Last week in our Making Digital Health Real series, Scott Nelson presented his research at Vanderbilt University on decision support and risk prediction for medication safety. This got me thinking about how we present information to enhance decision making.
In a previous blog post on clinical variation, I talked about closed loop decision support, where the computer makes all the decisions without the clinician. This is feasible and desirable for some tasks, but most tasks are guided by Chuck Friedman's Fundamental Theorem of Informatics, where the computer provides information that augments the human:
One of my favorite new papers by our colleagues Asher Lederman, Reeva Lederman, and Karin Verspoor describes how most clinical NLP research has tried to emulate clinicians and provide a decision to them for consideration (task as decision), such as "this patient probably has pneumonia" or "this patient is likely to be readmitted." They propose instead
a “tasks as needs” model representing a paradigm shift to focus on supporting clinicians rather than emulating them, and to bring clinical NLP in line with Friedman’s Fundamental Theorem.
In this new model, the NLP system will
identify the factors that are relevant to the clinical decision, surface them from the notes or other patient data, and present them concisely, providing the information that a clinician needs to make a decision. In short, we have transformed an end-to-end readmission prediction tool into a clinical tool supporting a discharge decision, and shifted the use of NLP as a feature extractor or end-to-end recommendation model in a fully automated tool to targeted evidence gathering, supporting human decisions.
Here is an example tool for presenting relevant information regarding a discharge decision:
Qing Zeng at George Washington University used NLP to alter the presentation of information to patients in discharge instructions from text to pictures and showed an increase in short-term memory of the discharge instructions.
One size may not fit all, though, as Qing discovered that cultural background played a large role in the interpretation of the images. For this image, many people understood it meant "chicken pox", but some cultures instead thought it was "bird flu" or asked "what did the chicken do to that person?"
Regional Linguistic Quirks (RLQ): I hiked the Werribee Gorge Trail this weekend and my Australian friend used the saying "touch wood", which we in the US render as "knock on wood." She was amazed to finally understand what was meant in the words to a song by that title.
Comments