Personal Interaction in the Vegetative State

Abstract

Background and purpose: Brain processing at varying levels of functional complexity and emotional reactions to relatives are anecdotally reported by the caregivers of patients in a vegetative state. In this study, computer-assisted machine-learning procedures were applied to identify heart rate variability changes or galvanic skin responses to a relative’s presence. Methods: The skin conductance (galvanic skin response) and heart beats were continuously recorded in 12 patients in a vegetative state, at rest (baseline) and while approached by a relative (usually the mother; test condition) or by a nonfamiliar person (control condition). The cardiotachogram (the series of consecutive intervals between heart beats) was analyzed in the time and frequency domains by computing the parametric and nonparametric frequency spectra. A machine-learning algorithm was applied to sort out the significant spectral parameter(s). For all patients, each condition (baseline, test, control) was characterized by the values of its spectral parameters, and the association between spectral parameters values and experimental condition was tested (WEKA machine-learning software). Results and comments: A galvanic skin response was obtained in two patients. The machine-learning procedure independently selected the nu_LF spectral parameter and attributed each nu_LF measure to any of the three experimental conditions. 69.4% of attributions were correct (baseline: 58%; test condition: 75%; control. 75%). In seven patients, attribution changed when the subject was approached by the test person; specifically, sequential shifts from baseline to test condition (“the Mom effect”) to control condition were identified in four patients (30.0%); the change from test to control was attributed correctly in seven patients (58%). The observation of heart rate changes tentatively attributable to emotional reaction in a vegetative state suggest residual rudimentary personal interaction, consistent with functioning limbic and paralimbic systems after massive brain damage. Machine-learning proved applicable to sort significant measure(s) out of large samples and to control for statistical alpha inflation.