CLUE.png

Clinical language understanding and extraction (CLUE) and electronic medical/health/patient record analytics (EMRA)

Creating cognitive insights from patient records at the point of care.

Overview

Using Natural Language Processing (NLP) and machine learning to provide intelligent insights from a longitudinal patient record for patient care.

Watson for Patient Record Analytics includes:

  • An abstracted patient summary centered around an automatically generated problem list
  • Semantic search that returns clinically meaningful matches on multiple dimensions
  • Disease-specific summaries with insights important to management of several common conditions

Background

Approximately 1.2 billion clinical documents are produced in the U.S. each year, comprising around 60% of all clinical data. Primary care physicians spend more than half of their workday, nearly 6 hours, interacting with the EHR. Nearly a third of EHR time by primary care physicians is spent reviewing the patient record and evidence-based resources. Around half of all questions arising during clinical care are not pursued by healthcare providers at the point of care.

The Patient Record

Patient records contain both structured and unstructured data. Unstructured data is free text – most parts of clinical notes are unstructured data. An older or sicker patient may have hundreds of clinical notes. Structured data includes medication orders, lab results, procedures, and vitals.

Diagram of a patient record. Unstructured at top includes Encounter (Clinical) Notes. Semi-structured includes medications, lab results, procedures, vitals
Schematic of the patient record.

It is typical for a doctor to spend 5 to 10 minutes to review the patient record in order to get a basic understanding of what’s going on with the patient before actually seeing the patient.

Watson Patient Record Analytics

Watson patient record analytics are used on top of the patient record to make sense of the data. Watson patient record analytics consist of:

  • NLP: Watson understands the content
  • Synthesis: Watson organizes the data in meaningful ways to help clinicians understand the data
  • Cognitive Insights: Combine patient data and general medical knowledge to provide deeper insight
Venn diagram with three circles. Labels in non-overlapping portions of circles - Cognitive Insights, Synthesis, NLP on patient record data

Problem-Oriented Medical Record (POMR)

POMR has become the de-facto record keeping standard in most US hospitals.

Problem list is also a mandatory section in the CCD (continuity of care), part of HL7's CDA (clinical document architecture) standard.

Diagram showing a "problems list" pointing out to procedures, vitals, clinical notes & timeline, lab tests, and medications

Quality Assessment of Automatically Generated Problem List

On average Watson found 1.2 very important or important problems missed by physicians per patient record (avg. 6 problems)

Graphic with three stacked histograms. Top - manually maintained, avg. rating 4.5, middle- physician generated, av. rating 8.4, bottom - Watson generated, avg. rating 7.4

Publications

Contributors