A Comprehensive Evidence Review for Clinicians
Prepared: 6 November 2026 Purpose: To place the September 2026 proposed rescissions of health spending in the context of what is actually known — from peer-reviewed and official sources — about how federal health funding instability affects research output, the research workforce, health-services infrastructure, service delivery, and population health. Citation method: Every factual claim carries an atomic source token that is resolved mechanically against NCBI records and the reference pool; no claim in this review rests on recalled figures.
Abstract
On 25 September 2026 the White House transmitted 11 proposed rescissions of budget authority totalling $810 million to Congress under section 1012(a) of the Congressional Budget and Impoundment Control Act of 1974 [1]. Three HHS accounts accounted for roughly $600.2 million of that total: Refugee and Entrant Assistance ($567,401,904), the Agency for Healthcare Research and Quality ($27,700,000), and the Office of the Secretary’s general departmental management ($5,098,000) [1]. Four days later GAO’s General Counsel concluded that all 11 accounts expire at the end of fiscal year 2026 on 30 September, that the President may not force the expiration of budget authority Congress enacted and did not rescind, and that Congress’s 45-day review period would not end before 9 November 2026 at the earliest; GAO stated that this was not a position on the merits of the programs [2]. OMB directed affected agencies to hold the funds [2]. The episode is best understood not as a single budget event but as a worked example of a structural property of American health funding: appropriations are time-limited, project-based, and competitive, so the timing and continuity of money are separate variables from its amount.
This review asks what the evidence base can and cannot support. It first describes what the three affected HHS accounts purchase — short-term medical coverage and medical screening for refugees and other eligible arrivals, care and custody of unaccompanied children, and the measurement and evidence infrastructure that AHRQ operates [3][4]. It then examines six literatures: whether grant money buys scientific output; what discontinuous funding does to research careers; whether public funding generates downstream innovation; whether health-services research changes clinical practice; whether public-health spending changes population outcomes; and what happens when coverage or services are withdrawn. Two consistent findings emerge.
First, the marginal productivity of research money is real but modest and hard to isolate. The best-identified study of NIH grants finds that receiving an R01 worth roughly $1.7 million produced about one additional publication over five years — a 7% increase — consistent with a competitive market in which investigators substitute other funding [5]. Across Canadian granting councils, impact rose with funding but decelerated, so impact per dollar was lower for large grants [6]. Peer-review percentile predicted later high-impact output poorly (area under the ROC curve 0.52) and did not predict output per million dollars [7].
Second, the empirical literature on spending and outcomes is dominated by observational designs whose direction of causation is contestable. A widely cited estimate from the local public-health literature is that a 10% increase in local public-health spending was associated with 1.1% to 6.9% lower mortality from preventable causes [8]; a meta-regression across 65 studies placed the mortality elasticity of health spending near −0.13 after correction for publication selection [9]. Medicaid expansion studies report all-cause mortality reductions from 6.1% in an early three-state comparison [10] to 11.8 deaths per 100,000 adults in a later national analysis, with state-level effects ranging from −63.8 to +30.4 deaths per 100,000 [11] — heterogeneity wide enough that the average should not be applied to any particular state.
What the evidence does not support is a claim that the September 2026 rescissions will produce measurable health harm. The affected sums are small relative to the accounts they sit in and small relative to total federal health spending; the specific consequences depend on unobligated balances, timing, litigation, and congressional action, none of which is yet observable. What can be said is narrower and more defensible: these service types have documented clinical need, the measurement infrastructure at issue has no ready substitute, and abrupt discontinuities in funding have repeatedly been followed by measurable changes in service volume and workforce in settings where data existed to observe them [12][13]. Distinguishing those three kinds of claim — what a program funds, what the literature shows, and what follows for a clinician — is the organising discipline of this review.
1. Introduction: What Happened, and What an Evidence Review Can Answer
1.1 The event, stated narrowly
On 25 September 2026 the President transmitted to both Houses of Congress 11 proposed rescissions of budget authority totalling $810 million, pursuant to 2 U.S.C. 683(a) [1]. The transmittal letter states that the proposals affect programs of the Departments of Commerce, Education, Health and Human Services, Homeland Security, Housing and Urban Development, and Justice, as well as International Assistance Programs [1]. The printed presidential communication confirms both the aggregate figure and the statutory basis [1]. Wire reporting described the package as “nearly $1 billion,” a rounding that is defensible only if the transmittal figure of $810 million is stated alongside it [14].
Three HHS accounts dominate the package:
| Account | Amount proposed for rescission | Stated appropriation | Stated justification in the transmittal |
|---|---|---|---|
| ACF, Refugee and Entrant Assistance (075-1503), FY2024/FY2026 | $567,401,904 | $6.3 billion (FY2024) | Account overfunded because executive actions sharply curtailed arrivals |
| AHRQ, Healthcare Research and Quality (075-1700), FY2026 | $27,700,000 | $345 million (FY2026) | Research judged wasteful or duplicative of NIH; specific AHRQ products and grants cited |
| HHS Office of the Secretary, General Departmental Management (075-0120), FY2026 | $5,098,000 | $625 million (FY2026), including $75 million in report language for the Office of Minority Health | Grants described as race- and sex-based and inconsistent with anti-DEI executive orders |
Source: transmittal letter and enclosure as printed [1]. The justifications are the administration’s own stated rationales, reported here as such.
