Rethinking the Architecture of Clinical Knowledge
From How Medicine Knows to How Medicine Learns, Thinks, Acts, and Improves
Medicine has become extraordinarily sophisticated at generating knowledge. We have clinical trials, systematic reviews, guidelines, diagnostic technologies, electronic health records, artificial intelligence, and increasingly powerful analytical tools.
Yet an important question remains:
Have we become equally sophisticated at organizing, teaching, evaluating, and operationalizing clinical knowledge?
The problem may not be a shortage of medical information. It may be that the different systems responsible for creating, interpreting, teaching, applying, and improving medical knowledge remain insufficiently connected.
Clinical education operates in one ecosystem. Evidence-based medicine operates in another. Clinical practice follows its own workflows. Health information technology frequently treats clinical knowledge as structured data to be stored and retrieved. Quality improvement measures outcomes, often without a sufficiently rich connection to the reasoning and knowledge that produced them.
What if these were not separate problems?
What if they were different dimensions of one larger architectural problem?
1. The epistemological question: How does medicine know?
Before designing better clinical systems, we must understand the nature of medical knowledge.
Medicine does not operate with absolute certainty. It operates with probabilities, imperfect observations, evolving evidence, contextual judgments, and decisions made under constraints.
A diagnosis is not simply a label. It is a provisional interpretation of available evidence.
A guideline is not an unquestionable instruction. It is an attempt to translate accumulated evidence into recommendations for particular clinical contexts.
A laboratory result is not knowledge by itself. Its meaning depends on the patient, clinical question, measurement characteristics, prior probability, and surrounding evidence.
This distinction matters enormously for health technology.
A system that stores medical facts without representing their context, provenance, certainty, and applicability may preserve information while losing much of its clinical meaning.
The philosophical question therefore becomes an engineering question:
How can clinical knowledge be represented without pretending that uncertainty, context, and interpretation do not exist?
This is where philosophy of science, clinical epidemiology, and health informatics must begin speaking to one another.
2. Evidence-based medicine: From information to justified decisions
Evidence-based medicine transformed modern clinical thinking by emphasizing the integration of best available research evidence, clinical expertise, and patient values.
But evidence-based medicine should not be reduced to displaying guidelines or linking research papers.
Its deeper contribution is a disciplined approach to clinical justification.
What is the clinical question? What evidence addresses it? How trustworthy is that evidence? How applicable is it to this patient? What uncertainties remain? What are the potential benefits and harms? What matters to the patient?
These questions form a reasoning process.
A mature clinical knowledge architecture should preserve that process rather than merely deliver its final recommendation.
It should distinguish:
Evidence from interpretation.
Association from causation.
Recommendation from mandatory action.
Population-level estimates from individual patient circumstances.
Established knowledge from emerging hypotheses.
Clinical certainty from computational confidence.
This is particularly important as artificial intelligence enters clinical workflows. A fluent answer is not necessarily a justified answer. A plausible recommendation is not necessarily an evidence-supported recommendation.
The next generation of clinical intelligence must become more accountable to the reasoning behind its outputs.
3. Medical pedagogy: Teaching clinicians how to think
Medical education has traditionally invested heavily in the acquisition of knowledge. But clinical competence requires much more than remembering facts.
Clinicians must recognize patterns, generate and revise hypotheses, interpret evidence, manage uncertainty, identify missing information, and understand when their initial reasoning may be wrong.
This makes clinical reasoning an educational challenge as much as a medical one.
The work of Geoffrey Norman, Kevin Eva, and Donald Schön offers important perspectives on clinical cognition, expertise, reflection, and professional learning.
Their intellectual contributions point toward a deeper educational possibility: systems should not merely provide answers to learners. They should help learners develop the ability to construct, examine, and improve their own reasoning.
Consider the difference.
A conventional educational system might explain why a diagnosis is correct.
A reasoning-oriented system might ask:
What findings support this diagnosis?
What alternative diagnosis could explain the same findings?
Which observation would most strongly discriminate between them?
What evidence would change your current judgment?
What important information is missing?
How would the decision change if the patient's context were different?
The objective is not simply knowledge transmission.
It is the development of better clinical thinkers.
This is one of the central motivations behind the VibeRounds approach: exploring how AI can facilitate questioning, reasoning, reflection, and intellectual engagement rather than merely generating answers.
4. Health IT: The missing architecture between knowledge and practice
Electronic medical records have dramatically improved the ability to store, retrieve, and exchange clinical information.
Standards such as HL7 FHIR have helped establish a common language for representing healthcare information. Terminologies such as SNOMED CT and LOINC support more consistent clinical meaning.
But interoperability of data does not automatically produce interoperability of clinical intelligence.
A FHIR-compatible system may exchange observations, conditions, medications, and encounters without understanding the clinical trajectory they collectively represent.
A clinical template may collect all the required fields without adapting to the patient's actual problem.
A decision-support alert may detect a condition without understanding the wider care context.
A population analytics system may identify a cohort without connecting its findings to the clinical reasoning and interventions that could improve outcomes.
This reveals a crucial distinction:
A healthcare information system is not necessarily a clinical knowledge system.
The opportunity is to move beyond static forms, isolated alerts, disconnected analytics, and fragmented workflows toward architectures that connect:
Clinical concepts and their relationships.
Patient-specific context and longitudinal data.
Evidence and clinical reasoning.
Dynamic clinical documentation.
Decision support and care pathways.
Monitoring and outcome evaluation.
Feedback into the knowledge system itself.
The architecture must preserve the distinction between clinical knowledge, computational rules, AI-generated suggestions, and authorized clinical decisions.
