此页面是自动翻译的,不保证翻译的准确性。请参阅 英文版 对于源文本。

Machine Learning Prediction of ADHD in Adolescents: Gaming, Sleep, and Sport

2026年11月15日 更新者:Yalin Aygun

Predicting Adolescent ADHD: Machine Learning Model of Internet Gaming Disorder, Sleep Quality, and Sport Participation

ADHD is one of the most common mental health conditions in teenagers. It can affect school performance, friendships, and overall well-being. Some everyday habits and lifestyle factors-how much a teenager plays internet games, how well they sleep, and whether they take part in sports-have each been linked to ADHD symptoms. However, these three factors have almost always been studied one at a time, and no study has yet tested whether looking at them together can help predict which teenagers have more severe ADHD symptoms.

This study aims to find out whether internet gaming, sleep quality, and sport participation, when considered together, can predict ADHD symptom severity in adolescents aged 12-17. We will invite 750 teenagers from secondary schools in Malatya, Türkiye, to complete a set of short, anonymous questionnaires about their attention and hyperactivity symptoms, gaming habits, sleep, and sport participation. We will then use machine learning-a type of computer analysis that can detect complex patterns and interactions between many factors at once-to build and test a prediction model.

The main question this study seeks to answer is: Do internet gaming, sleep quality, and sport participation jointly carry information that helps predict ADHD symptoms in teenagers, beyond what is already known from basic characteristics such as age, sex, and socioeconomic status? We are not testing a treatment, and the study does not involve any medication, biological samples, or changes to participants' usual routines. The findings may help schools, families, and health professionals identify teenagers who could benefit from early support and from healthier lifestyle habits. This is an exploratory study; it cannot show cause and effect, and any prediction model would need to be tested in other groups before being used in practice.

研究概览

地位

尚未招聘

详细说明

Background and Rationale Attention-deficit/hyperactivity disorder (ADHD) is a common neurodevelopmental condition affecting approximately 5.6% of adolescents worldwide, with substantial downstream consequences for academic achievement, social functioning, and psychological well-being. Although ADHD is multifactorial in etiology, a growing body of evidence indicates that modifiable lifestyle factors-particularly internet gaming disorder (IGD), sleep quality, and sport participation-are individually associated with ADHD symptom severity in adolescence. IGD co-occurs with ADHD at high rates and is hypothesized to share transdiagnostic mechanisms including impulsivity, emotion dysregulation, and maladaptive coping. Poor sleep quality and circadian delay are highly prevalent in adolescents with ADHD and appear bidirectionally linked to symptom severity. Conversely, structured sport participation has been proposed as a potentially protective lifestyle factor, although its effects appear to vary by sport type, dose, and measurement approach.

Critically, these three lifestyle factors have almost exclusively been examined in isolation, and conventional regression models impose linear and additive assumptions that may not reflect the interactive, threshold-dependent, and non-linear relationships likely to characterize them. Machine learning (ML) offers a complementary analytic framework that accommodates high-dimensional interactions, ranks feature importance without a priori constraints on functional form, and can identify subgroups at elevated risk. No study to date has applied an ML framework to test whether IGD, sleep quality, and sport participation jointly carry incremental predictive signal for ADHD symptom severity in adolescents. The present study addresses this gap.

Study Design and Setting This is an observational, cross-sectional, prediction-modeling study conducted among adolescents aged 12-17 years attending secondary schools in Malatya, Türkiye. Data will be collected during the [academic term] using standardized, self-administered instruments under the supervision of trained research assistants. The study is designed to develop and internally validate a multivariable ML model predicting ADHD symptom severity from IGD severity, sleep quality, and multidimensional sport participation, while adjusting for pre-specified sociodemographic, clinical, and lifestyle covariates. The study is exploratory in nature and is not designed to establish causal relationships. The protocol has been approved by the İnönü University Social Sciences Research and Ethics Committee [approval number, date], and written informed consent will be obtained from all participants and their legal guardians prior to data collection. The study will be conducted in accordance with the Declaration of Helsinki and applicable national regulations.

