A physician-scientist receives a K award. Five years of protected time, roughly a million dollars, a mentor, and a mandate to become an independent investigator.
She does everything correctly. She publishes. She attends the grant writing workshops. She refines her specific aims until they are clean. Her mentor is genuinely supportive and reads every draft.
Then she submits her first R01, and the reviewers ask a question that no amount of writing skill answers:
Where is the biostatistician with relevant expertise? Where is the second site? Where is the cohort large enough to power this? Who is the co-investigator with the methodological expertise this design requires?
She does not have them. Not because she does not know she needs them, but because her entire professional network consists of her mentor, her department, and the people her mentor happened to introduce her to.
Her competitor at a large institution with a deep research infrastructure has all of it, assembled almost passively, from proximity.
This is the K-to-R cliff, and it is usually described as a mentoring problem or a grant-writing problem. The evidence suggests it is substantially a team formation problem, which is a completely different thing and has a completely different solution.
The numbers on the transition
The conversion rate from career development award to independent funding is worse than most people outside the system realize.
An analysis of 1,778 NCI K awards from 1981 to 2013 found:
- About 43 percent ever received an R01-equivalent.
- Median time to that award: 5.5 years.
- Being in an independent hospital setting was associated with higher odds (OR 1.45), as were publications per year (OR 1.40).
So more than half of a heavily selected, heavily funded cohort never reaches independence, and those who do take more than five years past their K.
The broader picture is consistent:
- MDs are 15 percent of R01 awardees.
- Physician-scientists are roughly 1.3 percent of US physicians.
- Among 260 surveyed in 2024, 49 percent had considered leaving.
- The age at first R01 has shifted from investigators' thirties in 1981 to their forties by 2023.
- K08 and K23 recipients fell from 2,233 in 2005 to 1,642 in 2016, recovering to 2,070 by 2020.
Each failed transition writes off roughly a decade of training and around a million dollars of career development funding, and removes a clinician from the interface between laboratory and bedside that the entire mechanism exists to staff.
The network data is the striking part
Here is the evidence that reframes the problem.
An analysis of NIH multi-principal-investigator grants covering 30,127 principal investigators across 86,743 R01-equivalent grants from 2006 to 2023 mapped the collaboration network.
The findings describe a fragmented graph:
- The largest connected component contains 13,873 investigators.
- There are 3,383 isolated two-person components: pairs of investigators connected to each other and to nobody else in the network.
- The average investigator has roughly three collaborators.
NIH created the multi-PI mechanism explicitly to enable team science. What the network shows is that team formation is happening in small, isolated clusters rather than across a connected research community.
Three collaborators on average. Thousands of two-person islands.
That is not a scientific community assembling optimal teams. It is a large number of people working with whoever is nearby.
Collaboration capital is inherited, not built
The mechanism becomes clear once you ask where a junior investigator's collaborators actually come from.
From the mentor. The K mentor's network is the primary inheritance. If the mentor collaborates with a strong biostatistician, the awardee gets access. If not, they do not.
From the institution. A large research university provides methodologists, cores, cohorts, and colleagues by proximity. A community-affiliated or smaller institution provides fewer.
From the training lineage. Fellowship co-trainees who have dispersed, though these connections decay quickly without maintenance.
Almost nothing from any market. There is no mechanism to discover a willing co-investigator you do not already know.
Which produces two consequences that the conversion data reflects.
First, collaboration capital is distributed by accident. Awardees at institutions with deep infrastructure convert at higher rates, and the mechanism is not that they are better scientists.
And second, moving institutions can sever the graph. A physician-scientist who relocates between their K and their R may lose the collaborators, the cohort access, and the institutional relationships their application depended on. The mobility literature here is thin, which is itself notable given how consequential it appears.
Why the search tools do not solve it
Several tools attempt to address discovery and each addresses a different problem.
NIH RePORTER and its matchmaker function surface funded projects and program officers, which is genuinely useful for understanding a portfolio and identifying where a project might fit. It tells you who has been funded. It does not tell you who is available and willing to join your team.
