On The Healthcare Guys News this morning they write about predicting odds of a disease. Dr. Thomas McGinn is betting that he can predict strep, pneumonia and other ailments by calculating probabilities with a software program.

We here at Sentia Health have had all these thoughts before, during the first season of House in 2004. What Dr. McGinn is going to realize is that there isn’t a direct correlation between reported symptoms and correct diagnoses. After he realizes there is no direct correlation, he is further going to realize that correlation is not causality. There is a 100% correlation between wet sidewalks and rain, but wet sidewalks only cause rain in a very roundabout way and are only a small component of the larger engine that causes rain. Then there is the problem of correct diagnoses. Some of the time, and nearly all the time until about one hundred years ago, the patient gets better in spite of the doctor or care given. As a white male, when I go to the GP with the sniffles, I get told go to bed, drink plenty of fluids (Glenfiddich doesn’t count) and just wait. Another problem is that Dr. McGinn only has symptoms typed in by whatever practitioner, in whatever way, from which to pull. Common diagnoses like the ones he is mentioning aren’t worth doing the search for, much less writing the software. So we conclude the Dr. McGinn is barking up the wrong tree.

A better idea is to focus on the difficult diagnosis, like the fictional Dr. House. Week after week for eight years he racked his brain to come up with a diagnosis that fit the symptoms. This is where the predictive model shines. Pick out the odd symptom and do a search for it. Look at those diagnoses. Computers don’t discern the quality of the match (usually anyway, look up full text indexes in SQL Server) but they will give an odd match that maybe Dr. House (or Dr. McGinn back in the desert of the real) hadn’t considered. Maybe we could type in all the observed symptoms and see if there was one diagnosis that some brilliant clinician had come up with and cured in a patient in Botswana. This requires a database that correlates discrete symptoms with discrete diagnoses (discrete means values you could pick from a list). The only database we have today that will do that is SNOMED_CT (go back in the blog archives or do a search for it) and Dr. McGinn’s fiasco isn’t built on that. The second thing we need is a web based EMR (again do the search or go back in the archives if you don’t know the term) that has a lot of world wide data to mine. Maybe Dr. McGinn has one large hospital. Maybe he has several. He needs them all.

So yes, this vision is already a reality and Sentia already has a tool with which to do the searches we described above. It is targeted more at managing chronic disease, but is just as effective at finding Dr. House (not Dr. McGinn) style correlations. We can literally generate a list of patients who are pregnant, with systolic blood pressure above 120 triglycerides between 120 and 150 and diagnosed with Dengue Fever. Or any of these. Or all of these. Or who do NOT have Dengue Fever. We tested for that and ruled it out. In fact, here is a screen shot of the criteria generation tool:



Notice the concept text box with the ellipses to the side. That pops up a search window where you can do a SNOMED search right on the page. So, yeah, Dr. McGinn, good thought, bad execution. Too bad we’ve been doing this particular piece of magic since 2009 and doing it better than you can. Don’t feel badly, we are doing it better than anyone else can too. Once we cut the cost of health insurance by 1/3 (that again?!?) and open our EMR to anyone who wants to use it for free, we’ll have all the patients there are world-wide. We already have thousands.