms of abortion are possible. No one is doing them because no one is taking seriously both the evidence on harm (threatening to pro-choice interests) and the possibility that abortion still benefits many women while harming others (threatening to pro-life interests).
And perhaps nobody who’s interested in taking both these things seriously as a matter of science (not to mention feminism), wants the headache of upsetting both groups. But it really matters for women’s health that we do. One way to run experiments that are too legally tenuous to run centrally, is to collect data from people doing their own, decentralized Internet self-administration, like James Fadiman and Sophia Korb’s microdosing work. This might be a good method to work around institutional sexism to advance women’s health research. Progesterone is over-the-counter in the U.S., iron and folate are available everywhere, and these experiments could be done.
Drawing the DAG
Here’s what we might want to include in a DAG representing abortion and suicide:
A is for abortion — of which we can distinguish MA, medical abortion, from SA, surgical abortion;
S is for suicide;
D for self-death association;
P for poverty affecting child-rearing quality or real/perceived ability — of which we can imagine many subtypes including socio-economic disadvantage, SES; lack of supportive partner/family to help or isolation, I; low education, E; and St for stigma or perceived stigmatization of single motherhood or other special circumstances;
M for mental health problems — of which we can distinguish subtypes including depression, MD (medical depression); anxiety, MA; and traumatic stress syndrome/disorder, PTS;
N for compromised nutritional status — of which we can distinguish subtypes include iron-deficiency anemia, ID and folate deficiency, FD;
SU for substance use— of which we can distinguish subtypes including alcohol use, AU; cannabis use, CU; other psychedelics, PU; and other drug use, DU;
PS for poor sleep;
PH for physical health problems — of which we can distinguish subtypes including PHX for pain; PHI for inflammation; PHP for low progesterone; PHO for low oxytocin; PHM for metabolic problems;
VT for violence or trauma — of which we can distinguish V for violence exposure; T for trauma; and IPV for current interpartner violence;
PD for pressured decision — of which we can distinguish C for coercion to abort; PA for pressure to abort; and DA for difficult abortion decision;
DP for discrepancies between parents — of which we can distinguish DI for pregnancy intentionality; DW for pregnancy wantedness;
AG for women’s age group — within which we can distinguish at least three categories, YAG, youngest age group; CAG, mean childbearing age group; and OAG;
MM for moral misgivings;
GA for gestational age;
LO for low oxytocin;
OS for oxytocin sensitivity, assuming there’s a continuum;
LP low progesterone;
PC for especially painful, prolonged, or complicated abortion — of which we can distinguish SP for severe pain, PP for prolonged pain from abortion, and CA for abortion with complications
FD for family/dependents, assuming there’s something to Durkheim on suicide
Here’s DAGitty, a free online tool for drawing DAGs.
Here’s the DAG.
My first DAG! Abortion science’s first DAG? Testable implications and model code available on request.
Look complex? It is. It’s almost certainly not The Right Structure, either. The point is not that this is the one, but rather that we need to be drawing it out.
Summary: tl;dr - It’s a myth that abortion is proven safe. It may cause substantial harm including at least 2x elevated short-term suicide risk. This is a massive finding that pregnant women deserve to know about.
Correcting for confounds like social disadvantage and mental health problems explains considerably heightened suicide risk associated with abortion.
“One does not simply ‘correct for confounds.’ ”
Not when they’re colliders, because that risks introducing more bias than it corrects for (collider stratification bias). We have to draw out causality first (“science before statistics,” in McElreath’s parlance). Biological, psychological, and social mechanisms, separately and interacting, may help explain a possible causal relationship between abortion and substantially increased suicide risk.
Disclosure and Dialogue
No conflict: To the best of my knowledge and recollection, I’ve never been a part of pro-life or pro-choice efforts or organizations. Except I was briefly a member of the Green Party age 14, and the National Organization of Women (NOW) in college. No one paid me to do this work.
Humanning: Comments are open. Let’s be excellent to one another.
“Advancing statistics reform: ways to improve thinking and practice in the face of resistance,” Sander Greenland, July 2021 talk with June 2022 slides.
Statistical Rethinking, Richard McElreath, Bayesian statistics and causal inferences course (free lectures online) and book.
“Causal foundations of bias, disparity and fairness,” V.A. Traag and L. Waltman, 2022
“Rationality, Activity and Faith,” The Princeton Review, Vol. 2, July-Dec. 1882, p. 58-86, William James
Liberty, Isaiah Berlin, Oxford 2004
Mothers and Others: The Evolutionary Origins of Mutual Understanding, Sarah Hrdy, Harvard University Press, 2009
Thanks for reading Wilde Truth! Subscribe for free to receive new posts and support my work.
I should start by thanking you for writing this, although I am not sure what to make of it. This has to be one of the most difficult subjects to write about, and you seem to me to be a person of good will, so I don't mind engaging. I can't comment on the substantial points about abortion. I would not challenge the consensus judgement in this case without a better understanding of the subject. But whenever I run across a methodological critique of something, I try to think about "what would I have done differently."
I am trying to educate myself on causal inference. In my statistics program we were not taught about this - causality was a problem for scientists, inferences about correlations is our department. Which is fine I think. My ideal scenario is where a scientist comes up with a DAG, generates predictions which are tested to corroborate or falsify the entire structure. We still have the problem if identifying "what went wrong" in the case of a falsification, but we had that problem anyway.
I get the sense that causal modeling has a strong Bayesian leaning. This is more problematic for me. My #1 goal when using frequentist inference is stop people from coming up with stories that are driven by noise. Is designing the DAG after looking at the study's data allowed? If so, there's a big post-selection inference risk IMO. What I really want to do is to understand the sampling distribution of the whole causal modeling process.