It must also make provenance, versioning, permissions, and auditability foundational rather than optional.
5. The people whose thinking can help us get there
This intellectual undertaking cannot be developed from medicine alone.
It requires engagement with people who have transformed different parts of the problem.
Enrico Coiera offers a foundation in medical informatics, information flow, and clinical decision support.
Trisha Greenhalgh provides an essential perspective on digital health, complexity, implementation, and the realities of healthcare systems.
David Sackett and Gordon Guyatt represent the intellectual foundations of evidence-based medicine.
Geoffrey Norman and Kevin Eva offer important insights into clinical reasoning, expertise, and health professions education.
Donald Berwick demonstrates how systems thinking can translate into measurable improvements in healthcare.
Beyond medicine, John Dewey helps us think about learning through inquiry and experience. Donald Schön explores reflective professional practice. Nancy Cartwright challenges simplistic assumptions about evidence and causality. James Reason helps us understand human error and system safety. W. Edwards Deming offers a framework for understanding variation and continuous improvement.
These are not merely names to admire. They represent different intellectual tools for approaching one interconnected problem.
6. From isolated applications to a clinical knowledge flywheel
The long-term opportunity is to connect the entire lifecycle of clinical knowledge.
Imagine a system in which:
Medical knowledge defines clinical concepts, relationships, and evidence.
Patient context determines which knowledge is relevant.
Clinical reasoning interprets the available information.
Dynamic workflows collect the information needed for a particular clinical problem.
Decision support presents relevant, explainable suggestions.
Clinicians make and document decisions.
Longitudinal monitoring observes what happens afterward.
Population analytics identifies patterns, gaps, and opportunities.
Research and quality improvement evaluate whether the underlying knowledge and processes should change.
This creates a continuous cycle:
Understand → Compose → Recommend → Execute → Monitor → Learn → Improve.
The important idea is that the system should not stop when a recommendation is generated or a clinical encounter is completed.
The consequences of clinical decisions should inform subsequent learning, evaluation, and improvement.
However, this feedback must be governed carefully. Observed associations do not automatically establish causality. Clinical outcomes require appropriate risk adjustment, study design, and validation. AI-generated modifications to clinical knowledge should never become authoritative merely because an algorithm proposed them.
A genuine clinical knowledge flywheel requires scientific discipline, clinical governance, and human accountability.
7. The management and implementation problem
An intellectually elegant architecture can still fail in healthcare.
Healthcare organizations are complex environments. They involve competing priorities, resource limitations, professional boundaries, regulatory requirements, established habits, and the practical realities of clinical work.
This is why management experience matters.
A system must be technically sound, clinically useful, operationally feasible, financially sustainable, and acceptable to the people expected to use it.
Implementation is not a final stage after innovation. It is part of the design problem from the beginning.
The question is not simply whether something can be built.
It is whether it can be safely adopted, consistently operated, evaluated, maintained, and improved.
8. A research and engineering agenda
The emerging field of clinical intelligence needs more than another collection of AI applications.
It needs rigorous work across several interconnected questions:
Epistemology: How should clinical knowledge represent uncertainty, provenance, disagreement, and changing evidence?
Clinical reasoning: How can computational systems support hypothesis generation, diagnostic revision, and decision-making without replacing professional judgment?
Medical pedagogy: How can technology develop reasoning ability rather than dependence on generated answers?
Evidence-based medicine: How can recommendations remain connected to the strength, applicability, and limitations of their supporting evidence?
Health informatics: How can clinical concepts, longitudinal patient data, workflows, and decision support operate within coherent and interoperable architectures?
Evaluation: How should we measure improvements in reasoning, learning, clinical processes, patient outcomes, safety, and efficiency?
Governance: How can clinical rules, AI-generated content, and evolving knowledge be versioned, reviewed, validated, and held accountable?
These questions deserve sustained collaboration among clinicians, educators, philosophers of science, epidemiologists, informaticians, software engineers, researchers, and healthcare leaders.
Conclusion: Rebuilding the connection between knowing and doing
Medicine's next major transformation may not come from producing more medical information or generating more fluent AI responses.
It may come from improving the architecture connecting knowledge to reasoning, reasoning to action, action to outcomes, and outcomes to learning.
The ambition is to build systems that understand more of the clinical context, represent knowledge more responsibly, teach reasoning more effectively, support decisions more intelligently, and improve through scientifically governed feedback.
This is not a claim that technology can solve every problem in medicine. Nor is it an argument that clinical judgment should be automated.
It is an invitation to rethink the relationship between human expertise, scientific evidence, education, computation, and healthcare systems.
The ultimate objective is not to make medicine more computerized. It is to make the relationship between what medicine knows, how clinicians think, what healthcare systems do, and what patients experience more coherent, transparent, and continuously improvable.
That is the larger intellectual and engineering challenge.
And it is a challenge worth dedicating a career to.
Suggested intellectual starting points: Enrico Coiera's Guide to Health Informatics; David Sackett and colleagues' work on Evidence-Based Medicine; Geoffrey Norman's research on clinical reasoning; Trisha Greenhalgh's work on digital health and complexity; Donald Berwick's work on healthcare improvement; John Dewey's philosophy of inquiry; and Donald Schön's The Reflective Practitioner.
VibeRounds is one ongoing exploration of these questions, particularly at the intersection of clinical reasoning, medical education, clinical knowledge architecture, and health information technology. Its ideas and implementations should be evaluated through independent research, clinical validation, and real-world evidence.
No comments:
Post a Comment