Maternal Mortality, Death Certificates, and Churnalism

Celebrating great work analyzing data

I filled out my first death certificate in 1991. By 1993, I was the final arbiter of the certificates I completed. In 1997, I began to face the bewildering task of filling out death certificates for my outpatients. These were people I had not seen for weeks or months, who had died at home, for whom I had to come up with a cause of death.

I had never been trained to fill out a death certificate and often felt I was doing a rotten job of it. Even with the excellent guidance that exists (See this CDC page, a version of which I have referenced innumerable times), the cause of death that I enter on a death certificates is, sometimes, at best, an educated guess.

The consequences of this are not lost on me. I have been called by family members to discuss a loved one’s cause of death. The callers did not know why their relatives had died and wanted to learn “the facts.” Some of these people have been disappointed when I explain that “presumed pulmonary embolism” due to or as a consequence of “metastatic prostate cancer” was a guess based on what I had been told about the final minutes or hours of a person's life.

Furthermore, when I read articles, I am suspicious of any mortality outcome besides “overall mortality.” The difference between cardiovascular mortality and mortality related to an infectious disease or cancer might be definite or might be a barely considered decision made by a sleep-deprived 26-year-old cross-covering a colleague.

In addition to my conflicted relationship with the death certificate, anyone following Sensible Medicine’s churnalism coverage knows that I hold many of our health journalists in low esteem.¹ It should thus not be a surprise that I couldn’t help but click wildly when I saw this on X:

Huge congratulations to Saloni Dattani on her award and her outstanding work.² The award recognized Dr. Dattani’s article The rise in reported maternal mortality rates in the US is largely due to a change in measurement. Her work shows, fairly convincingly, that the rise in U.S. maternal mortality, which we have heard so much about, can be mostly explained (at least between 2003 and 2017) by a change in the way mortality data were collected. Here is the graph of the maternal mortality data from four countries, which appears to show a rise in death rates in the US.

To my read, and you should definitely read the article yourself, the big story is that the low maternal mortality rates we achieved between 1980 and 2002 were thanks to undercounting. The rise in the aughts and teens occurred as we gradually introduced a more accurate system. That a change in reporting had led to the apparent increase in mortality would have been obvious if the change had happened across the country at one time, but the change happened gradually as states changed their death certificates. The mortality rates thus appeared to be slowly rising.

This explains the rise in maternal mortality during this period. It also raises questions about comparisons across countries. Dr. Dattani’s article references the variability in reporting across different nations and ways in which the WHO tries to account for these differences in data collection.

Even as a total skeptic, I did not question reports I read about rising maternal mortality in the US. I should have. This probably reflects the fact that I am not only a skeptic but also a bit of a pessimist. It did not surprise me that mortality rates were rising; why would I question this? Reading this work reminds me, once again, that taking data at face value is not advisable. It also reminds me that our journalists often do not dive deep enough into the studies on which they report.

The lesson here is neither novel nor profound. When you hear a report or read an abstract, put aside your priors at first and look at the data as a frequentist. Consider, could these results represent type I error or type II error? Could this difference detected not actually exist, or have we missed a difference that does exist? Why might this have happened? Is there a problem with the data itself or with the sample size? Then, put on your Bayesian hat, look at the data in light of what you already know and expect.

1. Churnalism is the careless, incurious, and uncritical reporting of poorly done biomedical research. Churnalism trades a real story — why a study is unimportant or proves something other than it contends — for the easy headline. Sensible Medicine coverage of the topic is here.

2. Definitely dive into Dr. Dattani’s work here. It is super interesting and always presented clearly and concisely.