AIsupportshigh-quality,human-firstcare

This study presents the development and internal validation of a natural language processing (NLP) algorithm designed to detect and classify suicidal content in telehealth psychotherapy messages. Through a multi-phase approach, the researchers created a machine learning model that achieved an area under the curve (AUC) of 82.78 for accurately identifying suicide risk at the individual sentence level, demonstrating potential for real-time risk detection in therapeutic contexts. The findings suggest that this model could enhance clinical decision-making and improve understanding of patient-therapist communication related to suicide risk.

Bantilan, N., Malgaroli, M., Ray, B., & Hull, T. D. (2021). Just in time crisis response: suicide alert system for telemedicine psychotherapy settings. Psychotherapy research, 31(3), 289-299.

This study analyzed the patterns of patient use of a text chat-based psychotherapy system and found that the demographic characteristics of users align closely with those seeking traditional face-to-face therapy, although the median age of users is significantly younger. Most participants had prior therapy experiences that were unsatisfactory, indicating that they turned to this alternative due to perceived barriers such as cost and ineffectiveness of traditional therapy. The study highlights the appeal of text-based therapy's accessibility and flexibility, while also raising important questions about the implications of a virtual therapeutic connection, including the challenges of conveying emotional warmth and maintaining a strong therapeutic alliance without face-to-face interaction.

Nitzburg, G. C., & Farber, B. A. (2019). Patterns of utilization and a case illustration of an interactive text‐based psychotherapy delivery system. Journal of clinical psychology, 75(2), 247-259.

This study examined the impact of the COVID-19 pandemic on anxiety and depression symptoms among patients using a digital mental health platform, finding a significant increase in anxiety severity but no notable change in depression symptoms. Utilizing machine learning and natural language processing, the researchers identified a range of additional symptoms related to COVID-19, such as acute stress and insomnia, which traditional measures might overlook. The findings suggest that a broader, dimensional approach to assessing symptoms is necessary for understanding the pandemic's lasting psychological effects and for developing personalized treatment strategies.

Hull, T. D., Levine, J., Bantilan, N., Desai, A. N., & Majumder, M. S. (2021). Analyzing digital evidence from a telemental health platform to assess complex psychological responses to the COVID-19 pandemic: content analysis of text messages. JMIR formative research, 5(2), e26190.

Using machine learning, this study identifies two distinct trajectories of suicidal ideation in older adults undergoing treatment for major depression, with 31% experiencing an unfavorable trajectory and 69% achieving improvement. Key predictors of an unfavorable trajectory included baseline hopelessness, neuroticism, and low general self-efficacy, with hopelessness being the strongest predictor. The findings suggest that addressing these modifiable factors early in treatment may help improve outcomes for suicidal ideation, providing valuable insights for clinical assessment and intervention strategies.

Alexopoulos, G. S., Raue, P. J., Banerjee, S., Mauer, E., Marino, P., Soliman, M., ... & Areán, P. A. (2021). Modifiable predictors of suicidal ideation during psychotherapy for late-life major depression. A machine learning approach. Translational psychiatry, 11(1), 536.

This study explores the treatment of Somatic Symptom and Related Disorders (SSRD) using asynchronous messaging therapy. 173 participants received therapy for eight weeks, with symptoms assessed using the PHQ-9 and GAD-7. Unsupervised random forest clustering identified an Improvement group (41.62%) exhibiting significant symptom reduction and a Non-Response group (58.38%) showing persistent symptoms. The Improvement group expressed more positive emotions initially and showed a decline in negative emotions over time, while the Non-Response group consistently discussed negative feelings. These findings suggest that linguistic markers may be used to predict treatment outcomes in SSRD.

Malgaroli, M., Hull, T. D., Calderon, A., & Simon, N. M. (2024). Linguistic markers of anxiety and depression in Somatic Symptom and Related Disorders: Observational study of a digital intervention. Journal of Affective Disorders, 352, 133-137.