INTRODUCTION
In modern psychiatric inpatient settings, ensuring patient safety is paramount, particularly for individuals at high risk of self-harm or harm to others. Despite advances in psychiatric care, current risk assessment methods often rely on subjective evaluations or retrospective data, limiting their effectiveness in real-time crisis prevention. This challenge is compounded by limited staffing resources and the absence of tools capable of continuous, objective monitoring. As a result, incidents of self-harm or violence remain a significant concern in psychiatric wards.
Previous studies have identified various risk factors for mental instability, suicide, and aggression, including historical events, acute psychiatric symptoms, and environmental stressors [1-3]. However, these approaches often fail to integrate diverse risk factors at multiple temporal levels or provide timely predictions of imminent crises. This gap highlights the need for innovative early-warning systems that leverage non-verbal indicators, such as physiological changes and behavioral patterns, which frequently precede verbal expressions of distress.
Recent advancements in wearable sensor technology and machine learning have enabled real-time monitoring of digital phenotypes—health-related data passively collected via personal digital devices [4]. Digital phenotyping has shown promise in predicting health outcomes and psychiatric symptoms, particularly in outpatient or smartphone-based contexts [5-7]. However, its application in inpatient psychiatric settings remains underexplored, despite the potential to anticipate and prevent crises through continuous, real-time data analysis.
To address this unmet need, we developed the Mental Instability and Risk Early Detection Monitoring System (MIR-Med System), an innovative real-time monitoring platform designed to predict and prevent psychiatric crises among inpatients. The system integrates wearable devices and machine learning algorithms to collect and analyze digital phenotyping data, such as circadian rhythm patterns, heart rate variability (HRV), and sleep quality. It employs a three-layer risk assessment model that combines long-term historical factors, intermediate circadian rhythm disruptions, and short-term physiological indicators. This layered approach allows for a comprehensive and proactive framework for risk assessment and intervention.
This study aims to evaluate the feasibility, accuracy, and clinical utility of the MIR-Med System in psychiatric inpatient settings. By comparing system-generated alerts with verified clinical outcomes, this research seeks to establish the system’s potential to enhance patient safety, optimize clinical workflows, and improve outcomes in psychiatric care. Findings will contribute to the development of scalable real-time monitoring solutions applicable to diverse healthcare environments.
HYPOTHESIS
This study hypothesizes that the risk of self-harm or harm to others in psychiatric inpatients can be predicted using a three-layer risk assessment model that enables proactive, real-time monitoring. The three layers integrate long-term, intermediate, and short-term factors, creating a robust framework for risk evaluation.
Layer 1: Historical and initial assessment factors
The first layer includes historical risk factors (e.g., prior suicide attempts, self-harm, and violent behavior) and symptoms observed during the initial clinical evaluation. These well-established predictors provide insight into the patient’s risk profile over time-frames ranging from the past week to their lifetime [1,2].
Layer 2: Circadian rhythm disruption and psychological instability
The second layer is based on previous studies [8-13] demonstrating that disruptions in circadian rhythms contribute to psychiatric instability. Using data collected over extended periods of up to 4–5 years from hundreds of patients with major mood disorders, we leveraged digital phenotyping tools, including wearable devices and ecological momentary assessment apps, to capture psychological variables such as mood, anxiety, and irritability. These data were used to learn how these variables are associated with circadian rhythm patterns. Based on these findings, this layer evaluates risk factors within a timeframe of one day to one week.
Layer 3: Immediate physiological and behavioral indicators
The third layer focuses on short-term characteristics, typically within minutes to a day. Key indicators include: 1) quality and quantity of sleep from the previous night, 2) resting HRV upon waking, and 3) sudden changes in heart rate. These indicators are derived from data obtained through wearable devices.
By integrating these three layers, the study aims to validate the feasibility of a multi-dimensional monitoring system to predict and prevent self-harm and violent crises among psychiatric inpatients. This layered approach combines diverse temporal and physiological predictors, enabling real-time alerts and timely interventions to enhance patient safety and clinical care (Figure 1).
DESIGN AND METHODS
Purpose
The primary purpose of this study is to evaluate the feasibility and clinical utility of the MIR-Med System, a novel real-time monitoring platform designed to predict and prevent psychiatric crises, including self-harm and aggression, among psychiatric inpatients. The study aims to assess the system’s capability to collect reliable and consistent data, its accuracy in generating predictive alerts, and its integration into routine clinical workflows. Secondary objectives include evaluating the usability and scalability of the system, as well as its potential impact on improving patient outcomes and optimizing clinical efficiency.
Participant selection
Participants in this study will be recruited from the Department of Psychiatry at Korea University Anam Hospital. Eligible participants must be adults aged 18 years or older who have been diagnosed with a psychiatric disorder and are capable of providing informed consent. All participants must agree to wear a wearable device continuously throughout their hospitalization. Individuals will be excluded from the study if they have severe cognitive impairment that precludes their understanding of the study objectives or process, or if they have physical conditions, such as dermatological sensitivities, that prevent the use of wearable devices.
Ethical considerations
This study has been approved by the Institutional Review Board (IRB) of Korea University Anam Hospital [No. 2024AN0310]. Informed consent will be obtained from all participants prior to their enrollment in the study. To ensure confidentiality, all data will be anonymized and stored on encrypted servers, with robust cybersecurity measures in place to protect against unauthorized access or breaches. Participants will retain the right to withdraw from the study at any point without any impact on their standard treatment or care.
