TOPIC: Pulmonary Rehabilitation TYPE: Original Investigations PURPOSE: COPD is a common, preventable and treatable disease characterized by persistent respiratory symptoms and airflow limitation with a projected increase in prevalence over the coming decades. Periodic assessment of COPD is needed to determine the level of airflow limitation and assess impact on a patient’s health status, risk of future exacerbations, hospital admissions, mortality and to help guide therapy. Previously, COPD was viewed as a disease mostly characterized by dyspnea, now it is recognized that COPD impacts patients beyond this. As a result, a comprehensive assessment of symptoms is recommended. The COPD assessment test (CAT) is an 8 item measure of health status impairment and is commonly used to assess symptoms. Voice is an important biomarker of many medical conditions including diseases of the respiratory system which leave signatures in breath and speech. We build on our previous work of voice-based prediction of FEV1 and FVC (Ashraf, Obaid, et al. Chest 158.4 (2020): A1687, Walter, Kristin. JAMA 325.12 (2021): 1130-1131) to create a voice based early warning system for COPD exacerbation utilizing the CAT score. METHODS: We present the results of a prospective cohort study correlating voice and breath samples to CAT scores. We collected a total of 418 sessions from 9 participants aged 42 to 75 at a large suburban general hospital in Western Pennsylvania over the course of a year. Each participant provided biweekly voice and breath samples, along with self-reported CAT scores. We recorded audio samples on a commodity smart tablet using a proprietary software and analyzed recorded data off-line on the secure Telling.ai cloud. Based on input from health care providers, we weighted CAT questions pertaining to cough, sputum, wheezing and tightness of the chest more heavily than other questions. We extracted audio features from the recordings and developed a regression algorithm using k fold cross validation to predict deviation in the weighted CAT score from a baseline, and a binary classification retrospective prediction algorithm to predict a clinically important weighted CAT score increase (2 point or more) upto 7 days in advance. RESULTS: Our weighted CAT deviation from baseline regressor and retrospective exacerbation classification models had the following results, averaged across all folds, for all participants:(R squared = 0.972, Mean absolute error =0.449, Mean squared error = 0.572, Pearson Coefficient Correlation r = 0.998, two-tailed p-value < 0.0000001)and(Accuracy = 94.086%, Sensitivity = 94.229%, Specificity = 94.414%)respectively. CONCLUSIONS: We demonstrate that changes in voice and breath correlate strongly with changes in symptoms scores 0.97 R^2and MAE = 0.45 and furthermore, that signatures in voice can be detected even before the symptoms present themselves (2 point or more weighted CAT score increase upto 7 days in advance) with 94.23% sensitivity and 94.41% specificity . This technology requires no additional custom-built hardware, is cost effective, non-invasive and practical for ubiquitous and frequent use on mobile phones. CLINICAL IMPLICATIONS: The solution is ideally suited for personalized, predictive and frequent monitoring to enable the development of clinical decision support systems. This technique offers the promise of early interventions to reduce exacerbation frequency and severity, advancing personalized medicine for chronic and acute respiratory care. DISCLOSURES: Employee relationship with Telling.ai Please note: $20001 - $100000 by Mir Mohammed Daanish Ali Khan, source=Web Response, value=Salary Employee relationship with Telling.ai Please note: $20001 - $100000 by Rajat Kulshreshtha, source=Web Response, value=Salary Employee relationship with Telling.ai Please note: >$100000 by Prakhar Pradeep Naval, source=Web Response, value=Salary Speaker/Speaker's Bureau relationship with boeringer Ingleheim Please note: 2019-2020 Added 04/29/2021 by Anil Singh, source=Web Response, value=Honoraria Speaker/Speaker's Bureau relationship with astra zeneca Please note: 2019-2020 Added 04/29/2021 by Anil Singh, source=Web Response, value=Honoraria Chief Technical Officer relationship with Telling.ai Please note: >$100000 by Satya Venneti, source=Web Response, value=Salary