Ask the Expert: How can AI make online health information easier to understand?

Anjana Susarla, Omura-Saxena Professor in Responsible AI

In the United States today, around 93% of adults have online access, and about 80% of those go online seeking health-related information. However, low health literacy in the US means that even well-intentioned educational materials may remain opaque or overwhelming to users.

Anjana Susarla, the Omura-Saxena Professor in Responsible AI at The Broad College of Business, recently co-authored a study entitled “Promoting Health Literacy With Human-in-the-Loop Video Understandability Classification of YouTube Videos: Development and Evaluation Study” that developed an automated method to evaluate how understandable patient-education videos are for people seeking information on diabetes.

The findings expose gaps in user engagement and comprehension and point toward how health systems and content creators might adopt evidence-based digital health strategies to make information more accessible and behaviorally persuasive.

Below, Susarla shares most details about the findings and how artificial intelligence can improve health communication and digital literacy for patients seeking care information online.

It is estimated that almost 80% of the US adult population seeks online health information. YouTube, the largest video-sharing social media platform, hosts more than 100 million videos providing information on various medical conditions, and has become a major source of health information in the US and worldwide.

At the same time, only 12% of adults in the United States are considered proficient in their ability to meaningfully interpret health information. People searching for health information on digital media platforms do not always have the requisite health literacy to understand and interpret health information.

Our research aims to address that gap in developing AI-augmented methods to retrieve understandable and relevant video-based health education content.

Diabetes is among the most prevalent chronic conditions in the United States and many other parts of the world. More than 100 million US adults are now living with diabetes or pre-diabetes, according to the CDC 2020 report. Diabetes is a contributing factor to many other serious health conditions, such as heart disease, stroke, nerve and kidney diseases, and vision loss.

Evaluating health information in online videos is a multi-faceted problem where we needed to incorporate evidence-based guidelines from health literacy and patient education (which are designed for small group-based focus studies, etc.). However, there are thousands (or hundreds of thousands) of videos on each medical condition on online platforms.

So, any method used to automate the classification of videos also needs to be scalable. Using AI methods is a good way to achieve scale.  At the same time, cutting-edge AI algorithms need vast amounts of training data. In our case, we needed expert input from patient education experts to label the video, so it became prohibitively expensive to create very large training datasets.

We overcame that problem by combining a human-in-the-loop approach, where we can incorporate domain experts’ assessment when the results from machine learning (ML) models were insufficient.

Our results highlight that patients have different health information needs. With the vast amount of health information available in multimedia formats on social media platforms such as YouTube and Facebook, billions of people across the globe are accessing health care information via social media without any way to verify the accuracy, understandability, or relevance of the content. Our study offers a path to curate online health information using multiple criteria to meet the health literacy needs of a diverse population.

Our results can help healthcare organizations tailor education materials to enable patient education and empowerment and thereby increase adherence and self-care.

Well-designed patient education videos can be part of a holistic system of care encompassing disease prevention and lifestyle changes, along with resources for emotional support, better patient-physician interactions, and providing current and scientifically valid medical information to patients. The methods and principles we developed could be adapted to other chronic, acute, and infectious health conditions, such as cardiovascular disease and cancer, and to broader patient education contexts, such as medication adherence and patient safety.