Do Black Doctors Save More Black Babies?

Today’s post, for efficiency reasons, is another freestyle audio. For my devoted readers who are strongly allergic to audio, I’ve asked ChatGPT to write a short summary piece in my style, which I’ve lightly edited. That’s what you see below. Enjoy!

RT is totally reader-supported. To support its survival, please chip in by becoming a paid subscriber.

I am sorry to report that coffee is strongly associated with lung cancer.

True fact: Many studies from many countries have confirmed this finding.

That statement is therefore statistically true—but deeply misleading. Coffee drinkers develop more lung cancer because they are slightly more likely to smoke. Remove smoking from the equation, and the coffee–cancer relationship disappears.

Smoking is what epidemiologists call a confounder. I prefer the more cinematic term: lurking variable. It’s hiding in the fog, waiting to make two unrelated things look causally connected.

Keep that coffee example in mind.

In 2020, researchers published a provocative study in the prestigious journal Proceedings of the National Academy of Sciences. Using Florida hospital records, they examined roughly 1.8 million Black and white newborns and asked whether babies fared better when their physician was the same race.

The racial disparity was enormous. Black infant mortality was about 0.9%, compared with 0.3% among white infants.

The headline finding was even more explosive: Black newborn mortality was approximately 0.9% under white physicians but 0.4% under Black physicians. White infant mortality did not meaningfully change with physician race.

The researchers knew confounding was a problem. They adjusted for insurance, year, hospital and 65 commonly recorded diagnoses. Those adjustments eliminated roughly 80% of the original difference—but a smaller association remained.

Then the paper crossed a crucial scientific line.

This was an observational study. It could identify an association, but it could not establish cause and effect. Nevertheless, the authors wrote that Black physicians “systemically outperform” their colleagues when caring for Black newborns. They suggested that Black families might understandably seek Black physicians.

The media went wild. The finding even entered a Supreme Court affirmative-action case, where Justice Ketanji Brown Jackson cited it as evidence of the importance of Black physicians.

Given how the study was written and where it was published, believing its conclusion was entirely reasonable. The failure belonged to the scientific gatekeepers who allowed an observational association to be presented as a causal effect.

Then came the lurking variable.

In 2024, another research team obtained the original data and reproduced the analysis. They got essentially the same initial results. Then they added one variable the original study had omitted: very low birth weight.

This wasn’t an obscure technicality. Very low birth weight is the dominant predictor of newborn mortality.

Black infants weighing less than 1,500 grams were far more likely to have been treated by white physicians. Approximately 3.4% of the Black infants cared for by white doctors were in this extremely high-risk group, compared with 1.4% of those cared for by Black doctors.

The white physicians were treating substantially sicker babies.

Once researchers adjusted for birth weight, the supposed survival advantage associated with having a Black physician disappeared.

That does not mean racial concordance never matters. Randomized research has found that Black patients may communicate better with Black physicians and follow their recommendations more closely. It also does nothing to erase the appalling racial disparity in infant mortality.

What it means is that this particular study did not prove Black doctors save more Black babies. It found an association created by differences in how critically ill infants were distributed among physicians.

The larger failure was editorial. Scientists make mistakes. That is why prestigious journals have expert reviewers and editors. They are supposed to prevent association from being dressed up as causation—especially when the result is destined to shape headlines, public policy and Supreme Court arguments.

The lesson is simple: when you encounter a dramatic study, ask whether it was randomized or observational. If it was observational, you are looking at an association. Somewhere in the fog, a lurking variable may still be waiting.

When in doubt, remember the coffee.