Natural language processing system for rapid detection and intervention of mental health crisis chat messages.

Patients experiencing mental health crises often seek help through messaging-based platforms, but may face long wait times due to limited message triage capacity. Here we build and deploy a machine-learning-enabled system to improve response times to crisis messages in a large, national telehealth provider network. We train a two-stage natural language processing (NLP) system with key word filtering followed by logistic regression on 721 electronic medical record chat messages, of which 32% are potential crises (suicidal/homicidal ideation, domestic violence, or non-suicidal self-injury). Model performance is evaluated on a retrospective test set (4/1/21-4/1/22, N = 481) and a prospective test set (10/1/22-10/31/22, N = 102,471). In the retrospective test set, the model has an AUC of 0.82 (95% CI: 0.78-0.86), sensitivity of 0.99 (95% CI: 0.96-1.00), and PPV of 0.35 (95% CI: 0.309-0.4). In the prospective test set, the model has an AUC of 0.98 (95% CI: 0.966-0.984), sensitivity of 0.98 (95% CI: 0.96-0.99), and PPV of 0.66 (95% CI: 0.626-0.692). The daily median time from message receipt to crisis specialist triage ranges from 8 to 13 min, compared to 9 h before the deployment of the system. We demonstrate that a NLP-based machine learning model can reliably identify potential crisis chat messages in a telehealth setting. Our system integrates into existing clinical workflows, suggesting that with appropriate training, humans can successfully leverage ML systems to facilitate triage of crisis messages.

Digital bibliotherapy as a scalable intervention for suicidal thoughts: A randomized controlled trial.

OBJECTIVE: Suicide is a major public health concern in the United States, but few effective and scalable interventions exist to help those with suicidal thoughts. We hypothesized that reading first-person narratives about working through suicidal thoughts would reduce the desire to die among adults and that this effect would be mediated by increased perceived shared experience and optimism. METHOD: Using a randomized waitlist-controlled trial, we tested the effect of digital narrative-based bibliotherapy among 528 adults visiting a social media platform dedicated to providing mental health support. Participants were randomized to either a treatment condition (n = 266), in which they read one suicide narrative per day for 14 days or to a waitlist control condition (n = 262). The primary outcome was a measure of desire to die assessed daily for the 14-day trial period and at 2-week follow-up. RESULTS: Participants in the treatment condition reported lower desire to die than participants in the control condition during the 14-day trial period (ß = -0.26, p = .001) and at 2-week follow-up (t = -2.82, p = .005). Increased perceived shared experience (indirect effect b = -0.55, p < .001) and optimism (indirect effect b = -0.85, p < .001) mediated the effect of treatment on desire to die. CONCLUSIONS: Digital narrative-based bibliotherapy may be an effective intervention for those at risk for suicide, and may work in part by increasing feelings of perceived shared experience and optimism. Future research is needed to test the generalizability of these results to other platforms, groups, and conditions. (PsycInfo Database Record (c) 2022 APA, all rights reserved).

Consensus Statement on Ethical & Safety Practices for Conducting Digital Monitoring Studies with People at Risk of Suicide and Related Behaviors.

OBJECTIVE: Digital monitoring technologies (e.g., smart-phones and wearable devices) provide unprecedented opportunities to study potentially harmful behaviors such as suicide, violence, and alcohol/substance use in real-time. The use of these new technologies has the potential to significantly advance the understanding, prediction, and prevention of these behaviors. However, such technologies also introduce myriad ethical and safety concerns, such as deciding when and how to intervene if a participant's responses indicate elevated risk during the study? METHODS: We used a modified Delphi process to develop a consensus among a diverse panel of experts on the ethical and safety practices for conducting digital monitoring studies with those at risk for suicide and related behaviors. Twenty-four experts including scientists, clinicians, ethicists, legal experts, and those with lived experience provided input into an iterative, multi-stage survey, and discussion process. RESULTS: Consensus was reached on multiple aspects of such studies, including: inclusion criteria, informed consent elements, technical and safety procedures, data review practices during the study, responding to various levels of participant risk in real-time, and data and safety monitoring. CONCLUSIONS: This consensus statement provides guidance for researchers, funding agencies, and institutional review boards regarding expert views on current best practices for conducting digital monitoring studies with those at risk for suicide-with relevance to the study of a range of other potentially harmful behaviors (e.g., alcohol/substance use and violence). This statement also highlights areas in which more data are needed before consensus can be reached regarding best ethical and safety practices for digital monitoring studies.

Enhancing the scalability of the collaborative care model for depression using mobile technology.

