The Pathway

The problem is not that physicians resist evidence. In most cases, the evidence never arrives.

In 2000, Balas and Boren published a study in the Yearbook of Medical Informatics examining how clinical research reaches practice. Their finding: only 14 percent of original research ever translates into changes in patient care. For the fraction that makes it, the average time from discovery to routine clinical use is approximately seventeen years. For the rest — the large majority — the research exists in journals, is cited by other researchers, advances the field’s knowledge, and never enters the room where a physician makes a decision about a patient.

The seventeen-year figure has been cited thousands of times since. Morris et al., in a 2011 systematic review in the Journal of the Royal Society of Medicine, examined its provenance and found it averaging incommensurable measurements across different specialties, definitions, and time points. Their title was the critique: “The answer is 17 years, what is the question?” A 2025 analysis in Frontiers in Health Services went further, arguing the figure obscures rather than illuminates the current state of the field. The specific number is fragile. The structural observation it gestures at is not.

The observation: the default pathway between clinical research and clinical practice is passive. Journals. Conferences. Guidelines that update on institutional cycles. Passive dissemination does not reliably move evidence across the divide between the world that produces knowledge and the world that uses it. The physician’s mind is not the bottleneck. In eighty-six percent of cases, the pathway to the physician’s mind does not exist.

When the evidence does arrive, the problem changes form but not structure.

A physician reads a study. Understands the evidence. Does not change practice. This is not cognitive failure. It is organizational architecture. Fee-for-service reimbursement rewards volume of procedures, not integration of new evidence. Malpractice liability punishes deviation from established standard of care — even when the deviation is what the evidence supports. Clinical guideline update cycles run on bureaucratic timescales while the evidence moves on research timescales. The physician knows what the evidence says. The system within which the physician practices does not accommodate what they know.

The first failure: knowledge absent from the physician’s context. The second: knowledge present, behavior unchanged. Both are the same structural problem at different points in the same pathway. In the first, the pathway does not exist. In the second, it is blocked.

The proof that both are pathway problems is what happens when someone builds one deliberately.

Implementation science — the field that exists specifically because this gap is so persistent — has documented that active implementation with dedicated infrastructure achieves full adoption in two to four years. Not seventeen. Not fourteen percent adoption. Full implementation, in a fraction of the time, when someone constructs the transmission mechanism that the default architecture does not provide.

If the failure were fundamental — physicians resistant by nature, institutions too rigid to absorb new evidence, knowledge inherently difficult to apply — deliberate intervention would not work. It works. It works fast. The actors respond when the pathway exists. The failure is the default architecture, not the actors within it.

This is what I call a translation-mechanism failure. The input is real — the research, the evidence, the resources, the pain. The actor who should respond exists and, in the clean case, would respond if reached. What is absent is the mechanism to translate the input into a response by the decision-maker. The pathway between the input and the identifiable, addressable decision-maker is absent or blocked. Naming the input louder does not fix a transmission problem. Four decades of efforts to improve research dissemination in medicine have demonstrated this with uncomfortable clarity. The field did not need better evidence. It needed a pathway for the evidence it already had.

The same structure appears wherever an identifiable decision-maker is structurally insulated from an input that should change their behavior. Capital that exists without the deployment architecture to reach the projects that need it. Physical power grid interconnections that operate between nations without the regulatory framework to route electricity commercially across borders. Pain that accumulates in a population while the entity whose decisions determine their exposure operates behind structural insulation from that pain. In each case, the input is real. The pathway is not.

The concept has boundaries worth marking.

Translation-mechanism failure is a point-to-point problem. There is an identifiable node — the physician, the infrastructure bank, the military council — and the pathway fails to reach that node. The fix is building the pathway. This is structurally different from failures where the deciding actors are constitutively plural and dispersed — voters, markets, parliaments simultaneously — and there is no single addressable node. You cannot fix a pathway when the destination has no address. That failure requires a different architecture entirely: mechanisms that fire on a schedule rather than waiting for a dispersed collective to activate. The distinction matters because the prescriptions are different. One is an engineering problem. The other is a design-of-government problem.

The concept also has a limit condition. The healthcare case is clean: the physician, if reached, would prefer to respond. The pathway is the entire problem. In messier cases, the structural insulation is partly maintained by the actor whose response it prevents. The insulation is not accidental — it serves the insulated party. When the actor prefers the pathway’s absence, building the pathway is necessary but not sufficient. The pathway problem and the preference problem operate simultaneously. Attributing all non-response to pathway failure, in those cases, obscures the dimension of choice.

The diagnostic question, when something has the familiar shape — we have the research, the resources, the evidence, the pain, and nothing is changing — is not what is wrong with the input. The input may be exactly right. The question is what stands between the input and the entity with the authority to respond. If no pathway exists, no amount of input will produce a response.

Building one will.

Sources