Can Text Messages Identify Suicide Risk in Real Time? A Within-Subjects Pilot Examination of Temporally Sensitive Markers of Suicide Risk

A Scoping Review of Digital-Based Intervention for Reducing Risk of Suicide Among Adults

Suicide is a serious public health problem, especially among adults. Risk factors for suicide include the presence of mental health disorders, history of previous suicide attempts; substance or alcohol use and lack of social support. The impact of suicide risk includes psychological loss, as well as the trauma and emotional stress that can be felt by the families and communities left behind. Digital interventions have emerged as a promising alternative for suicide risk prevention. Previous research has focused on the findings of various designs, which did not provide clear intervention information to inform the implementation of the intervention. This study aims to describe a digital intervention to reduce the risk of suicidal behavior in adults. The design used in this study was a scoping review. The authors conducted a literature search from the Scopus, PubMed, and CINAHL databases. Inclusion criteria in this study were articles discussing digital interventions aimed at preventing suicide risk in adult populations, English language, full-text, RCT or quasi-experiment design, and publication period of the last 10 years (2014-2024). The major keywords used in the article search were suicide prevention, digital intervention, and adults. Data extraction used manual table and data analysis used descriptive qualitative with a content approach. The results showed that there were 9 articles that discussed digital-based interventions to reduce suicide risk in adults. The various types of digital interventions used were smartphone apps, online learning modules, and game-based interventions. These interventions offer significant potential in reducing the risk of suicidal behavior in adults. Digital interventions have an important role in reducing the risk of suicidal behavior in adults by considering aspects of suitability to individual needs and understanding digital literacy. Then, the development of mental health services and public health policies presented needs to be done with collaboration between stakeholders in suicide prevention efforts.

Perceived Burdensomeness and Thwarted Belongingness in Text Messages of Suicide Attempt Survivors

OBJECTIVE

Perceived burdensomeness and thwarted belongingness are considered interpersonal risk factors for suicide. Examining these themes in personal text messages may help identify proximal suicide risk.

METHOD

Twenty-six suicide attempt survivors provided personal text messages and reported dates for past periods characterized by positive mood, depressed mood, suicidal ideation (with no attempt), or the two-week period leading up to suicide attempt(s). Texts were then classified into the applicable period based on matching dates. Texts (N = 194,083; including n = 86,705 outgoing texts) were coded for perceived burdensomeness and thwarted belongingness by masked trained raters. Multilevel models were fit to examine whether the target themes (combined into one overall interpersonal risk variable due to low base rate) were more prevalent in texts sent during higher risk episodes (e.g., suicide attempt vs. depressed mood episodes).

RESULTS

0.57% of outgoing texts contained either target theme. As hypothesized, logistic models showed participants were more likely to send texts containing the target themes during suicide attempt episodes relative to suicidal ideation (with no attempt) episodes, depressed mood episodes, and positive mood episodes, and during suicidal ideation (with no attempt) episodes relative to positive mood episodes. All contrasts were robust to post-hoc correction except for suicide attempt episodes vs. ideation (with no attempt) episodes. No other significant pairwise differences for episode type emerged.

CONCLUSIONS

Despite the small sample size and low base rate of target themes in the texts, perceived burdensomeness and thwarted belongingness were associated with intra-individual suicide risk severity in personal text messages.

Prevention of Re-attempt Suicide Through Brief Contact Interventions: A Systematic Review, Meta-analysis, and Meta-regression of Randomized Controlled Trials

