Man v. Machine - Predicting Hospital Patient Discharges

A computer can do as good a job of predicting how many patients will be discharged from a hospital unit on a given day as doctors and nurses, according to new research coauthored by Sean Barnes, assistant professor of operations management. In some cases, the computer does even better.

Accurate estimates of patient departures can contribute to keeping health care costs down because they allow hospitals to make the most efficient use of resources. In recent years, the push toward sound management of available beds has led to the introduction of morning “huddles,” during which clinicians predict which patients will be ready to leave that day.

But that kind of logistical work might not be the best use of health care workers’ time. If the new model holds up, and can be improved, “we can minimize the time the clinicians spend thinking about operations and maximize the time they spend thinking about the treatment of patients,” Barnes says.

One notable feature of the model that Barnes and his coauthors devised is how few variables it uses: Fewer than 30. They include such factors as age, race, gender, how long a patient has been in the hospital, whether his or her status is “observational” and several symptoms.

The model’s simplicity would make it easy to reproduce in hospitals across the country.   

To test the model, the researchers looked at one general medical unit in a mid-Atlantic hospital. They examined 8,000 patient stays over a 34-month period stretching from 2011 through 2013.

Patients could be discharged at any time during the day; the model generated predictions both for 2 p.m. and for the end of the day.

Overall, the model tended to be more aggressive than clinicians in predicting which patients would be leaving. (Of course, humans made the final call on discharges.) But on what might be the most important measure, the total number of patients ready to go home by the end of the day, the computer beat the humans.

The research is scheduled to appear in the Journal of the American Medical Informatics Association. /CS/