A hospitalist at a 140-bed community hospital has been asked to update the alcohol withdrawal protocol, and she is starting from a blank document.
She has the last local version, written four years ago by a physician who left for a job across the state. She has the national guideline, which describes the pharmacology in general terms and says nothing about nursing escalation triggers, order-set logic, or what to do when a patient's CIWA score does not behave the way the textbook predicts. She has a friend from residency at a different hospital who might email her a PDF if she asks.
What she does not have is any way to find the pharmacist at a hospital her size, 150 miles away, who spent eighteen months moving that hospital from a benzodiazepine-first protocol to a phenobarbital-first one, and who could tell her in twenty minutes what took that hospital a year and a half to learn: which nursing triggers false-alarmed constantly, which dosing table caused problems in patients over 70, and why the first version of their order set failed.
She will spend the next six weeks rebuilding something that already exists, competently, at hundreds of other hospitals, because the document can travel and the person who wrote it cannot be found.
This is not an isolated inefficiency. It is the default condition of protocol development at 5,121 US community hospitals, repeated across every high-stakes condition, every few years, forever.
The heterogeneity is not subtle
Start with direct evidence that this reinvention is not merely duplicative but genuinely uneven in quality.
A study published in P&T surveyed 90 Northeast hospitals on their alcohol withdrawal management protocols and found two-thirds had no protocol at all for benzodiazepine-refractory alcohol withdrawal, a condition where the consequences of an inadequate plan are severe and where guidance exists but had evidently not reached most of the hospitals surveyed. The paper describes "considerable heterogeneity" among the hospitals that did have a protocol, meaning even the two-thirds with something in place were not converging on similar practice.
This is not a story about hospitals ignoring evidence. It is a story about evidence that exists in the literature failing to become a workable local protocol, because the gap between "we know what should happen" and "here is the order set that makes it happen at the bedside, with these dosing thresholds and these nursing escalation triggers" requires implementation knowledge that journals do not carry.
The stakes of getting that translation right are measurable. One reported implementation of a sepsis order set delivered antibiotics roughly 35 minutes faster, a difference that matters directly for a condition where mortality rises with every hour of delay. A well-built protocol is not paperwork. It is the mechanism by which a guideline actually reaches a patient.
The lag between knowing and doing is measured in decades
Layer in the second half of the problem: even where good evidence exists, it moves into practice extremely slowly.
Research published in the Journal of the Royal Society of Medicine puts the average research-to-practice lag at roughly 17 years. Some of that lag is inherent to how evidence accumulates and gets validated. But a meaningful share of it is a distribution failure that has nothing to do with the strength of the evidence and everything to do with the absence of any channel that carries implementation knowledge, not just conclusions, from the hospital that solved a problem to the hospital still working on it.
This is the same failure this series has documented in quality improvement generally: knowledge that works does not spread past the institution that proved it, because publication, conferences, and vendor content transmit conclusions, not operators. Protocol development is where that failure is most concentrated, because unlike a general quality initiative, a clinical protocol has to be rebuilt, in detail, at every single hospital that wants to use it, whether or not the underlying medicine has already been solved elsewhere.
The arithmetic of the tax
Here is a conservative estimate of what this actually costs, built from figures the dossier explicitly flags as containing an assumption rather than a hard measurement, so treat the total as directional.
There are 6,100 US hospitals, of which 5,121 are community hospitals and 3,567 belong to a health system, per AHA Fast Facts. If even a modest slice of those hospitals maintain something like 20 high-stakes condition protocols, sepsis, alcohol withdrawal, DKA, VTE prophylaxis, hyperkalemia, and refresh each roughly every three years as guidelines and drug shortages force updates, that works out to roughly 34,000 protocol rebuilds a year across the country.
Order-set development is not a quick task. Implementation consultants describe it as "a lifelong burden" rather than a one-time project, requiring leadership sponsorship, informatics build time, a pilot period, and ongoing maintenance, according to a 2013 Chest review of order-set practice. If a rebuild consumes something on the order of 40 hours of combined physician, pharmacist, and informatics time, an unverified but plausible assumption given the Chest description, and that time is billed at a blended rate near $150 an hour, the resulting figure lands near $200 million a year in labor alone, before counting the outcome cost of hospitals running an inferior local version because they never found the better one.
