The Uncertainty of Diagnoses

There was a sentinal event experienced when I matriculated into medical school at the age of 34.  The dean of the medical school stood up and quoted a study that showed in 1983 up to 40% of the time the working diagnoses at Johns Hopkins were found to be in error on autopsy results. He then went on to say that by the time we finished residency most of the knowledge we'd memorized in medical school would be obsolete. That's when I went out and spent over $3,500 for a computer only to be crushed that there was very little software that would help me 1) get through medical school and 2) help with diagnosing and treating a patient. 

A lot has happened since then but the promise of artificial intelligence, connected health information networks and computerization of health care hasn't really panned out the way I imagined over the last 25 years. But one thing that hasn't changed is the complexity of even simple diseases. Why?

Well, for one we don't practice medicine scientifically.  Take a simple sore throat.  We don't do viral and bacterial cultures on every one or even a random sample of patients to discover the exact pathogen with which we're dealing.  There are probably over 200 viruses and 50 bacteria and a multitude of mechanical and environmental agents that will produce almost identical symptoms.  We don't have instant tests with the exception of Rapid Strep, Rapid Influenza and Rapid Mono tests that can help us significantly.  Thus for the vast majority of cases we are practicing blind.  Fortunately most patient get well from this condition in 2 weeks no matter how we treat.  The evidence suggests that with the exception of a very small minority of conditions NOT treating is better than treating but it is really hard to convince patients no treatment is better than treating.

I purchased QMR, Iliad and a subscription to AMANet to access Octo Barnett's DxPlain to help me come to correct diagnoses. What I and my colleagues discovered is that there were about 1,500 signs and symptoms that covered almost all of the known diseases (over 20,000 in the databases).  That meant a large number of diseases presented with the same signs and symptoms.  There were very few diseases that had pathognomonic signs or symptoms.  Consequently we became pretty good at coming up the differential diagnosis (a list of the diseases that shared the same symptoms).  Our goal was to rule in or rule out the diseases by ordering tests and procedures.

In primary care we see patients every 10-15 minutes.  We go through the same process and usually have a relatively short differential list of diagnoses that we think we're treating.  We order tests that are returned to us over the next few days.  And yet we have to make a diagnosis for that visit at the time of the visit in order to get paid.  We usually pick the most probably diagnosis at the visit and that goes on the claims that is processed electronically.

The labs will come back and either rule this diagnosis in or out.  Guess what?  There isn't a process for us to go back and amend the visit diagnosis with the correct one if a test ruled the original one out and replace it with another one.  My guess is that over time up to 40% of the claims based diagnoses are totally bogus.

And then there are complicated diseases like Lupus Erythematosis or Fibromyalgia with no confirmatory tests.  A patient may be seen up to 10 times before these "diagnoses of exclusion" are made.  There is no systematic way to go back and amend the diagnoses of the previous 10 visits and change them to the final diagnosis made.

That's why most astute people will take claims based data in the ambulatory environment with a huge grain of salt.  What may be more valuable is a big picture of all of the diagnoses over time to get an idea of what's going on with a patient.  But that data is usually not available as it's hidden in many physician's charts, EMRs and insurance claims data.

We need a national system for experts to go through reams of claims based data, compare it with the symptoms documented and then systematically modify the diagnoses to improve their worth.  That's probably not going to happen.