ADHD

ADHD predicts the risk of developing attention-deficit/hyperactivity disorder in young adulthood using childhood demographic, clinical, and psychosocial predictors modeled with logistic regression.

Key Features:

• Logistic Regression Risk Model: Uses logistic regression models trained on data from the Avon Longitudinal Study of Parents and Children (ALSPAC) cohort comprising 5,113 participants followed from birth to age 17.
• Multivariable Predictor Integration: Incorporates predictors including sex, socioeconomic status, single-parent family structure, ADHD symptoms, comorbid disruptive disorders, childhood maltreatment, depressive symptoms, maternal depression, and intelligence quotient.
• External Cohort Validation: Validates model performance using independent datasets including the E-Risk cohort (UK), the 1993 Pelotas Birth Cohort (Brazil), and the Multimodal Treatment Study of Children with ADHD (MTA) clinical sample (USA).
• Machine Learning Benchmarking: Evaluates alternative models including Random Forest, Stochastic Gradient Boosting, and Artificial Neural Networks for comparison with logistic regression performance.
• Predictive Performance Metrics: Achieves predictive discrimination with reported AUC values of 0.82 for internal validation and approximately 0.75–0.76 in external UK and MTA cohorts.

Scientific Applications:

• ADHD Risk Prediction: Estimates the probability of developing attention-deficit/hyperactivity disorder in young adulthood based on childhood characteristics.
• Longitudinal Mental Health Research: Supports studies investigating developmental trajectories and early-life predictors of ADHD.
• Population Cohort Analysis: Enables comparative risk modeling across longitudinal cohorts from different geographic populations.

Methodology:

ADHD trains logistic regression models on longitudinal cohort data using demographic, clinical, and psychosocial predictors, evaluates alternative machine learning models including Random Forest, Stochastic Gradient Boosting, and Artificial Neural Networks, and validates predictive performance across multiple independent cohorts.