Arithmetic from those figures places the HHS share at approximately $600.2 million, or about 74% of the $810 million package, and places the three proposals at roughly 9.0%, 8.0%, and 0.8% of their respective stated appropriations. Those percentages are derived, not stated in the transmittal, and they matter for interpretation: none of the three proposals would eliminate its account.
1.2 The legal conclusion, stated as its own claim
On 29 September 2026 GAO’s General Counsel concluded that all 11 accounts expire at the end of fiscal year 2026, on 30 September 2026, and that the President may not force the expiration of budget authority that Congress enacted and did not rescind [2]. Under the Impoundment Control Act, GAO concluded, the President may withhold funds for up to 45 days of continuous congressional session, and Congress’s review period would not end until 9 November 2026 at the earliest [2]. GAO stated that its finding was not a position on the merits of the programs [2]. OMB directed affected agencies, including HHS, to hold the funds [2].
The White House characterised the mechanism as a “pocket rescission” and described it as the second such use, with the prior year’s use described as the first in 50 years [1]. Wire reporting describes that prior use as blocking $4.9 billion in congressionally approved foreign aid a year earlier, and reports that the Supreme Court declined to block those rescissions on foreign-affairs grounds [14].
Bipartisan congressional criticism was reported. Senate Appropriations Chair Susan Collins (R-Maine) said the action came without warning or consultation, described it as an encroachment on Congress’s constitutional power of the purse, and said she would work with colleagues to address what she called “these illegal actions” [14]. Those characterisations are Senator Collins’s own, reported as attributed criticism; this review does not adopt them and takes no position on the legal question they raise.
This review takes no position on the legal dispute, does not predict its resolution, and does not treat either the administration’s rationale or GAO’s conclusion as established fact about consequences. The dispute is relevant to clinicians for one practical reason: it determines whether money that has been appropriated reaches the organisations that spend it, and when.
1.3 Three questions an evidence review can answer
The event invites three questions, only two of which the literature can address.
Question 1 — descriptive. What do these accounts actually purchase? This is answerable from official program documents, and it is the necessary first step because a dollar figure means nothing clinically until it is attached to a service [3][4].
Question 2 — empirical. What does the peer-reviewed literature show about the relationship between funding and (a) research output, (b) research careers, (c) downstream innovation, (d) clinical practice, and (e) population health? This is answerable, but unevenly: some links are supported by well-identified quasi-experiments, others mainly by ecological correlations, and several by nothing stronger than plausibility.
Question 3 — inferential. What follows for a given clinician, clinic, or investigator? Here the literature provides analogues, not answers. Any statement about the consequences of the September 2026 rescissions for patients is inference, and is labelled as such throughout.
A fourth question — will this specific action cause health harm? — is not answerable now. The honest position, stated once here and repeated in the Limitations section, is that the evidence supports statements about funding volatility in general and about these service types in general, and that the specific consequences of this action are not yet observable.
1.4 Why a physician audience should care about a $27.7 million line item
A reasonable objection is that the amounts are trivial against a federal health budget measured in trillions. That objection misreads what is being cut. AHRQ’s appropriation is small precisely because its output is not services but the instruments by which services are measured: the Medical Expenditure Panel Survey, the Healthcare Cost and Utilization Project, the national healthcare quality and disparities reporting, the Consumer Assessment of Healthcare Providers and Systems surveys, and the evidence reviews that support preventive-services recommendations [4]. These are public goods with high fixed costs and no commercial substitute, and their value is not proportional to their budget. The same logic applies to the Refugee Medical Assistance and medical-screening functions within the refugee account, which are small line items attached to a legal admission pipeline [3].
Whether that structural argument is correct is an empirical question, and it is one this review addresses in Sections 8 and 12 rather than asserting here.
2. What the Affected Accounts Actually Fund
2.1 Refugee and Entrant Assistance
The Refugee and Entrant Assistance account funds the domestic side of refugee resettlement. ORR’s Refugee Program Bureau describes its purpose as helping refugees and other eligible individuals achieve economic self-sufficiency, well-being, and successful assimilation, and describes a sequence of services: initial resettlement upon arrival; time-limited cash and medical assistance; and longer-term resettlement services available for up to five years, including job readiness, case management, English language training, and specialised supports for children and youth, delivered through states, resettlement providers, and community-based organisations [3]. The Bureau lists its legislative authority as the Refugee Act of 1980, implemented through 45 CFR Parts 400-401 [3].
The clinically relevant components are specific:
| Program | What it provides, in ORR’s own description |
|---|---|
| Refugee Medical Assistance | Short-term medical coverage to refugees who are ineligible for Medicaid |
| Medical Screening | Initial domestic medical examination of arriving refugees, described as safeguarding public health and supporting healthy resettlement |
| Refugee Cash Assistance | Short-term financial assistance for refugees ineligible for TANF |
| Refugee Support Services | Employment services, case management, English language training |
| Preferred Communities | Support for refugees with unique vulnerabilities requiring special attention |
| Survivors of Torture | Services to restore health and dignity for people tortured in another country; eligibility does not depend on immigration status and there is no time limit |
Source: ORR Refugee Program Bureau program listing [3].
