Identification of Adolescents With Major Depressive Disorder Using Random Forest Based on Nocturnal Heart Rate Variability.

Abstract

Major depressive disorder (MDD) in adolescents is often underdiagnosed, with the current diagnosis predominantly relying on subjective assessment. Sleep disturbance and reduced heart rate variability (HRV) have been typically observed in adolescents with MDD. This study aimed to develop an automatic classification model based on nocturnal HRV features to identify adolescent MDD. Sixty-three subjects, including depressed adolescents and healthy controls, participated in the study and completed a three-night sleep electrocardiogram (ECG) monitoring, yielding 160 overnight RR interval time series and 7520 5-min short-term segments for analysis. Nineteen HRV features were extracted from the time domain, frequency domain, and nonlinear dynamics. The Bayesian-optimized random forest (BO-RF) algorithm was applied as the classifier, with performance evaluated using ten-fold cross-validation. The impact of data accumulation on the reliability of identification using short-term data and the importance of features were also examined. The BO-RF classifier based on long-term features achieved a noteworthy predictive accuracy of 80.6%, and the performance of the classifier using short-term data showed a significant improvement when more segment outcomes from the same night were included, ultimately achieving an accuracy of 75.0%. The Poincaré plot-derived features, especially heart rate asymmetry (HRA) features such as C1_d, significantly contributed to distinguishing depressed adolescents from healthy subjects. Nocturnal HRV features can effectively differentiate adolescents with MDD from healthy controls. This study provides a promising diagnostic approach for adolescent MDD, with the potential to be integrated into wearable devices for broader application.