Do Black Physicians Double The Survival Odds Of Black Newborns?

An extraordinary thing happened on Friday, July 7 (at 4:58 pm; h/t to Ted Frank, the Director of Litigation at Hamilton Lincoln Law Institute, for this detail), just over a week after Supreme Court Justice Ketanji Brown Jackson’s scathing dissent from the court’s ruling against racial preferences in college admissions.

The authors of an amicus brief submitted on behalf of the Association of American Medical Colleges filed a correction to the record owning up a statement in their brief (p.4} that turned out to be flagrantly false: “for high-risk Black newborns, having a Black physician is tantamount to a miracle drug; it more than doubles the likelihood that the baby will live.” Regrettably, this falsehood was quoted verbatim in Justice Jackson’s dissent.

An Elementary Statistical Error

To be clear, the claim that having a Black physician (a pediatrician in most cases) doubles the chances of survival for Black newborns was not invented out of whole cloth. It is based on a 2020 study of more than 1.8 million Florida births over the years 1992-2015, led by George Mason University business professor Brad Greenwood. One need not even have taken a basic introductory course in statistics to understand the error in logic underlying the specific claim made in the AAMC brief. The Greenwood et al. study showed that under the care of White physicians, Black newborns experience 430 more fatalities per 100,000 births than White newborns. However, under the care of Black physicians, this “mortality penalty” (as the study termed it) for Black newborns was reduced by 58% (i.e., 257 fewer deaths per 100,000).

At first glance, it might seem like common sense that reducing mortality risk by more than one-half would imply that the corresponding chances of survival would more than double. Sadly, intuition fails rather badly in this case. The overall mortality risk for Black newborns in this study was 894/100,000 births. But this, of course, implies that the average newborn had a 99.1% chance of survival! No intervention could have remotely doubled such a high likelihood of newborn survival: even a magic pill that eradicated all Black newborn deaths at best would have increased the likelihood of survival by less than 1%. If care by Black physicians indeed results in 257 fewer deaths per 100,000 infants, this implies a mere 0.26% increase in the likelihood survival, which obviously is quite far from doubling the odds of living for Black newborns.

Are Black Newborns More Likely to Survive if Treated by Black Physicians?

It is commendable that AAMC officially corrected the record on this matter, expressing their regret at “any confusion that may have been caused by the statement in the brief.” However, they took further pains to state “But however it is summarized, the study strongly supports the statement in Justice Jackson’s dissent that “the diversity that UNC pursues for the betterment of its students and society . . . saves lives.” 2023 WL 4239254, at *105 (Jackson, J., dissenting).” The statistics cited earlier from the Greenwood paper certainly would appear to support this revised claim.

Regrettably, this study has some sizable methodological flaws that have been extensively codified and explained in an August 2020 podcast by UCSF professor Vinay Prasad, MD, MPH (full podcast is 91 min. but relevant discussion runs from 1:31:00 to 0:52:00 mark). As he explains in the podcast (and here for those who do not have time to listen), Dr. Prasad is fully supportive of efforts to correct the underrepresentation of minorities in medicine. But he also believes that the claim made in this paper is very dubious in light of its methodological limitations and that corroborating evidence is essential before anyone relies on it to guide policy.

Although the Florida study observed nearly 2 million births—a very impressive sample size—at the end of the day, it is simply an observational study using billing records, not medical records (which would have been enormously more informative). The authors used statistical controls to make the groups being compared as statistically equivalent as feasible, but there was much that could not be controlled for, which inherently limits what can be reliably inferred from this 30,000 foot view of what actually was happening to care of infants in Florida. Put a different way, this was not a randomized controlled trial—which is the “gold standard” routinely required by the FDA to test drug efficacy and drug safety. And for that reason, we must be extremely cautious in interpreting its findings.

There are several major problems with this analysis.

Lack of NICU Data

The dataset used in this study has no indicator showing which infants were treated in neonatal intensive care units (NICUs) rather than receiving routine post-partum inpatient hospital care. This is a sizable limitation insofar as it is well-established that low birthweight (LBW) is one of the leading causes of infant mortality, meaning that a disproportionate share of newborn deaths (which are rare to begin with) will occur in NICUs whose principal purpose is to keep LBW infants alive. Moreover, Black mothers are more likely to birth low-weight infants (in this California study, for example, “African American women had a persistent 2.4-fold greater prevalence of having an LBW infant compared with white women.”). But it also is well-established that Blacks/African-Americans are grossly underrepresented among neonatologists, constituting only 3.8% of these MD specialists who work in NICUs.

The implications of all this, as Ted Frank red-flagged in a Tweet last fall, is straightforward: “The white docs aren’t seeing the same infants as the black docs. They’re more likely to get the NICU cases where all infants are less likely to survive, and study doesn’t control for that.” He followed with: “the study is confusing correlation with causation: if you have a black doctor, your baby is more likely to survive, but that’s because that means you’re less likely to be in the NICU, where there are fewer black doctors. It has nothing to do with the race of the doctor.”