The collaborative care model (CoCM) has substantial support for improving behavioral health care in primary care. However, large-scale CoCM adoption relies on addressing operational and financial implementation challenges across health care settings with varying resources. An academic medical center serving socioeconomically and racially diverse patients implemented the CoCM in seven practices. A smartphone application was introduced to facilitate CoCM care management during depression treatment (app-augmented CoCM). App features included secure texting, goal/appointment reminders, symptom monitoring, and health education material. A nonrandomized convenience patient sample (N = 807) was enrolled in app-augmented CoCM and compared with patients in standard CoCM (N = 3,975). Data were collected on clinical contact frequency, engagement, and clinical outcomes. App-augmented CoCM patients received more health care team contacts (7.9 vs. 4.9, p < .001) and shorter time to follow up compared with the standard CoCM sample (mean = 11 vs. 19 days, p < .001). App-augmented CoCM patients had clinical outcomes similar to the standard CoCM group (47% vs. 46% with ≥50% depression improvement or score <10), despite app-augmented patients having more prior depression treatment episodes. Further, the app-augmented group with greater app engagement demonstrated increased behavioral health appointment compliance, including more completed appointments and fewer no shows, and greater depression symptom improvement than those with less app engagement. App-augmented CoCM may improve patient engagement in treatment and provide opportunities to implement key CoCM elements without overburdening practice resources. CoCM sustainability and scalability in primary care may be enhanced by using this technology.

Proposed directions for suicide research: incorporating successful approaches from other disciplines.

Despite decades of suicide research, our ability to predict suicide has not changed. Why is this the case? We outline the unique challenges facing suicide research. Borrowing successful strategies from other medical fields, we propose specific research directions that aim to translate scientific findings into meaningful clinical impact.

Negative affect is more strongly associated with suicidal thinking among suicidal patients with borderline personality disorder than those without.

Patients suffering from borderline personality disorder (BPD) are at elevated risk for suicidal thoughts and behaviors (STBs), but this well-described and clinically important association is not well-understood. Prior research suggests that STBs often function as an attempt to escape aversive affect, and that people with BPD experience stronger emotion reactivity and greater discomfort with emotion than those without BPD. Here, we tested whether negative affective states are more likely to predict suicidal thoughts among those with BPD than those without this disorder. Data on affective states and suicidal thoughts were collected several times per day from 35 psychiatric inpatients using their smartphones to capture real-time associations between negative affect and suicidal thoughts. Results revealed that the association between negative affective states (e.g., abandonment, desperation, guilt, hopelessness, loneliness, rage, self-hatred, and upset), and severity of suicidal thinking was stronger among those with BPD than among those without BPD. This finding has implications for risk assessment and intervention in the clinical setting: for a given degree of reported negative affect, patients with BPD experience more suicidal ideation than those without. Further research needs to be done to elucidate the mechanism of this effect.

Guarding the Gate: Remote Structured Assessments to Enhance Enrollment Precision in Depression Trials.

BACKGROUND: Failed treatment trials are common in major depressive disorder and treatment-resistant depression, and remotely performed multifaceted, centralized structured interviews can potentially enhance signal detection by ensuring that enrolled patients meet eligibility criteria. METHODS: We assessed the use of a specific remote structured interview that validated the diagnosis, level of treatment resistance, and depression severity. The objectives were to (1) assess the rate at which patients who were deemed eligible for participation in trials by site investigators were ineligible, (2) assess the reasons for ineligibility, (3) compare rates of ineligibility between academic and nonacademic sites, (4) compare eligibility between US and non-US sites, and (5) report the placebo response rates in trials utilizing this quality assurance approach, comparing its placebo response rates with those reported in the literature. Methods included a pooled analysis of 9 studies that utilized this methodology (SAFER interviews). RESULTS: Overall, 15.33% of patients who had been deemed eligible at research sites were not eligible after the structured interviews. The most common reason was that patients did not meet the study requirements for level of treatment resistance. Pass rates were significantly higher at non-US compared with US sites (94.6% vs 83.3%, respectively; P < 0.001). There was not a significant difference between academic and nonacademic sites (87.8% vs 82.4%; P = 0.08). Placebo response rates were 13.0% to 27.3%, below the 30% to 40% average in antidepressant clinical trials, suggesting a benefit of the quality assurance provided by these interviews. CONCLUSIONS: The use of a remotely structured interview by experienced clinical researchers was feasible and possibly contributed to lower-than-average placebo response rates. The difference between US and non-US sites should be the subject of further research.