I'm currently preparing to record an episode of Death Panel about my most important niche interest: counting deaths in pregnancy and postpartum. This is in the news recently due to a new study in the American Journal of Obstetrics and Gynecology (hereafter AJOG) entitled "Maternal mortality in the United States: are high and rising rates due to changes in obstetrical factors, maternal medical conditions, or maternal mortality surveillance?" This study made the news because it purports to show, contrary to the popular narrative, that the rising rates of perinatal death we've been hearing so much about are not real increases, but rather artifacts of inaccurate perinatal death surveillance. It has been known since at least 2018 that the current method used to count these deaths is inaccurate and likely inflates death counts. The CDC pushed back on the study in a characteristically confusing way, taking issue with the alternative methodology the study's authors used to count perinatal deaths (known to result in substantial undercounting), even as it acknowledged that its own current method of counting results in overcounting.
So what's going on here? Indulge me as I pull back the veil on the social construction and social life of mortality data before getting into some more general (and more interesting) theoretical considerations. It all starts with the system of death reporting in the US. Death reporting is done at the state level; local officials or health authorities typically fill out death certificates then transmit those death certificates to the state vital statistics office, which then forwards them along to the National Vital Statistics System (NVSS), which tabulates mortality data from the whole country.
Accurately reporting pregnancy-related and pregnancy-associated deaths (more on the difference between these two presently) has been a huge challenge for the United States, even though it ain't rocket science to count deaths in pregnancy. The UK, for example, has a prospective obstetric surveillance system (called UKOSS for, straightforwardly enough, the UK Obstetric Surveillance System) that works very well. Not so in the US, which had problems harmonizing state-level mortality rates due to state-level differences in death reporting practices, especially around pregnancy. It was known through the 1980s and 1990s, furthermore, that the NVSS was severely undercounting pregnancy-related deaths. To remedy this problem, a so-called "pregnancy checkbox" was added to the 2003 Revised Standard Death Certificate. The purpose of the checkbox was to indicate whether the decedent was pregnant at the time of death or within 42 days of death (42 days post-delivery is the arbitrary cutoff we use to bracket the "postpartum period," although very good arguments can be made -- using so-called "late" pregnancy-associated deaths -- that this restriction leads to undercounting).
States didn't all adopt the revised death certificate at the same time, however. The US was unable to report a national maternal mortality rate (embarrassing, tbh) between 2003 and 2017 due to staggered adoption of the new death certificate across states. It is well-known that the checkbox estimates overcount perinatal deaths (this was highly publicized in Texas in 2018). Despite these issues, every state is now using the death certificate with the checkbox, so NVSS is once again able to tabulate national perinatal mortality rates. The results have been grim. NVSS documents a near-doubling of the already-high perinatal mortality rate from 17.4 deaths per 100,000 live births in 2018 to 32.9 deaths per 100,000 live births in 2021, with these overall rates masking emergency-level rates among Black and Indigenous women -- for example, 70 deaths per 100,000 live births among Black women that same year. For comparison, national perinatal mortality rates in the UK (not even one of the countries with the lowest rates) are on the order of 7-8 deaths per 100,000 live births annually. (COVID was definitely related to this increase, with 2021 appearing to be a high-water mark; provisional data seem to indicate these rates coming down, ever so slightly, in 2022.)
In the AJOG study, the authors count two ways: one using the checkbox only (known to overcount), and one counting only death certificates where pregnancy is indicated as an underlying cause of death (known to undercount). Because the method that undercounts delivers a lower count than the method that overcounts, the authors argue that the much-reported increase from 2018-2021 is not "real" -- that it reflects changing practices of obstetric surveillance and death reporting rather than worsening underlying trends in perinatal health or medical treatment of pregnant people.*
The authors explicitly framed their research question this way. They sought to investigate whether "the high and rising rates of maternal mortality in the United States reflect changes in obstetrical factors, maternal medical conditions, or maternal mortality surveillance." This formulation of the research question leads to several interesting considerations.
