Description
Documented non-adherence(NA) is the written record of a clinician’s concern that a patient may not be adhering to care plans, which are thought to results in poorer patient outcomes. The real impact of NA is difficult to measure at scale since such information is hidden in unstructured notes. Spending minimal clinical resource, we developed an expert-in-the-loop NLP pipeline (87% accuracy) to extract NA from notes and present analysis of patient characteristics across different dimensions.
Learning Objective: After participating in this session, the learner should be better able to:
1. Understand the problems faced by developeres of supervised algorithms versus a rule-based algorithm for NLP feature extraction in the medical domain
2. Be shown with the example of "documented non-adherence" (an impartant patient attribute often only recorded in unstructured notes) how experts-in-the-loop in developing concept extraction NLP can reduce the annotation workload for experts and still obtain reasonable results to conduct further clinical studies
Authors:
Joy Wu (Presenter)
IBM Almaden Research Center
David Grant, Barnes Jewish Hospital
Shrey Lakhotia, T.H. Chan Harvard School of Public Health
Patrick Tyler, Beth Israel Deaconess Medical Center
Daniel Gruhl, IBM Almaden Research Center
Chaitanya Shivade, IBM Almaden Research Center
Sebastian Gehrmann, Harvard University
Leo Celi, Beth Israel Deaconess Medical Center
Presentation Materials: