Machine learning-based discrimination of panic disorder from other anxiety disorders

Abstract

Panic disorder is a highly prevalent psychiatric disorder that substantially impairs quality of life and psychosocial function. Panic disorder arises from neurobiological substrates and developmental factors that distinguish it from other anxiety disorders. Differential diagnosis between panic disorder and other anxiety disorders has only been conducted in terms of a phenomenological spectrum. Through a machine learning-based approach with heart rate variability (HRV) as input, we aimed to build algorithms that can differentiate panic disorder from other anxiety disorders. Five algorithms were used: random forest (RF), gradient boosting machine (GBM), support vector machine (SVM), artificial neural network (ANN), and regularized logistic regression (LR). 10-fold cross-validation with five repeats was used to build the final models. A total of 60 patients with panic disorder and 61 patients with other anxiety disorders (aged between 20 and 65 years) were recruited. The L1-regularized LR showed the best accuracy (0.784), followed by ANN (0.730), SVM (0.730), GBM (0.676), and finally RF (0.649). LR also had good performance in other measures, such as F1-score (0.790), specificity (0.737), sensitivity (0.833), and Matthews correlation coefficient (0.572). Cross-sectional design and limited sample size is limitations. This study demonstrated that HRV can be used to differentiate panic disorder from other anxiety disorders. Future studies with larger sample sizes and longitudinal design are required to replicate the diagnostic utility of HRV in a machine learning approach.