By: Charlie Harp

Recently, we were working on a mechanism in Symedical designed to enable the creation and deployment of clinical insights based on the information in a patient’s electronic medical record.  The input included all manner of information: diagnoses, problems, vitals, lab results, medications, social history, procedures, family history and even elusive information like “ejection fraction” readings extracted from unstructured procedure notes.

The mechanism was working great and we were able to identify patients likely to be diabetic, have hypertension or heart failure that were previously missing information in the patient’s records.  It was pretty cool.

Here’s the thing that struck me as odd: At one point, when designing the congestive heart failure model, we wanted to identify whether or not the patient had a pacemaker.  Anyone that has worked in healthcare IT knows that the way you do that is to look for a procedure in the patient’s history that references the implantation of a pacemaker, right?

I was considering the semantics needed to identify terms that met this criteria, when I had one of those “wait a minute…” moments.  This was the moment when it dawned on me that we track two fairly significant categories of patient information almost exclusively using inductive reasoning.  These categories are missing patient anatomy and implanted medical devices.

Inductive reasoning is where you come to a probable conclusion based on what is observed.  It is often confused with deductive reasoning (I am looking at you Sherlock…) which is where you come to a certain conclusion based on what is observed.  Anytime we guess based on evidence, we are using inductive reasoning to reach that conclusion.

When it comes to devices that have been implanted or parts that have been removed, we identify them based on whether a surgical procedure was performed.  In some cases we might also track it as a “observation,” “problem,” or a “diagnosis”.  Another place where this information can commonly be found is in unstructured notes.  If you think about it, this is a blatant sign that our systems evolved from a billing focus, not clinical focus, because the presence or absence of biological or mechanical components is clinically significant.

It is also concerning, considering the relative quality and uncertainty of historical data in clinical systems today, that we might drive clinical decision making based upon the presence of something like a CPT code from five years ago.

The reason it struck me as wrong is because there is absolutely no reason why we should have to induce that the patient is missing their spleen.  This is something that was definitely known at some point and could be recorded if we had the proper mechanism to do so.

Is this type of information really that relevant?  If you’ve read my series on precision medicine you know that I believe for a clinical system to truly assist a provider, it must know as much as it can and it should have a level of understanding of the patient which approaches that of the provider.  I am fairly certain a provider treating a patient knows if the patient is missing a leg, or if the patient has an implanted cardiac defibrillator.  In fact, I would assume that if the provider sees a CPT code for having something removed, they will take a minute and verify with the patient that something was actually removed.  This information lives in the provider’s head and influences their decision about how to treat the patient.

Our current circumstantial approach only allows the applications and associated logic to guess based on the presence of any number of procedure codes (if, in fact the procedure was coded).  This is not critical unless we actually want the application to be useful to the provider.

For the avoidance of doubt, I am aware that there are concepts, in SNOMED CT for example, that represent medical devices (cardiac pacemaker) and missing anatomy (absence of ear).  What I am expressing is that (1) the use of these types of concepts are the exception, not the rule and (2) even when they are used they are not in a consistent, predictable place.

Imagine if we had a new category of information where we tracked this kind of data more like an inventory.  One would be a list of anatomic locations that have been surgically removed, transplanted or altered.  The other would be a list of devices that have been implanted into the patient.

The properties of each of these new categories of information would be unique, conceptually relevant and could prove to be very useful for disease management and analytics in general.  Once we have established these entities in the patient record we can relate them to procedures and complications which would further expand and empower the rich clinical tapestry of our understanding.

I always find it interesting when I encounter something that surprises a “salty dog” that has been around as long as I have.  It reminds me that there is much to learn and reevaluate if we hope to evolve healthcare IT.

If you know of a system that handles this type of information well or have any interesting epiphany stories of your own, please share them with me.  If it is too long or relevant for a comment, I am always looking for a guest blogger.