A system that asks patients about their symptoms and severity, and then orders appropriate tests, and generates an EMR could be GREAT for patients and doctors.
Too be honest, it's a billion dollar idea and I'm kind of sad I just gave it away on SDN.
I don't think you're the first person to have this idea; the idea of automating history/physical is probably the first thing somebody tried to do with a computer. The problem is history-taking relies on processes that, as of now, computers are pretty bad at.
It would be an easy problem to solve if a physical exam consisted only of questions about symptoms and severity (as well as a history), and followed a decision-tree based on answers. But from my (admittedly naive) understanding of the process, examination plays into the process as well, as do fuzzy, difficult-to-program variables like 'judgment.' Shortness of breath is an easy symptom to report, for example, but few patients will be reliable indicators of whether their SOB is due to wheezing or stridor. A person with a trained set of ears can make that determination pretty quickly and reliably, and adjust her hypotheses accordingly. A computer is forced to ask more questions, and rely on the patient's (dubious) responses. Furthermore, computers are likely to be limited to the patient's
complaints, which are not always commensurate with the patient's
symptoms; reports from patients often drive providers to look for other, non-reported symptoms that sometimes accompany certain diagnoses but not others. If you want to get a computer searching for patient's symptoms, you're back into the realm of a sophisticated AI system that can integrate spoken, visual, and auditory information with a diagnostic program.
This diagnostic program will also need some reliable 'judging' component that is able to include certain likely explanations for the finding but exclude others: you don't want a computer in Omaha ordering tests based on the suspicion that a patient has Malaria...except when you
do. Recognizing when something is plausible is pretty easy for a trained, experienced person, but formalizing this knowledge is very complicated and it's not the sort of ability that's easily transferred from one field to another. Plausibility of diagnoses isn't maintained across age, environment, condition, or case...how do you tell a computer the relevant information? I don't want to be the person who has to write the code that tries to formalize this sort of judgment.
I'm sure computers will play a larger and larger role in medicine in the coming decades (and I think that this is an exciting prospect), but don't fall prey to the computer geek's tendency to underestimate the abilities of a human being.