Institutional research profile systems were built explicitly for expertise discovery within universities. They largely stagnated for reasons this series has documented repeatedly: self-maintained profiles go stale, and nobody has an incentive to keep them current.
And the European experience is instructive. For Horizon consortium formation, which is the same problem at continental scale, a survey of national contact points described the official Partner Search Tool as "rarely used and only exceptionally successful" in identifying consortium partners.
That is a tool built specifically to solve partner discovery, funded and mandated, and reported as rarely used.
The recurring failure is that these systems index published work rather than current intent. They can tell you who did something. They cannot tell you who wants to do something next, which is the only question that matters for team formation.
Availability is the missing variable, again
This is the same structural gap that appears in cohort discovery, in peer review, in investigator recruitment for trials, and in nearly every other coordination failure in this series.
What a junior investigator needs to know is not who is qualified. It is who is qualified and available and interested in this specific project.
That state is private, changes constantly, and is published nowhere.
And the signaling problem runs in both directions:
The junior investigator cannot credibly signal what they bring. A senior methodologist receiving a cold email from an unknown assistant professor has no basis to evaluate it, and receives many such emails.
The senior investigator cannot see who needs them. A biostatistician with capacity and interest in a clinical area has no way to find the three K awardees within reach who need exactly that expertise.
So both sides wait, and the graph stays fragmented into two-person islands.
Why this matters more in the next few years
Two pressures make the current cohorts unusually fragile.
Funding turbulence raises the premium on team applications. Periods of budget uncertainty and policy change tend to advantage multi-site, multi-investigator applications with demonstrated infrastructure, which are precisely the applications that require a network to assemble. The investigator without collaboration capital is disadvantaged exactly when the environment tightens.
And the workforce is already strained. Physician-scientists at 1.3 percent of physicians, with 49 percent of a surveyed cohort having considered leaving, is a pipeline with limited resilience. Cohorts that fail to convert during a difficult period do not generally return.
What would actually work
A declared team-building intent signal. An investigator stating, visibly to appropriate peers, "assembling an R01 on this topic, need a biostatistician with these methods, a second site with this population, and a co-investigator with this expertise." That converts a private need into a discoverable one, and nothing currently does it.
Availability from the other side. A methodologist stating "I have capacity for one new collaboration this year in this area." The single most useful piece of information in the entire system and it is published nowhere.
Vouching through the training cohort. A junior investigator's fellowship classmates, now dispersed across institutions, are exactly the vouched, verified connections that could introduce them to collaborators outside their mentor's network. That graph exists and is not addressable.
Senior and retired investigators as co-investigators and advisors. An enormous and underused pool. Retired methodologists and senior scientists frequently retain deep expertise, have time, and are no longer competing for the same resources. They are also invisible to publication-based discovery for the reasons this series has covered.
And a record of role fulfilment. If someone commits to a collaboration and delivers, that should be visible. Team formation between strangers requires some basis for trust, and completed collaborations are the natural currency.
What you can do now
If you are on a K award
Audit your graph now, not in year four. Write down every collaborator you could credibly include on an R01. If the list is short and consists mostly of your mentor's contacts, that is your binding constraint, and it takes years to fix.
Build one relationship outside your mentor's network per year. Deliberately. A methodologist, a second-site investigator, a cohort holder. This is more consequential to your conversion odds than another writing workshop.
Reconnect with your fellowship cohort. They are now dispersed across institutions with access to different populations, methods, and resources. This is the highest-value network you already have and it is decaying.
Say what you need out loud. In talks, in emails, at meetings: "I am building toward an R01 on this and I need a collaborator with X." People cannot offer what they do not know you need, and most investigators are strangely private about this.
Ask about mobility before you move. If relocating, understand which collaborations, cohorts, and resources travel with you and which do not.
If you are senior
Say what capacity you have. "I can take one new collaboration this year in this area" is information that nobody publishes and that would immediately be useful to several early-career investigators near you.
