Can circadian rhythms of heart rate variability identify major depressive disorder? — A study based on support vector machine analysis

Abstract

Background

Major depressive disorder (MDD) is a prevalent and severe psychiatric condition for which objective diagnostic tools are lacking. Heart rate variability (HRV), an index of autonomic nervous system (ANS) function, has shown potential for distinguishing patients with MDD. This study aimed to improve classification performance by leveraging circadian rhythm features derived from multiple HRV indices using a support vector machine (SVM) approach.

Methods

Twenty-four-hour electrocardiographic recordings were collected from 116 patients with MDD and 63 healthy controls (HCs). Thirteen HRV indices, spanning the time domain, frequency domain, and nonlinear measures, were extracted, along with five corresponding circadian rhythm features (CRFs) for each index. To reduce feature dimensionality, a recursive feature elimination strategy based on the SVM was applied. The selected CRFs were then used to train the SVM classifier. Separate SVM models were constructed using CRFs derived from linear indices, nonlinear indices, and their combinations. Furthermore, their performance was compared across multiple metrics.

Results

The SVM model constructed using CRFs from all HRV indices achieved an accuracy of 97.80 %, sensitivity of 98.01 %, and specificity of 97.60 %. The models based on the CRFs from linear HRV indices outperformed those based on nonlinear indices, whereas the combination of all HRV indices yielded the best overall performance.

Conclusions

The results highlight the potential utility of HRV CRFs for objective MDD classification. Moreover, with an equal number of features, the CRFs derived from linear HRV indices proved to be more effective than those derived from nonlinear indices in discriminating patients with MDD.