Becker's Hospital Review

33 CIO / HEALTH IT Epic's Rover and Haiku apps on iPads. But the organization's active directory service and virtual desktop infrastructure network were down for 12 hours. All outpatient appointments and nonemergency surgeries were canceled July 19. Seattle Children's restored its systems by July 22. Aer the outage was discovered overnight on a ursday into Friday, health systems around the country similarly worked over the weekend to get operations back up and running by the next Monday morning. "rough a massive and nonstop all-hands-on-deck deployment of resources, hospitals and health systems rose to the occasion and have made tremendous progress in restoring mission-critical systems and patient care services," said John Riggi, national advisor for cybersecurity and risk at the American Hospital Association. "Some effects, although diminished, still continue — and the true impact to hospitals and health systems may not be known for weeks." In Arizona, Banner Health Chief Clinical Officer Marjorie Bessel, MD, received a call late the night of July 18 local time about the outage. Dr. Bessel is the Phoenix-based health system's incident commander for emergency response, a role previously utilized during the COVID-19 pandemic. e organization does tabletop drills for these types of cyber incidents. Banner had systemwide calls to update the situation every hour for the first eight hours, every two aer that. "is is not just your electronic medical record being down," Dr. Bessel said. "e tube stations that we use to send drugs up from the pharmacy and send specimens to the lab — down. Security alarm systems — down. All pharmacy systems — down. It's not the same as Epic or Cerner going down. It's way larger than that." Banner Health decided not to open its ambulatory locations July 19. e health system also delayed surgeries July 19 but was able to complete them later in the day, about three or four hours past schedule, aer getting systems up and running about 16 hours aer the outage was discovered. "You've got to be willing to make decisions that might turn out not to have been the best decision, retrospectively, but you make the best decision that you can, and you get everybody behind you," Dr. Bessel said. "I just want to say kudos to everybody in the healthcare industry, not just our team members at Banner Health, who, on the tails of very difficult pandemic a couple years ago, once again rose to the occasion, worked all night, and did everything they could to keep patients safe as well as provide the ability for them to get ongoing care." At Duke, IT staffers had to remedy the computers one at a time. It took about five to eight minutes to manually decrypt each machine, delete the CrowdStrike file and reboot. Operations leaders went around and put yellow sticky notes on the computers to prioritize. e goal was to get 50% of the workstations up and running as soon as possible, allowing patient care to be unaffected. Clinicians went back to paper records in the early going. Of the health system's over 60,000 computers that use CrowdStrike, 40,000 received the bad update. Of those, 18,000 had their screens go blue. e other 22,000 hadn't yet been restarted (the health system notified staff not to reboot their computers if they hadn't already). Less than 1,000 nonclinical machines remained unfixed as of July 24 (laptops belonging to people who have been on vacation, for instance). "e question every CIO in the country has is, 'Why was this pushed out so broadly'?" Dr. Ferranti said. "Why wasn't it pushed out to a smaller group to make sure everything's OK and do a rolling update? at's just blocking and tackling." Becker's reached out to CrowdStrike for comment. e company's CEO has been asked to testify before Congress. "ose are the kinds of things that are going to have to be commonplace in the industry," Dr. Ferranti added. "e stakes are just too high to be pushing things out at that scale. We saw over the last four or five days what can happen from that."n Can ChatGPT help with EHR interoperability? By Giles Bruce Can generative AI help improve healthcare data interoperability? Researchers from Evanston, Ill.-based Northwestern University think so. They authored a new study that found GPT-4, the large language model employed in ChatGPT, to be effective in translating EHR data into standardized form (known as Fast Healthcare Interoperability Resources, or FHIR). "This is going to greatly accelerate the pace of breaking down the walls between different health systems that hinder the aggregation of health data and the exchange of the data for performing large-scale research," said Yuan Luo, PhD, chief AI officer at Northwestern's Clinical and Translational Sciences Institute and Institute for Artificial Intelligence in Medicine, in an Aug. 7 statement. The researchers' FHIR-GPT model had a better match rate for EHR data than existing tools for medication statements, medication timing schedules, medical forms, reasons for medication administration, dose quantities, and medication administration routes, according to the study published July 19 in NEJM AI. The authors say they plan to further validate the tool before it can be adopted by health systems across the U.S. n "You've got to be willing to make decisions that might turn out not to have been the best decision, retrospectively, but you make the best decision that you can, and you get everybody behind you." — Marjorie Bessel, MD

• Cover