Our Leading Edge commentary, "The Impact of Artificial Intelligence-Enabled Tools on Electronic Health Record-Based Evidence Generation in Heart Failure," is out today in JACC: Heart Failure. The premise is simple: the EHR is an extraordinary repository of longitudinal data that we barely use as evidence, largely because roughly 80% of it is unstructured text and the variables that actually matter in HF (NYHA class, symptom trajectory, diuretic responsiveness, GDMT intolerance, frailty, reasons for referral) are rarely in a structured field.
We lay out where AI-enabled chart review is ready now and where it isn't, across observational research, registry abstraction, trial recruitment, endpoint adjudication, postmarket surveillance, and digital twins.
The near-term opportunity is deliberately unglamorous: clinician-supervised abstraction of shovel-ready tasks with defined criteria that can be audited against the source. We also take seriously the failure modes that don't show up in a validation paper: documentation bias, automation bias, and model drift. Deployment should be treated as the start of validation, not the end.
I am grateful to my co-authors (including mentors @ankeetbhatt , who introduced me to AI and NLP in medicine 4.5 years ago and @tremartyn ) and to the JACC:HF editorial team. @JonWCunningham @rcstarling @coconnormd @mpsotka @ProfMartinCowie @hfcollaboratory
#HeartFailure #AIinMedicine #HealthInformatics #LLM #RealWorldEvidence #JACCHF #HFCollaboratory #ClevelandClinic
We lay out where AI-enabled chart review is ready now and where it isn't, across observational research, registry abstraction, trial recruitment, endpoint adjudication, postmarket surveillance, and digital twins.
The near-term opportunity is deliberately unglamorous: clinician-supervised abstraction of shovel-ready tasks with defined criteria that can be audited against the source. We also take seriously the failure modes that don't show up in a validation paper: documentation bias, automation bias, and model drift. Deployment should be treated as the start of validation, not the end.
I am grateful to my co-authors (including mentors @ankeetbhatt , who introduced me to AI and NLP in medicine 4.5 years ago and @tremartyn ) and to the JACC:HF editorial team. @JonWCunningham @rcstarling @coconnormd @mpsotka @ProfMartinCowie @hfcollaboratory
#HeartFailure #AIinMedicine #HealthInformatics #LLM #RealWorldEvidence #JACCHF #HFCollaboratory #ClevelandClinic