MIR-Med System overview
The MIR-Med System is a cutting-edge monitoring platform that integrates wearable devices with machine learning algorithms to provide real-time prediction and prevention of psychiatric crises. Participants will wear Fitbit Charge 5 devices (Fitbit Inc.), which will continuously collect physiological and behavioral data, including circadian rhythms, HRV, and the quality and quantity of sleep. This data will be transmitted through Raspberry Pi 4B devices using LTE networks, ensuring uninterrupted communication with the MIR-Med System. Machine learning algorithms will process the data in real time to detect deviations from baseline indicators and predict the likelihood of psychiatric crises. When a predefined risk threshold is exceeded, the system will generate alerts, which will be communicated to medical staff via an intuitive web-based interface with visual and auditory notifications. Medical staff will also log clinical events, such as incidents of self-harm or aggression, to provide feedback for iterative refinement of the system’s predictive models.
Data management and quality control
To ensure the accuracy and reliability of the collected data, the study will implement a comprehensive data management and quality control framework. Data collected by the wearable devices will undergo real-time validation to detect transmission errors, missing values, or outliers. Any anomalies will be flagged by automated algorithms and resolved either through predefined correction protocols or manual review by the research team. Compliance with device usage will be monitored through daily synchronization logs, and participants will receive reminders or follow-ups if data irregularities are detected. All data will be securely backed up on encrypted servers to prevent loss due to system failures. In the event of device malfunctions, a contingency protocol will be enacted to minimize disruptions in data collection and ensure data integrity. These measures aim to uphold the validity of the study findings and provide a robust foundation for real-time monitoring.
Study flow
The study will be conducted in the following stages:
1) Enrollment and baseline assessment: Participants who provide informed consent will undergo an initial clinical assessment conducted by medical staff. Baseline risk factors, including historical clinical data and initial observations, will be entered into the MIR-Med System to establish individualized thresholds for risk prediction.
2) Continuous monitoring: Participants will wear Fitbit devices continuously throughout their hospitalization. These devices will collect real-time data on circadian rhythms, HRV, and sleep, which will be transmitted to the system for processing and analysis.
3) Real-time alerts: The MIR-Med System will analyze the incoming data in real time and generate alerts when risk thresholds are exceeded. Alerts will be communicated to medical staff to facilitate timely interventions.
4) Incident logging and feedback: Medical staff will record clinical incidents, such as self-harm or aggression, into the system, providing data for algorithm refinement and enhancing the predictive model’s accuracy.
5) Standard care: Throughout the study, participants will continue receiving standard treatment protocols, and no modifications to their care plans will be made based on the system’s alerts.
Performance metrics
The performance of the MIR-Med System will be evaluated using several key metrics. Sensitivity will measure the proportion of true positive cases correctly identified by the system, while specificity will assess the proportion of true negative cases accurately classified. Overall accuracy will reflect the correctness of all predictions, and the F1 score, representing the harmonic mean of precision and recall, will provide a balanced evaluation of the system’s performance in managing false positives and false negatives. These metrics will be calculated and compared to clinical outcomes to validate the system’s effectiveness in predicting psychiatric crises.
LIMITATIONS AND FUTURE DIRECTIONS
This study is limited by its single-site design, which may restrict the generalizability of the findings to other inpatient settings. Future research should include multi-center trials to validate the system across diverse clinical environments. Additionally, the study focuses solely on psychiatric inpatients, highlighting the need to explore the system’s applicability in outpatient and community-based care settings. Future studies should also investigate the long-term integration of the MIR-Med System into clinical workflows and consider the inclusion of additional physiological and behavioral markers to improve the system’s predictive accuracy.
CONCLUSION
The findings of this study are expected to contribute to advancing the field of psychiatric care by addressing critical gaps in current risk assessment practices. The MIR-Med System represents an innovative approach that integrates wearable technology and machine learning to provide real-time monitoring and prediction of psychiatric crises, including self-harm and aggression, among inpatient populations. By leveraging a three-layer model that incorporates historical risk factors, circadian rhythm disruptions, and immediate physiological indicators, the MIR-Med System offers a comprehensive and proactive method for enhancing patient safety and optimizing clinical workflows.
The potential of this system lies not only in its ability to improve early detection of psychiatric instability but also in its capacity to reduce the burden on healthcare providers by streamlining risk management processes. By enabling timely interventions and providing objective, data-driven insights, the MIR-Med System can foster a more efficient and responsive psychiatric care environment. Furthermore, the systematic validation of this technology in clinical settings serves as a foundation for its future application in outpatient and community-based care.
Despite the promising potential, this study acknowledges certain limitations, including its single-site design and the need for further evaluation in diverse clinical environments. Future research should focus on multi-center trials to establish generalizability, explore additional predictive markers to enhance accuracy, and investigate the long-term impact of integrating such systems into routine psychiatric care.
In conclusion, the MIR-Med System offers a novel, technology-driven solution to longstanding challenges in psychiatric risk assessment. Its development and validation mark a significant step forward in leveraging digital health tools to improve mental health outcomes. This research sets the stage for broader adoption of real-time monitoring technologies in mental healthcare, with the ultimate goal of delivering safer, more efficient, and patient-centered care.
PDF Links
PubReader
ePub Link
XML Download
Full text via DOI
Download Citation