A separate ORR bureau administers the care and custody of unaccompanied alien children [15]. The distinction matters clinically: the refugee account funds community-based services for people already resettled, whereas unaccompanied children’s care is a custodial and placement function with its own medical and mental-health obligations. Conflating the two is a common error in commentary, and it obscures which clinical services are actually at issue.
The transmittal’s stated justification for this rescission is that the account is overfunded because two executive actions — EO 14165, “Securing Our Borders,” and EO 14163, “Realigning the United States Refugee Admissions Program” — sharply curtailed arrivals [1]. That is a claim about admissions volume, not about the cost of serving people already in the country. Whether the two move together is an empirical question about the resettlement pipeline, and the transmittal does not present data on it.
2.2 AHRQ and the evidence infrastructure
AHRQ describes itself as a component of HHS dedicated to health services research — “the field of study that examines healthcare as it is delivered to patients, examining the quality, safety, and affordability of healthcare delivery” — and cites 42 U.S.C. § 299 et seq. as its statutory basis [4]. Its stated functions include patient-safety research, data analytics, evaluation of the strength of evidence for procedures, treatments, and preventive interventions, and training future health services researchers [4].
The agency’s product list is the best available description of what would be affected by a reduction in its appropriation [4]:
- Data infrastructure: the Medical Expenditure Panel Survey (MEPS), the Healthcare Cost and Utilization Project (HCUP), and the National Healthcare Quality and Disparities Report data tools.
- Measurement instruments: the Consumer Assessment of Healthcare Providers and Systems (CAHPS) programme, AHRQ Quality Indicator tools, and Surveys on Patient Safety Culture.
- Evidence synthesis: the Evidence-based Practice Center programme and its reports, which supply the analytic base for preventive-services recommendations.
- Patient-safety resources: the Network of Patient Safety Databases and related tools.
Two features of this list deserve emphasis. First, several items are not research projects in the ordinary sense but standing instruments — surveys, indicator sets, reporting systems — whose value depends on uninterrupted time series. An instrument that is paused and restarted does not simply resume; it loses the comparability that made it useful, a point developed in Section 12. Second, the transmittal’s stated justification is that the research is wasteful or duplicative of NIH and cites specific AHRQ-funded products and grants [1]. A claim of duplication is testable in principle — one could compare portfolios — but the transmittal does not present such a comparison, and this review found no peer-reviewed portfolio analysis that adjudicates it. Readers should treat “duplicative” as an assertion by the proposer, not as a finding.
2.3 The Office of Minority Health line
The third HHS proposal, $5,098,000 from the Office of the Secretary’s general departmental management account, concerns funds that the transmittal describes as $75 million included in report language for the Office of Minority Health within a $625 million appropriation [1]. The stated justification is that certain grants are race- and sex-based and inconsistent with anti-DEI executive orders [1]. OMH describes its work as population-specific data, program development, and grant resources directed at racial and ethnic minority health [16].
The proposed amount is small in absolute terms and small relative to the report-language figure — under one-tenth of it by arithmetic. The reason it belongs in this review is that it illustrates a general mechanism: when a program’s funding arrives through report language rather than explicit authorisation, its political durability is lower, and a rescission proposal can target it without amending the underlying statute. That is a structural observation about appropriations practice, not a judgment about the merits.
2.4 What the scale actually tells you
It is tempting to conclude either that the package is trivial or that it is catastrophic. Neither inference is supported. The package equals roughly 74% of $810 million in HHS funds and less than one-tenth of one percent of the three accounts’ stated appropriations by arithmetic [1]. What makes it consequential is not the amount but three design features: the funds sit in accounts with end-of-year expiration, the affected activities include standing measurement instruments with no substitute supplier, and the recipients are organisations — state health departments, resettlement agencies, academic centres — that cannot absorb a mid-year stop and restart without losing staff [4][3].
Those are inferences about mechanism. Section 13 examines historical episodes in which comparable discontinuities occurred and in which someone measured what followed.
3. The Legal Mechanics, Plainly
3.1 What budget authority is, and why the calendar matters
An appropriation gives an agency authority to incur obligations, and that authority is usually available for a defined period — one fiscal year for most annual appropriations, which for the federal government ends on 30 September. When the period of availability ends, unobligated balances expire and generally cannot be used for new obligations. This is why the date in the GAO letter is not a technicality: the dispute is over whether authority that would otherwise lapse on 30 September can be treated as having lapsed early, at the President’s direction, or whether it remains available to the agency until Congress acts.
3.2 What a rescission is
Section 1012(a) of the Impoundment Control Act, codified at 2 U.S.C. § 683(a), provides that when the President determines that all or part of any budget authority will not be required to carry out the full objectives or scope of the programs for which it is provided, or that the authority should be rescinded for fiscal-policy or other reasons, the President “shall transmit to both Houses of Congress a special message” specifying the amount, the account and project involved, the reasons, the estimated fiscal and budgetary effect, and the facts and considerations bearing on the decision [17].