Brief contact intervention (BCI) is a low-cost intervention to prevent re-attempt suicide. This meta-analysis and meta-regression study aimed to evaluate the effect of BCI on re-attempt prevention following suicide attempts (SAs). We systematically searched using defined keywords in MEDLINE, Embase, and Scopus up to April, 2023. All randomized controlled trials (RCTs) were eligible for inclusion after quality assessment. Random-effects model and subgroup analysis were used to estimate pooled risk difference (RD) and risk ratio (RR) between BCI and re-attempt prevention with 95% confidence intervals (CIs). Meta-regression analysis was carried out to explore the potential sources of heterogeneity. The pooled estimates were (RD = 4%; 95% CI 2-6%); and (RR = 0.62; 95% CI 0.48-0.77). Subgroup analysis demonstrated that more than 12 months intervention (RR = 0.46; 95% CI 0.10-0.82) versus 12 months or less (RR = 0.67; 95% CI 0.54-0.80) increased the effectiveness of BCI on re-attempt suicide reduction. Meta-regression analysis explored that BCI time (more than 12 months), BCI type, age, and female sex were the potential sources of the heterogeneity. The meta-analysis indicated that BCI could be a valuable strategy to prevent suicide re-attempts. BCI could be utilized within suicide prevention strategies as a surveillance component of mental health since BCI requires low-cost and low-educated healthcare providers.

Adolescent Social Communication Through Smartphones: Linguistic Features of Internalizing Symptoms and Daily Mood

The increasing use of smartphone technology by adolescents has led to unprecedented opportunities to identify early indicators of shifting mental health. This intensive longitudinal study examined the extent to which differences in mental health and daily mood are associated with digital social communication in adolescence. In a sample of 30 adolescents (ages 11-15 years), we analyzed 22,152 messages from social media, email, and texting across one month. Lower daily mood was associated with linguistic features reflecting self-focus and reduced temporal distance. Adolescents with lower daily mood tended to send fewer positive emotion words on a daily basis, and more total words on low mood days. Adolescents with lower daily mood and higher depression symptoms tended to use more future focus words. Dynamic linguistic features of digital social communication that relate to changes in mental states may represent a novel target for passive detection of risk and early intervention in adolescence.

The effect of universal school-based mindfulness on anhedonia and emotional distress and its underlying mechanisms: A cluster randomised controlled trial via experience sampling in secondary schools

This cluster randomised controlled trial examined the effectiveness of universal school-based mindfulness training (MT; vs. passive control) to lower anhedonia and emotional distress among mid-adolescents (15-18 years). It further examined three potential mechanisms: dampening of positive emotions, non-acceptance/suppression of negative emotions, and perceived social pressure not to experience/express negative emotions. Adolescents (ncontrol = 136, nintervention = 95) participated in three assessment points (before, after and two/three months after the in-class MT), consisting of Experience Sampling (ES) assessments and self-report questionnaires (SRQs) to corroborate the ES assessments. Analyses were based on general linear modelling and multilevel modelling. Overall, no evidence was found for a significant beneficial and long-lasting impact of the MT on adolescents' mental health. Importantly, some barriers inherently linked to universal MT approaches (low engagement in and mixed attitudes towards the MT) may have tempered the effectiveness of the MT in the current trial. Further research should prioritise overcoming these barriers to optimise programme implementation. Additionally, given the potential complex interplay of moderators at micro- (home practice), meso- (school climate), and macro-level (broader context), research should simultaneously focus on alternative ways of delivering MT at schools to strengthen adolescents' mental health.

Linguistic correlates of suicidal ideation in youth at clinical high-risk for psychosis

