When Best Practices Aren’t Enough: How Localized Thinking Dooms Healthcare Innovation
Joshua Tamayo-Sarver, MD, PhD, FACEP, FAMIA
I was super excited because I thought I had figured out the problem: we could identify who was going to have a major cardiac event within 30 days of an emergency department visit with incredible accuracy. But when we tried to scale the program, it fell flat on its face. My team’s perfect solution became another healthcare innovation casualty to an invisible enemy I’d later recognize as small-area variation. This phenomenon, where neighboring communities develop radically different standards of care, doesn’t just create treatment disparities; it systematically sabotages our attempts to improve healthcare at scale. Through painful lessons scaling technologies across 500+ hospitals, I’ve learned that overcoming this hidden adversary requires fundamentally reimagining how we design solutions, validate interventions, and measure success in complex health systems.
The Ghost in the Machine: How Geography Shapes Care Standards
Every medical student learns textbook treatment protocols, but real-world medicine operates in the shadow of what researchers call “practice style signatures.” The seminal Dartmouth Atlas Project revealed that a child’s likelihood of undergoing tonsillectomy had less to do with clinical need than their school district’s location. When Dr. John Wennberg discovered 400% variation in tonsillectomy rates between adjacent Vermont communities in the 1970s, he unearthed healthcare’s dirty secret: physician practice patterns adapt to local cultural norms, not just scientific evidence.
This phenomenon persists today in everything from antibiotic prescribing habits to cancer screening thresholds. The implications for health tech innovators become clear when you consider that most digital health startups test their solutions in single health systems, essentially baking one community’s practice style into their technology’s DNA.
The Innovation Trap: When Local Success Becomes National Failure
Early in my career, I watched a brilliant major acute cardiac event prediction algorithm fail spectacularly because we’d trained it on health system’s data. The model was amazing where we developed it but was difficult to deploy elsewhere. This wasn’t an isolated incident. Research shows AI/ML projects in healthcare fail to generalize beyond their initial deployment environment, often due to unrecognized local practice variations.
The root problem lies in how we validate innovations. Traditional pilot studies focus on proving efficacy in controlled settings, not robustness across diverse care environments. We’re essentially building solutions for healthcare’s “average” patient while operating in a system where average doesn’t exist.
Breaking the Cycle: A Blueprint for Variation-Resistant Design
1. Map practice style geographies before writing code
Our team now starts every project by analyzing care pattern variations across urban/rural, academic/community, and high/low socioeconomic status facilities. We are lucky to provide clinical services across virtually every healthcare setting (including virtual) and we learn to lean on that rich source of perspectives.
2. Build adaptive logic layers into core architectures
Modern machine learning techniques like mixture-of-experts models allow technologies to dynamically adjust decision thresholds based on local prevalence rates and resource availability. For our clinical workflow tool, this meant identifying how to standardize different customizations at the individual, department, hospital, health system, and global level.
3. Implement continuous validation loops
Static clinical validation fails in dynamic healthcare environments. We must monitor product performance across different practice styles in real-time. When variation exceeds acceptable thresholds, it is time for a deep dive.
From Variation to Transformation: A Call for Courageous Leadership
I’m reminded that healthcare’s greatest innovations emerge from its hardest truths. The small-area variation crisis isn’t just a technical challenge; it’s a moral imperative to confront medicine’s hidden plurality of truth.
To fellow innovators: Stop building solutions for one location and start building for heterogeneous systems. Demand evidence from five practice style clusters before declaring that you understand the problem. Partner with competitors to validate across divergent care models.
The variations we once saw as noise contain critical signals about healthcare’s complex reality. By designing innovations that embrace this diversity rather than ignoring it, we can finally create solutions worthy of scaling. Our patients don’t need another perfectly optimized tool, they need technologies resilient enough to heal medicine’s fractured landscape of care.
Dr. Joshua Tamayo-Sarver, MD, PhD, FACEP, FAMIA, develops and deploys technology solutions in the healthcare ecosystem as a clinician, business leader, software engineer, statistician, and social justice researcher. As the Vice President of Innovation at Inflect Health and Vituity, his unique formula of skills has helped develop over 35 solutions and scale multiple new healthcare products, including the first AI occult sepsis tool with FDA breakthrough designation. Dr. Tamayo-Sarver oversees corporate venture, internal incubation, and advisory services for AI-driven healthcare solutions, blending consumerism and clinical quality to fit the delicate balance of patient desire, user experience and quality medical care. A Harvard graduate, he holds degrees in biochemistry, epidemiology, and biostatistics, as well as a medical degree from Case Western Reserve University. He is a Mentor in the Emergence Program at Stanford University.
