Background: Mental health problems are a crucial global public health concern. Owing to their cost-effectiveness and accessibility, conversational agent interventions (CAIs) are promising in the field of mental health care. Objective: This study aims to present a thorough summary of the traits of CAIs available for a range of mental health problems, find evidence of efficacy, and analyze the statistically significant moderators of efficacy via a meta-analysis of randomized controlled trial. Methods: Web-based databases (Embase, MEDLINE, PsycINFO, CINAHL, Web of Science, and Cochrane) were systematically searched dated from the establishment of the database to October 30, 2021, and updated to May 1, 2022. Randomized controlled trials comparing CAIs with any other type of control condition in improving depressive symptoms, generalized anxiety symptoms, specific anxiety symptoms, quality of life or well-being, general distress, stress, mental disorder symptoms, psychosomatic disease symptoms, and positive and negative affect were considered eligible. This study followed the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines. Data were extracted by 2 independent reviewers, checked by a third reviewer, and pooled using both random effect models and fixed effects models. Hedges g was chosen as the effect size. Results: Of the 6900 identified records, a total of 32 studies were included, involving 6089 participants. CAIs showed statistically significant short-term effects compared with control conditions in improving depressive symptoms (g=0.29, 95% CI 0.20-0.38), generalized anxiety symptoms (g=0.29, 95% CI 0.21-0.36), specific anxiety symptoms (g=0.47, 95% CI 0.07-0.86), quality of life or well-being (g=0.27, 95% CI 0.16-0.39), general distress (g=0.33, 95% CI 0.20-0.45), stress (g=0.24, 95% CI 0.08-0.41), mental disorder symptoms (g=0.36, 95% CI 0.17-0.54), psychosomatic disease symptoms (g=0.62, 95% CI 0.14-1.11), and negative affect (g=0.28, 95% CI 0.05-0.51). However, the long-term effects of CAIs for the most mental health outcomes were not statistically significant (g=−0.04 to 0.39). Personalization and empathic response were 2 critical facilitators of efficacy. The longer duration of interaction with conversational agents was associated with the larger pooled effect sizes. Conclusions: The findings show that CAIs are research-proven interventions that ought to be implemented more widely in mental health care. CAIs are effective and easily acceptable for those with mental health problems. The clinical application of this novel digital technology will conserve human health resources and optimize the allocation of mental health services. Trial Registration: PROSPERO CRD42022350130; https://tinyurl.com/mvhk6w9p

Background: Depression has a high prevalence among young adults, especially during the COVID-19 pandemic. However, mental health services remain scarce and underutilized worldwide. Mental health chatbots are a novel digital technology to provide fully automated interventions for depressive symptoms. Objective: The purpose of this study was to test the clinical effectiveness and nonclinical performance of a cognitive behavioral therapy (CBT)–based mental health chatbot (XiaoE) for young adults with depressive symptoms during the COVID-19 pandemic. Methods: In a single-blind, 3-arm randomized controlled trial, participants manifesting depressive symptoms recruited from a Chinese university were randomly assigned to a mental health chatbot (XiaoE; n=49), an e-book (n=49), or a general chatbot (Xiaoai; n=50) group in a ratio of 1:1:1. Participants received a 1-week intervention. The primary outcome was the reduction of depressive symptoms according to the 9-item Patient Health Questionnaire (PHQ-9) at 1 week later (T1) and 1 month later (T2). Both intention-to-treat and per-protocol analyses were conducted under analysis of covariance models adjusting for baseline data. Controlled multiple imputation and δ-based sensitivity analysis were performed for missing data. The secondary outcomes were the level of working alliance measured using the Working Alliance Questionnaire (WAQ), usability measured using the Usability Metric for User Experience-LITE (UMUX-LITE), and acceptability measured using the Acceptability Scale (AS). Results: Participants were on average 18.78 years old, and 37.2% (55/148) were female. The mean baseline PHQ-9 score was 10.02 (SD 3.18; range 2-19). Intention-to-treat analysis revealed lower PHQ-9 scores among participants in the XiaoE group compared with participants in the e-book group and Xiaoai group at both T1 (F2,136=17.011; P<.001; d=0.51) and T2 (F2,136=5.477; P=.005; d=0.31). Better working alliance (WAQ; F2,145=3.407; P=.04) and acceptability (AS; F2,145=4.322; P=.02) were discovered with XiaoE, while no significant difference among arms was found for usability (UMUX-LITE; F2,145=0.968; P=.38). Conclusions: A CBT-based chatbot is a feasible and engaging digital therapeutic approach that allows easy accessibility and self-guided mental health assistance for young adults with depressive symptoms. A systematic evaluation of nonclinical metrics for a mental health chatbot has been established in this study. In the future, focus on both clinical outcomes and nonclinical metrics is necessary to explore the mechanism by which mental health chatbots work on patients. Further evidence is required to confirm the long-term effectiveness of the mental health chatbot via trails replicated with a longer dose, as well as exploration of its stronger efficacy in comparison with other active controls. Trial Registration: Chinese Clinical Trial Registry ChiCTR2100052532; http://www.chictr.org.cn/showproj.aspx?proj=135744

