How to Drive Department Collaboration and Physician Adoption: Dr Howard Dinh’s Quest at Kaiser Permanente
A validated decision tool only helps patients if clinicians open it, trust it, and act on it. Getting them there, and deciding which tools are worth building at all, is its own kind of work, and it gets far less attention than the equations themselves.
That work is the subject of the latest episode of MDAware: Upgrading Clinical Judgment. Dr. Graham Walker sits down with his former TPMG colleague Dr. Howard Dinh to talk about how Kaiser Permanente Northern California builds and maintains decision support across one of the largest integrated health systems in the country, what actually drives adoption, and where AI fits in.
Meet the Guest
Dr. Howard Dinh is Chair of Cardiology for Kaiser Permanente Northern California and leads the region's clinical decision support subgroup, which builds custom decision support workflows for clinicians across KP's Northern California medical centers. Before joining Kaiser, where he is now in his tenth year, he spent ten years in fee-for-service practice.
Starting With the Clinician, Not the Guideline
Dr. Dinh's group began about a decade ago with a specific problem. Roughly 70% of ischemic cardiovascular testing across KP Northern California was going to patients who didn't need it, and in a low pretest probability population, more testing means more false positives. The guidelines were already clear. Instead of assuming primary care doctors didn't know them, Dr. Dinh asked what was going on. Many said they knew the test wasn't indicated, but the patient or family was anxious, often after hearing about someone who died suddenly of a heart attack. Add short visits and a pretest probability calculation that, in Dr. Dinh's view, many cardiologists can't do reliably by gestalt either, and ordering the test becomes the path of least resistance. Dr. Walker points out how much of that decision has nothing to do with medicine: liability, quality metrics, and patient expectations all push in the same direction. The group's answer was a stable chest pain tool that does the calculation at the point of care and gives clinicians the confidence to hold back.
Meeting Clinicians Where They Are
Scaling that tool across 22 medical centers required a few hard lessons. Dr. Dinh started with early adopters and let results do the persuading. He put tools wherever physicians already work, whether that's inside Epic, a favorites list in KP's clinical library, or the emergency department dashboard. He also built in local reality: guidelines may recommend cardiac CT for certain chest pain presentations, but not every center has a scanner, so the algorithm had to account for that. The same flexibility paid off during the pandemic, when a contrast shortage hit and the group simply turned off the CT recommendation until supply returned, rather than trying to message thousands of clinicians. Asked what drives adoption most, Dr. Dinh points to ease of use. Physicians push back on six clicks, so those clicks have to pay off: the tool does the math, produces a clinical summary that can be pasted straight into the chart, documents the pertinent questions that get skipped in a rush, and carries the backing of the specialty chairs. His team is now working with IT to pull that data in automatically.
The Next Step with AI, Machine Learning, and Leading the Guidelines
Dr. Dinh uses AI himself, but he's wary of answers that change with the prompt. When he asks a cardiology question and gets something that sounds wrong, he pushes back and the model apologizes. A family physician asking a specialty question may not know to push back. Dr. Walker adds that no one would keep consulting a cardiologist who behaved that way. At KP, machine learning is already part of the research side, used to validate risk models such as PREVENT and to predict heart failure, but it is kept out of real-time chart use for now. Dr. Walker frames these models as the natural next step from the regression work behind tools like the Wells criteria, just run on far more data. Dr. Dinh then shares a new joint venture with an MIT-based AI company that will rebuild KP's decision tools on a platform with fully discrete data. That opens the door to running a tool retrospectively on 100,000 to 200,000 patients who presented with chest pain a decade ago, comparing its recommendations with what actually happened, and adjusting the variables to match. Dr. Dinh sees this as a way for the system to help shape guidelines rather than only follow them.
The Bottom Line on Driving Change
Dr Dinh notes that pressure to practice in a value-based way, especially for Medicare patients, is reaching clinicians regardless of how they're paid. For a physician at a health system without a decision support group, his advice starts with humility. Understand what your colleagues are dealing with, recognize that the problem is often the system and its incentives rather than the physician, and get everyone to agree on what's in it for them. Without that shared answer, he says, nothing changes.
MDAware: Upgrading Clinical Judgment is MDCalc's podcast dedicated to breaking down the science behind clinical scores and bringing transparency and confidence to your clinical workflow.