Study Uses Computer to Help Predict Death : Medicine: Program estimates time left for terminally ill patients. Critics caution that it could lead to rationed care.

The cold, calculating logic of computers has entered the emotional realm of death.

Using a new program, researchers say they are able to predict when a terminally ill person will die with more accuracy than doctors using only their own judgment.

The study could help doctors determine which treatments should be given to terminally ill patients and help decide when life-support efforts should be stopped.

“The computer remembers thousands and thousands of cases and keeps the different risk factors in perspective,” said Dr. William A. Knaus of George Washington University. Knaus led the study, published in the Jan. 31 issue of the Annals of Internal Medicine.

“And when we included the survival estimate from the patient’s own physician in the model, the two together predicted time until death more accurately than either alone,” he said.

The program was developed from June, 1989, to June, 1991, using information from 4,301 patients. It was tested from January, 1992, to January, 1994, on 4,028 patients, Knaus said.

The program, called SUPPORT (Study to Understand Prognoses and Preferences for Outcomes and Risks of Treatments), focused on nine diseases and conditions, such as liver disease, colon or lung cancer, heart or lung disease and multiple organ failure. Knaus said he is confident that SUPPORT will prove reliable and eventually be expanded to predict death rates for other diseases.

Seriously ill patients with a projected life expectancy of six months were entered in the study when they were hospitalized.

Their heart rate, respiration, blood chemistry and other measurements were recorded when the patients were admitted and repeated four times in a month. The results were used to calculate the life expectancy of a patient.

Pregnant patients, or those with AIDS, severe brain injury, serious burns and other acute problems were excluded.

How would the information be used?

“Most adults say that if they are going to die within a year, they want realistic estimates of their risks, both in the immediate future and during the next few months,” Knaus said. “This predictive tool is important for its use for counseling very sick patients and their families.”

But not everyone agrees.

Toby Gordon, vice president for planning and marketing at Johns Hopkins Hospital and Health Systems in Baltimore, said the program raises questions.

“Any information that helps us learn how to better take care of patients--in quality of care and quality of life--makes a contribution,” Gordon said. “But whether patients and their families will want to use it is questionable.”

He also questioned the ramifications of being able to accurately predict death.

“In the expansion of computer-assisted technology we will see a proliferation of these techniques, bringing into question ethics and rationing of care,” he said.

The authors warned that the project has not been tested outside the strictly controlled settings of teaching hospitals.