A Pipeline for Processing and Modeling Electrodermal Activity Data Collected in an Ambulatory Setting

Abstract

Background: Ambulatory assessment of electrodermal activity (EDA) is an emerging technique for capturing individuals' autonomic responses to real-life events. The data analytic pipeline for processing and modeling EDA data from laboratory assessments using a hydrating medium for electrodes is well described. Little work however has been devoted to developing an analytic pipeline for ambulatory assessment of EDA using dry electrodes. Decision points in the initial stages of the pipeline, such as the removal of motion artifacts, have important implications for assumptions further down the pipeline. Purpose: The primary goal of this presentation is to describe several methods for preparing and modeling ambulatory EDA data and to propose a pipeline that researchers may use as a guide when analyzing ambulatory EDA data. We use data from an ongoing study examining the effects of stressful tasks on adolescent mothers' EDA reactivity. Methods: In conjunction with a clinician-delivered mindfulness-based parenting intervention, 42 single adolescent mothers of young children used a biosensor band in their everyday life that continuously measured EDA and wirelessly connected to a smartphone app which delivered coping and parenting strategies previously learned in an in-person intervention. EDA was recorded 4 times per second. The biosensor band was also worn continuously during baseline and 3-month post-intervention assessments, each of which lasted approximately 2 hrs. Both assessments included two potentially stressful tasks: (1) a timed Stroop task and (2) a 10-min. mother-child videotaped interaction task where mothers were asked to teach their child a task slightly above the child’s developmental level. Results: We describe a variety of decision points as well as the advantages and disadvantages of each approach. Initial decisions included separating total EDA into phasic and tonic EDA, filtering noise such as movement artifacts or EDA values which may indicate poor conductance between the dry electrode and the skin. We followed these tasks by graphing the data, which showed that EDA varied both between-individuals and within-individuals during the Stroop and video tasks. We computed various other features of the EDA data using the EDA-explorer website developed by the MIT Media Lab as well as other algorithms developed for these data, including the number of peaks and their amplitude as well as EDA reactivity, quantified as the rate at which adolescent mothers returned to baseline EDA following an EDA peak. Conclusions: Although the pattern of EDA varied substantially across individuals, various features of EDA may be computed for all individuals enabling within- and between individual analysis and comparisons. EDA-explorer and the additional algorithms developed can be used to establish decision points for the preparation and analysis of dry-electrode, ambulatory EDA that can be used by other researchers.