Suicidal ideation (SI) is prevalent among individuals at clinical high-risk for psychosis (CHR). Natural language processing (NLP) provides an efficient method to identify linguistic markers of suicidality. Prior work has demonstrated that an increased use of "I", as well as words with semantic similarity to "anger", "sadness", "stress" and "lonely", are correlated with SI in other cohorts. The current project analyzes data collected in an SI supplement to an NIH R01 study of thought disorder and social cognition in CHR. This study is the first to use NLP analyses of spoken language to identify linguistic correlates of recent suicidal ideation among CHR individuals. The sample included 43 CHR individuals, 10 with recent suicidal ideation and 33 without, as measured by the Columbia-Suicide Severity Rating Scale, as well as 14 healthy volunteers without SI. NLP methods include part-of-speech (POS) tagging, a GoEmotions-trained BERT Model, and Zero-Shot Learning. As hypothesized, individuals at CHR for psychosis who endorsed recent SI utilized more words with semantic similarity to "anger" compared to those who did not. Words with semantic similarity to "stress", "loneliness", and "sadness" were not significantly different between the two CHR groups. Contrary to our hypotheses, CHR individuals with recent SI did not use the word "I" more than those without recent SI. As anger is not characteristic of CHR, findings have implications for the consideration of subthreshold anger-related sentiment in suicidal risk assessment. As NLP is scalable, findings suggest that language markers may improve suicide screening and prediction in this population.

Inferring sleep disturbance from text messages of suicide attempt survivors: A pilot study

OBJECTIVE

Identifying digital markers of sleep disturbance-a known suicide risk factor-may aid in the detection of imminent suicide risk. This study examined sleep-related communication and texting patterns in personal text messages (N = 86,705) of suicide attempt survivors.

METHOD

Twenty-six participants provided dates of past suicide attempts and 2-week periods of positive mood, depressed mood, or suicidal ideation. Linguistic Inquiry Word Count was used to identify sleep-related texts via a custom dictionary. Mixed effect models were fitted to test the association between suicide/mood episode type (e.g., attempt versus ideation) and three outcomes: likelihood of a text including sleep-related content, nightly count of texts sent from midnight to 5:00 AM, and sum of unique hour bins from midnight to 5:00 AM with outgoing texts.

RESULTS

Analyses with a sleep dictionary that was manually revised to be more accurate (but not the original unedited dictionary) showed sleep-related communication was more likely during depressed mood episodes than positive mood episodes. Otherwise, there were no significant differences in sleep-related communication or objective texting patterns across episode type.

CONCLUSIONS

Although we did not detect differences in sleep-related communication tied to suicidal thoughts or behaviors, sleep-related communication may differ as a function of within-person mood level.

Digital phenotyping in psychiatry: A scoping review

BACKGROUND

Digital phenotyping has been defined as the moment-by-moment assessment of an illness state through digital means, promising objective, quantifiable data on psychiatric patients' conditions, and could potentially improve diagnosis and management of mental illness. As it is a rapidly growing field, it is to be expected that new literature is being published frequently.

OBJECTIVE

We conducted this scoping review to assess the current state of literature on digital phenotyping and offer some discussion on the current trends and future direction of this area of research.

METHODS

We searched four databases, PubMed, Ovid MEDLINE, PsycINFO and Web of Science, from inception to August 25th, 2021. We included studies written in English that 1) investigated or applied their findings to diagnose psychiatric disorders and 2) utilized passive sensing for management or diagnosis. Protocols were excluded. A narrative synthesis approach was used, due to the heterogeneity and variability in outcomes and outcome types reported.

RESULTS

Of 10506 unique records identified, we included a total of 107 articles. The number of published studies has increased over tenfold from 2 in 2014 to 28 in 2020, illustrating the field's rapid growth. However, a significant proportion of these (49% of all studies and 87% of primary studies) were proof of concept, pilot or correlational studies examining digital phenotyping's potential. Most (62%) of the primary studies published evaluated individuals with depression (21%), BD (18%) and SZ (23%) (Appendix 1).

CONCLUSION

There is promise shown in certain domains of data and their clinical relevance, which have yet to be fully elucidated. A consensus has yet to be reached on the best methods of data collection and processing, and more multidisciplinary collaboration between physicians and other fields is needed to unlock the full potential of digital phenotyping and allow for statistically powerful clinical trials to prove clinical utility.

The relationship between text message sentiment and self-reported depression

BACKGROUND

Personal sensing has shown promise for detecting behavioral correlates of depression, but there is little work examining personal sensing of cognitive and affective states. Digital language, particularly through personal text messages, is one source that can measure these markers.