Recruitment and Data Collection Procedures Participants will be recruited through a two-stage sampling strategy. First, secondary schools in the target region will be stratified by school type (public vs. private) and randomly selected. Second, classrooms within participating schools will be randomly selected, and all eligible students will be invited to participate. Data collection will take place in a single session per classroom under standardized conditions, with research assistants available to answer clarification questions without influencing responses. Participants will complete a paper-and-pencil or tablet-based battery comprising the Conners-Wells Adolescent Self-Report Scale (CASS), the Internet Gaming Disorder Scale-Short Form (IGDS9-SF), the Pittsburgh Sleep Quality Index (PSQI), a multidimensional sport participation questionnaire, and a structured sociodemographic and lifestyle form. The full battery is expected to require approximately 30-40 minutes. To minimize social desirability and common-method bias, the order of instruments will be randomized across participants, and anonymity will be emphasized. No biological samples will be collected.

Measures and Instrumentation Primary outcome. ADHD symptom severity will be assessed using the ADHD-specific subscales (cognitive problems/inattention and hyperactivity) of the Turkish-validated CASS. Raw scores will be converted to age- and sex-specific T-scores; both continuous T-scores and a binary indicator (T ≥ 65) will be retained for modeling purposes.

Primary predictors. IGD severity will be measured with the nine-item IGDS9-SF, which maps directly onto DSM-5 criteria and yields a unidimensional total score. Sleep quality will be assessed using the PSQI global score and its seven component scores. Sport participation will be operationalized multidimensionally, including participation status, sport type (team, individual, or both), weekly training hours, years of participation, and competition level.

Covariates. Pre-specified covariates include age, sex, school grade, perceived socioeconomic status, maternal and paternal education, self-reported academic achievement, current ADHD medication use, presence of a chronic health condition limiting sport participation, weekday and weekend recreational screen time, and non-sport physical activity level. All covariates will be entered simultaneously with primary predictors to allow the models to capture interactive and non-linear relationships; no covariate will be dropped on the basis of univariable screening because such filtering can induce bias in predictive modeling.

Machine Learning Analysis Framework Preprocessing. Continuous variables will be standardized (z-scores); categorical variables will be one-hot encoded. Class imbalance for the binary outcome will be addressed during training using class weighting or stratified resampling, with resampling applied only within training folds to prevent information leakage. All preprocessing steps will be embedded within the cross-validation pipeline.

Model development. Several complementary algorithms will be trained to balance interpretability and predictive flexibility, including regularized logistic regression (elastic net), random forest, gradient boosting (XGBoost/LightGBM), and support vector machines. A multilayer perceptron will be considered as a sensitivity analysis. Hyperparameters will be tuned using nested cross-validation with Bayesian optimization or grid search.

Validation strategy. Internal validation will be performed using repeated stratified k-fold cross-validation (k = 5 or 10, with at least 10 repeats). Model performance will be reported as discrimination (area under the receiver operating characteristic curve, area under the precision-recall curve), calibration (calibration slope, intercept, Brier score), and, where relevant, decision curve analysis. For continuous outcomes, root mean squared error, mean absolute error, and R² will be reported. Optimism-corrected performance estimates will be obtained via bootstrap.

Interpretability. Feature importance will be quantified using permutation importance and SHAP (Shapley additive explanations) values. Partial dependence and accumulated local effects plots will be used to characterize non-linear and threshold relationships, with particular attention to potential interactions among IGD, sleep quality, and sport participation. Subgroup analyses will explore whether predictive performance or feature importance differs by sex, age group, and ADHD medication status.

Incremental predictive value. To test whether the three lifestyle factors carry incremental signal beyond covariates, a series of nested models will be compared: (1) covariates only; (2) covariates + IGD; (3) covariates + IGD + sleep quality; and (4) covariates + IGD + sleep quality + sport participation. Differences in discrimination and calibration will be evaluated using DeLong's test for correlated ROC curves, net reclassification improvement, and integrated discrimination improvement.

