The general approach to the differential diagnosis has remained unchanged since the dawn of modern medicine. A doctor sees a patient, gets some initial information, and then creates a list of what they think are the most likely and dangerous possibilities. This list of possibilities is created by comparing the present patient to others that the doctor has seen and/or querying his/her memory for disease patterns that they have read in textbooks.
This methodology can lead to errors for a variety of reasons. First, a doctor can neglect to consider disease possibilities that they have not seen or read about before. This happens often as there are thousands of diseases and it is just humanly impossible to remember them all. Furthermore, a single disease can present differently, and another common error is to not consider a less common presentation for a common disease.
For example, I was the nephrologist on call a number of years ago and was called to see a hemodialysis patient, Mr. P, who was having severe acute abdominal pain. The nurses had drawn some labs and he had a very elevated white count. His abdominal skin itself was acutely painful to even light touch. He also had a reticular rash over the area.
A CT was negative. I asked general surgery to assess, but they did not think he had an intraabdominal process. I, therefore, thought he had sepsis. He had a white count and sepsis is common in dialysis patients. His skin could be the source of infection and the cause of the pain. Or maybe it was SIRS from another infective source and he had mottling of the abdomen and ischemic pain from a low flow state. So I decided to give him antibiotics and admitted him to the hospital. But over the next few days, he wasn’t getting better and the abdominal pain remained excruciating despite high dose opioids.
What did Mr. P end up having? It was calciphylaxis, an acute/subacute vascular occlusive process that leads to severe skin necrosis from calcification within small vessels. His pain improved with appropriate therapy, but he was on inappropriate therapy and in pain on the ward for days before I was able to reexamine and modify my original diagnosis. Why had I not initially made the correct diagnosis? 1) I had never seen abdominal calciphylaxis before, I had only seen it in the legs and 2) I had always thought calciphylaxis presented with either necrotic skin lesions or subcutaneous nodules; I had never seen other presentations, such as a reticular rash.
Was I a bad doctor for not knowing these symptoms of calciphylaxis? Should I have clued in sooner? No, I’m human and it’s just impossible to remember all this stuff! Thousands of diseases multiplied by thousands of symptoms equals tens of thousands of unique data points that must fit into one, error-prone, usually sleep deprived and crunched for time brain. Diagnostic errors will inevitably occur and this is partly why patients die or are seriously harmed per year in the US alone from diagnostic error.
To really illustrate the impossibility of memorizing this data, look at the graph below (click here or the image to view it interactively). The nodes of the graph represent 2,904 diseases and symptoms, in addition to the 16,511 causal connections between them all. I manually collected this data over the past 7 years as I just couldn’t believe that there were better databases for baseball than there were for diagnostic data! The database is still being built, but, as you can see, even without all of human disease entered, there is an immense amount of information that is just impossible to keep in a human brain.
Only a searchable electronic database can allow human doctors to remember the information needed to avoid diagnostic error. The database I have built is called docLogica and is available on the web and on smartphone apps. It also contains test accuracies (likelihood ratios) as it’s equally impossible to remember the accuracies of all the tests! To see how this database could have helped Mr. P, observe the video from the iOS app below leading the user to the correct diagnosis. In a nutshell, you can give the app the patient’s parameters (age, sex, presentation acuity, symptoms and comorbidities) and the app will find the subset of disease that fits that criteria, arranged by population incidence. As you can see, calciphylaxis floats to the top when the patient has the renal failure comorbidity. The database adjusts for the fact that calciphylaxis is more common in renal failure than in the general population.
I cannot imagine diagnostic medicine evolving or improving without such a point of care tool. Give it a look and help bring diagnostic medicine into the 21st century.
This post was copyedited by Matthew Sem