A nurse-led fall prevention program at a 200-bed community hospital cuts falls with injury by more than half. The team presents it at a staff meeting. Someone writes an internal report. A poster goes to a regional conference, where forty people walk past it.
Eleven miles away, another hospital is running a fall prevention project from scratch, badly, for the third time in six years.
Neither hospital knows the other exists in this respect. The quality directors have never met. And the nurse who designed the intervention that worked will change jobs in three years, taking with her the specific knowledge of why the third step matters and what they tried first that failed.
This is the spread problem, and the number attached to it is one of the most damning statistics in health services research.
Fewer than 40 percent of healthcare improvement initiatives transition from local adoption to sustained spread beyond one area of one organization.
Not beyond the industry. Not beyond the region. Beyond one area of one organization.
The proof that it can work, boxed inside one state
Here is what makes this different from most complaints about healthcare fragmentation. We are not speculating about whether peer-to-peer quality knowledge transfer works. It has been built, it works extremely well, and its footprint is a state line.
Blue Cross Blue Shield of Michigan's Collaborative Quality Initiatives. Roughly 20 clinical collaboratives, connecting about 40 physician organizations and 20,000 physicians. Reported savings: an estimated $4.08 billion in health costs statewide, part of $6.3 billion saved by the broader Value Partnerships program since 2005.
Four billion dollars. In one state. Through a payer-funded structure that gets clinicians in a room to compare what actually works.
Solutions for Patient Safety, a pediatric hospital collaborative operating on an explicit "all teach, all learn" principle with more than 150 member hospitals, reports 35,000 children spared serious harm and $835 million in cost savings since 2012.
And the Keystone ICU project, one of the most celebrated quality successes in modern medicine, demonstrated dramatic reductions in central line infections through a structured collaborative approach in Michigan intensive care units.
So the mechanism is proven. Repeatedly. At scale. With enormous measured benefit.
Now look at how Keystone spread.
It spread through separate state-by-state replications. Rhode Island in 2010. Connecticut in 2013. Each requiring its own funding, its own leadership, its own program build. A federally funded national program was eventually required to extend it further.
The most successful patient safety intervention of its generation propagated by being rebuilt, state by state, over a decade. Not because the knowledge was secret, but because there was no channel that carried it.
The honest complication
Intellectual honesty requires stating a finding that complicates the enthusiasm.
A systematic review of quality improvement collaboratives published in BMJ Quality and Safety concluded that they "have proliferated internationally, but there is little empirical evidence for their effectiveness."
That is a serious finding from serious researchers and it should not be waved away.
How does it sit alongside Michigan's $4.08 billion and 35,000 children?
The most plausible reading is that "collaborative" describes a very wide range of things, from genuinely funded, data-driven, sustained programs with real accountability, down to a series of webinars with a shared logo. The successful examples share features that many collaboratives lack: sustained funding, mandatory data submission, a shared registry, real facilitation, and enough duration for practice to change.
Which produces the useful conclusion. The collaborative model works when it is genuinely resourced and fails when it is a label. And even where it works, it is bounded by whoever funded it: one payer, one state, one patient population.
That is the actual failure. Not that peer knowledge transfer does not work, but that it has only ever been built inside boundaries drawn by funders, never as a general capability.
Why knowledge does not travel on its own
Consider the routes by which a successful intervention might reach another hospital, and why each fails.
Publication. Slow, incomplete, and adversely selected. The paper appears two years later, describes what was done and rarely how, and is written by the minority of quality teams with academic support. The nurse manager who designed the best fall prevention program in the state will never publish it.
Conferences. A poster reaches the handful of people who stop, in a room where hundreds compete for attention, once a year.
Consultants. They do carry knowledge across institutions, which is precisely what you are paying for, at consultant rates, filtered through their commercial interest.
Professional societies. Publish guidelines, which describe what to do rather than how anyone actually made it work.
And personal networks. Which work, and are limited to whoever a quality director happens to know.
Notice what all of these share. They transmit conclusions rather than operators. A guideline tells you what to do. A publication tells you what happened. Neither connects you to the person who ran it and knows why step three exists.
The problem is a person, not a document
This is the same finding that recurs throughout this series, and quality improvement is where it is most costly.
The information that determines whether an intervention works in your hospital is almost entirely implementation knowledge:
- Which staff group resisted, and whether they were right.
- What the intervention looked like in month one versus month eight.
- Which metric looked good and was measuring the wrong thing.
- What you would need to have in place before starting.
- What made it stop working when the champion left.
- And whether it works in a 200-bed community hospital or only in the academic center that published it.
None of that is in the paper. All of it is in the head of the person who ran it.
Which means the question a quality director actually needs answered is not "what is the evidence for this intervention." It is "who has done this in a hospital like mine, and will they take my call?"
That question has no addressable answer anywhere in American healthcare.
Why this is getting worse
Three pressures are compounding.
Quality staff turnover. Post-pandemic turnover across healthcare has been elevated, and quality departments are not exempt. Every departure takes implementation knowledge that was never written down, which is the institutional amnesia problem this series has documented in other contexts.
Margin pressure. Fewer hospitals can afford to run a full improvement cycle from scratch, with the failed attempts that entails. The cost of independently rediscovering what someone else already knows has risen.
And accountability has risen. Value-based purchasing and public reporting mean the penalty for not knowing what already works elsewhere now appears directly in reimbursement.
More pressure to improve, less capacity to experiment, and the same absence of any channel to find out what has already been solved.
What would work
Person-keyed implementation records. Not "Hospital X reduced falls by 40 percent." Rather: "I led this, here, in a facility of this size, with this staffing model, and here is what happened, including the parts that did not work." Attached to a person who can be reached and who follows that record to their next employer.
