Hae Kook Lee
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Hae Kook Lee has a total of 144 co-authors. Here are the top co-authors by number of shared publications:
- Sun-Jin Jo
Department of Preventive Medicine, College of Medicine, The Catholic University of Korea, Seoul, Republic of Korea
10 shared publications - Hyeon Woo Yim
Department of Preventive Medicine, College of Medicine The Catholic University of Korea Seoul Korea
7 shared publications - Hyunsuk Jeong
Department of Preventive Medicine, College of Medicine, The Catholic University of Korea, Seoul, Korea
7 shared publications - Seung-Yup Lee
Department of Psychiatry, College of Medicine, The Catholic University of Korea, Seoul, Korea
7 shared publications - Yong-Sil Kweon
Department of Psychiatry, Uijeongbu St. Mary’s Hospital, College of Medicine, The Catholic University of Korea, Seoul, Republic of Korea
7 shared publications - Kyoung-Uk Lee
Department of Psychiatry, Uijeongbu St. Mary’s Hospital, College of Medicine, The Catholic University of Korea, Seoul, Korea
4 shared publications
Latest Publications
- Open Access
- 10 July 2026
- Published by JMIR Publications Inc. in JMIR mHealth and uHealth
- Vol. 14, e88223
Week-Ahead Prediction of High-Risk Drinking Episodes Among Young Adults Using Wearable Biosignals and Psychological Vulnerabilities: Prospective Observational Machine Learning Study
Background: Although machine learning has increasingly been used to predict mental health symptoms and maladaptive behaviors, real-world prediction of addiction-related risk remains limited. Emotional and temperamental vulnerabilities are established correlates of alcohol-related problems, yet few studies have integrated these factors with wearable-derived biosignals in alcohol-risk prediction models. Objective: This study evaluated whether machine learning models could predict weekly high-risk drinking episodes among young adults with elevated alcohol-use risk by integrating wearable-derived health data with baseline emotional and personality vulnerability indicators. Methods: In this prospective observational study, adults in their 20s completed weekly self-report surveys and wore Fitbit devices for 4 weeks. Features from week t were used to predict the Alcohol Use Disorders Identification Test-Korean version (AUDIT-K)–based high-risk drinking label at week t+1. Positive labels were defined using AUDIT-K high-risk drinking cutoffs, with scores of ≥20 for men and ≥10 for women. Extreme gradient boosting (XGBoost) and random forest models were evaluated across self-report-only, wearable-only, and integrated feature sets using 5-fold participant-level grouped cross-validation. Results: A total of 206 participants contributed 620 week-level observations, of which 85 (13.7%) were labeled as positive high-risk drinking episodes. In participant-level grouped cross-validation, the integrated random forest model showed the most favorable sensitivity-oriented performance, with a mean accuracy of 0.617 (SD 0.078), recall/sensitivity of 0.653 (SD 0.144), area under the receiver operating characteristic curve (ROC AUC) of 0.681 (SD 0.079), and area under the precision-recall curve (PR AUC) of 0.255 (SD 0.090). The integrated XGBoost model achieved an accuracy of 0.670 (SD 0.089), recall/sensitivity of 0.399 (SD 0.174), ROC AUC of 0.651 (SD 0.089), and PR AUC of 0.228 (SD 0.074). Shapley additive explanations analyses indicated that both baseline vulnerability indicators and wearable-derived weekly summaries contributed to model predictions. Conclusions: Integrating baseline emotional and personality vulnerability indicators with wearable-derived weekly health signals may provide useful information for week-ahead prediction of high-risk drinking episodes. These findings provide preliminary support for wearable-assisted alcohol-risk stratification, although the modest positive predictive performance indicates that external validation and more proximal within-person measures are needed before real-world early-warning or just-in-time adaptive intervention applications.
