Data-driven shared decision-making to reduce symptom burden in atrial fibrillation

项目总结

PROJECT SUMMARY Atrial fibrillation (AF) is the most common cardiac arrhythmia with symptoms that directly impair health-related quality of life (HRQoL). While catheter ablation is routinely performed to reduce AF symptoms and improve HRQoL, we lack evidence about which symptoms are likely to improve and for which patients. Ablations themselves may cause complications that lead to lower HRQoL. Shared decision-making (SDM) is a widely encouraged practice to navigate such complex choices by aligning treatment benefits and risks with the patient's stated values. However, no SDM interventions have focused explicitly on AF symptoms due to a lack of rigorous evidence about post-ablation symptom patterns and the decision aids necessary to communicate those findings. In this K99/R00 application, we propose to use data from electronic health records (EHRs) to characterize post-ablation symptom patterns, and display them in decision-aid visualizations to support personalized SDM about the best treatment modalities for an individual's patient's AF symptoms. In the K99 phase, we will use natural language processing (NLP) and machine learning (ML) to extract and analyze symptom data from narrative notes in EHRs. We will also employ a rigorous, user-centered design protocol created during my postdoctoral work to develop decision-aid visualizations. In the R00 phase, we will conduct a feasibility study in which the interactive decision-aid visualizations are introduced during consultations about ablation in clinical electrophysiology practices. Our specific aims are: (1) identify common symptom patterns in patients with paroxysmal AF post-catheter ablation (n>32,014); (2) develop and evaluate decision-aid visualizations of common AF symptom patterns (n=50); and (3) evaluate the feasibility of implementing the decision-aid visualizations in clinical practice (n=75). The training objectives of this project include mastering competencies in NLP, ML, human-computer interaction, symptom science, and implementation science. The long-term training goal is to assist Dr. Reading Turchioe to become a faculty member with an independent program of research. She seeks to lead an interdisciplinary team of scientists and clinicians committed to improving symptom management and HRQoL for individuals living with AF and other chronic cardiovascular conditions, with an eye towards health equity. To ensure success for the planned research and training activities, a multidisciplinary team of mentors with complementary expertise, established, well-funded programs of research, and a record of mentoring high-quality trainees will advise her. Moreover, this research will be conducted in a world-class academic medical center with exceptional resources for building and implementing technology and data science methods using EHR data. The proposed research is both significant and innovative: NLP and ML methods to extract EHR data for decision-aid visualizations are a novel approach to SDM in the understudied area of AF symptoms. Together, these techniques promise to enhance HRQoL for other AF treatment modalities (e.g. medications, lifestyle changes) and other chronic cardiovascular conditions. ! !