Researchers have known for years that African-Americans in the midst of a heart attack are far less likely than white patients to receive potentially life-saving treatments such as clot-busting drugs, a dramatic illustration of America’s persistent healthcare disparities. But the reasons behind such stark gaps in care for heart disease, as well as cancer and other serious illnesses, have remained murky, with blame fixed on doctors, hospitals, and insurance plans.

In the new study, trainee doctors in Boston and Atlanta took a 20-minute computer survey designed to detect overt and implicit prejudice. They were also presented with the hypothetical case of a 50-year-old man stricken with sharp chest pain; in some scenarios the man was white, while in others he was black.

“We found as doctors’ unconscious biases against Blacks increased, their likelihood of giving (clot-busting)) treatment decreased,” said lead author of the study Alexander R Green of Massachusetts General Hospital. “It’s not a matter of you being a racist. It’s really a matter of the way your brain processes information is influenced by things you’ve seen, things you’ve experienced, the way media has presented things.”

The study has been published on the Website of the Journal of General Internal Medicine.

“Years of advanced education and egalitarian intentions are no protection against the effect of implicit attitudes,” said Thomas Inui, president of the Regenstrief Institute Inc in Indianapolis, which studies vulnerable patient groups. “When do they surface? When we’re involved with high-pressure, high-stakes decision-making, when there’s a lot riding on our decisions but there isn’t a lot of time to make them, that’s when the implicit attitudes that are not scientific rise up and grab us.”

However, Green said he cannot explain why implicit bias would cause doctors to deprive patients of potentially life-saving therapy, and other researchers said they do not know how big a factor unconscious prejudice is in the far-reaching problem of disparities.