The demand for mental health services is increasing as more and more individuals are affected by mental health conditions. In the United States alone, one in five Americans lives with a mental health condition, and suicide rates have risen by over 30% in the past two decades. To address this growing need, organizations and healthcare providers are turning to digital tools and technology platforms such as crisis hotlines, text lines, and online chat lines.
However, these platforms face challenges, including high dropped call rates and lack of integration with clinicians. Organizations like the National Alliance on Mental Illness (NAMI) have reported a 60% increase in help-seekers between 2019 and 2021, leading to overwhelmed responders and long wait times for patients in crisis.
To address these challenges, a team of researchers at Stanford, led by medical students Akshay Swaminathan and Ivan Lopez, has developed a machine learning system called Crisis Message Detector 1 (CMD-1) using natural language processing. The goal of CMD-1 is to distinguish between urgent and non-urgent messages, improving the efficiency of crisis response and reducing patient wait times.
The team used data from Cerebral, a national online mental health company that receives thousands of patient messages per day. They labeled patient messages as crisis or non-crisis using a filter that considers key crisis words and patient IDs that had previously reported a crisis. Ambiguous messages were classified as crises to err on the side of caution.
CMD-1 was designed to complement human review rather than replace it. Crisis messages identified by the system are sent to a human reviewer for further attention. This approach ensures a balance between technological efficiency and compassionate care.
The results of the study were impressive, with CMD-1 detecting high-risk messages with 97% sensitivity and 97% specificity. The system reduced response time for help-seekers from over 10 hours to just 10 minutes, potentially redirecting high-risk patients away from suicide attempts. The team hopes that more machine learning models will be deployed in healthcare settings in the future, with a focus on addressing stakeholder pain points and integrating seamlessly into existing clinical workflows.
The development of CMD-1 highlighted the importance of involving healthcare professionals from the outset of AI model development. By addressing the challenges faced by clinicians and streamlining tasks, AI models can have a meaningful impact on patient care. This cross-functional approach, combined with CMD-1’s impressive results, demonstrates the potential of technology to augment the impact of clinicians and make healthcare delivery more efficient and human-focused.
The researchers believe that the success of CMD-1 is just the beginning of the integration of AI in mental healthcare. By leveraging data and technology, clinicians can deliver higher quality care and help more individuals in need. As the field continues to evolve, it is essential to prioritize collaboration between data scientists and healthcare professionals to ensure the successful deployment and integration of AI models in clinical settings.
*Note:
1. Source: Coherent Market Insights, Public sources, Desk research
2. We have leveraged AI tools to mine information and compile it
