Lately, the world is rife with legitimate concerns regarding the rise of and over reliance on AI in healthcare. Everything from biases and limitations to privacy concerns to inaccurate diagnoses loom large. However, AI is hardly the only source of errors. The humans who have been administering healthcare for centuries can hardly claim immaculate judgment. Instead of trying to design the perfect healthcare AI solution or build the perfect medical school, perhaps the optimal solution can be found in combination.
We’d like to draw your attention to a program at the University of Colorado, where there has been some groundbreaking research on the best ways to treat Sepsis, where AI and humans working together actually produced better patient outcomes than just AI alone. (Read original article here)
Sepsis is a dangerous blood infection. According to the CDC, in a typical year, 1 in 3 people who die in a hospital had sepsis during that hospitalization. And at least 1.7 million adults in America develop sepsis every year, with a high mortality rate. In a disease like sepsis, early and accurate detection is absolutely crucial. Every hour that treatment is delayed equates to a 10% drop in survival rates. Initial symptoms are quite generic (shortness of breath, increased heart rate, etc) and commonly associated with other ailments, thus presenting a unique set of challenges when it comes to early diagnosis. Human staff are likely to misinterpret early sepsis symptoms as something else, while automated AI predictive alerts would flood nurses with false alarms. Either solution on its own has undesirable ramifications.
After extensive research by UCHealth (details here), the best solution for patients ended up being a combination of human intellect and AI predictions. AI is used to generate medical alerts for staff, but these alerts are not reviewed by an overworked bedside nurse. Instead, they are evaluated by a virtual care nurse who has access to the patient’s chart and can monitor them remotely. This human nurse in front of a computer views the alert and evaluates the patient’s situation across many factors. If appropriate, they can contact the bedside nurse to start treatment immediately. This ensures that alerts are properly triaged while bedside staff are not overwhelmed.
This human/AI hybrid proved to be much more effective than previous solutions. UCHealth estimates its sepsis alert system is saving a few hundred lives a year within the UCHealth network alone. That’s quite substantial.
While predictive AI generated alerts were an important piece of the puzzle, their effectiveness didn’t shine until human judgment and innovative hospital procedure was added into the mix. The combination of AI and human involvement produced the best outcome.