The aim of this study was to test the Three-Step Theory (3ST) of Suicide in a sample of college students in China (N = 1097). All participants completed a battery of questionnaires indexing psychological pain, hopelessness, connectedness, suicide capacity, suicide ideation, and suicide attempt. The prevalence of suicide ideation and suicide attempt among the sample were 21.42% and 3.83%, respectively. The hierarchical regression analysis showed that psychological pain and hopelessness interacted to predict suicide ideation, and that connectedness was the most protective against ideation in those high on both pain and hopelessness. Suicide capacity differentiated attempters from ideators above and beyond current suicide ideation. These findings provide preliminary evidence for the validity of 3ST of suicide within the Chinese contexts. Implications for suicide prevention and intervention and future research directions are also discussed.

Background: Crisis hotlines serve as a crucial avenue for the early identification of suicide risk, which is of paramount importance for suicide prevention and intervention. However, assessing the risk of callers in the crisis hotline context is constrained by factors such as lack of nonverbal communication cues, anonymity, time limits, and single-occasion intervention. Therefore, it is necessary to develop approaches, including acoustic features, for identifying the suicide risk among hotline callers early and quickly. Given the complicated features of sound, adopting artificial intelligence models to analyze callers’ acoustic features is promising. Objective: In this study, we investigated the feasibility of using acoustic features to predict suicide risk in crisis hotline callers. We also adopted a machine learning approach to analyze the complex acoustic features of hotline callers, with the aim of developing suicide risk prediction models. Methods: We collected 525 suicide-related calls from the records of a psychological assistance hotline in a province in northwest China. Callers were categorized as low or high risk based on suicidal ideation, suicidal plans, and history of suicide attempts, with risk assessments verified by a team of 18 clinical psychology raters. A total of 164 clearly categorized risk recordings were analyzed, including 102 low-risk and 62 high-risk calls. We extracted 273 audio segments, each exceeding 2 seconds in duration, which were labeled by raters as containing suicide-related expressions for subsequent model training and evaluation. Basic acoustic features (eg, Mel Frequency Cepstral Coefficients, formant frequencies, jitter, shimmer) and high-level statistical function (HSF) features (using OpenSMILE [Open-Source Speech and Music Interpretation by Large-Space Extraction] with the ComParE 2016 configuration) were extracted. Four supervised machine learning algorithms (logistic regression, support vector machine, random forest, and extreme gradient boosting) were trained and evaluated using grouped 5-fold cross-validation and a test set, with performance metrics, including accuracy, F1-score, recall, and false negative rate. Results: The development of machine learning models utilizing HSF acoustic features has been demonstrated to enhance recognition performance compared to models based solely on basic acoustic features. The random forest classifier, developed with HSFs, achieved the best performance in detecting the suicide risk among the models evaluated (accuracy=0.75, F1-score=0.70, recall=0.76, false negative rate=0.24). Conclusions: The results of our study demonstrate the potential of developing artificial intelligence–based early warning systems using acoustic features for identifying the suicide risk among crisis hotline callers. Our work also has implications for employing acoustic features to identify suicide risk in salient voice contexts.

Social psychological studies show that people’s explicit attitude bias in public expression can differ from their implicit attitude bias in mind. However, the current automatic attitude analysis does not distinguish between explicit and implicit attitude bias. Simulating the psychological measurements of explicit and implicit attitude bias, i.e., self-report assessment (SRA) and implicit association test (IAT), we propose an automatic language-based analysis to distinguish explicit and implicit attitude bias in a large population. By connecting the criteria of SRA and IAT with the statements containing patterns of special words, we derive explicit and implicit attitude bias with the sentiment scores of the statements, which are obtained by pre-trained machine-learning methods. Extensive experiments on four English and Chinese corpora and four pairs of concepts show that the attitude biases obtained by our method on a large population are consistent with those of traditional psychological experiments in the costly small-scale experiments. The maximum gap between the sentiment scores of explicit and implicit biases reaches 0.9329. Furthermore, we achieve new findings on the difference between the evolution of explicit and implicit attitude bias. The maximum variance gap of sentiment scores in the dynamic changes between explicit and implicit biases reaches 0.249.