- 10 July 2026
- Published by JMIR Publications Inc. in JMIR mHealth and uHealth
- Vol. 14, e88223
Open AccessWeek-Ahead Prediction of High-Risk Drinking Episodes Among Young Adults Using Wearable Biosignals and Psychological Vulnerabilities: Prospective Observational Machine Learning Study
Background: Although machine learning has increasingly been used to predict mental health symptoms and maladaptive behaviors, real-world prediction of addiction-related risk remains limited. Emotional and temperamental vulnerabilities are established correlates of alcohol-related problems, yet few studies have integrated these factors with wearable-derived biosignals in alcohol-risk prediction models. Objective: This study evaluated whether machine learning models could predict weekly high-risk drinking episodes among young adults with elevated alcohol-use risk by integrating wearable-derived health data with baseline emotional and personality vulnerability indicators. Methods: In this prospective observational study, adults in their 20s completed weekly self-report surveys and wore Fitbit devices for 4 weeks. Features from week t were used to predict the Alcohol Use Disorders Identification Test-Korean version (AUDIT-K)–based high-risk drinking label at week t+1. Positive labels were defined using AUDIT-K high-risk drinking cutoffs, with scores of ≥20 for men and ≥10 for women. Extreme gradient boosting (XGBoost) and random forest models were evaluated across self-report-only, wearable-only, and integrated feature sets using 5-fold participant-level grouped cross-validation. Results: A total of 206 participants contributed 620 week-level observations, of which 85 (13.7%) were labeled as positive high-risk drinking episodes. In participant-level grouped cross-validation, the integrated random forest model showed the most favorable sensitivity-oriented performance, with a mean accuracy of 0.617 (SD 0.078), recall/sensitivity of 0.653 (SD 0.144), area under the receiver operating characteristic curve (ROC AUC) of 0.681 (SD 0.079), and area under the precision-recall curve (PR AUC) of 0.255 (SD 0.090). The integrated XGBoost model achieved an accuracy of 0.670 (SD 0.089), recall/sensitivity of 0.399 (SD 0.174), ROC AUC of 0.651 (SD 0.089), and PR AUC of 0.228 (SD 0.074). Shapley additive explanations analyses indicated that both baseline vulnerability indicators and wearable-derived weekly summaries contributed to model predictions. Conclusions: Integrating baseline emotional and personality vulnerability indicators with wearable-derived weekly health signals may provide useful information for week-ahead prediction of high-risk drinking episodes. These findings provide preliminary support for wearable-assisted alcohol-risk stratification, although the modest positive predictive performance indicates that external validation and more proximal within-person measures are needed before real-world early-warning or just-in-time adaptive intervention applications.
- Open Access
- 25 May 2026
- Published by Korean Neuropsychiatric Association in Psychiatry Investigation
- Vol. 23 (5) , 577-589
Evidence for Mental Health Promotion and Prevention in Korea: A Narrative Review
Objective This review critically evaluates internationally indexed evidence on mental health promotion and prevention interventions in Korea and examines whether the growing national infrastructure is underpinned by a scientific foundation that meets global standards.Methods A narrative review was conducted using PubMed, Embase, the Cochrane Library, and Web of Science to identify peer-reviewed quantitative studies published from database inception to August 8, 2025. Eligible studies evaluated programmatic or policy-level interventions explicitly designed to promote psychological well-being or prevent mental disorders among Koreans. Interventions were categorized according to the Institute of Medicine prevention framework and synthesized narratively.Results A total of 135 studies were included. Research targeting adolescents was relatively sparse, with a marked lack of promotion-focused interventions. Although research activity increased sharply after 2020, the evidence base remains uneven in methodological rigor. Only 23.7% of studies employed randomized controlled designs. A structural disconnect was evident in the research pipeline, where pilot studies rarely progressed to full-scale efficacy trials. Furthermore, interventions delivered through core community mental health institutions were frequently bypassed in research, with recent studies shifting heavily toward digital and web-based platforms.Conclusion While research volume has expanded, the field lacks the methodological and structural maturity required to guide national policy effectively. Establishing a robust scientific foundation is a prerequisite for transforming current national endeavors into a system delivering genuine population-level benefits. Future initiatives must prioritize rigorous evidence generation to align Korea’s mental health promotion and prevention strategies with global standards.
