Examining Passively Collected Smartphone-Based Data in the Days Prior to Psychiatric Hospitalization for a Suicidal Crisis: Comparative Case Analysis

BACKGROUND

Digital phenotyping has seen a broad increase in application across clinical research; however, little research has implemented passive assessment approaches for suicide risk detection. There is a significant potential for a novel form of digital phenotyping, termed screenomics, which captures smartphone activity via screenshots.

OBJECTIVE

This paper focuses on a comprehensive case review of 2 participants who reported past 1-month active suicidal ideation, detailing their passive (ie, obtained via screenomics screenshot capture) and active (ie, obtained via ecological momentary assessment [EMA]) risk profiles that culminated in suicidal crises and subsequent psychiatric hospitalizations. Through this analysis, we shed light on the timescale of risk processes as they unfold before hospitalization, as well as introduce the novel application of screenomics within the field of suicide research.

METHODS

To underscore the potential benefits of screenomics in comprehending suicide risk, the analysis concentrates on a specific type of data gleaned from screenshots-text-captured prior to hospitalization, alongside self-reported EMA responses. Following a comprehensive baseline assessment, participants completed an intensive time sampling period. During this period, screenshots were collected every 5 seconds while one's phone was in use for 35 days, and EMA data were collected 6 times a day for 28 days. In our analysis, we focus on the following: suicide-related content (obtained via screenshots and EMA), risk factors theoretically and empirically relevant to suicide risk (obtained via screenshots and EMA), and social content (obtained via screenshots).

RESULTS

Our analysis revealed several key findings. First, there was a notable decrease in EMA compliance during suicidal crises, with both participants completing fewer EMAs in the days prior to hospitalization. This contrasted with an overall increase in phone usage leading up to hospitalization, which was particularly marked by heightened social use. Screenomics also captured prominent precipitating factors in each instance of suicidal crisis that were not well detected via self-report, specifically physical pain and loneliness.

CONCLUSIONS

Our preliminary findings underscore the potential of passively collected data in understanding and predicting suicidal crises. The vast number of screenshots from each participant offers a granular look into their daily digital interactions, shedding light on novel risks not captured via self-report alone. When combined with EMA assessments, screenomics provides a more comprehensive view of an individual's psychological processes in the time leading up to a suicidal crisis.

A Machine Learning Approach to Passively Informed Prediction of Mental Health Risk in People with Diabetes: Retrospective Case-Control Analysis

Background

Proactive detection of mental health needs among people with diabetes mellitus could facilitate early intervention, improve overall health and quality of life, and reduce individual and societal health and economic burdens. Passive sensing and ecological momentary assessment are relatively newer methods that may be leveraged for such proactive detection.

Objective

The primary aim of this study was to conceptualize, develop, and evaluate a novel machine learning approach for predicting mental health risk in people with diabetes mellitus.

Methods

A retrospective study was designed to develop and evaluate a machine learning model, utilizing data collected from 142,432 individuals with diabetes enrolled in the Livongo for Diabetes program. First, participants’ mental health statuses were verified using prescription and medical and pharmacy claims data. Next, four categories of passive sensing signals were extracted from the participants’ behavior in the program, including demographics and glucometer, coaching, and event data. Data sets were then assembled to create participant-period instances, and descriptive analyses were conducted to understand the correlation between mental health status and passive sensing signals. Passive sensing signals were then entered into the model to train and test its performance. The model was evaluated based on seven measures: sensitivity, specificity, precision, area under the curve, F₁ score, accuracy, and confusion matrix. SHapley Additive exPlanations (SHAP) values were computed to determine the importance of individual signals.

Results

In the training (and validation) and three subsequent test sets, the model achieved a confidence score greater than 0.5 for sensitivity, specificity, area under the curve, and accuracy. Signals identified as important by SHAP values included demographics such as race and gender, participant’s emotional state during blood glucose checks, time of day of blood glucose checks, blood glucose values, and interaction with the Livongo mobile app and web platform.

Conclusions

Results of this study demonstrate the utility of a passively informed mental health risk algorithm and invite further exploration to identify additional signals and determine when and where such algorithms should be deployed.

Smart Helmet 5.0 for Industrial Internet of Things Using Artificial Intelligence

Information and communication technologies (ICTs) have contributed to advances in Occupational Health and Safety, improving the security of workers. The use of Personal Protective Equipment (PPE) based on ICTs reduces the risk of accidents in the workplace, thanks to the capacity of the equipment to make decisions on the basis of environmental factors. Paradigms such as the Industrial Internet of Things (IIoT) and Artificial Intelligence (AI) make it possible to generate PPE models feasibly and create devices with more advanced characteristics such as monitoring, sensing the environment and risk detection between others. The working environment is monitored continuously by these models and they notify the employees and their supervisors of any anomalies and threats. This paper presents a smart helmet prototype that monitors the conditions in the workers’ environment and performs a near real-time evaluation of risks. The data collected by sensors is sent to an AI-driven platform for analysis. The training dataset consisted of 11,755 samples and 12 different scenarios. As part of this research, a comparative study of the state-of-the-art models of supervised learning is carried out. Moreover, the use of a Deep Convolutional Neural Network (ConvNet/CNN) is proposed for the detection of possible occupational risks. The data are processed to make them suitable for the CNN and the results are compared against a Static Neural Network (NN), Naive Bayes Classifier (NB) and Support Vector Machine (SVM), where the CNN had an accuracy of 92.05% in cross-validation.

