In the past few months, I have read about two ideas that may revolutionize medicine in the coming decade. They both have to do with data and data analysis, on a large scale.
The first one goes by the name of “Watson”. That is the name of an IBM computer system which got famous recently for beating human contestants on the game show “Jeopardy”.
The most impressive thing about “Watson” is that it can make sense of information that is written in plain language, with all the ambiguities that this entails. In order to be successful on Jeopardy, it had to analyze a vast treasure of plain text information, and then be able to understand correctly the game’s questions and reply to them by scouring all its information.
Now here is where medicine comes in: While winning Jeopardy was a great public relations success for IBM, in the future they want to turn Watson into a kind of super-powerful omniscient medical assistant, among other uses. Watson will digest the medical literature and then be able to come up with diagnoses once you present it with symptoms.
I was listening to a talk about Watson by the boss of that IBM unit recently (at the American Physical Society March Meeting 2012 in Boston). He recounted the story of a patient who had been saved only at the last minute because the patient’s symptoms had puzzled doctors and actually were due to a very rare condition that only one (human) expert finally interpreted correctly. Upon feeding these very same symptoms in to Watson, the computer came up with a list of possible causes, and the correct guess was among the top few options that Watson listed. It is no wonder that the guys at IBM are very enthusiastic about the prospects.
People have tried to build “expert” databases in the past, by carefully entering information into a computer system. However, the total amount of unprocessed plain text information out there (e.g. in the medical literature) is vastly greater. Therefore, a system like Watson can potentially become a very powerful assistant. Probably no human expert could have the same overview over the whole literature. Of course, a human medical doctor will still have to look at the options that Watson presents, for a sanity check, and to order relevant follow-up tests on the patient.
Here is the second development that is potentially revolutionary. One serious shortcoming of medical diagnosis is the lack of quantitative data. By this, I mean that only when you go to the doctor because you are feeling ill they will start to analyze your blood sample. However, the resulting values will be informative only if they fall drastically outside of some “standard interval”. In other words, you probably have to be already very sick. It would be much better if one had available some standard values for you, and not only for the standard statistical average for the “typical” patient. Better yet, it would be great to have time series information that shows how the values are slowly changing over time. Of course, there will be harmless fluctuations and drifts, but you might spot dangerous trends (or sudden changes) very early. If concentration X had been hovering around some value for a long time with only minor fluctuations, and then one day suddenly begun to slowly drift upward, you may catch this trend and reply to it long before it really gets dangerous.
A first step into this direction has been made recently. Read more about it in this New York Times article:
The geneticist Michael Snyder, who directed the study, was also its subject. At first, they did a genetic analysis for him. Then, every two months or so extensive blood tests were performed. Sometime during the study, he developed diabetes. Due to the good time-resolution and the extensive tracking of many different proteins, the research was even able to tell that the illness likely started when he caught a cold. A connection of this type had not been observed before. Since the diabetes was detected early on, treatment could also start at an early stage, reducing the harm.
At the moment, of course, this approach is still much too expensive for the general public. Each blood sample analysis costed about $2500. However, in the future the costs will likely be reduced, and one may also focus on only a smaller subset of the 40,000 molecules that were tracked in this study.
If regular blood tests were combined with the capabilities of an expert computer system like Watson, it is very likely we can detect problems much earlier than now. Of course, this raises once again the question what to do when diagnosis outperforms treatment options, and that will likely (apart from costs) be one of the most hotly debated issues in the years to come.