- 25 May 2026
- Published by Korean Neuropsychiatric Association in Psychiatry Investigation
- Vol. 23 (5) , 577-589
Open AccessEvidence for Mental Health Promotion and Prevention in Korea: A Narrative Review
Objective This review critically evaluates internationally indexed evidence on mental health promotion and prevention interventions in Korea and examines whether the growing national infrastructure is underpinned by a scientific foundation that meets global standards.Methods A narrative review was conducted using PubMed, Embase, the Cochrane Library, and Web of Science to identify peer-reviewed quantitative studies published from database inception to August 8, 2025. Eligible studies evaluated programmatic or policy-level interventions explicitly designed to promote psychological well-being or prevent mental disorders among Koreans. Interventions were categorized according to the Institute of Medicine prevention framework and synthesized narratively.Results A total of 135 studies were included. Research targeting adolescents was relatively sparse, with a marked lack of promotion-focused interventions. Although research activity increased sharply after 2020, the evidence base remains uneven in methodological rigor. Only 23.7% of studies employed randomized controlled designs. A structural disconnect was evident in the research pipeline, where pilot studies rarely progressed to full-scale efficacy trials. Furthermore, interventions delivered through core community mental health institutions were frequently bypassed in research, with recent studies shifting heavily toward digital and web-based platforms.Conclusion While research volume has expanded, the field lacks the methodological and structural maturity required to guide national policy effectively. Establishing a robust scientific foundation is a prerequisite for transforming current national endeavors into a system delivering genuine population-level benefits. Future initiatives must prioritize rigorous evidence generation to align Korea’s mental health promotion and prevention strategies with global standards.
- 16 April 2026
- Published by Elsevier in International Journal of Drug Policy
- Vol. 152, 105280
Culturally mediated recovery: A cross-perspective qualitative study of psychosocial treatment for drug dependence in South Korea
- 16 April 2026
- Published by Elsevier in International Journal of Drug Policy
- Vol. 152, 105280
Culturally mediated recovery: A cross-perspective qualitative study of psychosocial treatment for drug dependence in South Korea
- Open Access
- 7 April 2026
- Published by Akademiai Kiado Zrt. in Journal of Behavioral Addictions
- Vol. 15 (2) , 756-765
Sex differences in the longitudinal association between high risk of internet gaming disorder and subsequent suicidality among adolescents: A cohort study
Objectives: Internet Gaming Disorder (IGD) has become a growing public health concern among adolescents. Although previous cross-sectional studies have linked problematic gaming with suicidality, longitudinal evidence remains limited. This study examined sex-specific associations between IGD risk and subsequent suicidal behaviors among adolescents who were not suicidal at baseline. Methods: Data were derived from the iCURE study, a prospective cohort involving 1,874 students (1,082 boys and 792 girls) aged 9–14 years. Participants with baseline suicidality were excluded. The Internet Game Use-Elicited Symptom Screen was used to identify adolescents at high risk of IGD (HRIGD). Suicidality was assessed at two-year follow-up using items adapted from the SCID-I. Relative risks (RRs) and 95% Confidence Intervals (CIs) were estimated using generalized linear models stratified by sex and adjusted for potential confounders. Results: At baseline, 8.0% of boys and 3.4% of girls were classified as HRIGD. At the two-year follow-up, boys in the HRIGD group showed significantly higher incidences of suicidality (17.4% vs. 5.0%), suicidal ideation (16.3% vs. 4.7%), and suicide plans or attempts (5.8% vs. 1.0%) than those at lower risk. In fully adjusted models, HRIGD remained associated with suicidality (RR = 3.05, 95% CI = 1.53–6.07), suicide ideation (RR = 2.99, 95% CI = 1.48–6.05), and suicide plans and/or attempts (RR = 4.03, 95% CI = 1.08–15.07) among boys, but not among girls. Conclusions: This longitudinal cohort study among adolescents without baseline suicidality indicates that IGD risk is related to subsequent suicidality in boys. Early identification and sex-specific preventive strategies targeting excessive gaming may help mitigate future suicide risk in adolescents.
