Confusing, at times, but generally well-reported. I expect it should be confusing due to the unfamilar terms (graphs and such) but as a person interested in keeping things honest I'm glad I read this. I think I'd rather be in the shoes of a person who ends up living longer than expected in this situation, but then again, asking doctors whether they'd be surprised whether someone lives past some arbitrary period of time is not the same as saying their life expectancy is that same period.

Semantics? Maybe. But think about how it actually gets reported. Is it:

"We expect you to live x months."

"We'd be surprised if you died before x months."

The difference is that the doctor is trying to give someone an estimation of likelihood, which is not a simple flip of a coin. To put it another way, SURPRISE is not something we expect to happen half of the time. Surprise is having a coin you believe to be weighted come up the "heavy" side half the time. Put still another way, if a doctor expects a patient to live between 2-8 months longer, that would still be the equivalent of "surprised if he died within 3 months" because on average, the prediction has 5 months as its midpoint. We can't expect that level of granularity from physicians, yet we DO.

I would never call myself surprised if either the heads or the tails side of a coin came up, nor should such an interestingly binary question be used for this.

What about just saying, "You might live past three months, but I wouldn't bet on it."

Is the doctor "surprised" or "not surprised" if a patient dies in 2 months in that case? This study seems an interesting example of Monday morning quarterbacking: we can model what it looks like for the cold and unfeeling computer to have accuracy or lots of false positives but not both, whereas with a doctor people will be litigating just as soon as the doctor estimates wrongly and the patient or patient's family see case for a lawsuit.

Next up: suing computers and the engineers who build the algorithms.