Introduce beyond your own group. The most valuable thing a mentor provides is not advice but access, and access to your own network is only the first hop. Introducing your mentee to someone outside it is worth more.
Take the cold emails seriously. A junior investigator writing to you has no other mechanism. Ten minutes and a redirect costs you almost nothing.
If you direct a career development program
Measure network growth, not just publications. Distinct collaborators outside the mentor's network is a leading indicator of conversion that nobody tracks and that is more actionable than publication counts.
Fund team formation directly. Travel, protected time, and structured introductions aimed at building collaborations outside the home institution. This is a smaller intervention than most program elements and plausibly more decisive.
Look at your own conversion rate by institution type. If awardees at resource-rich institutions convert better, that gap is largely infrastructure and network, and it is the thing your program could actually supply.
Frequently asked questions
What proportion of K awardees obtain an R01? An analysis of 1,778 NCI K awards from 1981 to 2013 found approximately 43 percent ever received an R01-equivalent, at a median of 5.5 years to that award, with higher odds associated with independent hospital settings and higher publication rates.
How many physician-scientists are there? Roughly 1.3 percent of US physicians, with MDs comprising about 15 percent of R01 awardees. Among 260 physician-scientists surveyed in 2024, 49 percent had considered leaving, and K08 and K23 recipient numbers fell from 2,233 in 2005 to 1,642 in 2016 before recovering to 2,070 by 2020.
When do investigators receive their first R01? Later than they used to. The age at first R01 has shifted from investigators' thirties in 1981 to their forties by 2023, extending the period of dependence on career development funding and institutional support.
Is NIH team science working? Partially and unevenly. Analysis of 30,127 principal investigators across 86,743 multi-PI grants from 2006 to 2023 found a fragmented network with roughly three collaborators per investigator on average and 3,383 isolated two-person components, indicating team formation clustering locally rather than across a connected research community.
Why can't investigators find collaborators through existing tools? Because those tools index published work rather than current availability and intent. NIH RePORTER surfaces funded projects and program officers, institutional profile systems rely on self-maintained data that goes stale, and in the European context a purpose-built partner search tool was described in a national contact point survey as rarely used and only exceptionally successful.
What most improves the odds of converting a K to an R01? The evidence points toward assembling a genuine team: co-investigators with complementary methodological expertise, access to adequate cohorts, and multi-site partnerships. These are network assets, typically inherited from a mentor or an institution rather than built deliberately, which is why conversion correlates with institutional setting.
The bottom line
The physician-scientist pipeline is described as a mentoring problem and treated with mentoring interventions: committees, workshops, writing programs, structured feedback.
The people failing to convert are not, for the most part, short of ideas, training, or feedback on their specific aims. They are short of a biostatistician, a second site, and a cohort, which are network assets that arrive by inheritance from a mentor or by proximity to a well-resourced institution.
The network data makes this visible. Thirty thousand investigators, three collaborators each on average, thousands of isolated two-person islands. That is what a research community looks like when there is no mechanism to find anyone you do not already know.
So more than half of a heavily selected, million-dollar-funded cohort never reaches independence, at a moment when physician-scientists are 1.3 percent of the profession and half of them have thought about leaving.
Somewhere near each of them is a methodologist with capacity, a retired investigator with time and expertise, and a colleague at another institution with exactly the population they need.
The tools built to find those people index what everyone has already published, which is a record of the past, when the only question that matters is who is free next year.
Part of a series on the missing professional infrastructure of healthcare. Previously: The GME Memory Reset
Evidence note: K to R conversion figures come from an analysis of 1,778 NCI K awards published in the Journal of Cancer Education (2025). Physician-scientist workforce figures come from National Academy of Medicine perspectives and FASEB Journal analysis, with the 49 percent considered-leaving figure drawn from a 2024 survey of 260 physician-scientists. Multi-PI network analysis comes from a 2026 preprint examining 30,127 investigators and 86,743 grants, which has not completed peer review. The characterization of the European partner search tool comes from a national contact point survey whose source page could not be independently retrieved and should be verified before republication.