First, this framing reflects a "levels of intervention" paradigm that is (unfortunately, in my opinion) so common to reproductive health research as to be taken for granted. It is awkwardly positivist, slicing the influences on pregnancy outcomes neatly into mutually-exclusive categories at distinct levels of organization. These are commonly the "patient," "provider," "system," and "community" levels. Unsurprisingly, "patient" factors in this literature tend to verge on victim-blaming (patients' "failure" to advocate for themselves or follow medical instructions). The literature ultimately attributes most deaths to "provider" factors, which is to say things that happen during the medical encounter for labor and delivery. This is unsurprising, because most of the data available to study these things is restricted to single-episode hospital visits for delivery -- a great object lesson in how the data systems and methods really determine what knowledge is produced by analyzing them.
Second and more interestingly, what does it mean for a death to be pregnancy-related (directly or indirectly caused by the pregnancy) or pregnancy-associated (merely occurring in the general time frame of pregnancy and postpartum)? Because most of these papers are being written by epidemiologists and (gulp) even worse, clinicians, there is a preoccupation with understanding pregnancy as a physiological/clinical state. Truly pregnancy-related deaths are those where the death is caused by, e.g., eclampsia (a severe hypertensive complication of pregnancy) or by, e.g., preexisting hypertension exacerbated by the physiological demands of pregnancy. The AJOG article zeroes in on these definitional distinctions as the major concern. The second counting method the authors use, the one known to undercount, tries to neatly exclude all of these non-physiological and non-clinical deaths during pregnancy. Deaths resulting from accidental injury, self-harm, and assault are categorically excluded from consideration. This seems clear-cut, but it isn't. While someone who died in a car accident while pregnant would be appropriately counted as a pregnancy-associated but not pregnancy-related death, a different type of accidental injury, unintentional overdose, is not as straightforward.
While clinical professional groups have issued gold-star statements about standards of care for treating substance use disorders in pregnancy, the actual reality does not match the standard-of-care documents. Many states criminalize not only drug use during pregnancy, but also the treatments for substance use disorders recommended by the professional associations of obstetricians (like methadone or buprenorphine treatment). Rapid opioid dose tapering and abrupt discontinuation are associated with increased rates of fatal overdose, mental health episodes, self-harm, and suicide. If someone's treatment is disrupted because they are pregnant, because of the legal environment being constructed around pregnancy, fetal personhood, and criminal liability, is their resulting death merely pregnancy-associated? Or would we be right to count it as a "true" pregnancy-related perinatal death?
Similarly, what about assault and homicide? Perinatal deaths from these causes are similarly not related to the physiological changes wrought by pregnancy itself. But to argue that they aren't related to pregnancy seems wrong. I worked for a number of years at a domestic violence hotline. It is well known, including from my professional experience, that abuse tends to begin and escalate during pregnancy, even to the point of murder. The reasons for this are cultural and social rather than physiological -- but are they any less salient? I would argue, especially in an environment where abortion and even birth control or sex education are increasingly criminalized, as they are in our country right now, that these causes are equally salient even if they are not as amenable to chopping up for a rote multivariable regression analysis.
All the talk of trends obscures what is really going on here -- the authors of the AJOG study want, probably because they are epidemiologists, clinicians, and very old hats in the realm of maternal mortality research, to restrict the counting to these physiological, clinical things; this undercount feels to them more "accurate" than the more expansive overcounting of the checkbox method. Rather than an intensification or diminution of existing trends, I think what we are experiencing is a qualitative transformation in how pregnancy is culturally, socially, and medically conceptualized. Pregnancy is increasingly a category or a state of suspicion, criminalization, dehumanization, and control as the state assumes the role of an abusive partner for all people capable of gestation. What's at stake in these arguments about how to count pregnancy-related deaths is how much or how expansively we want to politicize the public health problem of perinatal deaths.
*One question I had immediately: there weren't really any changes in surveillance practices from 2018-2021, so how could these surveillance practices alone account for the increase? Seems like an obvious question, but you should never be surprised that obvious questions elude a group of epidemiologists.