Healthcare has a knowledge gap that isn’t due to a lack of knowledge.
Medicine knows the early signs of heart failure, how to diagnose cancer, and how to manage chronic disease. And yet patients still wait – for appointments, for referrals, for someone with the right credentials to sign off on a decision that, in many cases, could have been made hours or days earlier if the right tools were available to the care team members they’d already seen.
The gap between what medicine knows and what reaches patients at the point of care is a gap in knowledge availability.
Specialists develop deep diagnostic capability within a domain, and the tools of their specialty often remain only available to them. Cardiology has echocardiography and advanced hemodynamic monitoring. Ophthalmology has retinal imaging. Oncology has increasingly sophisticated biomarker panels. These tools exist because they are needed to answer the questions specialists routinely face.
The problem is that the information those tools generate is often exactly what a broader set of clinicians – a generalist, a hospitalist, a nurse practitioner, or a specialist in an adjacent field – would need to make a confident decision, but isn’t available when and where that decision is being made.
The Information Gap
Consider what it takes to detect early-stage cardiac dysfunction, evaluate a suspicious lesion, or assess hemodynamic status in a deteriorating patient. The bottleneck is often access to a diagnostic tool that can generate the relevant signal at the point of care.
Without that tool, clinicians are working from an incomplete diagnostic picture. They often have the training to act, but they can’t act because the information isn’t there. Decisions that could be made confidently get deferred. Patients get escalated not because their condition clearly requires specialist intervention, but because the data needed to rule that out doesn’t exist yet.
This creates a compounding problem. Specialists are among the most capacity-constrained clinicians in medicine. When the referral pipeline mixes patients who genuinely need specialist-level intervention with patients who could have been managed earlier if better diagnostic tools were available upstream, everyone waits longer, including the patients whose outcomes depend most on timely decisions.
Giving more clinicians access to the tools that generate specialist-level information at the point of care can make the system more efficient. Treating clinicians can make better-supported decisions, specialists focus where their expertise has the most impact, and patients get the right level of care faster, wherever they first present.
The Next Step After Access
Much of the digital health conversation over the past decade has focused on improving access to care through telemedicine, virtual visits, and remote monitoring. These are meaningful improvements, but they largely preserve the underlying model: patients still depend on specialists to generate and interpret the diagnostic information needed to make confident decisions.
A complementary shift happens when diagnostic capability moves to wherever the patient is being seen, rather than moving the patient to the specialist. Better access gets more patients in front of the right clinicians. Better tools ensure more of those encounters generate the information needed to act.
This also helps enable “top of license” – every clinician operating at the full scope of their training, with the diagnostic information needed to act with confidence already in hand.
The same logic applies across specialties. An oncologist with access to real-time cardiac assessment doesn’t need to route every potentially cardiotoxic patient through a cardiologist before treatment. A hospitalist with point-of-care ultrasound can make volume management decisions without waiting for imaging. A primary care physician with the right screening tool can detect and triage conditions that would otherwise go unnoticed until they’re harder to treat.
The constraint in each case is the same: access to information, and the tool that generates it.
Where This Has Already Worked
Expanding diagnostic capability at the point of care isn’t a single playbook. Several examples demonstrate different mechanisms depending on the bottleneck:
Embedding the decision into the workflow
The most direct form embeds the diagnostic decision into the clinical encounter, removing the specialist from the interpretation loop entirely.
IDx Technologies did this for diabetic retinopathy screening. Historically, detecting the condition required routing patients to an ophthalmologist, a bottleneck that produced low screening rates and preventable vision loss. IDx changed the model by embedding an autonomous AI system directly into primary care. The output is simple: refer or don’t refer. No specialist interpretation required. Screening became a routine part of a primary care visit, disease was caught earlier, and ophthalmologists focused on patients who actually needed intervention. The key move wasn’t training more clinicians to read retinal images, but building a tool that removed that requirement.
Moving the hardware to the point of care
A second mechanism makes the underlying diagnostic tool available wherever the patient is, rather than replacing specialist interpretation.
Butterfly Network moved ultrasound out of centralized imaging labs and into the hands of frontline providers. Findings that previously required a cardiology consult or a scheduled echo (e.g., hemodynamic status, volume assessment, basic cardiac function) became available at the bedside in real time.
Eko Health approached the same problem from a different angle, layering AI-driven cardiac insight into the stethoscope – an instrument already present in most frontline clinical encounters. Rather than introducing a new tool, Eko upgraded an existing one, making conditions detectable that acoustic assessment alone would likely miss.
In both cases, the information didn’t change. What changed was where and when it could be generated.
Distributing the knowledge layer
The third mechanism operates on the level of structured clinical knowledge that specialists accumulate over careers – knowledge that has historically been difficult to transfer at scale.
Flatiron Health addressed this in oncology. Treatment knowledge has long been concentrated in academic medical centers, creating significant variation in care quality across settings. By structuring real-world outcomes data and integrating it into workflows, Flatiron gave community oncologists access to treatment patterns and clinical context that were previously difficult to operationalize.
Project ECHO demonstrated a version of the same mechanism in hepatitis C management, using structured remote mentoring to transfer specialist decision-making frameworks to primary care physicians. Specialists became knowledge hubs, equipping many providers rather than treating every patient themselves. The constraint those physicians faced wasn’t their capability, but access to the accumulated clinical knowledge that specialists carry. ECHO moved that knowledge without moving the specialist.
These mechanisms aren’t mutually exclusive, and the most robust implementations will likely combine elements of all three. In each case, the intervention expanded the information available to the clinician already in the room and outcomes followed.
AI as an Amplifier
AI-enabled diagnostic tools are significant because, for the first time, they make it economically and practically feasible to distribute specialist-level diagnostic capability broadly.
High-quality diagnostic tools have historically been expensive, complex to operate, and dependent on specialized training to interpret. Those constraints kept them concentrated in specialist settings. AI changes the interpretation layer: a model trained on hundreds of thousands of expert-labeled cases can generate a reliable, actionable output from a signal captured at the point of care, without requiring a specialist’s interpretive expertise.
This doesn’t make AI a replacement for clinical judgment. It makes AI the mechanism by which the information needed to exercise that judgment becomes available to more clinicians, in more settings, and at more points in the patient journey.
The cardiologist doesn’t become obsolete when an oncologist has access to AI-assisted cardiac screening. The cardiologist becomes more focused, seeing the patients who genuinely require their level of intervention, rather than serving as the gatekeeper for information that didn’t need to be gated.
In the near-term, the most impactful clinical AI tools are the ones that generate new diagnostic information at the point of care, not the ones that summarize or streamline information that already exists. The former changes what decisions can be made, and by whom; the latter changes how efficiently existing decisions are processed.
In the longer term, this reshapes how care is delivered. As specialist-level diagnostic tools become available to generalists, the distinction between specialist and generalist care becomes less about where you are in the system and more about the clinical complexity being managed.
Making Knowledge Available
The traditional model of care is linear: patient to generalist to specialist. Its vulnerability is also linear: every handoff is a potential delay, and every delay puts patient outcomes at risk.
A better model distributes access to specialist-level knowledge across the care team. Specialists remain essential, but their role shifts from being the sole source of diagnostic insight to being the clinical resource for cases that genuinely require their depth of intervention. Non-specialists operate with better information, make better-supported decisions, and escalate more selectively and accurately.
Technology does not create that model, but it is what makes it achievable at scale.
The limiting factor in modern medicine is not knowledge. It is the infrastructure for generating the right information, at the point of care, for the clinician who is already there. That is the problem AI-enabled diagnostic tools are primed to solve.
Leave a Reply