Sample Size Considerations and Power Analysis The target sample size is 750 adolescents. This target was determined a priori through a combination of conventional power analysis and contemporary guidance for prediction-model development. For the conventional power analysis (G*Power 3.1, linear multiple regression, fixed model, R² increase), assuming a moderate effect size (f² = 0.15), α = 0.05 (two-sided), 80% power, and 10 candidate predictors, the required minimum sample was approximately 118 participants. This confirms that the study is adequately powered for the primary association analyses.

More importantly, for prediction modeling, recent methodological guidance recommends minimum sample sizes of 237 to 694 participants for continuous outcomes to ensure low optimism and precise estimation of model performance. For binary outcomes, rules based on events per candidate predictor and balanced accuracy criteria suggest a need for approximately 500-790 participants. A sample of 750 therefore satisfies both continuous- and binary-outcome modeling requirements, supports repeated stratified k-fold cross-validation with stable performance estimates, and provides sufficient margin for an anticipated 10% data loss as well as for pre-specified subgroup analyses by sex, age group, and ADHD medication status.

Missing Data Management The extent and pattern of missing data will be characterized. If missingness is minimal (< 5%), complete-case analysis will be the primary approach, with multiple imputation by chained equations as a sensitivity analysis. If missingness exceeds 5%, multiple imputation will be the primary approach (m = 20 imputations), and results will be pooled according to Rubin's rules. Missing data will be handled within the cross-validation pipeline to avoid optimism.

Quality Assurance and Data Management Data will be entered into a secure, encrypted REDCap database with predefined range, consistency, and logic checks. A random 10% sample of records will be double-entered and compared to source documents. Data queries will be resolved by the study coordinator in collaboration with the principal investigator. All analytic code will be version-controlled and archived, and a statistical analysis plan will be finalized before database lock. The study will be reported in accordance with the TRIPOD+AI guideline for prediction models.

Reporting and Dissemination Results will be disseminated through peer-reviewed publication and conference presentation, with an emphasis on transparent reporting of model development, validation, and limitations. The final model will not be deployed clinically without external validation in an independent cohort.

研究类型

观察性的

注册 (估计的)

750

联系人和位置

本节提供了进行研究的人员的详细联系信息,以及有关进行该研究的地点的信息。

学习联系方式

研究联系人备份

学习地点

    • Battalgazi
      • Malatya、Battalgazi、土耳其(türkiye)、+90
        • Selahaddin Eyyubi AİHL (Science and Social Sciences Project School)
        • 接触:

参与标准

研究人员寻找符合特定描述的人,称为资格标准。这些标准的一些例子是一个人的一般健康状况或先前的治疗。

资格标准

适合学习的年龄

12年 至 17年 (孩子)

接受健康志愿者

是的

有资格学习的性别

全部

取样方法

概率样本

研究人群

The study population consists of adolescents aged 12-17 years who are currently enrolled in secondary schools in Malatya, Türkiye. Participants are recruited through a two-stage sampling strategy: schools are first stratified by type (public vs. private) and randomly selected, and classrooms within participating schools are then randomly selected, with all eligible students invited to take part. The sample is school-based and community-derived rather than clinically referred, and includes adolescents across the full range of ADHD symptom severity, from no or minimal symptoms to clinically significant levels. No clinical ADHD diagnosis is required for participation. The planned enrollment is 750 adolescents, which provides sufficient power for the planned prediction models and allows for an anticipated 10% data loss as well as pre-specified subgroup analyses by sex, age group, and ADHD medication status. Data are collected at a single time point under standardized conditions.

描述

Inclusion Criteria:

Adolescents aged 12-17 years at the time of data collection. Currently enrolled in a secondary school in Malatya, Türkiye. Written informed consent obtained (participant assent plus parental/legal guardian consent for minors).

Able to read and understand Turkish and to complete self-administered questionnaires.

No clinical ADHD diagnosis required; adolescents with and without ADHD symptoms are eligible.

Exclusion Criteria:

Age outside 12-17 years. Not currently attending a secondary school. Unable or unwilling to provide consent/assent, or no parental/legal guardian consent.

Visual, hearing, cognitive, or language impairment preventing independent completion of self-report questionnaires.

