Rapport: Pediatric Patient- and Family-Oriented Radiology Report


How can we design tools to improve medical communication about radiology imaging studies for patients and their family members? To answer this question, I applied human-centered design and data-driven computational methods to study and design for various communication needs of patients who are going through radiology imaging studies.


  • Project Date: May 2016 - May 2018
  • Affiliations:  Georgia Institute of Technology, Children's Healthcare of Atlanta, Emory University Hospital
  • Funding: IPaT seed grant, NSF #1464214
  • My Role:  Study design, relationship building with radiologist clinical partners, data collection, data analysis, prototype design, tool development, paper writeup and presentation
  • Collaborators:  Clayton Feustel (Research Software Engineer), Chaitanya Bapat (Research Software Engineer), Serena Tan (Designer), Meeshu Agnihotri (Graduate Research Assistant), Max Silverman (Graduate Research Assistant), Stephen Simoneaux (Lead Radiologist), and Lauren Wilcox (Principal Investigator)



Background

Results of diagnostic radiology imaging studies such as computerized tomography (CT) and magnetic resonance imaging (MRI) represent an important type of medical data, and play a critical role in diagnosing acute and chronic diseases. Since the early 1920s, just about 20 years after the invention of x-ray, radiology reports remain the center of communication between the interpreting radiologist and a referring physician, in a highly technical manner.




Radiologist communicates their interpretation of imaging data to a referring physician. This practice remains unchanged for over 100 years.



The highly technical nature of these unstructured text reports prevent lay patients to engage with their own health data.

Following the passage of the Health Information Technology and Clinical Health (HITECH) Act, diagnostic radiology reports are increasingly being made available to patients and their family members. However, these reports are not typically comprehensible to lay recipients, impeding effective communication about report findings. How do we make this data more accessible to lay patients so they can lead effective clinical conversations with their doctor?



Goal

The goal of this research is to understand how we can design tools to improve medical communication about radiology imaging studies for pediatric patients and their family members.



Methods
  • Qualitative analysis of 1600 online health forum posts
  • Text analysis of 205,250 de-identified MRI and CT reports
  • Expert knowledge elicitation with 7 radiologists
  • Field deployment and semi-structured interviews with 14 patient-parent pairs and 5 oncologists.


Analysis of Question Posts on Online Health Forums

Our first study is a content analysis of four online health forums: medhelp, health boards, cancer survivors network, and reddit. To build our corpus, we scraped about 1600 posts, using radiology specific keywords like radiology, mri and ct scans, and ended up using 480 question posts for analysis. Our thematic analysis of these posts yielded 4 major themes and 23 subthemes.



Posters turn to online health forums to make sense of their imaging data by sharing certain parts of the report content. An interesting observation is that posters did not have a problem with understanding just the terminology in the radiology report. Theirs needs were focused on inferencing meaning from whole sections.


This finding prompted us to aim for a higher level understanding and context of the language used in these radiology reports.



Expert Knowledge Elicitation Study

To understand the context, we first needed to understand the range of common concepts, or basic types of information, that could be communicated through a typical radiology report.


By extracting and sampling common phrases from large scale text analses, we were able to understand the nuanced context behind radiology specific expressions.

To do this, we mined and sampled 20 common n-gram phrases from 205,250 de-identified pediatric CT and MRI reports and solicited radiologists' expertise to understand common underlying concepts in radiology reports.

Example phrases included: "clinical correlation is needed", "cannot be completely excluded", and so on.



We turned these sampled phrases into a study packet. Each page contained one of the 20 phrases, 3 excerpts from reports containing the phrase, and a list of concept categories that we initially derived from our discussion with the radiologist partner.

We then validated the meaning of sampled phrases through hour long discussions with 7 radiologists, and discovered 13 concept categories that are commonly communicated in these reports.




Our focused discussions with radiologists informed the choice of language and information structure in the prototype. Here's what we learned:

  • Impression section contains the most important findings, often listed in the order of importance, effectively acting as a summary of the entire report.
  • Phrases such as “clinical correlation is needed” are used to communicate uncertainty and need for more clinical data to reach a conclusive diagnosis.