Including the failures, prominently. A quality director planning an intervention needs to know who tried it and abandoned it at least as much as who succeeded. Nobody presents that at a conference, which means the published record is systematically biased toward success in a field where most attempts fail.
Matched on setting, not on prestige. The academic medical center's implementation is frequently irrelevant to a 200-bed community hospital with a different staffing model. Setting comparability is the primary matching variable and no existing channel uses it.
Independent of payer and state boundaries. This is the specific gap. Michigan's model works and stops at the state line. Solutions for Patient Safety works and covers pediatric hospitals. A quality director in a general hospital in a state without a funded collaborative has access to neither.
And carefully bounded. Aggregate and de-identified outcome data only, no patient information, and explicit framing as peer experience informing rather than replacing local governance and review processes.
What you can do now
If you lead quality or safety
Write the implementation account, not the results summary. For every intervention you run, record what you tried first, what failed, who resisted, what you would do differently, and what preconditions mattered. Two pages. This is the document that would help another hospital and it almost never exists.
Name the operator in your internal reports. Not the sponsoring executive. The person who actually ran it. That is how your organization's own memory survives their departure.
Find one peer institution of similar size and trade honestly. A standing quarterly call with a comparable hospital, exchanging what worked and what did not, replicates most of the value of a formal collaborative at zero cost. Almost nobody does this and there is no reason not to.
Ask about failures explicitly when you inquire. "What have you tried that did not work?" produces far more useful information than asking about successes, and people answer it readily when asked directly.
If you are considering an intervention
Find someone who did it in your setting. Not the published academic center. A hospital of your size, with your staffing, in your market. The published effect size may not transfer at all.
Ask what the intervention looked like at month eight. Month one is enthusiasm. Month eight is whether it became normal work.
Ask what happened when the champion left. The most common failure mode of a successful intervention is that it was a person rather than a process, and this question surfaces it immediately.
If you fund or govern healthcare quality
Look at what Michigan proved. Roughly $4.08 billion in estimated savings from structured peer collaboration, bounded by a state line because a state payer funded it. The obvious question is what a general capability would be worth, and nobody has asked it seriously.
Fund the negative results. The single highest-value publication in healthcare operations would be an honest, attributed account of abandoned interventions. No conference will accept it and no journal will prioritize it, which is exactly why it needs funding rather than encouragement.
Support person-level records, not institutional ones. Institutional knowledge leaves when people do, which the turnover data makes unavoidable. A record that follows the operator is the only version that survives.
Frequently asked questions
How often do healthcare quality improvements spread beyond where they started? Rarely. Research published in Health Research Policy and Systems found fewer than 40 percent of healthcare improvement initiatives transition from adoption to sustained spread beyond one area of one organization, with manager turnover named among the barriers.
Do quality improvement collaboratives work? The evidence is mixed and the variation appears to be about implementation rather than the model. Michigan's Collaborative Quality Initiatives report an estimated $4.08 billion in statewide savings, and a pediatric collaborative reports 35,000 children spared serious harm since 2012. A systematic review in BMJ Quality and Safety nonetheless concluded that collaboratives have proliferated internationally with little empirical evidence of effectiveness, which likely reflects how widely the term is applied.
Why didn't the Keystone ICU project spread nationally on its own? It spread mainly through separate state-by-state replications, including Rhode Island in 2010 and Connecticut in 2013, each requiring independent funding and leadership, with a federally supported national program eventually needed to extend it further. The knowledge was public; the channel to carry it was not.
What information does a quality director actually need before starting an intervention? Implementation knowledge rather than effect sizes: which staff resisted and whether they were right, what the program looked like after eight months rather than one, which metrics misled, what preconditions mattered, and what happened when the original champion left. Almost none of this appears in published literature.
Why don't published papers solve the spread problem? Because they report conclusions rather than implementation, appear years after the work, come disproportionately from academically supported institutions, and are subject to strong publication bias toward success in a domain where the majority of attempts fail.
What is the main structural gap? Existing collaboratives that work are bounded by whoever funds them: one payer, one state, or one patient population. A quality director at a general hospital outside those boundaries has no equivalent access, and no general-purpose channel exists connecting a hospital's unsolved problem to the named person elsewhere who already solved it.
The bottom line
American healthcare has proven, repeatedly and at scale, that when you connect clinicians and quality leaders across institutions and let them compare what actually works, the results are extraordinary. Billions of dollars in one state. Tens of thousands of children spared harm in one hospital category.
And then it stopped. Not because the model failed, but because each instance was funded by a payer with a state, or a specialty with a population, and nobody ever built the general version.
So more than sixty percent of improvement work fails to spread past one unit of one hospital. The most celebrated safety intervention of its generation propagated by being rebuilt from scratch, state by state, over ten years. And a nurse who solved a problem elegantly in a community hospital will take that knowledge to her next job, where nobody will ask about it either.
The evidence is not the bottleneck. The evidence has been published for twenty years.
The bottleneck is that a quality director with a problem cannot find the person eleven miles away who already solved it, and there is nothing in the industry whose job it is to introduce them.
Part of a series on the missing professional infrastructure of healthcare. Previously: The Transnational Curbside
Evidence note: spread rates come from Cote-Boileau et al. in Health Research Policy and Systems (2019). Michigan savings figures are as published by Blue Cross Blue Shield of Michigan's Value Partnerships program and are payer-reported estimates. Solutions for Patient Safety figures are as published by that collaborative. The systematic review finding limited evidence for collaborative effectiveness is from Wells et al. in BMJ Quality and Safety (2018). Keystone ICU results are from Pronovost et al. in the New England Journal of Medicine (2006), with replication history drawn from subsequent state program reports. Savings estimates from payer and collaborative sources have not been independently audited.
