Leveraging YouTube Video Analytics for Patient Education: A Digital TherapyTool for Clinicians to Retrieve and Recommend Understandable Videos on Chronic Disease Management

点击翻译按钮获取中文摘要

Project Summary The easy availability of huge amount of user generated health information on social networks, blogs, YouTube, Twitter, and hospital review sites presents an unprecedented opportunity to investigate how social media can be a channel to inform and communicate healthcare information to patients and facilitate patient- centric health promotion and literacy improvement. YouTube hosts over 100 million healthcare related videos on a variety of medical conditions. This plethora of user-generated content can be leveraged by patients to improve adherence to clinical guidelines and self-care required for management of chronic diseases. In this project, we propose an augmented intelligence-based approach that effectively combines human input from domain experts and consumers with machine learning and natural language processing methods from computer science to winnow down and retrieve relevant, contextualized video materials that clinicians can recommend to patients. The problem of identifying the most relevant videos from a patient perspective is challenging, but provides an immense innovation space for this approach. We will leverage a co-training machine learning framework and incorporate inputs from patient education assessment tools and clinicians to assess diabetes- related videos on two dimensions: the amount of medical information encoded in the videos and video understandability. We will develop a user-centric patient education video recommender system by integrating these two dimensions with the YouTube video ranking results. Furthermore, we will apply a multi-dimensional evaluation strategy that combines computational evaluations, comparisons with YouTube baseline, and causal analysis methods to understand the performance of the automated methods and the relationship between video understandability and collective user engagement. Finally, we will integrate our computational approach in a modular research prototype technology platform that will accept health related YouTube videos as inputs (generated from patients' keyword searches on diabetes) and produce a ranked list of top 10 retrieved videos for further review by clinicians, and evaluated for barriers and facilitators of the technology usage. Recommending relevant educational materials in video format that leverage user-generated content is one way to deliver personalized and contextualized healthcare information, and resources for self-care management, to patients and consumers. As technology continues to advance and evolve, our methods can be refined further and evaluated via clinical trials to improve patient education, empower patients, caregivers and clinicians, and improve societal health and health literacy.