Detecting the onset and peak symptoms of exacerbations of COPD by analyzing speech

<bold>Introduction</bold> Early detection of exacerbations of COPD (ECOPD) improves treatment outcomes for people living with COPD. Since respiratory distress impacts voice characteristics, we hypothesized that speech analysis could be used to detect ECOPDs. <bold>Aim</bold> We evaluate how speech changes at onset and peak of symptoms of an ECOPD compared to baseline (stable COPD) speech using a daily monitoring mobile application. <bold>Methods</bold> Data were collected as part of the TACTICAS study, a prospective cohort study collecting voices from COPD patients daily for 12 weeks. ECOPDs were captured using the patient-reported, symptom-based events of the EXACT questionnaire. Changes between baseline, onset and peak of an ECOPD were analyzed using a GLM repeated measures with 25 Praat speech features extracted from sustained vowel /a/ in the morning and evening. <bold>Results</bold> 11 COPD patients completed daily EXACT and voice tasks, constituting 1691 voice recordings in total. 16 ECOPDs were captured during these 12 weeks, from which 6 were treated. Specific speech features were significantly different at onset and at peak of an EXACT-defined exacerbation (see Figure 1). <bold>Conclusions</bold> Speech of COPD patients changes at onset and peak of an ECOPD. Interpretable speech features could possibly be incorporated in future mHealth applications to detect ECOPDs. <fig><object-id>erj;64/suppl_68/PA4372/F1</object-id><object-id>F1</object-id><object-id>F1</object-id><graphic></graphic></fig>