Audio Generation and Optimization from Existing Resources for Patient Education

点击翻译按钮获取中文摘要

Project Summary/Abstract Health literacy is vital to achieving and maintaining good health. Several national programs have emphasized this goal and its importance. Text is generally much more efficient and cost-effective for presenting healthcare information on a large scale than interactive tools and videos. Over the past decade, therefore, most medical information has been provided as text, e.g., via printed pamphlets or on websites. We are entering a new era where a new similarly effective mode of information dissemination is becoming increasingly available: audio accessed with mobile devices. Millions of households have and use smart speakers and virtual assistants and they are increasingly used by patients and consumers to gather information. Hospitals also plan to gradually integrate them among their tools. However, there exist few if any guidelines on optimal generation and use of audio. The overall goal of this project is to discover how to support the creation of optimal audio from existing text sources for consumer and patient education. To accomplish this, four aims are proposed. The first aim is to identify audio features that affect information comprehension and retention. Here, features in audio content and style (e.g., word frequency or grammatical complexity) of the underlying information will be tested for impact. In addition, two groups of features specific to the audio medium will be tested: the delivery features (e.g., speed and pauses) as well as meta-features (e.g., speaker characteristics such as gender or accent and bias in listeners). This first aim will rely on large-scale datasets, semi-automatically generated and augmented with user scores for comprehension gathered using Amazon Mechanical Turk (MTurk). Statistical and machine learning approaches will be used to tease out the best features and combinations. The second aim focuses on discovering how to augment text for audio and finding the optimal combination of text and audio for information comprehension and retention. Different combinations will be tested online with MTurk participants using controlled user studies. The third aim is to update, test and provide the existing online free text editor to generate optimized audio. We will also start dissemination of the tool to potential users including API access to components. The project will conclude with a summative evaluation with representative consumers recruited at a local community health center and further dissemination of preferences, practical obstacles, and best practices for the medical community to help increase health literacy through this new, popular audio medium. If successful, this project will generate best practices for the medical community in using audio as an additional method for bringing healthcare information to the general public; it will provide an online, free tool to generate audio leveraging these best practices and will include API access so that other researchers can easily integrate tool components into their research and tools; and it will provide immediate practical lessons from working with consumers relevant for clinical practice.