Crucially, the statute then constrains what may follow. Subsection (b) provides that any amount proposed for rescission “shall be made available for obligation unless, within the prescribed 45-day period, the Congress has completed action on a rescission bill rescinding all or part of the amount proposed”; it adds that funds made available under this procedure “may not be proposed for rescission again” [17]. In other words, the ICA creates a proposal right, not a unilateral power: the President may ask, and may temporarily withhold while Congress considers, but the default is that the money must be released.
3.3 What GAO decided
GAO’s role under the ICA is to review proposed rescissions and report to Congress. In the 29 September 2026 letter, GAO’s General Counsel concluded that all 11 accounts expire at the end of fiscal year 2026; that the President may not force the expiration of budget authority Congress enacted and did not rescind; that the withhold is limited to 45 days of continuous congressional session; and that the review period would not end until 9 November 2026 at the earliest [2]. GAO expressly stated that this was not a position on the merits of the programs [2].
Two features of that conclusion deserve a clinician’s attention. First, the 45-day clock runs on continuous session days, not calendar days, which is why a message sent in late September can produce a review deadline in November. Second, the conclusion is about timing and authority, not about whether the programs are worth funding. A reader who supports the programs and a reader who opposes them can both accept GAO’s legal analysis without conceding the policy argument.
3.4 “Hold” versus “rescind”: what it means for a grantee
The distinction that matters operationally is between three states:
- Proposed rescission. Funds are identified in a special message; under the ICA they must be released unless Congress completes action within the 45-day period [17].
- Hold. An agency instructs programme offices not to obligate the identified balances pending resolution. OMB directed affected agencies, including HHS, to hold the funds [2]. A hold does not change the underlying appropriation; it changes what the agency is willing to do with it in the interim.
- Rescission enacted. Congress passes a rescission bill and the authority is permanently cancelled.
For a recipient organisation, the practical exposure differs by state. Funds already obligated under an executed award are legally committed; unobligated balances and not-yet-issued awards are where a hold bites. This review did not retrieve agency-level implementation guidance specifying how each affected programme would treat existing awards, and readers should not assume that a headline figure translates one-for-one into cancelled grants. That is a genuine gap in the public record as of this writing, and it is noted in the Limitations section.
3.5 The “pocket rescission” label
The White House described the mechanism as a “pocket rescission” and said this was the second such use, describing the prior year’s use as the first in 50 years [1]; wire reporting characterises that earlier use as blocking $4.9 billion in foreign aid and places it nearly 50 years after the last such action [14]. The term is not statutory. It describes the practical effect of transmitting a rescission message close enough to the end of a fiscal year that the 45-day review period cannot run before the balances would lapse anyway. Whether that characterisation is accurate as a matter of law is precisely what GAO’s letter disputes [2]; whether it is accurate as a matter of practice depends on what agencies do with the balances in the interim.
For an evidence review, the relevant point is narrower: the mechanism produces a period of legal uncertainty measured in weeks, at the end of a fiscal year, over accounts that fund services delivered by organisations operating on annual budgets. The literature on what such uncertainty does is examined in Section 13.
4. The Economic Structure of US Health Research Funding
4.1 Who pays, and through what instruments
Federal health research funding is not a single stream. The National Institutes of Health is the largest funder of biomedical research in the world and distributes most of its extramural money through project grants — the R01 being the canonical mechanism — awarded competitively after peer review, for defined periods, with renewal requiring a new application. AHRQ funds health-services research through a much smaller appropriation. The Centers for Disease Control and Prevention funds public-health programmes and some surveillance infrastructure. State and local health departments are funded through a mixture of federal grants, state appropriations, and local revenue.
Three structural features follow from that architecture and are independent of any administration:
- Time-limited authority. Awards have end dates; programmes have appropriation cycles. Continuity requires active renewal.
- Project-based competition. Money attaches to proposals, not to people or institutions, which means an investigator’s funding can go to zero between awards.
- Fixed-cost infrastructure financed by variable grants. Surveys, cohorts, and registries have costs that do not scale down proportionally when a grant ends.
4.2 Volatility as a design property, not an accident
Alberts and colleagues described the resulting system in 2014 as an “unsustainable hypercompetitive system” built on a long-held but erroneous assumption of never-ending rapid growth, one that discourages prospective students from entering the profession and makes it difficult for seasoned investigators to do their best work [18]. The argument is not that funding is too low in absolute terms but that the rate of growth the system was built to expect cannot be sustained, and that institutions and careers are structured as though it could.
The empirical signature of that structure is visible in application and award statistics. Across NIH from 2006 to 2016, applications rose 18.7% while the number of funded grants fell 6.7% and the number of R01s fell 17.0%; the mean success rate for grants with surgeons as principal investigators was 16.4%, below the 19.2% NIH mean [19]. A system in which fewer than one in five applications succeeds, and in which the ratio of applications to awards is rising, transmits volatility directly to investigators regardless of the total appropriation.
