All of these articles are completely wrong – utterly and stupendously wrong. They have failed at basic mathematics, which creates unneeded fear among expectant mothers. Maternal mortality in America is not high, it is not rising, and the risk factors for it are not mysterious and many of them can be controlled.
My claim here may seem outlandish given the popularity of apocalyptic narratives about maternal mortality. The reality, however, is that the seemingly escalating rates of maternal mortality have been driven by a simple change in a standard reporting form. We know this is the case because the form has been changed at various times by different states over the past 20 years, always with the same effect — a sudden spike in maternal mortality that was not driven by new deaths, but the revised form.
Recent scholarly analysis from the Centers for Disease Control and Prevention (CDC) suggests that the U.S. maternal mortality rate (MMR) is three times higher than it would be if we didn’t use the revised form. The situation is so severe the CDC has had to change how it handles maternal mortality-reported death cases. Although the CDC officially says the U.S. maternal mortality rate is a whopping 32.9 deaths per 100,000 births (vs. between four and 10 in Europe), a CDC report from 2020 notes that if we used the pre-2003 estimation method, based on the previous reporting form, the maternal mortality rate was 8.9 in 2002, 8.7 in 2015, and 8.7 in 2018. In other words, essentially unchanged.
It’s pretty open-and-shut. U.S. maternal mortality rose because of a change to the reporting form. That’s the entire story. The articles about crisis levels of maternal mortality are completely wrong. The new forms give us more information but they do not demonstrate either that the U.S. is an outlier when compared to Europe, or that something new and health-related is driving more maternal deaths.
Policymakers and the public should make decisions based on the best available data. The example of the misguided maternal mortality narrative shows how dangerous it can be when serious errors of interpretation and comparison get entered into the system.
I will explain why the data are in error in depth below. But to understand what’s going on, we must begin by understanding what maternal mortality is.
What Is Maternal Mortality?
Pregnancy and birth come with risks to the mother. It is a difficult, hard process on the body, with numerous side effects, and birthing a baby is not a medically trivial task. Prior to modern medicine, many women died during or immediately after childbirth due to various birth-related factors.
Because birth historically elevated women’s death risks so much, and because the death of a mother as a result of her child is so especially tragic and poignant a cause of death, maternal mortality has – entirely justifiably – attracted special and unique attention.
Society has extremely good reasons for having unique care for women’s likelihood of dying as a result of pregnancy or birth. As a result, statistical bodies take great pains to estimate “maternal mortality ratios” or “maternal mortality rates”: Basically, how many women die related to birth, compared to how many births occur.
This creates two analytic tasks: first, assessing how many births occur, and second, assessing how many women die related to birth. Counting babies is the easier of these two tasks and is done pretty reliably in most industrialized countries, at least (though below I’ll explain that it’s not really that simple). The harder part is identifying deaths.
The gold standard for identifying maternal deaths is known as vital record linkage. In this system, you take all the deaths of women under age 60 or so in a given year, and you use a unique identifier (such as a Social Security number), and you search for that identifier in a database of births in the same and prior year. When you get a match between an identifier marking a mother in the birth data and a death in the death data, you then check the cause of death. In principle, if it were “car accident,” that may not be maternal mortality, but if it was “infection” it might be. This yields an estimate of maternal deaths. Dividing by births yields the maternal mortality rate.
But this process is very laborious and data intensive. So the usual way maternal mortality is measured is simpler. For every death, somebody (such as a coroner or medical examiner) marks a cause of death. It might be something like “pneumonia” or “gunshot wound.” But there’s a category of causes specifically related to pregnancy and birth. When the examiner filling out the death form is reviewing the case, they look for those causes and, if present, mark the death as within the group of causes we classify as “maternal mortality.” That’s how most countries classify maternal deaths. And that’s what the U.S. used to do, until things began to change in 2003.
In 2003, the CDC became concerned that this approach was undercounting maternal deaths. They recommended that states update their death certificates to include a feature that a few states already had: a pregnancy checkbox. This checkbox would ask, on every death certificate, if the deceased had been pregnant in the last 42 days (the usual standard time for maternal mortality classification) and the last year.
