Lifetime Citizen Portal Access — AI Briefings, Alerts & Unlimited Follows
Health systems test AI, ambient listening and data standards to reduce clinician burden and identify rising risk
Loading...
Summary
Speakers from Intermountain, Cone Health and others described pilots of ambient listening, AI scribes, predictive models and a semantic data standard effort to improve real‑time visibility into patients, reduce clinician ‘pajama time’ and develop ‘next‑right‑action’ workflows for proactive care.
Panelists at the ASPE PTAC listening session described multiple AI and data initiatives aimed at reducing clinician administrative burden and improving real‑time detection of rising risk for patients.
Dan Wilchenquist, chief strategy officer at Intermountain Health, said the system has deployed ambient‑listening and AI tools to clinicians and is running roughly 70 AI projects. “Some of our doctors are really, really good at using macros. Haven't found a big difference in how they do coding. But for some of our doctors, it's saving them between 90 minutes and 2 hours of pajama time every day,” he said. Wilchenquist said Intermountain is piloting ambient listening for nursing that has reduced per‑shift documentation time and aims to lift bedside time from about 36 percent to roughly 41 percent.
Panelists described three technical aims: - Reduce clinician after‑hours documentation: Ambient scribes and AI drafting tools were credited with measurable time savings for physicians and nurses, and with improved documentation quality in pilots. - Real‑time situational awareness: Multiple speakers emphasized the need for near‑real‑time aggregation of clinical, claims and social‑determinant data rather than waiting for claims reconciliation weeks later. Intermountain described efforts with Graphite Health to create a semantic and syntactic clinical data standard and a “next‑right‑action” engine that translates many internal data tables to a cleaner dataset for near‑real‑time analytics. - Identify rising risk, not just historic high‑cost patients: Speakers warned that retrospective risk models often miss “rising risk” patients because they look in the rear‑view mirror; panelists said better, standardized, real‑time data are prerequisites for predictive models to identify those patients earlier.
Speakers also raised concerns and constraints: - Data quality and standards: Intermountain noted the difficulty of aggregating data across dozens of clinical systems and thousands of data tables; they described a workstream to normalize those tables into a common model inside their cloud environment. - Bias and representativeness: Committee members asked whether AI will underdetect needs for patients with limited prior system contact or whose data are sparse. Wilchenquist and others said addressing bias is an active concern, that clinician oversight is required, and that tool outputs will be subject to elevated scrutiny before deployment. - Risk and governance: Panelists said systems are piloting use cases (claims appeals, message drafting, triage guidance) and that clinical leaders must validate outputs; legal and governance review is necessary to control unintended consequences.
Ending: Panelists characterized AI as a necessary but nascent tool for reducing administrative burden and enabling proactive population health. They called for continuing investment in data standards, real‑time feeds and clinician engagement to ensure tools identify rising risk without amplifying bias.