- 7 April 2026
- Published by Akademiai Kiado Zrt. in Journal of Behavioral Addictions
- Vol. 15 (2) , 756-765
Open AccessSex differences in the longitudinal association between high risk of internet gaming disorder and subsequent suicidality among adolescents: A cohort study
Objectives: Internet Gaming Disorder (IGD) has become a growing public health concern among adolescents. Although previous cross-sectional studies have linked problematic gaming with suicidality, longitudinal evidence remains limited. This study examined sex-specific associations between IGD risk and subsequent suicidal behaviors among adolescents who were not suicidal at baseline. Methods: Data were derived from the iCURE study, a prospective cohort involving 1,874 students (1,082 boys and 792 girls) aged 9–14 years. Participants with baseline suicidality were excluded. The Internet Game Use-Elicited Symptom Screen was used to identify adolescents at high risk of IGD (HRIGD). Suicidality was assessed at two-year follow-up using items adapted from the SCID-I. Relative risks (RRs) and 95% Confidence Intervals (CIs) were estimated using generalized linear models stratified by sex and adjusted for potential confounders. Results: At baseline, 8.0% of boys and 3.4% of girls were classified as HRIGD. At the two-year follow-up, boys in the HRIGD group showed significantly higher incidences of suicidality (17.4% vs. 5.0%), suicidal ideation (16.3% vs. 4.7%), and suicide plans or attempts (5.8% vs. 1.0%) than those at lower risk. In fully adjusted models, HRIGD remained associated with suicidality (RR = 3.05, 95% CI = 1.53–6.07), suicide ideation (RR = 2.99, 95% CI = 1.48–6.05), and suicide plans and/or attempts (RR = 4.03, 95% CI = 1.08–15.07) among boys, but not among girls. Conclusions: This longitudinal cohort study among adolescents without baseline suicidality indicates that IGD risk is related to subsequent suicidality in boys. Early identification and sex-specific preventive strategies targeting excessive gaming may help mitigate future suicide risk in adolescents.
- Open Access
- 2 April 2026
- Published by Akademiai Kiado Zrt. in Journal of Behavioral Addictions
- Vol. 15 (1) , 195-208
Health response to problematic usage of the internet: A global survey on trends, available treatments and key challenges
Background and Aims: Problematic usage of the internet (PUI) is a growing global concern, emerging among more than 5.3 billion people who use the internet worldwide. While specific forms such as online gaming and gambling are recognized as disorders or conditions for further study in diagnostic manuals, global data on prevalence, treatment, and health responses to PUI remain limited. This study aimed to obtain perspectives from representatives of addiction medicine/psychiatry societies regarding the scope, treatment, and health responses to PUI and identify gaps. Methods: A global survey was conducted through the International Society of Addiction Medicine's Global Expert Network (ISAM-GEN), involving addiction societies from 38 countries across Europe, Asia/Oceania, the Americas, and Africa. The survey assessed responses to non-specific PUI and five subtypes: online gaming, gambling, pornography, social media, and online shopping. It included case scenarios and questions on the significance and severity of PUI, and country-level health responses. Results: Online gambling (94.8%) and gaming (86.9%) were the most frequently reported PUI forms, followed by social media (84.2%), pornography (68.3%), and online shopping (52.6%). Psychotherapeutic approaches, particularly cognitive behavioral therapy, were the most widely available treatments, reported as accessible by over 70% of country respondents. Despite growing awareness—reflected in the formation of PUI interest groups in 44.7% of societies—gaps were reported, including lack of professional certification (78.9%), insufficient practitioner education (68.4%), and inadequate expert training (63.2%). Notably, 65.8% rated the 10-year severity of PUI as extremely or very important. Discussion & Conclusion: Global attention to PUI is increasing, but more robust healthcare responses are needed. Addressing existing gaps requires enhanced training and sustainable international efforts.
