Abstract
Puberty is linked to mental health problems during adolescence, and in particular, the timing of puberty is thought to be an important risk factor. This study developed a new measure of pubertal timing that was built upon multiple pubertal features and their nonlinear changes over time (i.e., with age), and investigated its association with mental health problems. Using the Adolescent Brain Cognitive Development (ABCD) cohort (N ~ 9900, aged 9-13 years), we employed three different models to assess pubertal timing. These models aimed to predict chronological age based on: (i) observed physical development, (ii) hormone levels (testosterone and dehydroepiandrosterone [DHEA]), and (iii) a combination of both physical development and hormones. To achieve this, we utilized a supervised machine learning approach, which allowed us to train the models using the available data and make age predictions based on the input pubertal features. The accuracy of these three models was evaluated, and their associations with mental health problems were examined. The new pubertal timing model performed better at capturing age variance compared to the more commonly used linear regression method. Further, the model based on physical features accounted for the most variance in mental health, such that earlier pubertal timing was associated with higher symptoms. This study demonstrates the utility of our new model of pubertal timing and suggests that, relative to hormonal measures, physical measures of pubertal maturation have a stronger association with mental health problems in early adolescence.