Problem: Sometimes the difference between life and death is a quick and accurate diagnosis. With sepsis, a life-threatening reaction to an infection, there’s no definitive single test doctors can use to diagnose the condition.
Solution: Suchi Saria, an assistant professor at Johns Hopkins University, wondered: what if existing medical information could be used to predict which patients would be most at risk for sepsis? Algorithms that she subsequently created to analyze patient data correctly predicted septic shock in 85 percent of cases, by an average of more than a day before onset. That is a 60 percent improvement over existing screening tests.