Washington (AFP)

In thousands of US hospitals, algorithms are used to identify the most at-risk patients and provide them with extensive follow-up. But these algorithms, researchers reported Thursday in the journal Science, favor whites to blacks.

Ziad Obermeyer, from the University of California at Berkeley, says he almost stumbled across this discovery, analyzing with colleagues the data provided by a large unidentified university hospital.

The hospital calculated, using a standard algorithm in public health, a "risk score" and selected the 3% of the sickest or most vulnerable patients (diabetes, heart failure, emphysema ...), who could then call a number dedicated, get appointments the same day or be followed at home.

"On two patients, a Black and a White, with exactly the same score, the black patient was actually more likely to have more health problems in the next year than the white patient," says Ziad Obermeyer at AFP.

"White patients are doubling sicker black patients to get into the program," he says.

The algorithm did not take into account the "race", as the Americans say, the color of the skin.

The real problem with the program was that it underestimated the health status of black patients because it was based solely on the costs of care generated in the past by patients.

But "black patients, on average, generate less costs than white patients who have the same level of health," says Ziad Obermeyer. It is the result of deep-rooted inequalities that blacks go to the doctor less often, and when they go there, doctors order less testing or care.

"We have the illusion that we work with biological variables that describe physiology objectively," he says. In fact, all the data in the algorithm is based on financial transactions between the hospital and the insurance companies.

This is the original defect of these algorithms.

The company that markets the software has accepted a suggestion by researchers to reduce the racial imbalance by more than 80%.

But as Ziad Obermeyer points out, changing the code of algorithms is only the first step: databases should be built that are primarily concerned with patients' actual health status.

"It is really strange that the major source of data on medicine comes from financial transactions," concludes the researcher. "The health system does not take seriously the need to acquire quality health data, the information we have about people's health is almost accidental."

© 2019 AFP