Building Clinical Co-Pilots: Personalizing Healthcare Through Evidence and Experience

Healthcare is full of high-stakes choices, such as deciding what to do with a patient with a traumatic brain injury in the emergency department. These calls must be made fast—and with as much knowledge as possible.

Yet, as research shows, many of these decisions still rely on individual heuristics, leading to variability in outcomes. What if clinicians had co-pilots—AI systems designed to augment them, by distilling the most relevant knowledge from both clinical evidence and similar past patients, right at the bedside?

Why Similar Patients Matter

In our recent work on retrieval-augmented small language models (SLMs), we found that retrieving similar patients consistently improved predictive performance for ED disposition. By providing the model with real-world exemplars—patients who looked like the current one in terms of CT scans, vitals, and labs—we boosted sensitivity without sacrificing specificity.

This makes intuitive sense. Clinicians often lean on precedent: “I once saw a patient like this, and here’s what happened.” A system that can retrieve hundreds or thousands of prior cases, surface the most relevant ones, and present them in real time offers something human memory never could: scalable, evidence-aligned precedent.

And this value extends beyond clinicians. Imagine being a patient or caregiver, overwhelmed by the complexity of TBI. Seeing not only guideline-grounded recommendations, but also relatable case analogs (“Here’s what happened to others like you”) transforms medical information into something accessible and understandable. These systems can serve as educational tools, empowering shared decision-making and reducing the uncertainty that often defines serious diagnoses.

Evidence Meets Experience

But precedent alone isn’t enough. Evidence-based practice (EBP) anchors care in systematic research and guidelines. Our framework unifies EBP with case-based reasoning (CBR), showing how the combination balances safety with personalization.

• CBR (similar patients): boosts sensitivity, reducing missed admissions.
• EBP (guidelines): improves specificity, ensuring decisions align with standards.
• Together: they complement each other, balancing personalization with standardization.

This is the essence of a clinical co-pilot: not to dictate decisions, but to rapidly distill the best knowledge (population-level evidence and patient-level precedent) so that clinicians can make informed calls when every second matters.

Why a Co-Pilot? Learning From Coaches, Teachers, and Mentors

The NPR article “Pro Athletes Have Coaches. Why Not Everyone Else?” makes a compelling point: even at the highest level of performance, athletes rely on coaches to refine their craft. In music, too, we never stop learning, we continue to work with gurus and mentors throughout our journey.

Healthcare should be no different. Clinicians deserve co-pilots—systems that act as always-available coaches, offering perspective, evidence, and analogs that sharpen their judgment under pressure. They guide, contextualize, and support.

Building the Future of Clinical Co-Pilots

I’m inspired by the work of groups like OpenEvidence and Atropos Health, who are advancing real-world evidence and evidence-based AI to make medicine more adaptive and personalized. My vision builds on this momentum:

• For clinicians: Co-pilots that distill guideline passages and real-world evidence into clear, actionable insights.
• For patients and caregivers: Educational interfaces that provide relatable case examples alongside evidence, making complex prognoses more transparent and understandable.

The goal is simple but transformative: to move beyond one-size-fits-all care and toward a precision health system that is evidence-aligned, context-aware, and personalized at scale.

Closing Thoughts

As with music and sports, mastery in medicine is a lifelong pursuit. As engineers, we must work closely with patients, clinicians, and health systems to augment healthcare by designing human-centric tools and systems. I've realized that a huge part of my research should be in this intersection: qualitative and quantitative research, backed by human-centered design to build effective HealthAI systems.