Removing race from heart disease risk calculator does not impact ASCVD prediction
Click Here to Manage Email Alerts
Key takeaways:
- Race-free, sex-specific pooled cohort equations yield similar CVD risk discriminations as race- and sex-specific equations.
- The addition of social determinants of health did not improve discrimination.
Modeling data suggest using race stratification in pooled cohort equation risk calculators does not change a person’s risk for developing atherosclerotic CVD, whereas adding social determinants of health did not improve model performance.
The 2018 American Heart Association/American College of Cardiology Guideline on the Management of Blood Cholesterol recommends using the race- and sex-stratified ASCVD pooled cohort risk equations for 10-year ASCVD risk prediction to guide statin therapy initiation and risk factor management for primary prevention of ASCVD. However, evidence suggests that race-based clinical measures may be systematically biased, Arnab Ghosh, MD, MSc, MA, assistant professor of medicine at Weill Cornell Medicine and hospitalist at NewYork-Presbyterian/Weill Cornell Medical Center, and colleagues wrote in JAMA Cardiology.
“There were three findings from our paper; the first is that including race in the [pooled cohort] equations did not improve accuracy and did not reduce accuracy,” Ghosh told Healio. “The second was that, in place of race, using social determinants of health did not increase accuracy and prediction of the models, either. The third was that, compared with females, the prediction and calibration of CVD in males was worse, regardless of race. That is important, because there is an ongoing discussion in the field of preventive cardiology about where social determinants of health fit.”
Testing the role of race
Ghosh and colleagues analyzed data from 11,638 adults without ASCVD at baseline who participated in the prospective biracial REGARDS cohort study. The mean age of participants was 62 years; 58.1% were women. LDL ranged from 70 mg/dL to 189 mg/dL and non-HDL ranged from 100 mg/dL to 219 mg/dL at baseline (2003-2007). Researchers followed participants up to 10 years for incident ASCVD, including MI, CV death and nonfatal stroke.
Researchers assessed discrimination of the original pooled cohort equation, using C statistics and the net reclassification index, as well as calibration, using plots and the Nam-D’Agostino test statistic comparing observed with predicted events. Researchers then did the same for a set of best-fit, race-stratified equations including the same variables as in the pooled cohort equation (model C), best-fit equations without race stratification (model D), and best-fit equations without race stratification but including social determinants of health as covariates (model E).
“We thought it would be interesting to test whether race independently mattered in the context of these prediction tools that are so extensively used,” Ghosh said in an interview. “When the AHA released its 10-year risk prediction tools, they had specific equations for each race and sex: male, female, Black and white. This meant four separate equations and the numbers within them were all different. Using the REGARDS biracial cohort, which oversampled Black patients, we tested the cohort with the original equations, and then we simplified them so they were the same across all race and sex strata. Then, we compared the accuracy and calibration of each of those equations and then removed race, looking only at sex-specific equations. We took a subset of men vs. women and compared the accuracy within those subgroups.”
Across all strata (Black women, Black men, white women and white men), C statistics for the original pooled cohort equation did not change substantively compared with model C (Black women, 0.71; 95% CI, 0.68-0.75; Black men, 0.68; 95% CI, 0.64-0.73; white women, 0.77; 95% CI, 0.74-0.81; white men, 0.68; 95% CI, 0.64-0.71), model D (Black women, 0.71; 95% CI, 0.67-0.75; Black men, 0.68; 95% CI, 0.63-0.72; white women, 0.76; 95% CI, 0.73-0.8; white men, 0.68; 95% CI, 0.65-0.71) or model E (Black women, 0.72; 95% CI, 0.68-0.76; Black men, 0.68; 95% CI, 0.64-0.72; white women, 0.77; 95% CI, 0.74-0.8; white men, 0.68; 95% CI, 0.65-0.71).
Comparing model D with model E using the net reclassification index showed a net percentage decline in the correct assignment to higher risk for men but not women, according to the researchers. The Nam-D’Agostino test was not significant for all race-sex strata in each model series, indicating good calibration across all groups.
“Having removed race as an independent variable did not change the accuracy or the independent calibration,” Ghosh told Healio. “It is not that race does not matter. It is likely that the effect of one’s race probably plays through the other characteristics that are inherent in the model. What that means is whether one’s BP or cholesterol is controlled, or whether they have diabetes, the effect of race plays through those factors, through such things as whether you have access to health care and access to medications.”
New ways to calculate CV risk
The researchers noted that, if confirmed in other cohorts, the work suggests a need for “a broader and more nuanced discussion of the role of race in the assessment of ASCVD.”
As Healio previously reported, the AHA unveiled new PREVENT equations in November to evaluate 10- and 30-year absolute risk associated with cardiovascular-kidney-metabolic syndrome. The PREVENT equations were developed using real-world contemporary datasets including more than 6 million adults and includes HF risk in addition to risk for MI and stroke; omit race from CVD clinical care algorithms; include kidney function on top of traditional CVD risk factors for heart disease; and include components such as social determinants of health, blood glucose and kidney function, when clinically available.
“Race is a social construct; it is not a biological construct,” Ghosh told Healio. “We do not want to build in the idea that race is a biological construct unassumingly.
“The new PREVENT equations talk about introducing the social determinants of health into models to help physicians in the clinic make decisions about risk factor modification,” Ghosh said. “Our findings suggest we must be very careful about this, because we found [social determinants of health] did not play a role, likely for the same reason race does not play a role.”