I graduated med school 1976 did internal medicine mostly solo private practice for 43 years retired 2022 never had electronic records as I thought they wasted my time Im glad that AI is replacing the tedium of electronic records Im concerned re docs loosing their clinical skills due to AI Im more concerned that doctors will never get the clinical skills to begin with
what a great conversation! @erictopol you discussed how AI performs best within highly integrated systems such as UCSF, where longitudinal patient data sit inside a single unified record.
You both also noted that in much of the US (and globally) health data remain fragmented across multiple providers, limiting the interpretive reliability of AI agents embedded within any one institutional EHR.
This raises a related architectural question.
In many parts of the world without integrated health systems — for example India — patients effectively act as the integrators of their own medical history, traditionally carrying paper records between providers.
While imperfect, this highlights that the patient, rather than any institution, is the only continuous element across a lifetime of care.
As AI agents become capable of longitudinal reasoning, would a patient-centred model make more sense — where the primary health repository is controlled by the patient (a secure personal health record aggregating data from multiple providers), with AI operating at that layer rather than inside a single health system?
In other words, could AI shift the “source of truth” in medicine from institution-centred records toward patient-centred longitudinal data ownership, potentially improving completeness, portability, and personalization of AI-driven insights?
What's your take on whether this architecture is technically and clinically realistic, or whether issues such as data provenance, validation, and liability make institutional systems the unavoidable anchor point for medical AI
PhD Scientist here - found this wonderfully interesting and enlightening. Currently I work as a Scientific Review Officer (DoD) and we are only just beginning to use AI. Still, the personal touch in recruiting qualified reviewers and training them to write appropriate and through reviews is quite time consuming and could be streamlined. Really enjoyed this, thank you.
I graduated med school 1976 did internal medicine mostly solo private practice for 43 years retired 2022 never had electronic records as I thought they wasted my time Im glad that AI is replacing the tedium of electronic records Im concerned re docs loosing their clinical skills due to AI Im more concerned that doctors will never get the clinical skills to begin with
I'm no longer a clinician, but I found this podcast enormously interesting and enlightening.
what a great conversation! @erictopol you discussed how AI performs best within highly integrated systems such as UCSF, where longitudinal patient data sit inside a single unified record.
You both also noted that in much of the US (and globally) health data remain fragmented across multiple providers, limiting the interpretive reliability of AI agents embedded within any one institutional EHR.
This raises a related architectural question.
In many parts of the world without integrated health systems — for example India — patients effectively act as the integrators of their own medical history, traditionally carrying paper records between providers.
While imperfect, this highlights that the patient, rather than any institution, is the only continuous element across a lifetime of care.
As AI agents become capable of longitudinal reasoning, would a patient-centred model make more sense — where the primary health repository is controlled by the patient (a secure personal health record aggregating data from multiple providers), with AI operating at that layer rather than inside a single health system?
In other words, could AI shift the “source of truth” in medicine from institution-centred records toward patient-centred longitudinal data ownership, potentially improving completeness, portability, and personalization of AI-driven insights?
What's your take on whether this architecture is technically and clinically realistic, or whether issues such as data provenance, validation, and liability make institutional systems the unavoidable anchor point for medical AI
PhD Scientist here - found this wonderfully interesting and enlightening. Currently I work as a Scientific Review Officer (DoD) and we are only just beginning to use AI. Still, the personal touch in recruiting qualified reviewers and training them to write appropriate and through reviews is quite time consuming and could be streamlined. Really enjoyed this, thank you.