Thread by @lymanstoneky

Your regular reminder that a major reason the US is high on both of these is that *we measure them more inclusively* than many other countries.

https://twitter.com/_cingraham/status/1461387993516687362

There are many reasons to not correct! One reason is that it's not always clear which method is better so the correct direction of correction is unclear.

Another is that we sometimes have no idea what effects may exist, because we don't have parallel measurement.

That is, sometimes two measurement techniques exist in parallel so we can see precisely how they differ in aggregate effects. But often we don't have this, so all we know is "it's different measurements so may not be comparable."

The main source of the chart is the OECD, and they literally flag the US data as a break in series and possibly unreliable!

So why is the US different?

Well, you have to think about *how we measure maternal mortality*.

There are two ways we can do this in rich countries with fairly complete registration of deaths. Poorer countries are a different story we'll ignore for now!

In rich countries, the USUAL way, done in most countries, is that there are certain *causes of death* which we know are associated with pregnancy and delivery. So you add up all those causes, divide by recent births, and bam, maternal mortality!

That's more-or-less what most countries do. There's some extra finessing and sometimes there are some efforts made to catch "maternal mortality from non-perinatally-related causes" but generally we ID delivery-related causes and divide by births.

But this method isn't actually very good! It has a big problem!

Ya know what that problem is?

Delivery could impact your body such that you die of *non-pregnancy* causes. Earlier today I tweeted about postpartum depression. We know depression alters suicidality too.

So in the very-conventional way of measuring maternal mortality, postpartum-depression-induced-suicide would (generally, tho there are exceptions!) not *tend* to be included in maternal mortality.

But this seems wrong, because this kind of death actually *is* related to delivery. Likewise, delivery could alter women's odds of dying from a whole host of facially-normal causes that don't scream "PREGNANCY AND DELIVERY!" to a statistician.

To correct for this, the US *began to add*, in 2002, a "pregnancy checkbox" on death paperwork, where you check it if the woman was pregnant. The trouble: different states have written their checkbox in different ways!

Some ask "pregnant in last 42 days" some ask "pregnant in last year" and there are other permutations. But the point is it's something of a crapshoot.

But the upshot of it is that a HUGE share of US maternal mortality is deaths *more than 42 days after delivery*.

Why does this matter?

Because the WHO definition used by *every other rich country* cuts off the data at 42 days.

So the US data is "deaths of any cause within a year of being pregnant." Other countries are either "deaths of pregnancy-related causes within 42 days of being pregnant" or "deaths of any cause within 42 days of being pregnant."

SPOILER:

One year is 365 days, which is almost NINE TIMES AS LONG as the 42 day window being compared against.

Yikes!

Here's a nice CDC report laying a lot of this out: cdc.gov/nchs/data/nvsr…

Now, for the record, I think that the one-year standard is actually a *better* standard than 42 days. I think that partly because, uh, *duh*, but also because using year-increments is easier to use for historic demographic replication.

So for example, in historic demography we often only have YEAR of births/deaths. So using year-basis is a nice thing, since we can replicate it going back very long time periods, whereas replicating 42-day windows is harder.

Careful analysis using consistent measures has found that 1) the US maternal mortality rate has NOT increased on an age-adjusted basis, 2) the US maternal mortality rate probably IS understated, but 3) probably other countries understate even more

Nice example of this here: journals.lww.com/greenjournal/F…

Moreover, the "pregnancy checkbox" is wrong! A nice study showed that *21%* of reported maternal deaths *did not have a pregnancy* in the last year! ncbi.nlm.nih.gov/pmc/articles/P…

So while we may underestimate maternal mortality, the reality is we are also capturing a very large amount of *fake* maternal mortality.

The point isn't that the true number is higher or lower, just that measuring this with any precision is extremely difficult.

Unfortunately, CDC WONDER does not, as best I can tell, explicitly code which deaths are ID'd from the checkbox. However, we can make a decent guess! Some maternal death codes rise in frequency a lot in recent years, while others do not.

Moreover, some maternal death codes relate to very specific maternal causes of death, whereas others are vaguer or less specific: these are *more likely* to be from the checkbox.

And of course, there *is* a death code for later-than-42-days, so we can see what role that's playing!

And what do we find?

Here's maternal mortality rates by code using the multiple-cause files and the underlying-cause files.

The multiple-cause files mean a death could be included in *multiple* of those lines, whereas in the underlying cause file each line is fully separate.

So you can see that maternal mortality by this measure has risen, but that the black line's % increase is small compared to the blue and red lines. Indeed blue is ZERO early on.

This has a very big effect!

Here's how maternal mortality has increased since 1999, if you include or exclude the codes which have seen the most change in measurement.

At the state level, for privacy reasons, it's hard to disaggregate all this stuff. But we can look at total maternal mortality for states with enough data and ask how much their mortality rose after a pregnancy checkbox was added, since checkboxes were added at different times.

Checkboxes were added between 2001 and 2017. Here's how mortality changed before/after them:

As you can see, when pregnancy checkboxes are added, maternal deaths measured tend to rise by AN ABSOLUTELY HUGE AMOUNT. On average, pregnancy checkboxes DOUBLE the number of reported maternal deaths.

Now, this doesn't mean they're wrong! Remember, we know from detailed case studies that only about 21% of these deaths are actually misreported! So we were clearly *missing* lots of deaths beforehand.

But it means that data systems WITH a pregnancy checkbox are completely incomparable to data systems WITHOUT a pregnancy checkbox.

And since AFAIK the presence or absence of such a checkbox is not one of the factors used by e.g. WHO/OECD to guess at true maternal mortality rates, any comparisons are going to be extremely fraught.

So does the US have an abnormally bad maternal mortality rate?

.... I dunno! Because until we have a clear idea of which European peer countries actually have *identical* data collection processes, it's clear that even *tiny* measurement differences have *huge* effects!

Standardizing statistics across different systems is *extremely* difficult, and in the US you've got 51 state/DC systems to standardize AND THEN you have to try to figure out how to make it fit with international definitions.

Beyond that, international "definitions" are often unclear: whether you have a pregnancy checkbox or not does not impact whether what you report *meets the WHO's definition*! But it has a huge impact on rates!

So standardizing to a *definition* is not sufficient. You have to standardize *to a data-generating process*. The correct way to standardize is not to a standard like not "mothers who die in X period" but something like "mothers who die in X period as measured by X method"

Where you not only adjust to fit the definition, you make adjustments to guess what the best method *would* show if applied in that context: and on that, the pregnancy checkbox is a huge issue.

And note that my suspicion here is the checkbox is actually a good improvement in the data, and more countries should get more inclusive in counting maternal deaths.

Final note:

I have no objection to doulas or even financial support for doulas. Doulas can be great and increasing access to their services might be good. But the data cited is still not very good even if the cause is just!