Getting at quality: What’s involved in knowing the difference between appropriate and inappropriate care

May 12th, 2017 / By L. Gordon Moore, MD

Implantable cardioverter-defibrillators (ICDs) can save a person’s life when they detect a dangerous heart rhythm and correct it by delivering a shock to the heart. For another person it might be a very expensive device that leads to little more than a lump under the skin near the right collar bone and special screening at the airport.  The difference between these scenarios could be a quality-of-care indicator, but figuring out that difference can be difficult and costly.

A common method of knowing is to require that the clinician submit data to prove that the device is appropriate for the person. We could require that each time one of these devices is implanted, the information required for validation of appropriateness is submitted to an external review process.  The guideline for appropriate use of ICDs requires knowing the answers to a lot of questions.

In a nice study on finding the people most likely to benefit from an ICD insertion, Wijers et al describe some of the required data elements:

Obtained variables included patient demographics, indication for implantation, New York Heart Association (NYHA) functional class, left ventricular ejection fraction (LVEF), renal clearance, history of diabetes, documented rhythm disorders, QRS duration, medication and device settings.[i]

And here they describe some of the required clinical elements:

Ischaemic cardiomyopathy as underlying heart condition was defined as either the presence of coronary artery disease, myocardial infarction or both. A history of atrial tachyarrhythmia was documented when patients suffered from atrial fibrillation or atrial flutter, or had experienced episodes of these arrhythmias in the past.[ii]

The problem with all of this reporting is that it takes health professionals away from their work of helping people and forces them into the administrative task of reporting on that work, thus adding cost to the health system in an attempt to improve quality and reduce unnecessary cost.

This is where computers are supposed to step in and help.

The problem is that computers are literal.  If a word is spelled correctly and in the right place, a computer might find that data element handily.  If electronic medical records (EMRs) had a check box for every data element, a mouse-click or two could tell us if a particular patient met the guidelines for appropriate ICD insertion.

If EMRs had a click-box for every data element (a.k.a. structured data), we might also expect doctors and nurses to spend too many hours documenting all the elements that are typically dictated or written into a person’s chart.  EMRs have some structured data, but the bulk of information is lost to us as unstructured data in text notes.  This is the trade-off EMRs make so that doctors and nurses can actually get through their day and take care of patients.

Barbara Zellerino RN, MHA (one of my colleagues and expert in unstructured data mining) tells me that “heart block” (one of the essential data elements in determining appropriateness for ICD insertion) can show up in many ways in an EMR:

  • third degree heart block
  • Complete Heart Block
  • high-grade heart block
  • total heart block
  • CHB
  • 3rd degree heart block
  • third-degree AV heart block

If a natural language processing engine fails to understand that these are all the same, that engine’s reports will be erroneous.

For computers and EMRs to live up to their potential to support care delivery, we need advanced natural language processing that has the capacity to understand and link the fundamental concepts in spite of variability in the way the data presents.

Given advances in machine learning and ability to map nomenclatures and concepts, we should be nearing the time when we can use smart engines to free health professionals from administrative work so they can care for their patients.

L. Gordon Moore, MD, is senior medical director for Populations and Payment Solutions at 3M Health Information Systems.

[i] Wijers, S. C., B. Y. M. van der Kolk, A. E. Tuinenburg, P. a. F. Doevendans, M. A. Vos, and M. Meine. “Implementation of Guidelines for Implantable Cardioverter-Defibrillator Therapy in Clinical Practice: Which Patients Do Benefit?” Netherlands Heart Journal 21, no. 6 (June 1, 2013): 274–83. doi:10.1007/s12471-013-0407-x.

[ii] ibid.