Sherlock Holmes would be proud of IBM's Watson computer. No longer the bumbling sidekick portrayed by Arthur Conan Doyle, the supercomputer has managed to master natural language skills, defeat Jeopardy contestants, and wow medical school professors with its potential to diagnose esoteric diseases. I suspect it would even make the late IBM President Thomas J. Watson, for whom the computer is named, proud.

Working with medical researchers at Columbia University, IBM is inputting data from medical textbooks and journals to create a diagnostic engine unlike any in the world. But it's likely we won't see a commercially available product for perhaps two more years.

Isabel Healthcare, on the other hand, has had an impressive, commercially available diagnostic tool up and running for many years and already has extensive research to show what it's capable of.

But before exploring those capabilities, it's worth stepping back a moment and asking: Why do physicians need a diagnostic aid? For one thing, there's just too much information out there for any one doctor to master. But there's also a less obvious reason.

In How Doctors Think, Jerome Groopman, MD, from Harvard Medical School, points out that physicians are prone to certain cognitive errors, and several of those can be addressed by information technology. In fact, one study that looked at 100 incorrect diagnoses found that only one in four resulted from a lack of medical knowledge. The other mistakes were the result of these cognitive traps.

"Availability" bias is one common mistake. A doctor who works in an inner city medical clinic that handles lots of alcohol abuse and the accompanying DTs (delirium tremens) is likely to jump to the conclusion that the next trembling alcoholic who enters the clinic has the DTs, even though a careful IT-assisted analysis would reveal several other possible causes for his shakes. In other words, doctors tend to lean toward diagnoses that are most available to them in their day-to-day routine.

Inside Watson, IBM's Jeopardy Computer

Similarly, an "attribution" error can lead to misdiagnoses. It's the kind of mistake that clinicians make when they stereotype patients--or attribute certain characteristics to patients without good evidence. Dr. Groopman tells the story of a retired merchant marine who was hospitalized with fatigue, a swollen abdomen, legs, and feet, and the smell of alcohol on his breath. A physical exam revealed an enlarged, hard, nodular liver.

The attending clinician diagnosed alcoholic cirrhosis, but Groopman points out that such snap judgments are often the result of doctors' repulsion for groups of patients they find offensive or indifferent about their health. In the case of the merchant marine, a more objective, in-depth diagnostic workup revealed the man actually had Wilson's disease, a genetic disorder that also causes liver damage and the rest of his symptoms. The man wasn't an alcoholic; he just enjoyed a glass of rum at dinner.

Enter Isabel's Diagnostic Engine

Isabel's diagnostic tool can help sidestep many of the cognitive errors clinicians make by providing a dispassionate analysis of a patient's record, pulling data directly from the electronic health record (EHR). Like Watson, it's capable of understanding ordinary English sentence constructions through the use of statistical natural language processing.

The software has the ability to analyze unstructured medical information and recognize patterns that correspond to specific concepts. The medical data is further processed using 40 algorithms, and the software can talk to the EHRs because it has the ability to speak the SNOMED language. SNOMED, or Systematized Nomenclature of Medicine, is the standardized clinical terminology recognized by most EHR programs.

According to Isabel Healthcare, the program has worked effectively with EHRs from NextGen, PatientKeeper, A4 Health Systems, Cerner, Misys, and Eclipsys.

Researchers from Stony Brook University have shown that Isabel's diagnostic skills generate tangible results. When they compared Isabel's final diagnoses of 50 complex cases to the actual diagnoses made by experienced clinicians, they found 74% of the computer's conclusions were correct when the patient's entire case--including medical history, physical examination findings, and lab test results--was simply copied and pasted into the software's data entry field. When each patient's information was entered into the program's key findings section, a slightly more time-consuming process, the success rate jumped to 96%.

My guess is Sherlock Holmes would be proud of both Dr. Isabel and Dr. Watson.

Recommended Reading: IBM Boosts Health Analytics With 'Watson' Supercomputer Capabilities Can IT Really Nail A Complex Medical Diagnosis? Dare To Question The Status Quo Is BI The Key To Personalized Medicine? Three Mobile Medical Apps That Turned My Head Does IT Enhance Patient Safety? Let's See Proof See more by Paul Cerrato

The Healthcare IT Leadership Forum is a day-long venue where senior IT leaders in healthcare come together to discuss how they're using technology to improve clinical care. It happens in New York City on July 12. Find out more.