Correlations between patterns of activity and the response to treatment yield distinct signatures for different antidepressive treatments

Abstract

Wrist actigraphy is a non-invasive technology to monitor activity over extended periods of time. Specific alterations in circadian patterns of activity have been described in several psychiatric disorders and are associated with symptom severity in major depressive disorder (MDD). The aim of this study was to investigate the correlations between activity patterns and the response to different MDD treatments. We used actigraphy recordings acquired for 3 independent clinical trials investigating the response to internet-delivered cognitive behavioral therapy (iCBT), escitalopram, or ketamine. We designed a feature extraction procedure applicable both to recordings spanning a single circadian cycle, and to average circadian profiles. First, we investigated the effects of MDD treatments independently from magnitude of response, and found that each treatment altered a specific subset of features. We next asked whether features of activity patterns before treatment correlate with the magnitude of response to MDD treatment, and whether the correlations are treatment-specific. We found distinct signatures in the correlations between longitudinal changes in individual features and the response to different treatments. Next, we trained linear regression ensembles to model the response to treatment using the patterns of activity during the depressive episode and ran in-depth analysis of coefficients to characterize the differences between treatments. We found that better response is associated with higher between-days variability for iCBT, and with lower between-days variability for ketamine, while lower within-day fragmentation was common to both treatments. Our data suggests that the analysis of circadian patterns of activity can potentially be used for predicting the response to different MDD treatments, however larger confirmatory studies are required for future clinical applications.

Individual patterns of activity predict the response to physical exercise as an intervention in mild to moderate depression

“…Weaker circadian entrainment, in general, predicts better response to low intensity PE, but worse response to moderate and high intensity PE. We have applied a similar data analysis pipeline for the modelling the response to iCBT or ketamine [15]. For iCBT, weaker circadian entrainment correlated with better response, similar to low intensity PE.…”

“…For iCBT, weaker circadian entrainment correlated with better response, similar to low intensity PE. In contrast to iCBT and PE, the most relevant correlates of the response to ketamine were descriptors of circadian profile and ultradian patterns of activity [15]. This further emphasizes the differences in feature subsets correlating with the response to different antidepressive therapies.…”

“…The raw data was exported using the proprietary software and processed in Matlab® (version 9.x; The MathWorks Inc, Natick, MD, USA) using custom scripts (see also [15,16]). The quality control was performed by the same observer, blind to group belonging.…”

“…The MathWorks Inc, Natick, MD, USA) using custom scripts (see also [15,16]). The quality control was performed by the same observer, blind to group belonging.…”

“…Therefore, we believe that our data analysis pipeline provides a sound basis for further applications based on activity monitoring as objective measurement. While we provide proof-of-concept for prediction of response to treatment (see also [15,16]) training ML models to support personalized treatment for depression requires standardized actigraphy recording procedure and antidepressive interventions.…”