Neural networks have made some limited headway in specific situations, and in theory, it's possible, but not in the forseeable future. It actually has less to do with the doctor than with the patient. The patient with a complete, concise and sensible history is a rare bird indeed, and history-taking is a lot more art than you might imagine.
Computers would still be OK, if tedious, if the diseases played by the rules, but common illnesses often masquerade as uncommon ones, and vice versa. It would take one heck of a computer program to diagnose gonorrhea based on a few red bumps on the skin in the fingerwebs (I've done this more than once). Just yesterday, a mother was worried about her child's rash. He'd been treated with antibiotics for a sore throat. I looked at and felt the rash, and my first question was whether he'd had diarrhea in the previous few days. Bingo. Viral exanthem, easy for a human, cumbersome for a computer program.
Prescriptions are a lot less a part of the process than you think. That could be left to software once the diagnosis is clear.
More important is a description of what to expect. Many hospitals use computer-generated discharge instructions in their emergency departments, as an example. They tend to print out several pages that nobody reads. We know you're only going to hear about one word in five, and doctors are pretty good at getting across most of the major points individualized to the patient. Give us highly educated, literate, curious patients who think clearly when they're ill and stressed, and we'll write computer programs that do this better. Already one of the big problems with health care in the US is that insurers think the way you do, and that makes them think doctors should/could be "more efficient." Twenty years ago, family physicians were trained with the expectation they'd treat three or four patients per hour. Now they're expected to see six. How do you get fifteen minutes' conversation into ten minutes? I'd argue the human factor is much more important than the part that can be written in code.
· 1 decade ago