METHODS

We correlated privacy-preserving sentiment analysis of text messages with self-reported depression symptom severity. We enrolled 219 U.S. adults in a 16 week longitudinal observational study. Participants installed a personal sensing app on their phones, which administered self-report PHQ-8 assessments of their depression severity, collected phone sensor data, and computed anonymized language sentiment scores from their text messages. We also trained machine learning models for predicting end-of-study self-reported depression status using on blocks of phone sensor and text features.

RESULTS

In correlation analyses, we find that degrees of depression, emotional, and personal pronoun language categories correlate most strongly with self-reported depression, validating prior literature. Our classification models which predict binary depression status achieve a leave-one-out AUC of 0.72 when only considering text features and 0.76 when combining text with other networked smartphone sensors.

LIMITATIONS

Participants were recruited from a panel that over-represented women, caucasians, and individuals with self-reported depression at baseline. As language use differs across demographic factors, generalizability beyond this population may be limited. The study period also coincided with the initial COVID-19 outbreak in the United States, which may have affected smartphone sensor data quality.

CONCLUSIONS

Effective depression prediction through text message sentiment, especially when combined with other personal sensors, could enable comprehensive mental health monitoring and intervention.

Digital Technologies for Emotion-Regulation Assessment and Intervention: A Conceptual Review

The ability to regulate emotions in response to stress is central to healthy development. While early research in emotion regulation predominantly employed static, self-report measurement, the past decade has seen a shift in focus toward understanding the dynamic nature of regulation processes. This is reflected in recent refinements in the definition of emotion regulation, which emphasize the importance of the ability to flexibly adapt regulation efforts across contexts. The latest proliferation of digital technologies employed in mental health research offers the opportunity to capture the state- and context-sensitive nature of emotion regulation. In this conceptual review, we examine the use of digital technologies (ecological momentary assessment; wearable and smartphone technology, physical activity, acoustic data, visual data, and geo-location; smart home technology; virtual reality; social media) in the assessment of emotion regulation and describe their application to interventions. We also discuss challenges and ethical considerations, and outline areas for future research.

Robust suicide risk assessment on social media via deep adversarial learning

OBJECTIVE

The prevalence of social media for sharing personal thoughts makes it a viable platform for the assessment of suicide risk. However, deep learning models are not able to capture the diverse nature of linguistic choices and temporal patterns that can be exhibited by a suicidal user on social media and end up overfitting on specific cues that are not generally applicable. We propose Adversarial Suicide assessment Hierarchical Attention (ASHA), a hierarchical attention model that employs adversarial learning for improving the generalization ability of the model.

MATERIAL AND METHODS

We assess the suicide risk of a social media user across 5 levels of increasing severity of risk. ASHA leverages a transformer-based architecture to learn the semantic nature of social media posts and a temporal attention-based long short-term memory architecture for the sequential modeling of a user's historical posts. We dynamically generate adversarial examples by adding perturbations to actual examples that can simulate the stochasticity in historical posts, thereby making the model robust.

RESULTS

Through extensive experiments, we establish the face-value of ASHA and show that it significantly outperforms existing baselines, with the F1 score of 64%. This is a 2% and a 4% increase over the ContextBERT and ContextCNN baselines, respectively. Finally, we discuss the practical applicability and ethical aspects of our work pertaining to ASHA, as a human-in-the-loop framework.

DISCUSSION AND CONCLUSIONS

Adversarial samples can be helpful in capturing the diverse nature of suicidal ideation. Through ASHA, we hope to form a component in a larger human-in-the-loop infrastructure for suicide risk assessment on social media.