This change may seem subtle, but it’s vitally important. Suddenly, death certificates started showing up with extra data about maternity. A woman dies of pneumonia and has a box checked saying she “was pregnant” in the last 42 days – she’s maternal mortality now, even if her cause of death might not have previously been considered “maternal.” In general, causes that are extremely distal from birth (like car accidents) are still supposed to be excluded, but there are many health conditions that can cause maternal mortality, and now U.S. death certificates have gotten way broader in capturing births.
As a result, whenever states adopt the new checkbox, their maternal mortality rates skyrocket, because tons more maternal deaths are discovered, and that’s because tons more pregnancies among dead people are discovered.
Checking the Checkbox
Compare the trendline for maternal deaths using the prior method to headline MMR data, which reflects the checkbox:
(Lyman Stone)
Was this difference really caused by the checkbox? The answer is a resounding yes. We can know this because states didn’t all adopt the checkbox at the same time, but staggered in different years between 2003 and 2018. Maternal mortality is sufficiently rare, and many states are so small, that we can’t get valid state-specific estimates of their maternal deaths to assess. And a few states had checkboxes or similar tools before adopting the official CDC checkbox.
But there are 18 states which were large enough to have consistently reported maternal mortality, had no prior checkbox before adopting the CDC-recommended one, and which adopted the checkbox all at once for a full year instead of rolling it out within the year. For those 18 states, we can see clearly what happened when they adopted the checkbox.
(Lyman Stone)
In general, MMR rises sharply when a checkbox is adopted. The average increase, in fact, is 104 percent in the year of adoption, which means the checkbox led estimated MMR to double.
Because states rolled out the checkbox in different years, the national data obscures the sharp changes around the adoption of the checkbox. But in state data, it’s crystal clear. U.S. maternal mortality is rising because we changed how we measured it. If we measured it the way we did before 2003, there would have been no increase at all.
In a formal panel model with a variety of control variables and state and year fixed effects, I confirm that the year of adoption of the checkbox is associated with a doubling of maternal mortality.
Careful readers may note that this creates a new question: Why does the checkbox increase maternal mortality estimates?
Where Babies Come From
The gold-standard way to assess maternal mortality is to link birth records to death records. The usual method is to look at death causes related to live birth. However, the checkbox doesn’t ask about birth. It asks about pregnancy. That turns out to be very different.
It’s pretty easy for a country to count its births, but counting pregnancies is much, much harder, because there’s no official record of them. Thus, in France, for instance, if a woman has an abortion and then dies, that may not be counted as maternal mortality because there’s no birth record. But in the United States, she would be counted, because she was pregnant. This is the key intuition: The U.S. checkbox is actually capturing pregnancy-related mortality rather than birth-related mortality.
The U.S. approach captures women who die within 42 days of birth or abortion or miscarriage, from classically birth-related causes. And it turns out, there are quite a few of those women. Perhaps a fifth to a third of U.S. maternal mortality consists of women who were indeed pregnant, but did not have a live birth: They had a miscarriage or abortion, and then died within 42 days of pregnancy-related causes. Those cases are ignored by European maternal mortality.
They shouldn’t be. Pregnancy-related death arising from the lack of a birth is an important part of maternal mortality. It’s absolutely crazy that the “gold standard” approach ignores this component of maternal mortality. The U.S. has higher maternal mortality partly because our broader measure of mortality is correct and the narrower European approach is wrong. If they adopted our more correct approach, their maternal mortality rates would rise quite a bit – very likely by two to three times, as ours did.
So part of the reason the checkbox raises maternal mortality estimates is because it captures deaths resulting from pregnancies that didn’t yield births. The other reason is that the checkbox is sometimes just wrong.
Careful study by state vital statistics offices and the CDC has found that the pregnancy checkbox also leads to misclassification, especially when nonprofessionals complete the death form. Anywhere from five to 20 percent of women classified as pregnant within the last year were not, in fact, pregnant in the last year. This misclassification is 40 percent higher among Black women.
Thus, the U.S. maternal mortality rate not only actually includes a wide range of pregnancy mortality divided only by births rather than all pregnancies, but it also includes a five to 20 percent overcount of pregnancy mortality due to misclassification, and this overcount is more severe among Black women.
