There’s smart, and then there’s Watson smart. IBM Watson, the computer system that IBM has invested over a billion dollars developing, started with the promise of solving all the world’s problems quickly and efficiently. The IBM Watson Health program, which started three years ago, was finally unveiled just a few months ago with the capacity to provide health and management recommendations for patients with a dozen or so oncology conditions. The MediOrbis team was excited to preview, quite literally, the best that computer science had to present to the medical community.
Our preview of IBM Watson Health’s platform was both inspiring – seeing the depth and complexity of computer learning – and disappointing, as we witnessed the limitations that even the most sophisticated computer systems possess.
To start with, the platform is impressive in its scope of input data. When provided with electronic medical records, Watson scans and filters the data, distilling it down to only the most important and clinically relevant information. Watson can analyze the myriad of patient specific variables, including demographics, location, and gender/race, in conjunction with every single important medical component of the patient and disease in question, and map it against a database of thousands of articles in the published literature. Watson then provides a list of the preferred diagnostic steps in sequential order, and concomitantly suggest first, second and third line treatments for the specified malignancy. All recommendations are fully supported by literature citations, which can be easily accessed via links provided on the page. Wow. Imagine a personalized clinical summary and UpToDate article for each patient!
This is quite the accomplishment and I do not intend to minimize its importance. It is the result of the highest level of collaboration between technology and software programmers, and the top clinical and medical minds. At the same time, there are some shortcomings to the program. First, it is useless in the absence of electronic records. While I personally despise hand – written notes, I wouldn’t say that they render my clinical skills non-existent! Additionally, the computer’s algorithm requires that all clinical variables be entered and accounted for. There is no reliable way to obtain a recommendation or second opinion if components of the workup are missing. Lastly, and perhaps most importantly, the recommendations do not account for patient specific social or economic factors. Patients have their own specific needs, desires and considerations. Their appetite for a multitude of issues varies: their level of aggressiveness regarding their care, their willingness to be treated with experimental or novel therapies, or whether to consider a clinical trial are just a few.
Another piece to consider is that the clinical data Watson analyzes is derived exclusively from peer reviewed articles and journals. Conference abstracts and presentations are ignored.
The IBM Watson Health Platform is extraordinary in its breadth and depth of knowledge and its ability to streamline, simplify, assist and guide clinical care. Ultimately, its value lies in its potential to continue to be developed into an even more powerful clinical tool, to increase its scope of patient specific considerations and diseases for which it is relevant, and to be incorporated routinely into clinical care. As is often the case with innovative medical programs, the collaboration between man and machine is crucial, and the input of both is essential to best care for our patients.