As a result of efforts to reduce healthcare costs and improve quality, patients with heart failure are increasingly receiving treatment in community-based programs, such as managed long-term care programs which support older adults in remaining independent in their community. Managed long-term care aims to reduce unplanned hospitalizations and emergency department (ED) visits, but heart failure is still the leading reason for these avoidable events. In managed long-term care, the care coordinator (i.e., a registered nurse or social worker) maintains regular contact with patients by telephone to ensure that they receive care congruent with their medical needs. From the linguistic perspective, verbal communications between patients and healthcare providers are information-seeking and sharing behaviors, as they include problem-focused communication. From the acoustic perspective, heart failure can affect patients’ voice and speech characteristics due to swelling caused by fluid retention or compression of the laryngeal nerve due to enlarged heart structures. While verbal communication between patients with heart failure and their care coordinators can provide insight into hospitalization and ED risks, it is largely untapped in managed long-term care. To address this gap, we aim to examine whether audio-recorded verbal telephone communication (hereafter called verbal communication) can be utilized to improve risk prediction. In the K99 phase, we will focus on identifying information in verbal communications between patients with heart failure and their care coordinators. We will extract the following potential risk factors for hospitalizations or ED visits from verbal communications: (1) conversational characteristics to analyze interactions in patterns of communication, (2) language phenotypes based on a list of the language of risk factors, including heart failure symptoms, poor self-management, and other hospitalization risks, and (3) acoustic features by analyzing voice signals. In the R00 phase, we will focus on developing risk prediction models for hospitalizations or ED visits for patients with heart failure in managed long-term care. We will develop several machine learning-based risk prediction models for hospitalizations or ED visits using information derived from: a) structured electronic health records, b) care coordination notes, and c) verbal communications between patients with heart failure and their care coordinators (identified during the K99 phase). We will evaluate if the risk prediction performance of machine learning algorithms can be improved by integrating information from different data sources. This proposal is aligned with the Strategic Vision key area of the National Heart, Lung, and Blood Institute (NHLBI), "Leverage emerging opportunities in data science to open new frontiers in heart, lung, blood, and sleep research." This study will be an important step toward achieving my long-term career goal of developing risk predi...