51ÔÚÏß

With 94.5% accuracy, algorithm could improve response to 198,000 mental health and substance abuse calls in NYC alone, significant long term savings possible for city

NEW YORK, April 12, 2024 /PRNewswire-PRWeb/ -- A local high school student concerned about the growing public health crisis related to mental illness and substance abuse has developed an award-winning algorithm that predicts what resources are needed in response to 911 calls. And, the algorithm is more accurate than human operators. Usage of the AI could be a game-changer for cities whose emergency medical services (EMS) are being inundated with 911 calls for mental health support.

Pierce Wright, a junior and volunteer Emergency Medical Responder (EMR), spent the past year teaching himself Python to code the AI, training it with the NYC emergency response database of over 24 million calls. Wright's results: the algorithm achieved a 94.5% accuracy rate. In comparison, human operators accurately predict the resources needed for a given 911 call only 92.3% of the time, making the algorithm 2.2% more effective than human operators.

On Sunday, April 7, the project was awarded first place for his Medicine and Health Sciences project in the TerraNYC STEM fair held at NYU's Tandon School of Engineering.

Wright, who has volunteered more than 500 hours at an EMS station, says he was inspired to create the algorithm based on his observations while fielding calls on shift, and living in New York City (NYC).

"I noticed a majority of the calls I responded to were related to mental health or substance abuse concerns, not always actual medical emergencies," Wright says. "In these cases, the patient typically ends up being taken by ambulance to a local emergency department where they may wait hours for appropriate treatment, or simply to be transferred to a separate mental health facility. These calls also can divert us away from other urgent medical crises."

Wright, who lives and attends school in Manhattan, says headlines about the city's exploding mental health crisis weighed heavily on his mind as he thought about what he could do to help. The increase in the number of mental health 911 calls is particularly acute in NYC where there are close to 200,000 mental health crisis 911 calls per year—a number that has tripled since 1999.

"When responding to mental health calls, I felt like we were just passing the buck by transporting the patient to a hospital, where they weren't really getting the help they needed," Wright continues. "We aren't trained to provide mental health support and I wondered if there was a better use of resources."

With nine million 911 calls in NYC annually, the algorithm has the ability to more effectively allocate resources for 198,000 calls each year in NYC alone, including identifying cases where a mental health counselor would be a more appropriate response than an ambulance or police. EMS teams have few other options than transporting a patient to the closest hospital, using up ER capacity, while a mental illness call may have likely been better served by a counselor. If Wright's algorithm is successful in preventing unnecessary hospital admissions, an estimated savings of $123 million nationally may be possible. In 2017, $5.6 billion was spent on mental health and substance use disorder visits to emergency departments nationwide.

After researching how artificial intelligence has been used in EMS settings and how cities are addressing the country's burgeoning mental health crisis, Wright felt he could make a contribution to the space by creating and training an AI model to identify the appropriate response to incoming 911 calls.

Twenty-four million 911 calls from NYC's EMS Incident Dispatch Data from a seventeen-year time span of January 2005 to March 2022 were used to establish the algorithm. Of the data's thirty-one variables, Wright selected seven, including an incoming call's time of day, zip code, police precinct, and initial severity level upon which to train the model, since those variables had the greatest impact on the final call type.

"There was a lot of trial-and-error that went into developing this model," Wright says. He recounts that one of the biggest challenges was in handling such a large amount of data.

"My machine kept crashing," Wright recalls.

Undeterred, Wright experimented with different approaches and methodologies that ultimately proved better able to handle the enormous 5.6GB data set, even when that meant re-coding significant portions of the algorithm.

At the recent New York State Science and Engineering Fair held at the New York Hall of Science in Queens, N.Y., Wright's work was additionally honored with a second-place trophy.

"It is rewarding that others saw value in my research, but my real hope is that municipalities will use the algorithm to build on pilot programs that pair teams of EMS with mental health professionals, such as NYC's B-HEARD," Wright says. "Initial data from B-HEARD shows that with the right resources, about fifty percent of patients with mental health concerns can be treated at home without needing to go to the hospital."

He adds that the model is customizable and scalable. For instance, additional data such as new call types specific to different areas would make the model more robust and customizable to local needs.

Wright, 17, is a student at The Browning School in Manhattan; he continues to volunteer as an EMR with the Westport (CT) Volunteer EMS on weekends. A paper about his project titled "Utilizing AI to Optimize EMS Response to Acute Mental Illness and Resulting ER Resource Allocation" is publicly available at

Media Contact

M. Wright, Wright Family, 1 (212) 860-3204, [email protected]

SOURCE Wright Family