- 2 April 2026
- Published by Akademiai Kiado Zrt. in Journal of Behavioral Addictions
- Vol. 15 (1) , 195-208
Open AccessHealth response to problematic usage of the internet: A global survey on trends, available treatments and key challenges
Background and Aims: Problematic usage of the internet (PUI) is a growing global concern, emerging among more than 5.3 billion people who use the internet worldwide. While specific forms such as online gaming and gambling are recognized as disorders or conditions for further study in diagnostic manuals, global data on prevalence, treatment, and health responses to PUI remain limited. This study aimed to obtain perspectives from representatives of addiction medicine/psychiatry societies regarding the scope, treatment, and health responses to PUI and identify gaps. Methods: A global survey was conducted through the International Society of Addiction Medicine's Global Expert Network (ISAM-GEN), involving addiction societies from 38 countries across Europe, Asia/Oceania, the Americas, and Africa. The survey assessed responses to non-specific PUI and five subtypes: online gaming, gambling, pornography, social media, and online shopping. It included case scenarios and questions on the significance and severity of PUI, and country-level health responses. Results: Online gambling (94.8%) and gaming (86.9%) were the most frequently reported PUI forms, followed by social media (84.2%), pornography (68.3%), and online shopping (52.6%). Psychotherapeutic approaches, particularly cognitive behavioral therapy, were the most widely available treatments, reported as accessible by over 70% of country respondents. Despite growing awareness—reflected in the formation of PUI interest groups in 44.7% of societies—gaps were reported, including lack of professional certification (78.9%), insufficient practitioner education (68.4%), and inadequate expert training (63.2%). Notably, 65.8% rated the 10-year severity of PUI as extremely or very important. Discussion & Conclusion: Global attention to PUI is increasing, but more robust healthcare responses are needed. Addressing existing gaps requires enhanced training and sustainable international efforts.
- Open Access
- 19 February 2026
- Published by JMIR Publications Inc. in JMIR Research Protocols
- Vol. 15 (1) , e82782
Identifying Predictors of Gambling Episodes and Craving Using Ecological Momentary Assessment and Smartwatch-Based Physiological Measures: Protocol for a Longitudinal Observational Study
Background: Like other addictive behaviors, problem gambling is often chronic and relapsing. While digital interventions offer low-threshold treatment and support, their effectiveness is often limited by small effect sizes, low adherence, and high dropout rates. Progress in digital technology has enabled the development of ecological momentary interventions (EMIs), which provide just-in-time support tailored to users’ needs. However, EMIs for addictive behaviors have hardly been developed so far. Objective: This study aimed to explore and identify relevant predictors of gambling episodes assessed by ecological momentary assessments and physiological smartwatch (Apple Watch) data, which in turn may be used for the further development of EMI-based interventions. Methods: A total of 109 at-risk gamblers were recruited online in a collaborative study between Switzerland and Korea. Over a period of 28 days, participants were asked to complete brief ecological momentary assessment surveys 3 times a day (morning, afternoon, and evening) asking about their gambling behavior and their levels of craving intensity, sleep quality, physical activity, boredom, vitality, depression, and anxiety. They were instructed to wear an Apple Watch that continuously and passively recorded several physiological indicators (eg, heart rate [variability], sleep metrics, and physical activity). Machine learning techniques and multilevel modeling approaches will be used to develop prediction models for gambling episodes and to identify relevant predictors. Results: Data collection has been completed since April 2025. In total, 109 participants have been enrolled in both countries (52 in Switzerland, 57 in Korea), and datasets are currently being prepared for analysis. The collected data are expected to enable the development of prediction models for gambling episodes. Conclusions: Incorporating the relevant predictors found in this study into digital intervention programs and providing just-in-time individually tailored intervention elements could improve program engagement and effectiveness. The approach used in this study is transferable to other digital interventions for addictive behaviors and holds promise to exploit their potential.
