Google’s New AI System Can Predict Risk of Death Among Admitted Patients


A doctor using AI and big data for research. / Photo by: Elnur via Shutterstock


Prediction of death is something people would normally consult with fortune tellers, but with the advantage of AI and deep learning, future algorithms might be able to generate the life expectancy of a person. Right now, scientists at Google are training a computer system to predict whether or not a hospital patient will pass away within 24 hours after admission.

A woman with late-stage breast cancer came to the hospital to get medical attention. Her lungs were filled with fluids. The computers at the hospital analyzed her vital signs and estimated the patient’s chance of dying during her hospital stay was 9.3 percent. That percentage is quite low compared to the patient's actual condition.

However, Google’s new AI system that studied the 175,639 data points of the patient resulted in a death risk of 19.9 percent. Google published the account of the unidentified woman in a study who died within days after hospital admission in May to highlight the potential of neural networks in the healthcare industry.

The system they recently developed has the capacity to forecast the outcome of patients admitted in the hospital, the duration of their hospital stay, the possibility of re-admission, and the estimated risk of dying. The forecast capacity of the system is based on analyzed relevant data points. Compared to current tools that mine data from electronic health records, the tool created by the tech giant can extract information quickly and precisely, leading to conclusions that doctors may use.

“Because we were interested in understanding whether deep learning could scale to produce valid predictions across divergent healthcare domains, we used a single data structure to make predictions for an important clinical outcome (death), a standard measure of quality of care (readmissions), a measure of resource utilization (length of stay), and a measure of understanding of a patient’s problems (diagnoses),” scientists involved in the study noted.

In addition to improving and optimizing the forecasting tool, Google’s research team will test it in clinics to predict symptoms and diseases. While the new AI can be helpful in healthcare, the use of AI, deep learning, and neural networks to read and analyze health records remains a sensitive matter. It requires a careful approach to ensure privacy and safety of patients, a well-programmed system to avoid misinformation, and a large amount of funding to support the innovation.