4.3 The grant cycle as a transmission mechanism
The cycle has five points at which a funding discontinuity can bind: application timing, review, award issuance, obligation, and renewal. Each has a different lag. Review takes months; awards are issued on a fiscal calendar; obligations can be constrained by a hold; renewals depend on the next year’s appropriation. A rescission proposal in late September therefore lands at the point of maximum leverage — after awards for the closing year have been issued but before the next year’s cycle is funded.
This is why the size of a rescission is a poor guide to its effect. A $27.7 million proposal against a $345 million appropriation [1] is 8% by arithmetic, but if it is concentrated in a programme that supports a standing survey or a training cohort, the discontinuity can exceed its share.
4.4 What the structure implies for appraisal
Two implications shape the rest of this review.
First, studies of “funding and productivity” typically estimate the effect of additional money at the margin, not the effect of loss. The two are not symmetric: a lab that loses a grant may dissolve, whereas a lab that gains one may simply hire. Almost all of the strongest evidence concerns marginal gains [5][6], and that asymmetry should temper any extrapolation to contraction.
Second, because funding is allocated by peer review, and because peer review is only modestly predictive of subsequent impact [7], variation in funding is partly quasi-random with respect to quality. That is bad for the allocation of science but useful for causal inference, and it is why the grant-funding literature can say something credible at all. Section 5 takes up that literature and its confounding problem directly. ## 5. Does Money Buy Science? Grant Funding, Productivity, and the Confounding Problem
5.1 The question, precisely
“Does funding buy science?” is really three questions: does an additional dollar to an investigator produce additional output; does it produce better output; and would removing a dollar remove output? The literature answers the first with a qualified yes, the second with a qualified no, and the third almost not at all.
5.2 The best-identified estimate: Jacob and Lefgren
The strongest study in this literature exploits a feature of NIH review that approximates random assignment near the funding threshold. Jacob and Lefgren assembled all applications — unsuccessful as well as successful — to NIH standard research grants from 1980 to 2000 and estimated the effect of receipt using both ordinary least squares and instrumental variables [5]. Receipt of an NIH research grant worth roughly $1.7 million led to one additional publication over the next five years, a 7% increase [5].
The authors interpreted that small effect as consistent with a competitive market for research funding in which the loss of an NIH grant simply causes researchers to shift to another source [5]. That interpretation is the single most important sentence in this literature for present purposes, because it implies that the marginal NIH dollar is partly a substitute for a dollar from somewhere else, not a pure addition to scientific effort.
Three limitations should be stated plainly. The estimate concerns marginal awards to investigators who applied and were near the threshold; it says nothing about investigators who never apply, nor about the disappearance of an entire programme. The outcome is publication count, which weights a case report and a practice-changing trial equally. And the sample period ends in 2000, before the doubling of the NIH budget was fully absorbed and long before current success rates.
5.3 The shape of the funding–impact relationship: Fortin and Currie
If marginal dollars have small effects, one might still expect large grants to outperform small ones, because scale should permit ambitious work. Fortin and Currie tested that proposition directly in three disciplines funded by the Natural Sciences and Engineering Research Council of Canada, relating four impact indices — article counts, citation counts, most-cited article, and number of highly cited articles, each measured over four years — to the amount of funding received [6].
Impact was positively but only weakly related to funding; impact was generally a decelerating function of funding, so impact per dollar was lower for large grant-holders; investigators who received additional funds from a second federal council, CIHR, were not more productive than those funded only by NSERC; and the impact of researchers who received funding increases did not predictably increase [6]. The authors concluded that scientific impact, as reflected by publications, is only weakly limited by funding, and suggested that strategies targeting diversity rather than “excellence” might prove more productive [6].
This is a cross-sectional analysis of funded investigators, not a randomized comparison, and it cannot rule out that the most able investigators are both better funded and more productive. Its value is in the shape of the relationship: even under confounding that should bias the estimate upward, the curve bends down.
5.4 What a grant actually produces, and how variable that is
A different way to ask the question is descriptive: given an award, what comes out? Agarwal and Tu followed all 4,451 initial R01 grants awarded by NIH in 2000 for 20 years. Those grants generated 55,053 publications and 3,705,553 citations; the median grant produced 8 publications over two decades (interquartile range 4 to 17), with a median of 441 citations and a median grant-level h-index of 7 [20]. The authors emphasised that trajectories were highly variable among awards and that the metrics captured different domains of success [20].
The distribution matters more than the median. A system in which the median R01 yields eight papers over twenty years, and in which the top decile yields many times that, is a portfolio system: it depends on a large number of bets, most of which return modestly. Portfolio logic is robust to the loss of any single bet and fragile to the loss of a whole class of bets — a distinction that becomes relevant when a programme rather than a project is cut.
5.5 Peer review as a weak predictor
If peer review identified the best science reliably, then funding and impact would be tightly linked. They are not. Lauer and colleagues analysed 6,873 de novo cardiovascular R01 grants funded by the National Heart, Lung, and Blood Institute between 1980 and 2011, examining top-10% articles. The grants yielded 62,468 articles, of which 13,507 (22%) were top-10%. Better percentile ranking was modestly associated with more top-10% articles, but discrimination was poor — area under the ROC curve 0.52 (95% CI 0.51–0.53) — and, critically, there was no association between percentile ranking and top-10% articles per million dollars spent [7].
