3 ms·

Based on threads like yesterday's https://news.ycombinator.com/item?id=34172092 https://news.ycombinator.com/item?id=34172092 , some vocal members of the HN cro

Based on threads like yesterday's https://news.ycombinator.com/item?id=34172092 https://news.ycombinator.com/item?id=34172092 , some vocal members of the HN crowd might prefer the opinion of "ChatGPTMD" to a board certified US physician (and perhaps with good reason in some cases).

rafaelero 4y ago
We should use both. Hopefully physicians will start integrating these technologies in their practice.
- willnonya 4y ago
  How could that be bad?
- Jerrrry 4y ago
  Doctors will soon use mdGPT just like they use Google today.
  - n8henrie 4y ago
    Definitely. I've sometimes dreamed of leaving clinical medicine to pursue a career helping integrate some "smart" helpers into EHR (ones designed to help physicians and patients instead of just billers and coders). With 10 minutes and scikit-learn I can get pretty good test-set accuracy for length of stay based on pretty bare bones input (time of day, mode of arrival, age / sex, bigram / trigrams of chief complaint). Given access to vitals, labs, historical diagnoses, a provider's (and their colleagues') historical practices, and a little PyTorch / HF transformers I bet there would be some low hanging fruit for suggested diagnoses and quick pick orders. Could in some cases streamline getting to the same end result, in others possibly make recommendations to bring practice more in line with current recommendations or to prompt investigation into uncommon presentations or rare diseases. And certainly would have some sharp edges and potential harm too. But based on HN reports of experiences with US medicine, seems like the potential upside may be worthwhile.
    - nradov 4y ago
      Go ahead and do it. The SMART on FHIR platform is now supported by most major EHRs and should make it fairly straightforward to write that app. If the data elements you listed have been entered into the patient's chart then they should be available through the API. https://smarthealthit.org/ https://smarthealthit.org/ The hard part is demonstrating the value of such apps to provider organizations, and getting them to change their workflow. Clinicians are extremely busy and don't want to click more buttons unless it improves patient care (and they get paid for it).
sinenomine 4y ago
Honestly, the board of M.D.s - the ensemble of upper-tier experts - is going to be the real tough benchmark for the years to come. Expecting to compete with this with only a few forward passes through a 0.5T-scale LLM is almost too hopeful.

rafaelero 4y ago

We should use both. Hopefully physicians will start integrating these technologies in their practice.

willnonya 4y ago
How could that be bad?
Jerrrry 4y ago
Doctors will soon use mdGPT just like they use Google today.
- n8henrie 4y ago
  Definitely. I've sometimes dreamed of leaving clinical medicine to pursue a career helping integrate some "smart" helpers into EHR (ones designed to help physicians and patients instead of just billers and coders). With 10 minutes and scikit-learn I can get pretty good test-set accuracy for length of stay based on pretty bare bones input (time of day, mode of arrival, age / sex, bigram / trigrams of chief complaint). Given access to vitals, labs, historical diagnoses, a provider's (and their colleagues') historical practices, and a little PyTorch / HF transformers I bet there would be some low hanging fruit for suggested diagnoses and quick pick orders. Could in some cases streamline getting to the same end result, in others possibly make recommendations to bring practice more in line with current recommendations or to prompt investigation into uncommon presentations or rare diseases. And certainly would have some sharp edges and potential harm too. But based on HN reports of experiences with US medicine, seems like the potential upside may be worthwhile.
  - nradov 4y ago
    Go ahead and do it. The SMART on FHIR platform is now supported by most major EHRs and should make it fairly straightforward to write that app. If the data elements you listed have been entered into the patient's chart then they should be available through the API. https://smarthealthit.org/ https://smarthealthit.org/ The hard part is demonstrating the value of such apps to provider organizations, and getting them to change their workflow. Clinicians are extremely busy and don't want to click more buttons unless it improves patient care (and they get paid for it).

sinenomine 4y ago

Honestly, the board of M.D.s - the ensemble of upper-tier experts - is going to be the real tough benchmark for the years to come. Expecting to compete with this with only a few forward passes through a 0.5T-scale LLM is almost too hopeful.