Design of an Interactive Radiology Report

We were convinced that, to support communication, simply focusing on "technical-to-lay" language translation of the radiology report would be a misguided approach. Instead, we decided to support some aspects of the translation, just enough to scaffold patients and their parents’ participation in clinical conversations about radiology reports.

Informed by the prior studies, our prototype design was guided by the following design goals:

  • Identify medical concepts and important sections of interest to the patient.
  • Structure and organize report content to simplify navigation.
  • Clarify medical jargon within the clinical context.
  • Enable contextualized selection of report content to associate content fragments with questions and discussion topics.
Rapport Design Features

Rapport is a novel hand-held tablet application designed to support patient families in reviewing and communicating about radiology studies with their physician. It automatically organizes report content through automated detection and segmentation of section headers. We designed three features to support collaborative review and discussion.



Summary: the summary tab contains text coming from the impression section of a radiology report, which often lists diagnostic findings in the order of importance.

Information cards: Upon touching each word, information cards display lay-friendly explanations and visualizations on-the-fly.

My Notes: based on user interaction data, the My Notes feature generates a set of discussion topics to help patient families apend notes or questions they would like to discuss with the doctor.



Video Demo


Pilot Deployment Study

Our pilot deployment involved qualitative usability evaluation studies in a hospital setting with 14 cancer patient families and 5 oncology physicians using the patients' actual radiology data. The pilot study generated the following design insights:



Families valued the Summary feature, knowing that doctors also pay attention to it. Many parents commented on prior experiences encountering difficulty navigating the report content, but changing the label from Impressions to Summary seemed to provide reassurance to patient and parents' understanding of the section content.

"I was telling her that this [summary] is my favorite page right here [...] I want to know what's the most pressing based on the doctors that looked at the report."

Parent of teen cancer patient

"It makes it a lot easier, rather than taking the effort to look it up and then find a reliable website to look at, that has factual information."

Teen cancer patient


The expediency of retrieving medical term definitions afforded an overall improved reading experience. Besides the convenience of faster information retrieval, providing definitions in our self-contained application ensured that definitions came from a trusted source.




Finally, the ability to write notes already referencing relevant parts of the report can be particularly helpful when trying to remember what was said about a specific term. Building on this history of note-taking behaviors, the My Notes feature supported families' coordinated participation (via turn-taking) in medical conversations.




Improved Design Features after Pilot Feedback



Integration of web services for real-time querying: The application uses a natural language processing (NLP) system known as Apache cTAKES to identify clinically relevant terms and pulls 3D visualizations of the human anatomy as well as definitions from ontology databases, such as MedlinePlus, Wikipedia, and RadLex.




Tablet optimized design: We designed the user interface to reflect recent design trends in tablet-based mobile applications.





Tumor size-object conversion chart: To support lay understanding of tumor sizes, the application visually shows how the tumor size (notated in millimeters or centimeters) compares to common everyday objects that lay people are likely familiar with.



Inches Centimeters       Object
0.04 in 0.1 cm Grain of sugar
0.08 in 0.2 cm Crayon (tip)
0.2 in 0.5 cm Pea
0.2 in 0.6 cm Pencil (width)
0.4 in 0.9 cm Ladybug
0.5 in 1.2 cm Lipstick
0.6 in 1.4 cm M&M candy
0.6 in 1.6 cm Jean button
0.7 in 1.8 cm Dime
0.8 in 1.9 cm Penny
0.8 in 2.1 cm Nickel
1.0 in 2.4 cm Quarter
1.1 in 2.6 cm Bottle cap
1.4 in 3.5 cm Film roll (width)
1.6 in 4.0 cm Ping pong ball
1.7 in 4.3 cm Golf ball
2.1 in 5.4 cm Credit card (height)
2.5 in 6.4 cm Tennis ball
3.0 in 7.6 cm Baseball
3.4 in 8.5 cm Credit card (width)




Publications
ACM Digital Library Supporting Families in Reviewing and Communicating about Radiology Imaging Studies.
Matthew K. Hong, Clayton Feustel, Meeshu Agnihotri, Max Silverman, Stephen F. Simoneaux, and Lauren Wilcox. Proceedings of the 35th Annual ACM Conference on Human Factors in Computing Systems (CHI 2017), Denver, CO, USA, 2017 (25% acceptance rate).