cancer care and computer diagnosis

A new NHS computer system will use a simple questionnaire to predict the probability of a patient having early cancer symptoms, allowing doctors to send them for screening and hopefully early detection. http://www.guardian.co.uk/society/2009/dec/29/cancer-diagnosis-computer-programme is interesting, and the system promises to save up to 10,000 lives a year in the UK. Great, roll it on, but let's be careful we don't throw the baby away when we change the nappy.

The risk in using computer systems for diagnosis is that the computer generally only gives a likelihood, but because the result came from the computer, people trust it more, and because of poor education, many people equate high probability with certainty. But a probability is just that. 90% certain is not certain, it is just very likely. If a technician interprets a negative as a no, then we have a problem. I have first hand experience of this problem in action. I picked up a DVT on a transatlantic flight a few years back. At the hospital, a nurse asked me a fixed list of lifestyle-based questions to determine whether I was in a high risk group. I wasn't, therefore as far as she was concerned, I couldn't have a DVT. Never mind that I had all the classic symptoms of a DVT, the computer said there was only a 7% probability I would get one, so I couldn't have. End of story, go home and die. I went home, a few days later the pain in my leg disappeared very suddenly, followed a short while later by symptoms that I know indicated an embolism - I've had one before. That confirmed it to me, that there must have been a clot. As far as I'm concerned, I definitely had a DVT, and should have had treatment for it, but the hospital wouldn't treat me because my lifestyle suggested I wasn't in a group likely to get one. I could easily have died for no reason other than being one of the 7% rather than the 93%. If  even get one again, it is unlikely to be any better, because of course my health record says I didn't have one, just because some doctors and nurses are crap at maths.

If computers are used properly, they can greatly improve health care, and this new system will certainly save many lives. But I would bet that the other side of that same story is that some other people will die because of it who would otherwise have lived, because some doctors will equate being in a low-risk group to not having the disease.