- 19 February 2026
- Published by JMIR Publications Inc. in JMIR Research Protocols
- Vol. 15 (1) , e82782
Open AccessIdentifying Predictors of Gambling Episodes and Craving Using Ecological Momentary Assessment and Smartwatch-Based Physiological Measures: Protocol for a Longitudinal Observational Study
Background: Like other addictive behaviors, problem gambling is often chronic and relapsing. While digital interventions offer low-threshold treatment and support, their effectiveness is often limited by small effect sizes, low adherence, and high dropout rates. Progress in digital technology has enabled the development of ecological momentary interventions (EMIs), which provide just-in-time support tailored to users’ needs. However, EMIs for addictive behaviors have hardly been developed so far. Objective: This study aimed to explore and identify relevant predictors of gambling episodes assessed by ecological momentary assessments and physiological smartwatch (Apple Watch) data, which in turn may be used for the further development of EMI-based interventions. Methods: A total of 109 at-risk gamblers were recruited online in a collaborative study between Switzerland and Korea. Over a period of 28 days, participants were asked to complete brief ecological momentary assessment surveys 3 times a day (morning, afternoon, and evening) asking about their gambling behavior and their levels of craving intensity, sleep quality, physical activity, boredom, vitality, depression, and anxiety. They were instructed to wear an Apple Watch that continuously and passively recorded several physiological indicators (eg, heart rate [variability], sleep metrics, and physical activity). Machine learning techniques and multilevel modeling approaches will be used to develop prediction models for gambling episodes and to identify relevant predictors. Results: Data collection has been completed since April 2025. In total, 109 participants have been enrolled in both countries (52 in Switzerland, 57 in Korea), and datasets are currently being prepared for analysis. The collected data are expected to enable the development of prediction models for gambling episodes. Conclusions: Incorporating the relevant predictors found in this study into digital intervention programs and providing just-in-time individually tailored intervention elements could improve program engagement and effectiveness. The approach used in this study is transferable to other digital interventions for addictive behaviors and holds promise to exploit their potential.
- Open Access
- 1 January 2026
- Published by XMLink in Journal of Korean Medical Science
- Vol. 41 (39)
Drug Use Patterns and Characteristics by Drug Type Among Drug Abusers: Findings From the Survey of Drug Abusers in Korea 2021
- 1 January 2026
- Published by XMLink in Journal of Korean Medical Science
- Vol. 41 (39)
Open AccessDrug Use Patterns and Characteristics by Drug Type Among Drug Abusers: Findings From the Survey of Drug Abusers in Korea 2021
- Open Access
- 26 December 2025
- Published by Akademiai Kiado Zrt. in Journal of Behavioral Addictions
- Vol. 14 (4) , 1619-1624
Unintended positive behavioral changes during self-report and wearable data collection on digital media use in young adults
Background and aims: The use of mobile devices is increasingly recognized as a valuable tool for interventions targeting mental health issues. This brief report aims to share insights into the unexpected positive changes observed during the data collection process for predicting addictive behaviors using mobile technology. Methods: We analyze self-reported baseline and 4-week follow-up data collected to develop an algorithm for predicting addictive digital media use. Two hundred and six adults in their 20 s completed a smartphone self-report survey on digital media use behaviors and emotion, while their step and heart rate data was passively collected via Fitbit. Results: Despite the absence of any interventional feedback beyond the data collection process, positive changes were observed, including time spent on digital media use ( p = 0.031), a reduction in high-risk digital media use score ( p < 0.001), and depressive symptom score ( p = 0.004). Conclusions: When collecting data on digital media use and related emotion via mobile devices, it is important to recognize that positive changes may occur in participants even without planed intervention. Failing to consider this phenomenon could lead to an overestimation of the effects of interventions provided through smart devices.
- 26 December 2025
- Published by Akademiai Kiado Zrt. in Journal of Behavioral Addictions
- Vol. 14 (4) , 1619-1624
Open AccessUnintended positive behavioral changes during self-report and wearable data collection on digital media use in young adults
Background and aims: The use of mobile devices is increasingly recognized as a valuable tool for interventions targeting mental health issues. This brief report aims to share insights into the unexpected positive changes observed during the data collection process for predicting addictive behaviors using mobile technology. Methods: We analyze self-reported baseline and 4-week follow-up data collected to develop an algorithm for predicting addictive digital media use. Two hundred and six adults in their 20 s completed a smartphone self-report survey on digital media use behaviors and emotion, while their step and heart rate data was passively collected via Fitbit. Results: Despite the absence of any interventional feedback beyond the data collection process, positive changes were observed, including time spent on digital media use ( p = 0.031), a reduction in high-risk digital media use score ( p < 0.001), and depressive symptom score ( p = 0.004). Conclusions: When collecting data on digital media use and related emotion via mobile devices, it is important to recognize that positive changes may occur in participants even without planed intervention. Failing to consider this phenomenon could lead to an overestimation of the effects of interventions provided through smart devices.