Social media use and self-injurious thoughts and behaviors: A systematic review and meta-analysis

Despite considerable public and scholarly debate about the role of social media in self-injurious thoughts and behaviors (SITBs), no comprehensive, quantitative synthesis of this literature has previously been undertaken. The current systematic review and meta-analysis examines associations between social media use and SITBs, including suicidal ideation, suicide plans, suicide attempts, and nonsuicidal self-injury (NSSI). A range of social media behaviors and experiences were identified, including cybervictimization and perpetration, exposure to and generation of SITB-related content, problematic use, sexting, social media importance, and frequency of use. A systematic search of PsycINFO, Medline, CINAHL, and the references of prior reviews yielded 61 eligible studies. Results largely suggested medium effect sizes for associations between specific social media constructs (cybervictimization, SITB-related social media use, problematic social media use) and SITBs. There was no association between frequency of social media use and SITBs; however, studies on this topic were limited. The majority of studies identified focused on cybervictimization, and results suggested positive associations with all SITBs, with the association between cybervictimization and suicidal ideation stronger for adolescents than adults. Overall, findings highlight the utility of examining specific social media behaviors and experiences, and point to the need for more research in this area.

Tweet Classification to Assist Human Moderation for Suicide Prevention

Social media platforms are already engaged in leveraging existing online socio-technical systems to employ just-in-time interventions for suicide prevention to the public. These efforts primarily rely on self-reports of potential self-harm content that is reviewed by moderators. Most recently, platforms have employed automated models to identify self-harm content, but acknowledge that these automated models still struggle to understand the nuance of human language (e.g., sarcasm). By explicitly focusing on Twitter posts that could easily be misidentified by a model as expressing suicidal intent (i.e., they contain similar phrases such as "wanting to die"), our work examines the temporal differences in historical expressions of general and emotional language prior to a clear expression of suicidal intent. Additionally, we analyze time-aware neural models that build on these language variants and factors in the historical, emotional spectrum of a user's tweeting activity. The strongest model achieves high (statistically significant) performance (macro F1=0.804, recall=0.813) to identify social media indicative of suicidal intent. Using three use cases of tweets with phrases common to suicidal intent, we qualitatively analyze and interpret how such models decided if suicidal intent was present and discuss how these analyses may be used to alleviate the burden on human moderators within the known constraints of how moderation is performed (e.g., no access to the user's timeline). Finally, we discuss the ethical implications of such data-driven models and inferences about suicidal intent from social media. Content warning: this article discusses self-harm and suicide.

Neurophysiological Markers Related to Negative Self-referential Processing Differentiate Adolescent Suicide Ideators and Attempters

Background

Adolescent suicide is a major public health concern, and presently, there is a limited understanding of the neurophysiological correlates of suicidal behaviors. Cognitive models of suicide indicate that negative views of the self are related to suicidal thoughts and behaviors, and this study investigated whether behavioral and neural correlates of self-referential processing differentiate suicide ideators from recent attempters.

Methods

Adolescents with depression reporting current suicidal ideation and no lifetime suicide attempts (suicide ideators, n = 30) and past-year suicide attempts (recent attempters, n = 26) completed a self-referential encoding task while high-density electroencephalogram data were recorded. Behavioral analyses focused on negative processing bias (i.e., tendency to attribute negative information as being self-relevant) and drift rate (i.e., slope of reaction time and response type that corresponds to how quickly information is accumulated to make a decision about whether words are self-referent). Neurophysiological markers probing components reflecting early semantic monitoring (P2), engagement (early late positive potential), and effortful encoding (late late positive potential) also were tested.

Results

Adolescent suicide ideators and recent suicide attempters reported comparable symptom severity, suicide ideation, and mental disorders. Although there were no behavioral differences, compared with suicide ideators, suicide attempters exhibited greater P2 amplitudes for negative versus positive words, which may reflect enhanced attention and arousal in response to negative self-referential stimuli. There were no group differences for the early or late late positive potential.

Conclusions

Enhanced sensory arousal in response to negative stimuli-that is, attentional orienting to semantic, emotional, and self-relevant features-differentiates adolescent suicide attempters from ideators and thus may signal risk for suicidal behavior.