Your browser version is outdated. We recommend that you update your browser to the latest version.

Frailty Decision Making

This work is being done as a consultant to a research team led by Dr. Charlene Weir, Dr. Bruce Bray, Dr. Qing Trietler-Zeng, and Dr. Jennifer Garvin. The team includes Yijun Shao, Dr. Rashmee Shah, and Yan Cheng.

The topics of frailty and cardiac interventions are not mine. The team was existent and developing these ideas before I joined. My contributions include a frailty ontology, a representation of physician decision making, and a pilot investigation into information presentation preferences.

Frailty Ontology

Our first task has been developing an ontology of frailty. The goal is to discovery entities that can be used to reconcile the multitude of frailty instruments currently in use. This work was assisted by Villiammai Chidambaram, Damian Borbolla, and Parveen Banu Ghani.

First we collated all of the terms/ideas referred to in 6 commonly used frailty assessment instruments. We matched these terms to their SNOMED-CT concepts, as near as we could. We are using a unique methodology, originated by Dr. Weir, to insure that our ontology reflects physician thinking. We are interviewing physicians as they review complicated cardiac cases, with multiple types of frailty indicators. By reviewing the interview recordings we can discover the frailty information that they are using to make decisions. Once we have those instrument and physician topics we are looking at the terms found in patient records, to make sure that the entities are able to found and that the terms are represented among the entities. Terms are found using a topic modeling algorithm, which searched 100,000 clinical notes (Shao, et al., 2016), and by human annotation of 1000 text snippets from clinical notes.

I setup the ontology with elements from the Ontology of General Medical Science (OGMS):

Clinical finding: (def.) A representation that is either the output of a clinical history taking or a physical examination or an image finding, or some combination thereof.

Clinical history: (def.) A series of statements representing health-relevant qualities of a patient and of a patient's family.

Two other entities are included for their relationship to frailty.

Clinical picture: (def.) A representation of clinically significant bodily components, dispositions, and/or bodily processes of a human being that is inferred from relevant clinical findings.

Diagnosis: (def.) The representation of a conclusion of a diagnostic process.

Note that each of these entities is a representation (L2). They are interpretations made by physicians. We discern the corresponding representation from an entry in the clinical record.

Part of our frailty ontology.Part of our frailty ontology.

 

By matching each of the entities to a concept in SNOMED-CT we make it easier for the NLP developers to find synonyms to include in the list of EHR entries that indicate the existence of a representation in the author's mind. Some hierarchical elements from SNOMED-CT are also included in the frailty ontology.

Clinician Decision Making

The next part of this study will be to review the interviews to create a representation underlying physician decisions.