ARISE-CARE: Advancing Rehabilitation for Stroke Patients with AI to Elevate Therapy Care - Project Summary Stroke stands as a leading contributor to disability, with an estimated 89.1 million adults worldwide, and 9.4 million adults in the United States living with the residual effects of stroke. Rehabilitation interventions are effective in mitigating stroke-related disability, greatly enhancing the quality of life, and reducing life-time health care expenses. However, significant variance exists in the quality of interventions delivered by various rehabilitation therapists, at various rehabilitation facilities, making the evaluation of rehabilitation quality a national imperative. Although our clinical research team has developed a standardized and validated coding scheme to assess the quality, our study shows that this manual assessment is both arduous and time- consuming. In addition, the integration of rehabilitation activities and outcomes into electronic health records (EHRs), while offering considerable advantages, has introduced unique challenges. Notably, clinicians often report burnout due to burdensome documentation requirements, frequently requiring additional time outside intervention sessions and encroaching on personal time. This not only escalates the workload for therapists but also compromises the accuracy of the documentation. Recent initiatives in generative artificial intelligence (AI) have emerged to automate and streamline clinical documentation. Yet, a comprehensive evaluation of generative AI techniques for clinical documentation remains scarce. Furthermore, although many AI approaches to enhance rehabilitation have been successful in computational experiments, they often face challenges when being implemented in real-world rehabilitation practice. A fundamental question that demands addressing is: Are healthcare facilities, directors, therapists, and payers ready to intergrade AI-based approaches into rehabilitation practice? In this project, we aim to address the above issues through collaborative and innovative approaches. The study outlines three specific aims: 1) Develop advanced natural language processing (NLP) models to identify important indicative cues of rehabilitation quality from a unique dataset of transcripts from over 3,000 rehabilitation session recordings. 2) Assess generative AI models for automated clinical documentation and compare the AI-generated notes, manually created “ground truth” notes, and the notes from the real world EHR. 3) Use the Consolidated Framework for Implementation Research (CFIR) framework to assess the readiness of facilities, directors, therapists, and payers in integrating AI-based approaches into rehabilitation practices. The project will significantly enhance rehabilitation therapy care for stroke survivors, as it addresses the pressing need to improve the quality of rehabilitation interventions and to evaluate the effectiveness of generative AI in clinical documentation. Furthermore, by assessing the readiness of key stakeholders in real-world rehabilitation settings, the project lays the groundwork for informing future studies on the implementation of AI-based approaches in rehabilitation.
