As a senior clinician, I am aware of the gaps in the care of older adults with serious illness. This was brought home recently when I began caring for a woman in her early 90s who was transferred to our facility for stabilization of recently decompensated heart failure, so she could have a Watchman procedure, stop her DOAC (Direct-Acting Oral Anticoagulant), and monitor a recent rectal bleeding problem. The cause of her rectal bleeding was not clear from review of the recent hospital records, and this patient and her family did not know whether her prior cancer treatment of radiation and chemotherapy was palliative or curative. She did acknowledge she was experiencing worsening incontinence with bloody diarrhea, and our nurse noticed a similar discharge from her vagina. Fortunately, I was able to reach her oncologist, who agreed this was terminal cancer. Her cardiologist agreed it was time to stop the DOAC, and she was discharged home with the support of hospice.
Fast forward two years: will it be possible to use a large language model in EPIC to ask for an update on the status of this patient’s colorectal cancer? The CCCC (Coalition for Compassionate Care California) has hit a “Home Run” on the current and future role of AI to ensure patients with serious illness receive the care that matters to them. They have created a 4-part series that set up the final presentation in July, which focused on AI in Advance Care Planning (ACP).
Charlotta Lindvall at Harvard highlighted the importance of creating structured common language data in the EHR to permit filtering of important ACP information from the medical record. However, most clinical notes are free text. She cited work from LLMs (large language models) showing their superiority in identifying ACP in patients with cancer, and doing it far more efficiently and accurately when compared to clinicians. Some of these LLMs are now able to create summaries of patient illness understanding, shared prognostic information, what is important to patient/family, psychological support, and family support.
Matthew Gonzales, MD, CMO at Providence Institute for Human Caring, shared how AI had dramatically improved GOC (Goals of Care) conversations in Epic and had improved the quality of those conversations. Using Microsoft Azure OpenAI GPT 4.0 with chunking, they have been able to aggregate information into the front page of EPIC under 4 tiles (POLST, Advance Directive, Code Status, and GOC). AI audits the quality, accuracy, and equity of this information with a high degree of specificity, precision, and accuracy. Their success in implementing AI at 4 hospitals has gained approval to roll this out at all 52 hospitals this September.
City of Hope is an NCI Comprehensive Cancer Consortium of 4 hospitals with 37 locations and 1 million outpatient visits annually for > 255,000 patients. Finly Zachariah, MD presented on AI Scribes for ACP, Hospice, and Palliative Care. They have piloted DAX Copilot (Microsoft AI) and have found that > 85% of clinicians save time, and more than 60% saved > 30 minutes/day. He provided multiple resources for improving the quality of their AI recordings with the goal of training it to “think like me.” Their ACP model has ACP prompts to quickly allow a revised, improved note. By 6 months’ use, he found DAX Copilot reduced his documentation time per note by 50%.
Jonathan Handler, MD, Senior Fellow, Innovation at OSF Healthcare in Chicago, presented on the use of AI to promote the best possible end-of-life care for each patient. They identified 13 inputs to help them predict the 90-day mortality for their inpatients. Using AI, they were able to abstract this data from the EHR, which then helped their clinicians adjust their care plans to be concordant with prognosis.
As a semi-retired clinician in private practice, I am not ready yet to jump into the water as deeply as these presenters’ health systems have, but I can certainly see the value and am open to participation as it becomes more accessible and user-friendly. The gaps in our current care capacity are obvious and demand improvement.
I am pleased that CCCC has partnered with CALTCM to make this series available at a very reasonable price to non-members, and free for CCCC members.
As an older clinician, I am cautious about adopting AI. With that said, I am enjoying exploring what is has to offer. My training (psychologist) in charting was mostly OTJ. Depending on the location I was providing services, the expectation of what to include often varied. And, as an older clinician, I can still remember the days of hand-written notes. Charting evolved, and I along with it. What I also noted was the shift toward charting to meet the needs of payors rather than charting to meet the needs of sharing patient data with other professionals. AI has capacity for bridging these extremes (and is legible!). But the challenge remains in how the data are gathered. Checklists and prompts are only as good as the programmer. Without acknowledging the clinician's input (clinical intuition?), capturing nuances can be lost. But I am really enjoying seeing what possibilities lie ahead for implementing this tool clinically.