According to a presentation I heard two month ago it's not good enough yet. They can make accurate predictions in many cases, yet not in all/enough. At the moment they see more potential in analyzing health insurance claims, as falling back to a human expert is easier.

Watson is already better at diagnosing some cancers better than human doctors.

http://www.wired.co.uk/news/archive/2013-02/11/ibm-watson-me...

Do you still have a link to the presentation ? It sounds interesting.

I'm sorry, it was a presentation of an IBM researcher during the BioASQ workshop. Perhaps you can find the slides/a summary on the BioASQ.org website.

My father's life was save by early cancer diagnosis a couple years ago so it is a topic that I care about. But it's a complex and difficult topic.

I find it fascinating how information technology continually seems like something that could save/improve/revolutionize healthcare but which so far consistently fails and not for a "we couldn't implement the right algorithm" kind of way.

Just in the cancer topic, as I understand it, cancer now is either found by blood-test plus further examination method or by a doctor or patient "just feeling something is wrong". In the case of a lab result any improvements are going to be marginal until there are tests which taking into individual physiological/genetic differences.

In the case of "something is wrong", once the flag is set, I believe you have a reasonable number of ways to check things (though without enough symptoms, you often have too many things to check, something that AI may be able to "do better at" without doing significantly better since you can't information from a brick wall any more than you can squeeze blood from a stone). The "flagging something as wrong" thing may be the most important but I strongly suspect that this part comes from doctors and patients simply having more information than is communicated by any list of symptoms.

I mean, just in general, consider that "first world medicine" might be less useful for third world people who have a widely different constellation of health problems.

And the discussion could go on and on.

Has anyone tested whether a crack team of human doctors who only get their information discreet database formats could be useful?

Imagine making a portable device that use watson to analyze your blood and health, you could give your self an medical exam every week.

Well, even allowing that its feasible in the medium term it would give you worldwide 'first-world' diagnosis, not treatment.

For most maladies, the treatment is some combination of pharmaceuticals, lifestyle changes, or diet changes. The latter two can be undertaken anywhere, so that leaves pharma.

Name-brand pharmaceuticals are expensive, but in almost all cases, there are less expensive options available that are almost as good. Most of these are off-patent, can be manufactured anywhere on earth, and can be acquired for pennies.

An intelligent diagnostic / treatment computer would know where the patient was located, their access to medical facilities, their access to drugs, and make a recommendation based on those qualifiers.