Pier and colleagues showed why, at the level of the reviewer. Replicating all aspects of NIH peer review, they had 43 reviewers rate the same 25 applications. There was no agreement among reviewers on quality, whether measured qualitatively or quantitatively, and no agreement on how a given number of strengths and weaknesses translated into a numeric score; the outcome appeared to depend more on which reviewer was assigned than on the proposal [21].
Together these findings have an uncomfortable implication for both advocates and critics of funding agencies. If review cannot reliably distinguish the top proposals, then a funding cut of 8% removes a set of grants that is close to a random sample of the applicant pool in expected quality — which limits the scientific damage of a marginal cut, and simultaneously undermines the meritocratic justification for any particular allocation. Fang and Casadevall drew the logical conclusion and proposed a modified lottery: peer review to identify meritorious proposals, then random selection among them, on the grounds that review panels cannot accurately stratify proposals and that bias is magnified by low paylines [22].
5.6 The natural experiment that speaks most directly to volatility: ARRA
The American Recovery and Reinvestment Act of 2009 gave NHLBI an unusual opportunity: it funded 165 de novo R01 grants that fell below the institute’s payline, alongside 458 funded by meeting the payline [23]. This is close to the experiment a contraction literature would want, in reverse — an exogenous expansion that let reviewers’ threshold drop.
The ARRA grants received less money (median $1.03 million vs $1.87 million) over a shorter period (median 3.0 vs 4.9 years), and generated fewer publications (996 vs 3,895 through May 2014, consistent with their smaller size and number). But on the outcome that matters for efficiency — normalized citation impact per $1 million spent — the two groups were statistically indistinguishable (median 2.15, IQR 0.73–4.68 vs 2.03, IQR 0.75–4.10; P=0.61) [23]. Grants that peer review had ranked below the payline produced, per dollar, output of comparable impact.
The inference for contraction is indirect but important: it suggests that the marginal grant is not dramatically worse than the average grant, so the quality cost of funding fewer grants may be smaller than intuition suggests. It says nothing about the capacity cost — the lab that closes, the cohort that disperses, the survey that stops — because those are not measured in citations per dollar.
5.7 Correlation studies and what they cannot show
A parallel literature reports that funded investigators outperform unfunded ones on bibliometric indices. Among academic otolaryngologists, faculty who received NIH funding had significantly higher h-indices than non-funded peers, and h-indices increased with funding level, although there was no correlation between average h-index and funding totals at the departmental level [24]. Among orthopaedic surgeons, 22% of faculty-level foundation grant awardees later received NIH funding, rising to 46% among career-development award winners, and basic-science awards were more than four times as likely to convert (OR 4.70, 95% CI 1.66–13.33) [25].
These studies are useful for describing the career pipeline and useless for causal inference. Funded investigators differ from unfunded ones in training, institutional resources, mentorship, and prior output — all of which predict both funding and productivity. The authors of the otolaryngology study are explicit that the h-index “may be predictive of” grant awards, which is the reverse of the causal claim often attributed to such data [24].
5.8 What the cluster supports, and what it does not
Supported. Additional grant money produces measurable but modest additional output at the margin; the relationship decelerates; peer review predicts later impact poorly; a grant funded below the payline produces comparable impact per dollar; and the distribution of output across grants is heavy-tailed.
Not supported. That a specific percentage cut to a specific programme will reduce scientific output by a proportional amount. Nothing in this literature estimates that, and the substitution interpretation in Jacob and Lefgren implies the reduction could be smaller than proportional [5].
Not studied. The effect of losing funding on the composition of science — which questions stop being asked, which populations stop being sampled. Publication counts cannot detect that, and it is the outcome most relevant to a clinical audience.
6. The Research Workforce: Early-Career Attrition, Soft Money, and Discontinuous Funding
6.1 The demographic shift is well documented
The clearest workforce finding in this literature is compositional: the funded investigator population has aged, and entry has slowed.
Antman and colleagues analysed the National Cancer Institute’s R01 principal investigator workforce from fiscal years 1990 to 2016. The workforce expanded 1.4-fold and aged; the number of PIs in the three oldest age groups increased dramatically while younger groups did not; sustained increases in funded older PIs stemmed from increases in the number of older PIs submitting applications rather than higher funding rates for them; and the decline in funded younger PIs was driven in part by a marked increase in time from PhD to first R01 application and award, and in part by decreased retention of PIs in the funded R01 workforce beyond their first R01 [26].
That last clause is the one that speaks to volatility. Retention beyond the first award is not primarily a story about age; it is a story about what happens to investigators after their initial support ends.
6.2 The physician-scientist pipeline
The 2014 NIH Physician-Scientist Workforce Working Group report, as summarised by Milewicz and colleagues, identified high rates of attrition among young investigators, failure to maintain a robust and diverse pipeline, and a marked increase in the average age of physician-scientists, with older investigators continuing to work and too few younger investigators entering to replace them [27]. The authors proposed shortening time to independence by at least five years, among other measures [27].
A parallel consensus report from internal-medicine chairs described the physician-scientist workforce as having both declined and aged, and warned that the reduced inflow and outflow render the system vulnerable to sudden collapse as the senior workforce retires [28]. Both documents are consensus statements by professional bodies, not primary analyses, and should be read as informed advocacy for a position their authors hold.
