Background: Concern for underrepresentation of minorities in clinical trials has focused on enrollment proportions and generalizability. The interplay of trial event-rates with diversity has not been emphasized.
Methods: The Carotid Revascularization Endarterectomy vs Stent Trial (CREST) randomized 2502 patients and compared them by race, ethnicity, baseline characteristics, and primary outcome (any peri-procedural stroke, death or MI and subsequent ipsilateral stroke up to 10 years); those with unknown race or ethnicity were excluded. Proportional hazards models adjusting for age, sex, symptomatic status and treatment were used to test for a treatment by race/ethnicity interaction.
Results: One-hundred-nine patients (4.4%) were black, 32 (1.3%) Asian, 2332 (93.4%) white, 11 (0.4%) other by self-report, and 18 (0.7%) unknown; 90 (3.6%) were Hispanic, 2377 (95%) non-Hispanic, and 35 (1.4%) unknown. Compared to whites, racial minorities were younger (mean age 67±8.9 vs 69±8.8, p=0.004), more often female (44% vs 34%, p=0.01), symptomatic (63 vs 52%, p=0.01), and diabetic (51% vs 29%, p<0.0001), but less often dyslipidemic (76% vs 85%, p=0.004), current smokers (19% vs 27%, p=0.04), or had a history of cardiovascular disease (34% vs 46%, p=0.007). Hispanics were more often diabetic (48% vs 30%, p=0.0002). The rate of the primary endpoint was 10.9%±0.9% at 10 years, and did not differ by race or ethnicity (pinter>0.24). In the context of this low rate, even if minority recruitment were increased to represent 50% of study participants, and if the treatment difference in one race were a greater hazard of HR = 1.49 (anticipated alternative hypothesis), then the hazard ratio in the other group would need to be <0.73, or >3.31, to have 90% power of detection.
Conclusions: The proportion of racial and ethnicity participation in CREST was suboptimal at < 10%. Primary outcomes did not differ by minority or ethnic status. However, in low event-rate trials very high and even unrealistic enrollment goals for diversity, for example ≥50%, may still be insufficient for detection of outcome differences by race or ethnicity.