A Study of Risk Relevance Reasoning Based on a Context Ontology of Railway Accidents

With the application of risk management and accident response in the railway domain, risk detection and prevention have become key research topics. Many dangers and associated risk sources must be considered in collaborative scenarios of heavy-haul railways. In these scenarios, (1) various risk sources are involved in different data sources, and context affects their occurrence, (2) the relationships between contexts and risk sources in the accident cause mechanism need to be explicitly defined, and (3) risk knowledge reasoning needs to integrate knowledge from multiple data sources to achieve comprehensive results. To express the association rules among core concepts, this article constructs two ontologies: The accident-risk ontology and the context ontology. Concept analysis is based on railway domain knowledge and accident analysis reports. To sustainably integrate knowledge, an integrated evolutionary model called scenario-risk-accident chain ontology (SRAC) is constructed by introducing new data sources. The SRAC is integrated through expert rules between the two ontologies, and its evolution process involves new knowledge through a new risk source database. After three versions of the upgrade process, potential risk sources can be mined and evaluated in specific contexts. To evaluate the risk source level, a long short-term memory (LSTM) neural network model is used to capture context and risk text features. A model comparison for different neural network structures is performed to find the optimal evaluation results. Finally, new concepts, such as risk source level, and new instances are updated in the context-aware risk knowledge reasoning framework.

Learning Process During Risk Detection in Adolescents With ADHD

OBJECTIVE

To evaluate and describe the performance during the learning process of risk-detection versus risk-benefit processing in adolescents diagnosed with ADHD.

METHOD

Thirty-five adolescents with ADHD and 26 paired controls participated. The tests applied are Iowa-type children version paradigm and Stroop test.

RESULTS

Adolescents with ADHD exhibited lower risk-benefit processing capacity and lower ability to detect risk selections; main findings also indicate that adolescents with ADHD were slower to learn to avoid risk choices. In addition, they also presented a deficient inhibitory control.

CONCLUSION

Results confirm the presence of a deficit in advantageous choice in adolescents with ADHD. By providing a measure of risk choice-and not only a net score-we show that adolescents with ADHD also fail to avoid risk choices. This deficit is mainly because they are slower in learning how to avoid risk choices, and not simply deficient. Literature is scarce concerning studies with Iowa-type paradigms in samples intregated exclusively by adolescents. More research is needed to clarify the nature of these deficiencies.

Enhancing Risk Detection Among Homeless Youth: A Randomized Clinical Trial of a Promising Pilot Intervention

Homeless youth frequently experience victimization, and youth with histories of trauma often fail to detect danger risks, making them vulnerable to subsequent victimization. The current study describes a pilot test of a skills-based intervention designed to improve risk detection among homeless youth through focusing attention to internal, interpersonal, and environmental cues. Youth aged 18 to 21 years ( N = 74) were recruited from a shelter and randomly assigned to receive usual case management services or usual services plus a 3-day manualized risk detection intervention. Pretest and posttest interviews assessed youths' risk detection abilities through vignettes describing risky situations and asking youth to identify risk cues present. Separate 2 (intervention vs. control) × 2 (pretest vs. posttest) mixed ANOVAs found significant interaction effects, as intervention youth significantly improved in overall risk detection compared with control youth. Post hoc subgroup analyses found the intervention had a greater effect for youth without previous experiences of indirect victimization than those with previous indirect victimization experiences.

Facial emotion identification and sexual assault risk detection among college student sexual assault victims and nonvictims

OBJECTIVE

High rates of sexual victimization among college students necessitate further study of factors associated with sexual assault risk detection. The present study examined how social information processing relates to sexual assault risk detection as a function of sexual assault victimization history.

PARTICIPANTS

225 undergraduates (M_age = 19.12, SD = 1.44; 66% women).

METHODS

Participants completed an online questionnaire assessing victimization history, an emotion identification task, and a sexual assault risk detection task between June 2013 and May 2014.

RESULTS

Emotion identification moderated the association between victimization history and risk detection such that sexual assault survivors with lower emotion identification accuracy also reported the least risk in a sexual assault vignette.

CONCLUSIONS

Findings suggest that differences in social information processing, specifically recognition of others' emotions, are associated with sexual assault risk detection. College prevention programs could incorporate emotional awareness strategies, particularly for men and women who are sexual assault survivors.

Exploring revictimization risk in a community sample of sexual assault survivors

Previous research points to links between risk detection (the ability to detect danger cues in various situations) and sexual revictimization in college women. Given important differences between college and community samples that may be relevant to revictimization risk (e.g., the complexity of trauma histories), the current study explored the link between risk detection and revictimization in a community sample of women. Community-recruited women (N = 94) reported on their trauma histories in a semistructured interview. In a laboratory session, participants listened to a dating scenario involving a woman and a man that culminated in sexual assault. Participants were instructed to press a button "when the man had gone too far." Unlike in college samples, revictimized community women (n = 47) did not differ in terms of risk detection response times from women with histories of no victimization (n = 10) or single victimization (n = 15). Data from this study point to the importance of examining revictimization in heterogeneous community samples where risk mechanisms may differ from college samples.