The study itself offers a variety of clues that support this explanation of the observed “mortality penalty:”

• The study replicated its finding using hospital length of stay—found in other studies to be a consistent indicator for quality of care in healthcare settings. But since NICU lengths of stay obviously will be much longer than for healthy newborns, it should be no surprise that LOS for Black infants treated by White physicians are higher than for their counterparts treated by Black physicians, because again this is an apples-and-oranges comparison of two very different groups of infants.
• Of equal significance, the study failed to replicate its finding when examining maternal deaths among the same cohort of births. That is, Black mothers do not experience the same mortality penalty as their newborns. If indeed White physicians are less capable than their Black peers of rendering the appropriate care to Black patients due either to lack of knowledge of, worse communication or unconscious bias, why don’t these same limitations produce adverse mortality consequences for moms? What coherent explanation could account for why physicians treating infants would be susceptible to such effects while their OB counterparts evidently are immune? This paradox disappears once we recognize that infants are differentially distributed across NICUs and regular OB beds whereas their mothers are not.

Attributing Infant Outcomes to a Single Physician is Problematic

Dr. Prasad’s podcast describes in depth how misleading it is to attribute an infant death to a single physician. The average NICU stay, for example, is over 13 days. During that time, an infant will be seen by rotating cast of doctors and medical staff. The AHCA database used captures the “NPI number of the medical doctor, osteopathic physician, dentist, podiatrist, chiropractor or advanced practice registered nurse who had primary responsibility for the patient’s medical care and treatment.” For infants who end up dying in the NICU, the lion’s share of the care will have occurred in the NICU—whether it lasted for days or weeks—and the responsible doctor much more likely to be a neonatologist than if the infant died in the course of a routine inpatient stay (most premature babies enter the NICU within 24 hours of birth).

Thus, even the most highly skilled neonatologist is likely to have a higher infant mortality rate among his/her newborns than a pediatrician treating healthy newborns. But of equal importance, there is going to be a much larger team of individuals over a much longer period of time caring for that infant, introducing a multiplicity of ways in which one human error or misjudgment can culminate in death for highly fragile infants. Even the most conscientious and skilled practitioner cannot control everything in such an environment. And this study has no way of identifying the race of whatever caregiver might have been the proximate cause of an infant death. And most infant deaths will not be the result of error on anyone’s part. So this study provides a rather crude and hyper-simplified look at a very complex reality.

Other Unmeasured Selection Problems

It is certainly true that the study carefully controlled for some key variables of interest, including insurance status (e.g., Medicaid, self-pay), 65 most-prevalent comorbidities, quarter-year fixed effects, hospital fixed effects, hospital-year fixed effects and physician fixed effects. Statistical experts will immediately recognize and understand the nature of the analysis being done here: the authors conscientiously did the best they could with the data available to make the comparison groups statistically equivalent. The average reader need not know the meaning of “fixed effects” or exactly how such variables are controlled for statistically. Suffice it to say that the inclusion of these variables greatly matters for interpreting the observed mortality differences by race of patient and physician.

Specifically, without taking into account all these variables, Black newborns treated by White physicians have triple the in-hospital mortality rate (894/100,000) of White newborns treated by White physicians (290/100,000). So the mortality penalty is 604 deaths/100,000 (894-290). But once all these other variables are taken into account, the mortality penalty shrinks to 318/100,000—nearly a 50% reduction. The first model, taking into account only patient and physician race, has an R-squared of 0.001, meaning it explains less than 1/10 of one percent of all the variation in observed mortality! The second model, with all controls variables included, explains 13.7% of mortality variation. So it certainly does much better, but still leaves a whole lot unexplained.

One factor that might further explain mortality differences, elaborated at length in Dr. Prasad’s podcast, is the possibility that wealthier Black families may be more likely to have selected a Black pediatrician in advance rather than relying on luck-of-the-draw once the mother arrives at the hospital for delivery. For a variety of reasons, wealthier families are likely to have better birth outcomes—few premature/LBW births, fewer adverse outcomes related to maternal smoking/alcohol/drug use, etc. And the mere fact of having selected a pediatrician in advance implies a level of communication and understanding that might be absent in infant-patients pairings that occur under emergency conditions in the middle of the night. So this selection effect will result in Black pediatricians having better birth outcomes than they might otherwise, but would be attributable to the characteristics of the infant’s family, not anything to do with the quality of care provided by the physician.

Some Inconvenient Contradictory Findings

Ted Frank has pointed out in a recent Wall Street Journal op-ed that strictly speaking, when analyzing a very low probability dichotomous event (i.e., death either happens or it doesn’t), the proper analytic tool is a logistic model rather than the linear regression model used in the main paper. Greenwood et al. did indeed run conditional logit estimates in an appendix (Table S9). These findings still show that the mortality penalty shrinks for Black infants treated by Black physicians, basically confirming the main paper’s results.

