http://www.pubmedcentral.nih.gov/articlerender.fcgi?artid=1361135

The above link goes to a 2005 article titled “Testing for statistical discrimination in health care” that appeared in the journal Health Services Research.

The article describes a study by Balsa, McGuire, and Merdith that was motivated by a 2002 report by the Institute of Medicine. The Institute of Medicine report, titled “Unequal Treatment,” had described a discrepancy in between the healthcare provided for ethnic minorities and the healthcare provided for whites. Even after controlling for such “social factors” as socioeconomic status, insurance, and geography, there still is a significant discrepancy in the quality of healthcare provided to patients of different ethnicities. Thus, there appears to be a prevalence of racial discrimination by doctors during the “clinical encounter.” The Institute of Medicine noted three possible explanations for a doctor’s ethnic discrimination: 1.) straight-up racism, 2.) stereotypes about the “health-related behavior of minorities,” or 3.) statistical discrimination in diagnosis based on the prevalence of the illness within the racial group.

Balsa, McGuire, and Merdith (2005) attempt to determine which of the three alternative explanations best explains most cross-racial differences in healthcare. To do this, the authors looked at medical records from doctor visits of 11,000 patients. While the patients were of various ethnic backgrounds, all the doctors were white family physicians. Each record included the patient’s symptoms and the doctor’s diagnosis.

The researchers found that there were systematic differences in doctors’ diagnoses depending on the race of the patient. With regard to certain illnesses, differences in diagnosis were predicted by regression models that involve Bayesian statistics and either of two different forms of statistical discrimination. For racial differences in the diagnosis of hypertension and diabetes, a model that incorporates Bayesian priors about the prevalence of the illness among people of a particular race simulated doctors’ differences in diagnosis. For racial differences in the diagnosis of depression, a model that accounts for miscommunication (by just adding noise) between doctors and patients with different language and cultural backgrounds accounted for the discrepancies among diagnoses.

Differentiating between statistical discrimination and outright racism is important. In the case of racism, doctors are not acting the best interest of her patient. But in the case of statistical discrimination as described above, the doctors are “doing the best they can, given the information available.”