This helps explain those articles linked above pointing to rising maternal mortality among Black women: Mere misclassification bias alone has caused a major increase in Black women’s maternal mortality. And since Black women also have elevated rates of abortion, the checkbox would also tend to push their maternal mortality up more.
On the whole, then, the checkbox – both its virtues (tracking a broader definition of maternity) and its vices (misclassification with a racial bias) – can explain not only “rising maternal mortality,” but also some part of the racial gap in maternal mortality.
Maternal mortality is not rising in America, and the racial gap in maternal mortality is probably much smaller than headline data suggests.
How Much Does Pregnancy Really Increase Death?
Maternal mortality is an important indicator. But in a world where maternity-related medicine has advanced leaps and bounds beyond where it was a century ago, it’s worthwhile to wonder if using cause-specific data is actually appropriate.
Rather, what if we try to measure “excess mortality” among pregnant women? Do pregnant women have different death rates than non-pregnant women?
This is a bit difficult to assess. In principle, every death certificate with the pregnancy checkbox checked should be counted for this basis, and the CDC has this data. It is not entirely clear, however, if the CDC has actually preserved all pregnancy-checkbox data in public multiple-cause mortality files.
But assuming that it has done so, that yields about 6,500 pregnancy-checkbox deaths from 2018 to 2021, vs. about 3,500 core “maternal mortality” deaths. To estimate pregnancies, I use published birth data for 2018 to 2021. Assuming pregnancies have nine-months gestation, and that the checkbox could be flagged up to 12 months after a birth, each birth yields 21 months of exposure time.
I also use data from the Guttmacher Institute, estimating that the average abortion happens at three months, and also that checkboxes could detect that pregnancy up to 12 months later. Finally, I assume that miscarriages represent between 20 percent total human conceptions and occur, on average, at one month of gestation, thus creating 13 months of exposure. I calculate deaths per exposure year from 2018 to 2021.
I then compare this to all non-pregnant deaths of women for the same period, to estimate mortality rates for each group; each exercise is conducted for five-year age bins.
(Lyman Stone)
It’s very likely the CDC is not capturing 100 percent of checkbox deaths in the publicly reported data. Nonetheless, all deaths with any pregnancy-related condition as any cause of death reveal a much lower death rate than women generally, except for women under age 15.
That is to say, assuming CDC is faithfully assigning at least one pregnancy-related condition as a cause for each pregnancy-related death, then it appears that maternal mortality is actually lower than mortality for non-pregnant women.
This is somewhat shocking since pregnancy and birth are serious medical conditions with serious risks, but the possible reasons why pregnancy may be associated with lower death odds are fairly straightforward:
- Very sick, unhealthy or death-prone people are probably less likely to become pregnant in the first place.
- Very sick people are likelier to miscarry or abort if they do become pregnant, shortening exposure time.
- Pregnancy causes many Medicaid-ineligible women to become presumptively eligible for Medicaid.
- Pregnancy may induce women to reduce tobacco, alcohol or other substance use.
- Murder, a major cause of death for young women, might decline during pregnancy due to a possible reduction in intimate partner violence (though this is much debated).
The first two factors suggest that women who become pregnant are just a different, incomparable population, and so estimates of pregnancy-related mortality are just not going to capture their real risk exposure. The third through fifth reasons all point to possible changes pregnancy could cause to women’s environment and behavior, and the behavior of those around them.
All are speculative, but in principle, measuring excess mortality during and after pregnancy is probably a better way for countries with good record-keeping to measure maternal mortality, rather than mere cause-counting. Large, longitudinal data sets could be used to estimate the effect of pregnancy and birth on odds of death, for example, giving a much better idea of the real risks of pregnancy for women.
Until then, we are left with maternal mortality data as it is. If we use a consistent reporting measure for the last two decades, what we find is no major long-run increase in maternal mortality. U.S. maternal mortality is very similar to other rich countries if we measure them apples to apples. Our official rate is higher because we have adopted a better and broader measurement standard for pregnancy risks. Thus, our high numbers do not reflect a public health failure, but rather a statistical measurement success.
Lyman Stone is director of research for the consulting firm Demographic Intelligence, a research fellow at the Institute for Family Studies and a PhD candidate at McGill University.
Governing’s opinion columns reflect the views of their authors and not necessarily those of Governing’s editors or management.
Related Articles