However, the most complete version of the model (which controls for insurance, co-morbidity, time and hospital), White newborns treated by Black physicians experience a higher mortality rate that is statistically significant. In none of the many variations of the main paper’s linear regression results was the impact of Black physicians on White infant mortality even close to being statistically significant. This inconsistency in results depending on how the data are modeled—which the authors did not even attempt to explain—provide one additional strong reason to treat the results with caution.

Even if the Greenwood Study Results Are Correct, How Many Lives Could Be Hypothetically Saved by 100% Physician-Patient Racial Concordance?

Interestingly, the AAMC brief focuses on a single set of showing that 257 lives per 100,000 births potentially could be saved were all Black infants treated by Black physicians—representing a 58% reduction in the mortality penalty. But these results are for an intermediate model that does not control for the full set of variables and only explains about 5% of the variation in mortality rates. A better model reported in Table 1 splits the sample into infants treated by White doctors compared to infants treated by Black doctors showing that the mortality penalty is 318 deaths/100,000 Black infants treated by White doctors, but only 169/100,000 for their counterparts treated by Black doctors (these models explain roughly 15% of variation in mortality rates).

That represents a relative reduction of 47% rather than the 58% figure used by AAMC and an absolute reduction of only 149/100,000 rather than the 257/100,000 AAMC figure. That’s still arguably a “miracle drug” for those inclined to characterize it in that fashion, but also likely to be directionally closer to the true effect size given all the ways in which the study’s methodological limitations lean in favor of over-estimating the impact of racial concordance on infant survival odds. So for purposes of discussion, I will treat this figure as the best possible outcome since for all the reasons described above, I certainly do not view it as the most likely outcome.

In the Florida sample (covering 22.8 years), 362,597 Black infants were treated by White physicians and 97,304 were treated by Black physicians. Thus, hypothetically, had all infants treated by White physicians instead been treated by Black physicians, then hypothetically 540 lives could have been saved among this sample, or about 24 per year. Note there were 521,740 additional Black infants omitted from the sample either because the race of their physician was neither Black nor White or could not be determined. But I have no way of determining whether the observed mortality penalty reduced would apply to them, so have excluded them from consideration in terms of calculating potential life-years saved. That said, together with the other Black births described, amount to 981,641 total births (20.4% of all Florida births), which averages to 43,000 Black births per year.

Currently, Black life expectancy at birth is 70.8 years. So assuming full physician-patient racial concordance was even possible for Black infants in Florida, the saving of 24 lives would add 1,699 years (=620,208 days) of life expectancy to the 43,000 annual Black births, or about 2 weeks of added life expectancy for each birth cohort. Thus, it would shrink the current 5.6 year life expectancy gap between Blacks and Whites by 0.7%. But that assumes every Black baby now treated by a White physician could be treated by a Black physician, which is obviously completely unrealistic.

Currently, 5.7% of U.S. pediatricians are Black or African-American, vs. 13.9% of those under age 18. Assuming this same racial imbalance exists in Florida, the current supply of pediatricians would need to increase by 140% to reach racial parity. There really would be no compelling policy justification for going beyond this point. But that would imply that the number of Black infants treated by Black physicians would increase only 5,974 per year, reducing predicted infant deaths by 9 rather than 24. This implies adding 5.25 days to average Black infant life expectancy each year which means shrinking the racial mortality gap by about 0.25%. No one should sneer at such an achievement, but I myself would hesitate to characterize these as “miracle drug” results.

Even if Physician-Patient Racial Concordance Improves Outcomes, Should This Be Used to Justify Race-based Admission Preferences?

Those willing to read the entire Greenwood et al. study may be astonished to discover a passage that somehow did not make it into the AAMC brief or Justice Jackson’s dissent: “For families giving birth to a Black baby, the desire to minimize risk and seek care from a Black physician would be understandable. However, the disproportionately White physician workforce makes this untenable because there are too few Black physicians to service the entire population. Moreover, it avoids the foundational concern of resolving the disparities in care offered by White physicians. Finally, it is important to note that physician performance varies widely among physicians of both races, suggesting that exclusively selecting on physician race is not an effective solution to mortality concerns” [emphasis added].

This point was reinforced in a talk Greenwood gave a year later in which he stated “One thing that usually comes up about this time in the conversation is, like, “Okay, well, this suggests that Black moms should only have their babies be treated by Black doctors and vice versa for Whites.” And I think that that is worst possible interpretation of the paper. Any person should be able to go into any hospital, in these United States, and receive equal treatment.” Greenwood’s point is that more research is needed to discover the specific reasons White physicians produce worse infant outcomes and correct these through better training and education, not segregate patients by race.

Imagine the counterfactual, that the Greenwood study had shown treatment by White doctors reduced mortality among Black infants. Would anyone be using those results to argue for tilting medical school admissions policies in favor of White applicants in order to save Black lives?

The Supreme Court has spoken. We should see how things play out. But as the foregoing suggests, even in a world of entirely race-neutral medical school admissions policies, we need not worry about there being an explosion in Black infant mortality. In the meantime, health services researchers have plenty of further work to do to more accurately measure whether racial concordance indeed produces disparities in birth outcomes and if so, how specifically these arise and how they can be fixed.