Does heart rate variability change over acute episodes of bipolar disorder? A Bayesian analysis

Bipolar disorder (BD) is a severe psychiatric condition featuring autonomic nervous system dysfunctions, detectable with abnormal heart rate variability (HRV). This is a promising biomarker, but its dynamics over an acute episode of mania or depression, the two polarities of BD, are poorly understood. Studies of intra- individual HRV changes in BD cannot afford to recruit more than only few dozen patients, as collecting this kind of data is very resource-intensive. This makes the ground treacherous for frequentist statistical analyses. Unlike previous studies, we therefore took a Bayesian approach, as more suitable for quantifying uncertainty in small samples. Specifically, we developed an interpretable probabilistic model of HRV changes over the course of an acute BD episode, using the natural logarithm of the Root Mean Square of Successive RR interval Differences (lnRMSSD) as HRV measure. Patients with BD were recruited at the onset of an acute episode, either mania or depression, and had three-to-four follow-up assessments up to euthymia. Conversely, previous studies used just two assessments, which prevents adequately modelling change. During each assessment, lnRMSSD sleep values were recorded with a wearable device and symptoms’ severity was assessed with clinician-administered questionnaires. A model allowing the lnRMSSD rate of change with respect to symptoms’ improvement to vary across polarities was no better fit than a model disregarding polarities. Strong evidence emerged for a positive lnRMSSD change over the episode’s resolution (probability of positive direction = 95.175%), but sample size limited ascertaining the magnitude of this effect (95% Highest Density Interval of [-0.0366-0.4706] with a Region of Practical Equivalence of [-0.05, 0.05]). In conclusion, improvements in BD symptoms are associated with an increase in lnRMSSD but these changes do not appear to be influenced by the episode polarity.