Can Medical Schools Funnel More Doctors Into the Primary Care Pipeline?

(Spoiler - no.)  Medical students respond to both money and regard, and that makes specialization much more attractive than primary care.  Specialization is also, in a lot of fields, a guarantee of employment - hello IR - an important factor when you are graduating with a pile of nondischargeable debt.  Tuition-free medical schools can help with the financial motivations behind specialization, but not the regard factor. 

That the CMS and other insurers have been ratcheting down payment for primary care for years now, the loss of private practice as the ACA encouraged consolidation and ended privilege and feeder systems for hospitals has had a big impact as well.  If you want more primary care physicians, you need to change how they are paid, and change legislation and regulation to encourage private practice.  You also have to start recruiting people into the primary care pathway in college (or even high school).

Can Medical Schools Funnel More Doctors into the Primary Care Pipeline?

‘Striking Paradox’: AI Doctor Passes Medical Exams, Fails with Real-Life Diagnosis

If you’ve been reading our AI summaries, this is not a striking paradox. The failure of the AI to ask appropriate questions, expand diagnoses, or deal with extraneous unrelated information is well-known. At this point, most of the basic LLM models can pass a written USMLE exam and score well on standard certifications. They do not handle all sorts of imaging diagnostics well—those are CNN models (convolutional neural networks) rather than LLMs (large language models). 

It’s another example of how not to use an AI, which, at its little digital heart, is a search and summarization engine. We shouldn’t be looking at AI to compete with humans on diagnostic capabilities; we should be looking at these early models to exclude diagnoses and answer questions like “Should this lesion be biopsied?” or “Look at these 10,000 records in 5 minutes and see if the correct panels were pulled for the listed condition.” 

Conversationally, the LLMs are also not well-suited to explore diagnostic conversations with patients (except for the fact that they speak dozens of languages near-perfectly). That particular lack of curiosity is something that can be overcome with prompting or programming. They do an excellent job of explaining follow-up care or laying out conversations about upcoming procedures. 

Bad thinking about how to implement AI as a diagnostic and training tool is a pet peeve of ours, as you can see. 🙂