AI assisted medicine

In 1963, a famous neurologist, Professor Shigeo Okinaka of Tokyo University has published at his retirement lecture that the misdiagnosis rate throughout his career as a doctor was 14.2%. He had been eager to have made autopsy for the patients in charge, when they died, to confirm his own diagnosis for them. Of course, the diagnosing means, such as laboratory exams or diagnostic imaging, were limited those days and the concepts of diseases were different from now. But his misdiagnosis rate was accepted by doctors with much surprise. It was a surprise that such an excellent neurologist as Okinaka could bring on such a rate. In my med student days, this anecdote was often told as a lesson to study hard of the diagnostics.

Reaching correct diagnosis is always the very beginning of medicine. It determines whether a patient would go dead or alive in most serious cases. The more important it is for a patient, even the more essential it is for a doctor. 

 

In the recent issue of Science, Eric J. Topol of Scripps Research Institute has written how accurate diagnoses could be made when generative AI is introduced in the diagnostics in medicine. 

https://www.science.org/doi/10.1126/science.adn9602

As introduced in this article, there are possibilities of misdiagnosis in two aspects of clinical medicine. Clinical doctors at the front of medical service are always diagnosing their patients in System 1 thinking way. Without laboratory tests or diagnostic imaging, the doctors should give a diagnosis, even if tentative, to the patient before him/her. Automatic, near-instantaneous, reflexive and intuitive manner. In Japan, under the system of the socialistic medical economics, most of the doctors could not spend much time for each patient. Generative AI may benefit the doctor as well as the patient since there would be less chances for the bias of the doctor in charge. 

On the other hand, deliberate, thoroughly exmined consultation, which only few patients could receive at exceptionally specialized institutions at present, may depend on System 2 thinking. This could be supported by generative AI as well. It may give the updated knowledge of medicine to the doctors which is essential for the correct diagnosis. This may happen less frequently than the former cases but may determine the prognosis of the patient.  

Generative AI sounds very helpful to us in clinical services. There could be, however, two issues, so far as I understand, when generative AI is introduced into the clinical medicine. 

The first is who would pay for the cost of AI. At least, the present system in medical economics would not afford that for routine use. Could only limited number of wealthy patients pay for that? In our country, MHLW seems to replace the area the publich health insurance covers to that of private insurance. The people is told they would have more chances for better medicine in that private insurance covering area. They don't know how much private insurance costs everyone.

The second issue is who would take the responsibility of misdiagnosis by generative AI. In our country, they won't bring such a case to court so often. With system being americanized, there should be more cases of law suits. Who should be responsible for such a case?

As a whole, however, it should be a progress in the medicine. It would be accepted and used properly in clinical medicine in the near future. At least, much better than development of the AI controlled arms.