That $200 million estimate should be read as an order-of-magnitude illustration, not an audited number. What is not in dispute is the underlying mechanism it is illustrating: thousands of hospitals independently paying skilled clinical staff to solve a problem that other skilled clinical staff, at other hospitals, have already solved.
Every attempt at a document commons has already failed, and the pattern is instructive
This is not a problem nobody has tried to solve. It has been attempted at least twice, at national scale, and both attempts converged on the same failure mode.
The Morningside Initiative, launched around 2007, set out explicitly to enable sharing of clinical decision support knowledge across organizations. Its own assessment, published in the Open Medical Informatics Journal in 2010, stated that sharing of CDS knowledge across organizations was "currently virtually non-existent" and that the investment required to build and maintain a shared artifact was one "only larger organizations have been able to undertake." The initiative did not survive as an active, widely used resource.
AHRQ's CDS Connect, a federally funded attempt at a public repository of clinical decision support artifacts launched around 2016, has quietly faded. As of this writing, its dedicated URL redirects to a general AHRQ clinical decision support page rather than an active, browsable repository of shared protocols. The public commons attempt, backed by a federal agency with no commercial incentive to hoard, still did not achieve durable adoption.
Meanwhile the private-sector substitutes that did survive, Zynx and Elsevier's order-set content libraries, Epic's Foundation System and Community Library, are licensed content products, not people networks, and Epic's version is customer-only, useless to the roughly half of US hospitals running a different EHR vendor. UpToDate answers "what does the evidence say," not "what did the order set look like after the first version failed at the bedside." None of these substitutes carries the one thing that actually determines whether a protocol works locally: the account of what was tried, what broke, and who is willing to explain it.
The pattern across every failed and surviving substitute is the same. The document can be copied. The provenance cannot.
Why the document is not the scarce resource
This is the structural diagnosis, and it is the same finding this series has made about quality improvement generally, sharpened here to its most concrete form.
A protocol document, the flowchart, the dosing table, the order-set logic, is genuinely useful, and it is also genuinely easy to obtain today. A hospitalist with an internet connection can find several published or vendor-licensed versions of a sepsis protocol within an hour. What she cannot find is the answer to the question that actually determines whether adopting one will work: who built this, what did they reject along the way, what broke at go-live, and will they take a twenty-minute call about it.
That information exists nowhere as a searchable object. It lives in the memory of the pharmacist who spent eighteen months on the phenobarbital transition, in the head of the nurse educator who knows which escalation trigger caused false alarms, in a departed physician's personal notes that left with her when she changed jobs. Every substitute this article has named, Zynx, Epic Community Library, UpToDate, AHRQ CDS Connect while it existed, is keyed to the artifact rather than to the person, which means every one of them stops exactly where the useful knowledge begins.
Why nobody owns the fix
Run through the plausible builders and the reason each one stalls.
Content vendors like Zynx and Elsevier monetize licensed protocol documents. Their business model depends on selling the artifact, not on connecting a customer to a competitor's staff member who could explain it better for free.
Specialty societies publish guidelines, which describe what should happen in general clinical terms, and stop at the hospital door. A guideline cannot tell a 140-bed community hospital's pharmacist what dosing threshold caused problems in patients over 70, because that is implementation detail a society-level document was never designed to carry.
Regional quality collaboratives, the model that this series has documented working extremely well within Michigan's Blue Cross Blue Shield Collaborative Quality Initiatives and the pediatric Solutions for Patient Safety network, are bounded by whoever funds them: one payer, one state, one patient population. A hospitalist at a general community hospital in a state without a funded collaborative has access to neither.
Epic could expand its Community Library, but only for Epic customers, and only as artifacts rather than as a routed network of reachable authors.
No party currently has both the reach across institutions and vendors that a general solution requires, and the incentive to build a people network rather than a document library.
The pressures making this worse, not better
Three forces are compounding the cost of this gap rather than easing it.
CMS quality measures raise the price of an inadequate local protocol. SEP-1 and its successor sepsis measures tie reimbursement and public reporting to how quickly and completely a hospital's protocol gets a patient through defined steps. A hospital running an inferior local version now pays for it directly in a way it did not a decade ago.
