Speech markers of depression dimensions across the spectrum of cognitive functions

Abstract

AbstractBackgroundDepression is a common among patients with cognitive impairment and encompass different dimensions, which may vary in their association with cognitive outcomes and biological substrates. Previous studies have shown a shift toward the use of more negative words in written documents and social media posts of depressed patients with intact cognitive function. To objectively characterize depression dimensions across the spectrum of cognitive functions, we explored their association with the emotion‐based measures of speech in participants of memory clinic.MethodsThis study capitalized on semi‐structured digitized speech data collected at the hospital of University of Pennsylvania. Subjects (n = 302, 40% female, mean age = 67.13±9.59 years, education = 15.82±2.84 years, Mini‐Mental Status Exam (MMSE) = 24.36±5.04) were native English speakers with cognition ranging from normal cognition to mild dementia. Participants were instructed to describe the Cookie Theft scene using complete sentences and their descriptions were digitally recorded. Emotion‐based measures of the content words, including valence (i.e., the pleasantness), arousal (the intensity of emotions), and dominance (the word power), were rated based on published norms. Each emotional measure was averaged per speaker. Depression dimensions were defined based on itemized Geriatric Depression Scale – 15 item (GDS‐15) including Dysphoria, Withdrawal‐Apathy–Vigor (WAV), anxiety, hopelessness and Subjective Memory Complaint. Non‐parametric partial correlations examined associations between speech emotion‐based measures and depression scores adjusting for age, sex, education and MMSE.ResultsOverall depression score (total GDS) did not correlate with the speech measures. However, when examining specific depression dimensions, higher WAV and hopelessness correlated with reduced valence (WAV: r = ‐0.21, P = 0.00, hopelessness: r = ‐ 0.14, P = 0.01) and dominance (WAV: r = ‐ 0.18, P = 0.001, hopelessness: r = ‐ 0.16, P = 0.004). We did not find a significant association between emotional word scores and other depression dimensions.ConclusionPatients with high apathy and hopelessness select words that carry more negative and less powerful emotional valence. Our findings highlight the potential applicability of speech biomarkers in characterization of depression dimensions across spectrum of cognitive and are consistent with previous findings that reported more negative bias and more submissive speech in depressed patients.