hero image

The prevalence of mental health needs is increasing, with one in five Americans experiencing a mental health condition, and a 30% rise in suicide rates over the last two decades. Organizations like the National Alliance on Mental Illness (NAMI) have observed a 60% surge in individuals seeking crisis support between 2019 and 2021.

Digital tools becoming popular in mental health care

To cope with the rising demand for mental health support, organizations and healthcare providers are increasingly relying on digital tools, including crisis hotlines, text lines, and online chat services. Despite these efforts, the dropout rates for calls to such services can be as high as 25%. Additionally, there is a significant lack of integration between these digital support platforms and the clinicians responsible for the callers’ care.

The substantial drop in response rates in crisis helplines, such as the National Suicide Prevention Lifeline, is primarily attributed to the overwhelming demand surpassing available responders. In 2020, the response rates for chat and text messages were reported at 30% and 56%, respectively, leaving many patients in crisis without adequate support.

It is worth noting that the current queuing method, which operates on a first-come-first-served basis rather than prioritizing urgency, further exacerbates the inefficiency. Therefore, an alternative approach could involve distinguishing between urgent and non-urgent messages, potentially enhancing the triage process and overall system efficiency.

Using machine learning to identify urgent cases

A team of Stanford medical students, led by Akshay Swaminathan and Ivan Lopez, collaborated with clinicians and operational leaders at Cerebral, an online mental health company, to develop a machine learning system named Crisis Message Detector 1 (CMD-1). The interdisciplinary team, utilized natural language processing to create CMD-1, which efficiently identifies and auto-triages concerning messages. The system significantly reduces patient wait times from 10 hours to under 10 minutes.

CMD-1, in particular, proves beneficial in scenarios where speed is crucial, improving crisis response team efficiency and enabling them to address a greater number of cases more effectively. This accelerated triage process allows for more efficient resource allocation and prioritization of urgent cases.