- Open Access
- 21 November 2025
- Published by JMIR Publications Inc. in JMIR Preprints
Week-Ahead Prediction of High-Risk Drinking Episodes Among Young Adults Using Wearable Biosignals and Psychological Vulnerabilities: Prospective Observational Machine Learning Study (Preprint)
BACKGROUND Although machine learning has increasingly been used to predict mental health symptoms and maladaptive behaviors, real-world prediction of addiction-related risk remains limited. Emotional and temperamental vulnerabilities are established correlates of alcohol-related problems, yet few studies have integrated these factors with wearable-derived biosignals in alcohol-risk prediction models. OBJECTIVE This study evaluated whether machine learning models could predict weekly high-risk drinking episodes among young adults with elevated alcohol-use risk by integrating wearable-derived health data with baseline emotional and personality vulnerability indicators. METHODS In this prospective observational study, adults in their 20s completed weekly self-report surveys and wore Fitbit devices for 4 weeks. Features from week t were used to predict the Alcohol Use Disorders Identification Test-Korean version (AUDIT-K)–based high-risk drinking label at week t+1. Positive labels were defined using AUDIT-K high-risk drinking cutoffs, with scores of ≥20 for men and ≥10 for women. Extreme gradient boosting (XGBoost) and random forest models were evaluated across self-report-only, wearable-only, and integrated feature sets using 5-fold participant-level grouped cross-validation. RESULTS A total of 206 participants contributed 620 week-level observations, of which 85 (13.7%) were labeled as positive high-risk drinking episodes. In participant-level grouped cross-validation, the integrated random forest model showed the most favorable sensitivity-oriented performance, with a mean accuracy of 0.617 (SD 0.078), recall/sensitivity of 0.653 (SD 0.144), area under the receiver operating characteristic curve (ROC AUC) of 0.681 (SD 0.079), and area under the precision-recall curve (PR AUC) of 0.255 (SD 0.090). The integrated XGBoost model achieved an accuracy of 0.670 (SD 0.089), recall/sensitivity of 0.399 (SD 0.174), ROC AUC of 0.651 (SD 0.089), and PR AUC of 0.228 (SD 0.074). Shapley additive explanations analyses indicated that both baseline vulnerability indicators and wearable-derived weekly summaries contributed to model predictions. CONCLUSIONS Integrating baseline emotional and personality vulnerability indicators with wearable-derived weekly health signals may provide useful information for week-ahead prediction of high-risk drinking episodes. These findings provide preliminary support for wearable-assisted alcohol-risk stratification, although the modest positive predictive performance indicates that external validation and more proximal within-person measures are needed before real-world early-warning or just-in-time adaptive intervention applications.