6.3 Why postdocs leave, and what that implies
Lambert and colleagues surveyed 1,284 biomedical postdocs in the United States about career choices, focusing on underrepresented minority and female postdocs. They reported reasons why some choose not to pursue academic research and suggested interventions early in postdoctoral training to prevent attrition of underrepresented groups [29]. The survey design cannot establish that funding instability caused the choices, but it identifies the decision point at which the pipeline narrows.
6.4 The soft-money problem
A structural feature of US academic biomedical research that receives less attention than it deserves is that a large share of investigators’ salaries and laboratory costs are paid from grants rather than from institutional budgets. In that arrangement, a funding discontinuity is not a reduction in research activity; it is a reduction in income, with the investigator personally exposed. The literature retrieved for this review documents the aggregate consequences — aging, reduced retention, longer time to independence [26][27] — but I did not find a peer-reviewed study that estimates the causal effect of a funding gap on individual retention with a credible counterfactual.
That absence is itself a finding. The workforce literature is rich in description and thin in identification, and the policy claims made from it are therefore stronger than the evidence.
6.5 What discontinuous funding plausibly does — and how confident to be
Three mechanisms are plausible and partly supported:
- Timing mismatch. A gap between awards of 12 to 24 months exceeds the runway most soft-money investigators can bridge, so departure is rational rather than a failure of commitment. The retention finding in the NCI analysis is consistent with this [26].
- Selection against risk. Investigators who expect discontinuity choose safer projects with faster returns. This is a prediction of the Alberts argument about hypercompetition [18] and is not directly measured.
- Loss of institutional memory. When a lab disperses, tacit knowledge — how a cohort was recruited, why an assay was changed — leaves with the people. This is not captured in any bibliometric study retrieved here, and its absence from the literature is a measurement gap rather than evidence of unimportance.
The honest summary: discontinuities in funding are associated with attrition and delayed independence at the population level, the association is consistent across several datasets, and the causal estimate for an individual investigator is unknown.
7. Public Funding and Downstream Innovation: Drugs, Devices, and the Limits of Patent Linkage
7.1 The strongest claim in the literature
The most frequently cited estimate of the public sector’s contribution to drug development comes from Galkina Cleary and colleagues, who examined NIH funding associated with 210 new molecular entities approved by the FDA from 2010 to 2016. They identified more than 2 million publications related to the 210 drugs or their 151 known biological targets; more than 600,000 (29%) were associated with NIH-funded projects in RePORTER, representing more than 200,000 fiscal years of project support from 1985 to 2016 and project costs exceeding $100 billion from 2000 to 2016, roughly 20% of the NIH budget over that period [30].
The headline finding is that NIH funding contributed to every one of the 210 new molecular entities approved in that window, and that the contribution was concentrated on drug targets rather than on the drugs themselves; 84 first-in-class products were associated with more than $64 billion of NIH-funded projects, and for targeted drugs, funding related to targets preceded funding related to the drugs [30].
7.2 The patent-linkage literature and its assumptions
Sampat and Pincus examined patent citations linking NIH-funded academic medical centre research to patents. In prospective analyses, over a third of AMC grants awarded between 1990 and 1995 resulted in publications cited in patents, most of them drug and biotechnology patents assigned to private firms; patents citing NIH-funded AMC publications were associated with 106 new FDA-approved drugs, half of them new molecular entities and a quarter priority NMEs. In retrospective analyses, about half of new molecular entities approved between 2000 and 2009 had citation links to NIH-funded AMC research [31].
The authors are careful about what this means, noting that more research is needed to understand the types of links these citations represent and their policy implications [31]. That caution is warranted, because citation linkage has three well-known failure modes:
- Citation is not contribution. A patent cites prior art whether or not that art was necessary. The linkage establishes a knowledge flow, not a causal share.
- Attribution is asymmetric. Public funding that produces a tool, a target, or a negative result may be essential and yet generate no citation.
- The counterfactual is missing. Neither study can say what would have been discovered had the funding been 10% lower. The claim “NIH contributed to every NME” is compatible with a world in which a 10% reduction would have delayed some drugs by a year and left the rest unchanged.
7.3 What a clinician should take from this
The innovation literature supports a claim about architecture: public funding builds the target-validation and tool-building base on which private development depends, and that base is poorly captured by ordinary productivity metrics because its outputs are public goods [30][31]. It does not support a claim about elasticity: no retrieved study estimates how much downstream innovation would be lost per dollar withdrawn, and the long lags (decades) make such an estimate very difficult to identify.
For present purposes the implication is modest. A reduction in basic and services research funding is unlikely to change drug approval counts in the next two years, because the pipeline is long. If it has an effect, it will appear in the 2030s, in a counterfactual that no one will be able to observe. That is a reason for humility in both directions.
8. Health-Services Research as Clinical Infrastructure
8.1 The claim to be tested
The implicit claim behind concern about AHRQ reductions is that health-services research is clinical infrastructure: that it changes what clinicians do, and that removing it would change practice for the worse. That claim is testable in pieces. The evidence is strongest for measurement and for implementation support, weaker for evidence synthesis alone, and weakest for the proposition that a specific agency’s budget determines practice.
