Predicting Falls After Discharge from the Emergency Department

For America’s aging population, preventing falls is crucial for maintaining independence in their golden years. Brian Suffoletto, MD, and his team are using AI and digital technology to predict and prevent falls in older patients after leaving the emergency department (ED).

By 2030, 20% of the US population will be 65 or older. Around 20 million patients over 65 visit EDs each year, and 20% of Stanford Hospital’s ED patients fall into this vulnerable group. Suffoletto, associate professor of emergency medicine, uses ED visits to assess the future fall risk of older patients. He uses body-worn sensors and AI to identify those at risk of falls and provide digital interventions for them to maintain independence at home.

Suffoletto’s team used ED accelerometer data and a neural network model to predict post-discharge fall risk in a recent study. The neural network model can detect subtle nuances in movement that humans might miss, improving prediction accuracy. It doesn’t merely add elements; it may introduce complex, hidden features. Early results confirmed the effectiveness of this approach in assessing and predicting post-ED fall risk.

Going beyond predictive diagnostics, Suffoletto and his team are also designing digital interventions to reduce fall risk in the crucial months following discharge.

In a recent study involving 150 older adults, Suffoletto gathered data on their step counts after leaving the ED. During their rehabilitation period, these patients tended to lead sedentary lives, translating into a higher risk of falls due to factors like sarcopenia muscle loss, decreased range of motion, and more.

Suffoletto is enrolling elderly patients assessed as fall risks. These patients will receive accelerometer-based pedometers and participate in a text message program designed to encourage daily step count reporting.

Suffoletto has spent more than 10 years developing digital behavioral interventions for various medical risks from young adult binge drinking to distracted driving. Recent studies include machine modeling to predict intoxication based on measurable changes in voice, and the use of smartphone accelerometer sensors to detect alcohol intoxication by analyzing an individual’s gait.


Updated Spring 2024