- 21 November 2025
- Published by JMIR Publications Inc. in JMIR Preprints
Open AccessWeek-Ahead Prediction of High-Risk Drinking Episodes Among Young Adults Using Wearable Biosignals and Psychological Vulnerabilities: Prospective Observational Machine Learning Study (Preprint)
BACKGROUND Although machine learning has increasingly been used to predict mental health symptoms and maladaptive behaviors, real-world prediction of addiction-related risk remains limited. Emotional and temperamental vulnerabilities are established correlates of alcohol-related problems, yet few studies have integrated these factors with wearable-derived biosignals in alcohol-risk prediction models. OBJECTIVE This study evaluated whether machine learning models could predict weekly high-risk drinking episodes among young adults with elevated alcohol-use risk by integrating wearable-derived health data with baseline emotional and personality vulnerability indicators. METHODS In this prospective observational study, adults in their 20s completed weekly self-report surveys and wore Fitbit devices for 4 weeks. Features from week t were used to predict the Alcohol Use Disorders Identification Test-Korean version (AUDIT-K)–based high-risk drinking label at week t+1. Positive labels were defined using AUDIT-K high-risk drinking cutoffs, with scores of ≥20 for men and ≥10 for women. Extreme gradient boosting (XGBoost) and random forest models were evaluated across self-report-only, wearable-only, and integrated feature sets using 5-fold participant-level grouped cross-validation. RESULTS A total of 206 participants contributed 620 week-level observations, of which 85 (13.7%) were labeled as positive high-risk drinking episodes. In participant-level grouped cross-validation, the integrated random forest model showed the most favorable sensitivity-oriented performance, with a mean accuracy of 0.617 (SD 0.078), recall/sensitivity of 0.653 (SD 0.144), area under the receiver operating characteristic curve (ROC AUC) of 0.681 (SD 0.079), and area under the precision-recall curve (PR AUC) of 0.255 (SD 0.090). The integrated XGBoost model achieved an accuracy of 0.670 (SD 0.089), recall/sensitivity of 0.399 (SD 0.174), ROC AUC of 0.651 (SD 0.089), and PR AUC of 0.228 (SD 0.074). Shapley additive explanations analyses indicated that both baseline vulnerability indicators and wearable-derived weekly summaries contributed to model predictions. CONCLUSIONS Integrating baseline emotional and personality vulnerability indicators with wearable-derived weekly health signals may provide useful information for week-ahead prediction of high-risk drinking episodes. These findings provide preliminary support for wearable-assisted alcohol-risk stratification, although the modest positive predictive performance indicates that external validation and more proximal within-person measures are needed before real-world early-warning or just-in-time adaptive intervention applications.
- 28 October 2024
- Published by Springer Nature in European Child & Adolescent Psychiatry
- Vol. 34 (5) , 1637-1647
Recurrence of internet gaming disorder in Korean adolescents: a 24-month follow-up study
Many adolescents use internet games, and some of them experience psychological and social difficulties due to excessive gaming. Although there have been many studies on the onset and associated factors of internet gaming disorder (IGD) , research focusing specifically on its relapse is scarce. Considering that addictive disorders often have a chronic course with frequent relapse, this study prospectively observed adolescents from the general population to explore IGD recurrence rate and associated factors of recurrence. In this prospective cohort study, from 1587 middle school gaming users, 1452 (91.5%) were followed up at 12 months and 24 months. The results showed that the cumulative recurrence rate of IGD was 16.1%, which was higher than the initial incidence rate of 3.6%; the adjusted incidence rate ratio was 2.43 with 95% CI 1.21–4.87 (P = 0.012). In addition, starting internet gaming before entering primary school was associated with a higher risk of relapse (P = 0.004). Limiting internet gaming for children before they enter primary school may reduce the risk of relapse of IGD that they may encounter during their adolescent years.
- 28 October 2024
- Published by Springer Nature in European Child & Adolescent Psychiatry
- Vol. 34 (5) , 1637-1647
Recurrence of internet gaming disorder in Korean adolescents: a 24-month follow-up study
Many adolescents use internet games, and some of them experience psychological and social difficulties due to excessive gaming. Although there have been many studies on the onset and associated factors of internet gaming disorder (IGD) , research focusing specifically on its relapse is scarce. Considering that addictive disorders often have a chronic course with frequent relapse, this study prospectively observed adolescents from the general population to explore IGD recurrence rate and associated factors of recurrence. In this prospective cohort study, from 1587 middle school gaming users, 1452 (91.5%) were followed up at 12 months and 24 months. The results showed that the cumulative recurrence rate of IGD was 16.1%, which was higher than the initial incidence rate of 3.6%; the adjusted incidence rate ratio was 2.43 with 95% CI 1.21–4.87 (P = 0.012). In addition, starting internet gaming before entering primary school was associated with a higher risk of relapse (P = 0.004). Limiting internet gaming for children before they enter primary school may reduce the risk of relapse of IGD that they may encounter during their adolescent years.