Any condition or circumstance that, in the investigator's judgment, would compromise participant safety or data integrity.

学习计划

本节提供研究计划的详细信息,包括研究的设计方式和研究的衡量标准。

研究是如何设计的?

设计细节

  • 观测模型:COHORT
  • 时间观点:CROSS_SECTIONAL

队列和干预

团体/队列
Adolescents Aged 12-17 Years

Single observational cohort of adolescents aged 12-17 years attending secondary schools in Malatya, Türkiye. All participants complete the same self-administered battery at one time point: the Conners-Wells Adolescent Self-Report Scale (CASS) for ADHD symptoms, the Internet Gaming Disorder Scale-Short Form (IGDS9-SF), the Pittsburgh Sleep Quality Index (PSQI), a multidimensional sport participation questionnaire, and a demographic/lifestyle form. No intervention, medication, biological sample, or change to usual routine is involved. Sport participation, ADHD medication use, and other characteristics are treated as covariates or subgroup variables in the machine learning prediction models, not as separate groups.

研究衡量的是什么?

主要结果指标

结果测量
措施说明
大体时间
ADHD Symptom Severity: Conners-Wells Adolescent Self-Report Scale (CASS) Inattention and Hyperactivity Subscales
大体时间:Single time point (baseline assessment). All participants complete the CASS once during the data collection session; no follow-up assessments are planned.

ADHD symptom severity will be assessed using the ADHD-specific subscales (cognitive problems/inattention and hyperactivity) of the Turkish-validated Conners-Wells Adolescent Self-Report Scale (CASS) for adolescents aged 12-17 years. The scale is self-administered under standardized instructions. Raw scores are converted to age- and sex-specific T-scores, with T ≥ 65 indicating clinically significant symptom burden. Both continuous T-scores (primary) and a binary indicator (T ≥ 65 vs. < 65) will be retained for machine learning modeling. The CASS demonstrates acceptable to good internal consistency (Cronbach's α = 0.76-0.87) and established criterion validity against the Conners Parent and Teacher Rating Scales-Revised in Turkish adolescents.

Single time point (baseline assessment). All participants complete the CASS once during the data collection session; no follow-up assessments are planned.

合作者和调查者

在这里您可以找到参与这项研究的人员和组织。

赞助

合作者

调查人员

  • 首席研究员:Yalin Aygun, Prof. Dr.、Inonu University

出版物和有用的链接

负责输入研究信息的人员自愿提供这些出版物。这些可能与研究有关。

一般刊物

研究记录日期

这些日期跟踪向 ClinicalTrials.gov 提交研究记录和摘要结果的进度。研究记录和报告的结果由国家医学图书馆 (NLM) 审查,以确保它们在发布到公共网站之前符合特定的质量控制标准。

研究主要日期

学习开始 (估计的)

2027年1月1日

初级完成 (估计的)

2027年1月1日

研究完成 (估计的)

2027年2月1日

研究注册日期

首次提交

2026年11月15日

首先提交符合 QC 标准的

2026年11月15日

首次发布 (实际的)

2026年11月19日

研究记录更新

最后更新发布 (实际的)

2026年11月19日

上次提交的符合 QC 标准的更新

2026年11月15日

最后验证

2026年11月1日

更多信息

与本研究相关的术语

计划个人参与者数据 (IPD)

计划共享个人参与者数据 (IPD)?

不

IPD 计划说明

De-identified individual participant data will not be shared. This is an investigator-initiated, school-based observational study without dedicated funding or data-sharing infrastructure. The dataset contains self-reported information from minors, and the informed consent/assent procedures did not include permission for broader data sharing beyond the study team. Aggregate results will be reported in peer-reviewed publications.

药物和器械信息、研究文件

研究美国 FDA 监管的药品

不

研究美国 FDA 监管的设备产品

不

此信息直接从 clinicaltrials.gov 网站检索,没有任何更改。如果您有任何更改、删除或更新研究详细信息的请求,请联系 register@clinicaltrials.gov. clinicaltrials.gov 上实施更改,我们的网站上也会自动更新.

订阅