EPtalk by Dr. Jayne 6/6/24

ONC is seeking feedback on a recent white paper that highlights its vision for Health Equity by Design. The organization is formulating strategies to reduce healthcare disparities by including health equity throughout the creation and implementation of health IT policies, programs, and workflows. The approach aids in identifying gaps and disparities and creates an environment where “technology itself anticipates, avoids, and reduces, not exacerbates, health disparities.” Additional outcome goals include mitigating systemic inequities and improving person-centered decision-making, implementing population health interventions, and strengthening public health. Comments can be provided through June 10 at 11:59 p.m. ET.

Having worked in emergency and urgent care settings for the majority of my career, I’ve seen how the lack of easily accessible mental health services adds to overcrowded waiting rooms and delays patients from receiving appropriate care. A recent initiative in Oklahoma equips police officers and Certified Community Behavioral Health Clinics with IPads for telehealth visits, helping reduce the number of emergency visits and hospitalizations for mental health issues. More than 30,000 devices have been deployed to the field, allowing faster patient interventions in a less resource-intensive manner.

Police officers no longer have to wait with patients at the emergency department, and patients have a shorter wait time for lower-acuity care that better matches their needs. The statistics are impressive – a 93% reduction in inpatient hospitalizations for mental health crisis over a six-year period, and $62 million in savings. The program has expanded to provide IPads directly to behavioral health patients, further reducing the need for costly interventions. I hope other states, counties, and cities take note of this program and consider implementing it in their own areas.

A project for one of my clients led me to dig into the new final rule from the Office for Civil Rights (OCR) and CMS that is designed to prohibit discrimination in AI based on data points such as age, gender, race, and ethnicity. It’s great to have regulations, and this one in particular brings 558 pages of PDF joy, but it’s unclear how this will be enforced. AI bias can be difficult to detect, and when identified, there’s a chance that organizations will be subject to their own biases in determining how to address it. The presence of a formal rule opens the door for whistleblowers and reports of problems from end users, which should help keep the industry honest.

One of the important elements in the rule is the definition of the term “patient care decision support tool” as “any automated or non-automated tool, mechanism, method, technology, or combination thereof used by a covered entity to support clinical decision-making in its health programs or activities.” That definition encompasses everything from EHR-embedded AI clinical decision support to paper checklists found at the bedside, and everything in between. Hopefully this will serve as a catalyst for organizations to ask some questions about tools they have in place or are considering, including reviewing the data being used to trail the model or validate the tool, making the tool’s decision-making process transparent, identifying how people will be involved in the implementation and monitoring of tools, and describe the steps that will be taken if there is a suspicion that harm has occurred.

With that in mind, it’s timely that Epic Systems has released a new “AI Trust and Assurance Suite” that is designed to help clients test and monitor their AI models. According to announcements from Epic, the software is designed to automate data collection and mapping and to ensure consistency. Since one can’t really see Epic’s documentation unless one is an Epic customer (or someone violates all kinds of rules by slipping one a copy,) it’s unclear how this tool will work for the numerous Epic clients who have custom fields and their own unique ways of using data.

Epic says it will release the tool’s monitoring templates and data dictionaries as open-source software this summer, which should help clients who have custom AI models or who are using tools from third parties. Still, that’s a significant burden on clients who will have to analyze the tool and its functions carefully. I doubt many organizations have analysts budgeted to address it, so we’ll have to see what the speed of uptake looks like.

I’ve been mentoring a resident physician who is considering a fellowship in clinical informatics. One of our recent conversations was around the role of generative AI in academic pursuits. Most organizations I have encountered have come to the conclusion that they can’t prevent students from using the latest and greatest digital tools, but that guardrails need to be in place to preserve academic integrity. Oregon Health & Science University Professor William Hersh, MD has a clear policy for the “Introduction to Biomedical & Health Informatics” course that provides guidance to students. Key points include that generative AI systems can be useful tools but should not be used to substitute one’s own knowledge; and students can ask generative AI systems for content, but final responses — including those in discussions, quizzes, tests, and term papers — should reflect the student’s “own thinking, judgment, and language.” It is also noted that students shouldn’t shortchange their learning by relying on generative AI, and that the need for a fundamental core of knowledge and understanding is needed by practitioners in the field.

Pediatric dermatology researchers are celebrating their victory over ChatGPT as detailed in a recent study that looked at accuracy rates on board-type questions. They tested ChatGPT versions 3.5 and 4.0 using questions from the American Board of Dermatology as well as “Photoquiz” questions from the journal Pediatric Dermatology. Although ChatGPT 4.0 gave human pediatric dermatologists a run for their money in some areas, the humans outperformed both versions overall. Researchers call on clinicians to understand the tools and how they might be helpful in practice.

The Institute for Safe Medication Practices has released additional guidance on actions needed to prevent drug name selection errors when facilities are using automated medication dispensing cabinets. Although vendors have taken steps to improve their products, some features require a customer to opt in to the newer safety features through manual configuration or software updates. The Institute is calling on vendors to support dynamic search functions and standardized medication names, and for care delivery organizations to educate staff, analyze workflows, and require indications for certain overrides.

I spent some time this week helping teach CPR to a local youth group. Attendees were amazed at how much easier it looks in movies and television. Some of the participants were smaller physically and couldn’t generate the force needed to do effective compressions, but they were great at recognizing the signs that CPR is needed and demonstrating how to take charge of a scene. If you’re not certified in CPR and use of the AED (automated external defibrillator), consider taking a class. At minimum, consider learning about hands-only CPR from our friends at the American Heart Association and identifying where AEDs might be kept in your daily travels. Bystanders recently initiated CPR at my local Costco. Will you be ready if the time comes?

Email Dr. Jayne.