How would you feel if your fate was in the "hands" of a computer? The idea may not be as far-fetched as you think: A new study by the National Institutes of Health (NIH) says that computers, using a mathematical formula, could determine the wishes of incapacitated patients as accurately as?if not better than?their family members or close friends.

"We thought with the surrogate decisions, is that the best we got? And are we stuck with it?" says NIH bioethicist David Wendler, co-author of the study published in this week's PLoS Medicine. "Or could we do better? So we started working on what alternatives could be better."

In the study, Wendler and his colleagues proposed this scenario: a 70-year-old Native American male, with a PhD and severe Alzheimer's, develops a life-threatening infection. If the patient's treatment or non-treatment preference is unknown, doctors would typically turn to his designated next of kin surrogate for guidance. But under Wendler's plan, a proposed computer software program, dubbed the "population-based treatment indicator," would make the choice by locating the profiles of similar patients?in terms of age, race, education, illness and other factors?and matching the most popular treatment choice for that group with one of the current patient's options.

The researchers analyzed nearly 20,000 surrogate and patient decisions, and discovered that surrogates are imperfect predictors of what a patient might want, getting it right only about two thirds of the time. Wendler says the proposed treatment indicator could add a layer of accuracy to the process.

But the idea of a computer program making life-or-death decisions is controversial, to say the least. It not only raises ethical questions, such as whether or not a mathematical formula can be a substitute for emotional human connections, but critics say that it would be a Herculean?if not impossible?task to compile the database required from which to draw information to make such monumental decisions.

"Just because a computer algorithm is a better predictor, it doesn't then follow that we should use it," says Dan Brock, director of medical ethics at Harvard Medical School. There are many other reasons for human surrogate decision makers, he adds, such as the fact that many patients want their family to have a say in what is perhaps the most important decision of their lives, regardless of accuracy. In other words, the patient trusts the family?or a close friend?to make the right decision.

Wendler and his colleagues built a preliminary "population-based treatment indicator" around one simple variable: Pooling stats from surveys done with cancer patients, they discovered that most Americans would choose a life-saving treatment if there was at least a 1 percent chance it would bring them to an acceptably healthy state, considered to be the ability to reason, remember and communicate. Conversely, they learned that people would nix life-saving treatments if there was more than a 99 percent chance that they might survive but be left unable to reach a healthy state as defined by reasoning, remembering and communicating.

NIH researchers used this preliminary treatment indicator to predict patients' wishes and compared its accuracy with that of surrogate decision makers. Their findings: they were both accurate just over 78 percent of the time.

But that's based on a single factor, and Wendler says that a database that could tailor predictions for an unlimited number of individuals would have to take into account a whole host of other factors like religion, gender, age, ancestry and others that come into play in life-or-death decisions. "A 95-year-old severe diabetic might say 'no' to a 2 percent chance of recovery after invasive treatment, while a 20-year-old might say 'yes' to a 1 percent chance of recovery," he notes.

He believes, however, that if researchers could achieve 78 percent accuracy based on a single variable, it follows that accuracy would increase substantially if more personal details?gathered through thousands of public surveys?were included in the database.

Comments