Scientists at MIT’s McGovern Institute for Brain Research have developed a specialized language-processing tool that evaluates text conversations to predict suicide risk during mental health crises. Published in the Journal of Psychopathology and Clinical Science, the study demonstrates that analyzing specific words and phrases used by individuals in distress can accurately flag imminent danger and clarify which risk factors matter most.
Mapping the Anatomy of a Crisis
Predicting suicide attempts remains a major challenge for clinicians because dozens of interacting factors contribute to suicidal ideation and behavior. While psychiatric conditions like depression, post-traumatic stress disorder, and borderline personality disorder elevate risk, environmental stressors such as poverty, loneliness, and discrimination also play complex roles.
To determine which symptoms signal the highest immediate danger, researchers collaborated with Crisis Text Line, a global nonprofit providing free, confidential, 24/7 support via text. The research team analyzed de-identified data from approximately 16,000 text conversations handled by trained volunteer crisis counselors.
Capturing Distress in Real Time
The conversations were categorized into three risk levels: non-suicidal, suicidal ideation without imminent risk, and imminent risk. Researchers focused heavily on the imminent risk group, which included individuals who expressed a specific plan or an intent to die within 48 hours.
Daniel Low, a former graduate student in Senior Research Scientist Satra Ghosh’s Senseable Intelligence Group, noted that typical studies rely on patients recalling symptoms long after a crisis passes, whereas Crisis Text Line data captures symptoms live as distress unfolds. Low is now a research scientist at the Child Mind Institute and a visiting scholar at Harvard University.
Constructing the Custom Lexicon
To process the text data, the research team constructed a custom suicide-risk lexicon. They used artificial intelligence to generate a preliminary list of words and phrases tied to 49 established suicide risk factors, including those linked to ideation, attempts, and death.
Expert clinicians then manually reviewed and refined the list, resulting in approximately 60 terms for each of the 49 risk factors. The team trained a machine learning model to scan crisis conversations for these lexicon terms to estimate an individual’s risk level. Because the lexicon ties specific terms directly to distinct risk factors, the tool also revealed which factors correlated most strongly with imminent danger.
What the Algorithm Uncovered
The analysis revealed patterns that align with previous research while offering new clarity on immediate crisis indicators. While depression is widely recognized as a major risk factor for suicidal ideation, the machine learning model found that mentions of lethal means and substance use appeared more frequently in the highest-risk group than expressions of depressed mood or fatigue.
Active suicidal ideation and self-injury also emerged as strong predictors of imminent risk. Meanwhile, anxiety, post-traumatic stress disorder, and general emotional pain registered as intermediate predictors. Researchers suggest that with further validation, this tool could assist clinicians and support staff in assessing risk more accurately during critical interventions.