Generative AI makes drafting cheap and makes discernment scarce. A hospitalist can now generate a plausible-looking sepsis order set in minutes with an AI assistant. What she cannot generate is knowledge of which of the forty plausible variants actually worked at the bedside somewhere else, which makes the question "who has actually run this" more valuable precisely as drafting itself becomes commoditized.
Distressed hospital consolidation strips small hospitals of committee capacity. With 43.5 percent of 2025 hospital M&A deals involving a financially distressed party, per Kaufman Hall, the small and rural hospitals least able to staff a full protocol committee are also the ones losing independent capacity fastest, at the same time CMS measures raise the cost of getting the protocol wrong.
What would actually work
Register the author, not just the artifact. A protocol shared without the name of the person who built it and led its implementation carries a fraction of its value. The missing primitive is a searchable record connecting condition and setting to a specific, reachable person who actually ran the implementation.
Capture what failed, not only what succeeded. The most valuable information in any protocol's history is the version that was tried and abandoned, and why. No current substitute records this, because no one presents failed versions at a conference or publishes them in a journal.
Match on setting, not on prestige. A protocol built at an academic medical center with a full-time informatics team is frequently unusable, as written, at a 140-bed community hospital with none. Setting comparability, bed size, staffing model, EHR vendor, should be the primary matching variable, and no current substitute uses it.
Build in a reciprocity obligation, not just a directory listing. A list of names who theoretically built a sepsis protocol somewhere is not the same as a norm that the named person actually answers a twenty-minute call. The value of this system depends entirely on whether reachability is a real, enforced expectation rather than an aspiration.
Treat de-identified institutional documents as shareable by default, with local approval preserved. Any shared protocol needs to be explicit that it represents peer experience, not a clinical recommendation, and that adopting it remains subject to the receiving hospital's own medical staff and P&T committee approval. This is what makes the exchange legally comfortable rather than an unbounded liability exposure.
Prioritize the highest-stakes, most commonly rebuilt conditions first. Sepsis, alcohol withdrawal, and DKA are not arbitrary examples; they are conditions nearly every acute-care hospital manages, where guidance exists in the literature but implementation detail does not travel, making them the highest-leverage starting point for any shared registry.
What you can do now
If you are building or updating a protocol
Before you start from a blank page, find someone who already built this at a comparable hospital. A hospital your size, with your staffing model and your EHR vendor, not the academic center that published the guideline. The published version may not transfer at all.
Ask explicitly what failed the first time. Most implementers will answer this readily when asked directly, and it is far more useful than asking what the final version looks like, because the final version omits everything that had to be learned to get there.
Write your own implementation account as you go, not after. What you tried first, what broke, what you would do differently. This two-page document is the single most valuable thing you will produce and it almost never gets written down anywhere durable.
If you lead pharmacy, quality, or a clinical practice committee
Name the person who actually built each protocol in your internal records, not just the sponsoring committee. That is how your hospital's own knowledge survives when the author leaves, which the community-hospital turnover pattern makes a near certainty over any multi-year horizon.
Find one peer hospital of comparable size and trade honestly. A standing relationship with a comparable hospital, exchanging what worked and what did not on your highest-stakes protocols, replicates a meaningful share of what a funded regional collaborative provides, at no cost beyond the relationship itself.
Prioritize the conditions where heterogeneity is documented and stakes are highest. The P&T finding that two-thirds of surveyed hospitals had no protocol for benzodiazepine-refractory alcohol withdrawal is a specific, addressable gap; do not wait for the next audit to discover whether your hospital is one of them.
If you fund or lead rural and critical access hospital networks
Recognize protocol committee capacity as a resource that consolidation is actively stripping away. With over 43 percent of 2025 hospital deals involving distressed parties, the hospitals least able to staff a protocol committee are losing independent capacity at the same time CMS measures raise the cost of an inadequate one.
Fund reachable-author networks over document libraries. The Morningside Initiative and AHRQ CDS Connect both demonstrate that a well-funded document repository does not solve this problem on its own; the missing piece is connecting a struggling hospital to a specific person who has already solved its exact problem.
