Which questions does machine learning answer well or badly? ML models have lacked the ability to label confidence intervals or quantify statistical certainty when producing results. Michael Jordan and colleagues introduce “prediction-powered inference,” a standardized protocol for constructing valid confidence intervals and P values to support responsible and reliable inference.

A. Jay Holmgren and colleagues find time by physicians in the Electronic Medical Record is up sharply, both during and after patient scheduled hours. UCSF providers spent nearly 5.5 of 8 patient scheduled hours working in the EMR. Health systems and policymakers may need to adjust both productivity expectations and reimbursement policies.

Jonathan Kolstad and Berkeley Haas colleagues have created Healthpilot–a free, online platform that uses machine learning to weigh complex personalized factors and provide Medicare consumers with recommendations for better, and often lower cost, coverage options than those offered by insurance brokers.

With a $17.5M grant from the CDC, CPH faculty and colleagues are partnering with Kaiser Permanente Southern California to launch a new center boosting California’s capacity to respond to infectious disease outbreaks. The California Center for Outbreak Readiness will develop cutting-edge analytics, tools and platform for emergency health response.

As medical AI tools are deployed in the hospital, their accuracy may falter as patients do better. Adam Yala comments on the limitations of predictive models and the needed work to improve them.

In this Q&A, Irene Chen describes how machine learning could help clinicians identify and support victims of intimate partner violence, her work on machine learning for equitable healthcare, and a related upcoming event at UCSF, Toward Algorithmic Justice in Precision Medicine.

JAMA Editor in Chief Dr. Kirsten Bibbins-Domingo, in conversation with Ida Sim, Professor of Medicine, UCSF and CPH co-Director, about the promise and perils of AI in the clinical encounter, and how to keep humanity at the center.

Host Dr. Christine Sinsky, AMA Vice President of Professional Satisfaction, and guest A Jay Holmgren, Assistant Professor of Medicine at UCSF, and the Center for Clinical Informatics and Improvement Research, discuss how assessing Epic Signal data and other EHR metadata can be a useful tool for evaluating team-based interventions to reduce inbox volume.