Amr Mohamed

Objectives: To derive sex-specific Relative Body Mass Index (rBMI) cut-offs for classifying adolescents into four adiposity categories defined by dual-energy X-ray absorptiometry (DXA)-based body fat percentage (BF%), quantify their stability, and evaluate their classification performance against current BMI%-based classification and previously published ROC-derived rBMI cut-offs. Materials and Methods: Data from 567 observations in adolescents aged 11–19 years were analyzed. Adiposity categories (normal, mildly elevated, moderately elevated, severely elevated) were defined by sex-specific BF% thresholds. Classification and Regression Tree models with inverse class-frequency weighting were fitted in two formulations: sex-stratified, and a pooled-threshold model with shared lower boundaries and sex-specific upper boundaries. Threshold stability was assessed by bootstrap resampling. Performance was estimated by repeated nested cross-validation, with the full derivation pipeline repeated within each training fold; the previously published ROC-derived thresholds were re-derived within each fold so that both approaches carried the same correction for optimism. Results: Sex-stratified thresholds were 103, 118, and 162 in males and 100, 118, and 176 in females; three of six agreed with the previously published ROC-derived values to within 0.5 units and a fourth to within 3.2 units. Five of the six published values fell within the corresponding bootstrap intervals, the exception being the female moderately/severely elevated boundary (160.0; 95% CI: 161.4–218.7). The male boundary separating normal from mildly elevated adiposity was identified in 76.0% of bootstrap resamples; estimating the lower boundaries from the pooled sample raised this to 99.9% and improved ordinal agreement among males (quadratic-weighted kappa 0.812 vs. 0.759). Out-of-fold accuracy was 0.697, 0.694, and 0.694 for the sex-stratified, pooled-threshold, and ROC-derived approaches, respectively, with all paired differences including zero in the overall sample. All three exceeded BMI% classification (accuracy 0.608), which identified only 15.5% of adolescents with mildly elevated adiposity and produced more than twice the proportion of errors spanning two or more categories (6.0% vs. 1.8–2.1%). Conclusions: rBMI cut-offs aligned more closely with DXA-defined adiposity than BMI% classification, particularly in the intermediate categories. Tree-derived and ROC-derived thresholds converged and performed equivalently, indicating that threshold location does not depend on the derivation procedure. A pooled-threshold formulation provided a simpler and more stable rule. These are derivation-stage estimates requiring external validation before clinical adoption.

This study examined associations of perceived restedness, sleep duration, and emotion regulation difficulties with mental health outcomes among university students. Data from 1192 Canadian college students were analyzed using multiple regression models. Mental health outcomes included distress (DASS), perceived stress (PSS), anxiety (GAD), and depression (PHQ), alongside indicators of emotion regulation, physical activity, mindfulness, and social support. Hierarchical models assessed the incremental value of these predictors. Across models, perceived restedness was consistently associated with better mental health after accounting for sleep duration and other covariates, whereas sleep duration showed less consistent associations across outcomes. Difficulties in emotion regulation were the strongest and most consistent correlates across outcomes, while physical activity and mindfulness showed inconsistent effects once other variables were considered. Adding health behavior variables improved model performance, whereas sociodemographic factors contributed little additional explanatory value. Interaction effects were generally small and inconsistent, although students who felt somewhat rested appeared to benefit more from stronger social support. These findings suggest perceived restedness may serve as a practical and efficient indicator of mental health screening, but results should be interpreted as hypothesis-generating given measurement asymmetry, self-report design, and pandemic-era data.

As data complexity and volume increase rapidly, efficient statistical methods for identifying significant variables become crucial. Variable selection plays a vital role in establishing relationships between predictors and response variables. The challenge lies in achieving this goal while controlling the False Discovery Rate (FDR) and maintaining statistical power. The knockoff filter, a recent approach, generates inexpensive knockoff variables that mimic the correlation structure of the original variables, serving as negative controls for inference. In this study, we extend the use of knockoffs to Light Gradient Boosting Machine (LightGBM), a fast and accurate machine learning technique. Shapely Additive Explanations (SHAP) values are employed to interpret the black-box nature of machine learning. Through extensive experimentation, our proposed method outperforms traditional approaches, accurately identifying important variables for each class. It offers improved speed and efficiency across multiple datasets. To validate our approach, an extensive simulation study is conducted. The integration of knockoffs into LightGBM enhances performance and interpretability, contributing to the advancement of variable selection methods. Our research addresses the challenges of variable selection in the era of big data, providing a valuable tool for identifying relevant variables in statistical modeling and machine learning applications.