It’s all in the algorithm!

Two Dartmouth Institute student-researchers are developing a machine-learning model to evaluate and improve patient education materials

A person newly diagnosed with cancer is suddenly faced with need to educate themselves about all aspects of the disease, including the risks and benefits of various treatment options. Most clinicians offer their patients educational materials, such as informational pamphlets, to help them learn about their diagnosis and available treatment options. However, few techniques exist for systematically reviewing the quality of written health information designed to help patients make decisions about which sources to trust. In frustration, many patients turn to Google, where the information they find can be inconsistent or unreliable. 

In an effort to improve what can often be a frustrating and confusing search for information, two PhD students at The Dartmouth Institute, Katie Saunders MPH'16, and Curtis Petersen MPH'14, are using machine-learning technology to rate and evaluate whether the information patients receive is clear and helpful. In a recent paper they co-authored in JCO Clinical Cancer Informatics, Saunders and Peterson, along with faculty co-authors Glyn Elwyn, Marie-Anne Durand, investigate the untapped potential for applying machine-learning technology to analyze a variety of patient education materials, including handouts, decision aids, and brochures. Find out more about their work below.

Could you describe for those of us who aren’t familiar with the term, what is machine learning and how long has it been in existence?

KS: When computer algorithms learn new information and improve – kind of like we do – that’s machine learning. There’s complicated math involved,  but the concept itself is straightforward. 

CP: Machine learning is the method in which a person gives a computer program data and it gives predictions or insights back. This general idea has been around since the 1950s; however, new computing technologies in recent years have made machine learning popular across many industries for automating tasks like analyzing language and spotting readability issues. 

What are the benefits to using machine-learning technology to improve medical information quality – for both providers and patients?  

KS: Researchers have developed ways to score whether a document is trustworthy, unbiased, or user friendly. Most of these rating systems are manual, so a real person must do the assessment. It is time-intensive and not particularly reliable. 

We think we can use machine-learning technologies to automate these processes. Technologies like natural language processing can perform even more nuanced assessments on larger quantities of text than human raters can. What we found in our paper is that some—but not many—groups have started doing just that. 

What are you working on next as a result of your research?

KS: We are working to build our own machine-learning model to assess patient education materials. The first step is training human raters to reliably identify nuances in language and tone in the written materials. We welcome graduate students who want experience in machine learning and want to be part of this project to contact us! 
                  
CP: From the patient’s perspective, the goal is to produce material that conveys detailed and actionable information. From the provider’s perspective, the goal is to provide them with material about which they feel confident can help patients understand their health better. 

At The Dartmouth Institute, students often have the opportunity to collaborate with faculty who are also leading health services researchers. Can you talk about what it means to work directly with experts in patient engagement and shared decision making, such as Glyn Elwyn, MD, PhD, MSc, and Marie-Anne Durand, PhD, MSc, MPhil? 

KS: Marie-Anne and Glyn have dedicated their careers to studying ways to improve communication between people who are ill and their clinicians. I’ve learned so much from them over the past few years. 

The idea for how to use machine learning to improve patient communication emerged out of conversations Curtis and I had been having as I learned about hands-on training he was doing with machine-learning algorithms. Glyn and Marie-Anne supported us and were excited about the idea. 

Then last summer, JCO Clinical Cancer Informatics reached out to Glyn about doing a paper about technology and shared decision making for cancer. Instead of going with one of the many other projects he is working on, he encouraged us to further develop the machine learning idea. I appreciate his mentorship and feel privileged to work with this group every day. I see this type of work shaping my career as a researcher. 
 

POSTED 1/7/2019 AT 08:52 AM IN #education #home