The AHA's "Scientific Statement" on Why They Are Hamstringing Their Heart Attack Risk Algorithm a Priori
Print Friendly and PDF

iSteve commenter res follows up on that New York Times article I blogged about earlier about how the American Heart Association has stopped asking for race as an input into its algorithm that predicts your risk of heart attack and stroke:

Here is the link to the paper itself.

Novel Prediction Equations for Absolute Risk Assessment of Total Cardiovascular Disease Incorporating Cardiovascular-Kidney-Metabolic Health: A Scientific Statement From the American Heart Association

If anything I think the paper authors topped the NYT in their rhetoric (this is a scientific paper, right?).

Here’s a key part of the “scientific statement from the American Heart Association:”


The Work Group discussed the role of race in CVD [cardiovascular disease] risk prediction. Because race is a social construct and an historically fraught proxy representing various lived experiences, there is the potential for the harmful interpretation that it represents a biological risk factor when included in risk prediction, which may result in race-specific treatment decisions.

The point of the 2013 American Heart Association algorithm was to give doctors a tool to judge an individual patient’s risk of heart attacks and strokes so the doctor can decide better what to advise (e.g., time to cut back on salt). The fact that including race helped doctors give more accurate and thus better advice to black patients was a good thing, even if nobody yet understands the precise combo of nature and nurture behind why including self-identified race as a factor improved how the algorithm worked.

But 2013 was at the beginning of the Great Awokening. Many optimists claim we are past Peak Wokeness, but big slow-moving institutions like the American Heart Association keep announcing new policies following out the demented logic of the post-George Floyd racial reckoning.

A decade later:

Therefore, it was decided a priori not to include race as a predictor in the development of PREVENT and to use the recently developed race-free equations for eGFR on the basis of serum creatinine.

Should medical decisions with life and death implications really be made on a priori grounds?

This is consistent with the growing consensus to remove the use of race from clinical algorithms broadly in medicine.

Racism, and not race, structures our social and individual lived experiences, is associated with adverse SDOH, and represents a key driver of adverse CVD outcomes. Therefore, many have advocated for the measurement and inclusion of measures of structural racism or other SDOH [social determinants of health] (eg, education, income, social deprivation index) that may be able to be intervened on.

Do you remember back in the 1990s when the Clinton Administration spent a fortune on the Human Genome Project on the assumption that genetics was linked to medicine?

Well, that was racist. Today, we know racial differences in heart attacks and strokes are due to structural racism and other social determinants of health.

How do we know this, you might ask?

The American Heart Association’s answer: Don’t ask. We just know it.

For example, QRISK, a UK-based prediction model for CVD, incorporates a postcode–level deprivation index (Townsend deprivation score). Furthermore, the inclusion of race in risk prediction may imply that differences by race are not modifiable and may reify race as a biological construct, which may worsen health disparities.

Alternatively, not allowing doctors to use the fact that blacks tend to be at more risk of heart attack and stroke could kill off more blacks, just like how Black Lives Matter got more blacks killed in homicides and car crashes during both the Ferguson Effect (2015-16) and the Floyd Effect (2020-?). But the Establishment has kept that hushed up.

Dead men tell no tales.

In this regard, it is important to note that there continue to be disparities in CVD risk factors and CVD incidence, with Black individuals having higher levels and rates, respectively. Thus, it is of crucial importance to assess and address the SDOH that underlie racial differences. However, most contemporary datasets do not routinely include comprehensive measures of SDOH, limiting the ability to integrate these factors in risk prediction. Furthermore, it should be highlighted that tools and measures to assess the direct effect of racism are currently limited.

In other words, the evidence that Social Determinants of Health and only Social Determinants of Health drive racial differences in health is lacking. Most obviously, poor mestizos in the United States tend to have remarkably good health relative to their income and education. It’s called the Hispanic Paradox.

But only infidels and heretics ask questions like that. True believers don’t need data to believe. They have faith in the teachings of the great scientist Ibram X. Kendi.

Therefore, perhaps most critically, concerted research efforts are needed to determine the nonbiological factors that underlie racial differences in CVD risk

But NOT the biological factors, which are just social constructs.

and continue to update and revise risk prediction models to enhance assessment with these measures. In the current model development of PREVENT, the social deprivation index at the zip code level was included in derivation among the subsets where available. However, despite interest in inclusion of measures that more directly reflect risk related to racism (eg, residential segregation, perceived racial discrimination) and additional individual- and place-based measures of social drivers (eg, income, education, residential green space),

Mexican-American neighborhoods are famous for their abundance of income, education, and residential green space.

the lack of standardized assessment and capture in data sources was a key limitation.

Therefore, although the PREVENT equations represent a critical step forward, integration of the social deprivation index is only a first step; the inclusion of relevant measures that represent individual experiences of discrimination, structural and systemic racism, and individual- and place-based SDOH should be a priority in risk prediction moving forward. As we move forward and strive to transform care delivery to equitably improve CKM health, we must acknowledge the contributions of structural and systemic racism in CVD risk.

We must.

res continues his comment:

The supplemental material is available here.

Perhaps most useful is “Supplemental Table. Comparison of Demographic Factors, Predictors, and Outcomes in the newly developed 2023 PREVENT equations and the ACC/AHA 2013 Pooled Cohort Equations.”
Which compares the variables used in the two models.

I find it notable that there is no comparison of the predictive power of the two models. As ic1000 notes, surely that is of primary importance? An easy way of doing that would be to include a version of Figure 3 for the older model. Or they could be really radical and measure the performances with something like AUC of the ROC.

Looking at those variables I wonder if they are trying to use eGFR (note they used a race free version of that though) as a predictive proxy for race. Thoughts?

My thought is that rather than deleting from the risk algorithm whether or not the patient is black, they should be going in the opposite direction and adding categories for, say, East Asian, South Asian, mestizo Hispanic, and mulatto Hispanic.

Sure, we don’t know exactly why mestizo Hispanics live so long other than it’s, no doubt, some combination of nature and nurture, like most things in the human sciences.

After all, my a priori assumption is that as Animal House taught us: “Knowledge Is Good.” And the more the better.

But ever more very important people these days suspects facts are racist.

iSteve commenter candid-observer adds:

candid_observer says:
November 15, 2023 at 2:13 pm GMT

It’s not just that, obviously, the predictive power of the algorithm goes down if race is omitted. It’s that it is blacks who will suffer from the loss of predictive power: they, and their doctors, will be less likely to be informed of significant risk of heart disease and so less likely to take corrective measures. In short, they will die more often.

The dishonesty of the [NYT] writer here is remarkable. She even brings up the case of an adjustment in how a kidney metric was calculated that originally employed race which, apparently, put blacks at greater risk of not receiving correct treatment. Clearly, she wants the reader to infer that that’s what’s going on in the case of the cardiac algorithm which includes race, when it is certainly the opposite.

Blacks might ask themselves, with allies like this, who needs hostile racists?

[Comment at]

Print Friendly and PDF