For many physicians, the workday does not end when clinic hours are over. Late at night, they log back on to finish charts, respond to patient messages and tackle backlogs of administrative tasks that follow them home. The profession even has a nickname for it: pajama time.
At New York City-based Weill Cornell Medicine, AI is starting to take aim at that burden.
Virtual scribes are drafting clinical notes, and generative AI tools are preparing responses to the flood of patient portal messages that surged during the COVID-19 pandemic and never disappeared.
“Providers have spent a lot of time documenting clinical encounters, which is tedious and often forces them to take work home,” Adam Cheriff, MD, chief medical information officer for the Weill Cornell Physician Organization, told Becker’s. “That’s starting to change with technology.”
Dr. Cheriff frames Weill Cornell’s approach to AI in three lanes. The first is workflow automation: scribes, summaries, prior authorizations and other tasks that can be handled by machines instead of human hands. The second is diagnostics: algorithms that scan radiology images for patterns invisible to the eye, measure tumors with more precision or sift through genomic and wearable data to flag early warning signs of disease. The third is decision support: using predictive analytics to identify which patients are most likely to be readmitted or who might benefit most from care management.
According to Dr. Cheriff, the most immediate payoff is not financial but cultural.
“We have some early signals that providers adopting these technologies are capable of seeing a higher volume of patients or are able to take less work after hours,” he said.
Dr. Cheriff said future iterations may expand further — automating coding and billing at the proper intensity — changes that could directly affect revenue. Still, he is measured in his outlook.
“Our mandate in medicine is to do no harm,” Dr. Cheriff said. “That means being deliberate about which AI tools make it into practice.”
He recalled a recent example in which a clinician used a public-facing generative AI tool to answer a clinical question and the output was incorrect.
“The use case involved a fairly straightforward dosing calculation: converting milligrams into micrograms,” he said. “The AI delivered it incorrectly. And that obviously can lead to profound harm.”
Weill Cornell has been “reasonably conservative in the use cases” it pursues, he said, focusing on technologies that can be shown to be safe and effective. The health system also maintains a process of ongoing evaluation, ensuring that tools continue to perform as intended once deployed.
“You have to have a well-designed reference and evaluation process — and the discipline to keep evaluating — to make sure you’re choosing the right, safe use cases,” he said.
Weill Cornell is beginning to build AI literacy into physician training, teaching clinicians to keep a human “in the loop” and to apply judgment before acting on machine outputs. Dr. Cheriff sees AI not as a replacement for physicians but as a partner that can free them from drudgery.
“We feel this is going to be a transformative technology, and we want to leverage it,” he said.
The post The tech that’s shrinking ‘pajama time’ appeared first on Becker’s Hospital Review | Healthcare News & Analysis.
Health IT