Frequently asked questions
How many US hospitals are duplicating the same clinical protocols? There are 6,100 US hospitals, including 5,121 community hospitals, per AHA Fast Facts. A study of 90 Northeast hospitals found two-thirds had no protocol at all for benzodiazepine-refractory alcohol withdrawal, and the ones that did showed "considerable heterogeneity," published in P&T (2018).
Is there a shared library of hospital clinical protocols? Not a durable, actively used one. AHRQ's CDS Connect repository, launched around 2016, now redirects to a general clinical decision support page rather than an active shared library. The Morningside Initiative, an earlier attempt, concluded in 2010 that cross-organization sharing of clinical decision support knowledge was "currently virtually non-existent." Private options like Zynx and Epic's Community Library exist but are licensed content or customer-only artifacts, not networks of reachable authors.
How long does it take a hospital to build a clinical protocol or order set? There is no single audited figure, but implementation experts describe order-set development as "a lifelong burden" requiring leadership sponsorship, informatics build time, piloting, and ongoing maintenance, per a 2013 Chest review, and an illustrative estimate of roughly 40 hours per rebuild across physician, pharmacist, and informatics time is plausible though unverified.
Why does evidence-based guidance take so long to become an actual hospital protocol? Research indicates an average research-to-practice lag of roughly 17 years, published in the Journal of the Royal Society of Medicine (2011). Part of this reflects how evidence accumulates, but a substantial share reflects the absence of any channel that carries implementation knowledge, the practical detail of how a guideline becomes a working order set, from the hospital that solved it to the hospital still working on it.
What happened to AHRQ's CDS Connect repository? As observed in 2026, the CDS Connect URL (cds.ahrq.gov) redirects to a general AHRQ clinical decision support informational page rather than an active, browsable repository of shared protocols, indicating the public commons attempt has faded from active use.
Do sepsis order sets actually improve care speed? Evidence points that way. One reported implementation of a sepsis order set delivered antibiotics roughly 35 minutes faster than the prior process, a meaningful difference for a condition where delay is directly linked to worse outcomes, per a report cited in CHEST Physician.
The bottom line
Every few years, thousands of hospitals across the country sit down to write the same sepsis protocol, the same alcohol withdrawal escalation pathway, the same DKA order set, almost entirely alone. The national guideline exists. The document is not hard to find. What is missing, every single time, is the person who already built a working version and could explain in twenty minutes what took them a year and a half to learn.
The scale of that reinvention is not small: roughly 34,000 protocol rebuilds a year across US hospitals, an illustrative $200 million in labor before counting the cost of the inferior local versions that result when a hospital never finds the better one. Two federally and institutionally backed attempts at a document commons, Morningside and AHRQ's CDS Connect, both stalled, because a document library was never the missing piece.
The missing piece is provenance. Who built this, what did they abandon along the way, what broke at go-live, and will they answer a call about it. That knowledge exists in a specific person's memory at a specific hospital, and today it disappears the moment that person changes jobs, because nothing tracks the author, only the artifact.
That hospitalist at the 140-bed community hospital is going to spend the next six weeks rebuilding a protocol that already works well, somewhere, 150 miles away. The pharmacist who solved it would very likely take her call. Nothing in American healthcare currently tells either one of them the other exists.
Part of a series on the missing professional infrastructure of healthcare. Previously: The Shadow Consult Network
Evidence note: hospital counts are from AHA Fast Facts (2024 data, published 2026). The alcohol withdrawal protocol heterogeneity finding is from P&T (2018), a survey of 90 Northeast hospitals. The Morningside Initiative assessment is from the Open Medical Informatics Journal (2010). The status of AHRQ CDS Connect reflects a direct observation of the site made in September 2026 and may change. The research-to-practice lag figure is from the Journal of the Royal Society of Medicine (2011). The order-set burden characterization is from a 2013 Chest review. The $200 million labor estimate is this article's own illustrative calculation built on an explicitly unverified assumption (roughly 40 hours per rebuild) applied to hospital counts from AHA data; it should be read as directional, not as an audited industry figure. The distressed hospital M&A share (43.5 percent of 2025 deals) is from Kaufman Hall. The sepsis order-set speed improvement (roughly 35 minutes faster antibiotic delivery) is reported in CHEST Physician and reflects a single implementation rather than a controlled multi-site study. The opening scenario is a composite illustration built from the patterns documented in this evidence, not a specific reported case.