8.2 Evidence synthesis and practice: the USPSTF natural experiments
The US Preventive Services Task Force issues recommendations that are graded, and grade A and B recommendations trigger coverage requirements for many plans. Changes in those recommendations therefore function as natural experiments in the effect of authoritative evidence synthesis on practice.
When the Task Force issued a grade B recommendation for colorectal cancer screening in adults aged 45 to 49 in May 2021, Siddique and colleagues examined claims for 10,221,114 commercially insured beneficiaries. Mean screening uptake rose from 0.50% to 1.51% between the pre- and post-recommendation periods — an absolute change of 1.01 percentage points (95% CI 0.62–1.40) — but the relative change was not statistically significant (202.51%, 95% CI −30.59% to 436.87%), and the authors noted disparities by area-level characteristics [32]. In other words: a measurable but very small absolute movement, from a very low base, in a population with good access.
The reverse experiment is more informative. After the 2012 Task Force recommendation against prostate-specific antigen screening, PSA testing, biopsy, and prostate cancer incidence declined significantly in a cohort of about 6 million privately insured men followed from 2008 to 2014, with the biopsy rate per 100 patients with a PSA test falling from 1.95 to 1.52 and the proportion of new diagnoses managed with definitive local treatment falling from 69% to 54% [33]. De-implementation, in other words, happened — and at a scale that guideline authors might envy.
National trend data contextualise both findings. Across eight recommended clinical preventive services between 2011 and 2019, utilisation rose for four, fell for cervical cancer screening (annual percentage-point change −1.01), and was consistently lower among the uninsured [34]. The pattern is not one of guideline-driven transformation; it is one of slow, uneven, insurance-dependent drift.
8.3 Patient-safety programmes: large effects, fragile designs
The strongest evidence that a structured, externally supported programme can change clinical outcomes comes from the central line-associated bloodstream infection (CLABSI) literature. The national “On the CUSP: Stop BSI” collaborative enrolled 1,071 adult intensive care units in 44 states, the District of Columbia, and Puerto Rico, contributing 27,153 ICU-months and 4,454,324 catheter-days. The mean CLABSI rate fell from 1.96 cases per 1,000 catheter-days at baseline to 1.15 at 16–18 months, with an adjusted incidence rate ratio of 0.57 (95% CI 0.50–0.65) — an overall 43% decrease [35].
A hospital-wide high-reliability programme at a paediatric hospital reported that the serious safety event rate fell from 1.15 to 0.19 events per 10,000 adjusted hospital-days (an 83.3% reduction), preventable harm events fell 53%, and observed hospital mortality fell from 1.0% to 0.75%, although severity-adjusted expected mortality rose slightly over the same period [36].
Both are before–after or collaborative cohort designs without concurrent randomised controls, and both are vulnerable to secular trend, regression to the mean, and reporting changes. The paediatric result illustrates the hazard concretely: observed mortality fell while expected mortality rose, which is exactly the pattern that severity adjustment exists to detect. These studies support the claim that programmes can reduce measured harm; they do not establish that the underlying rate fell as much as the reported metric.
8.4 Awareness is not implementation
The most consistent finding in the implementation literature is that distributing evidence does not change practice. Rangachari and colleagues, reviewing central-line bundle implementation in two intensive care units, concluded that awareness of evidence-based practices alone does not translate to implementation, and that organisational variables — leadership, safety culture, teamwork, engagement — determine whether a bundle is used [37]. This is why the AHRQ-style model pairs evidence products with implementation tools and collaboratives rather than publishing reviews alone [4].
De-implementation evidence points the same way. When one institution empowered respiratory therapists to discontinue nebulised hypertonic saline and N-acetylcysteine that did not meet clinical practice guideline criteria, monthly treatments fell from 3,565 to 548 and associated full-time equivalents from 5.1 to 0.8, while non-targeted treatments remained stable [38]. The intervention was not new evidence; it was a policy that acted on existing evidence.
8.5 Measurement instruments are also clinical infrastructure
A less obvious contribution of health-services research is the validation — and invalidation — of the measures clinicians and payers use. A medical-record review of 16 AHRQ Patient Safety Indicators across nine Swiss hospitals found that positive predictive values ranged from 90–99% for indicators such as iatrogenic pneumothorax, postoperative acute kidney injury, postoperative respiratory failure, postoperative sepsis, wound dehiscence, and birth trauma, but only 18–49% for pressure ulcer, retained surgical item, catheter-related bloodstream infection, fall with hip fracture, and accidental puncture or laceration [39].
That is a remarkable spread, and it carries a direct clinical implication: an indicator with a positive predictive value near 20% will mislabel most of the events it flags, and a hospital judged on it is being judged partly on coding practice. The value of the validation work is that it tells you which indicators to trust. Removing the institution that does that work does not remove the indicators; it removes the check on them.
8.6 Appraisal
Supported. Structured programmes with implementation support can reduce measured harm substantially [35][36]; authoritative evidence synthesis can move practice, though the absolute effects in screening are small [32][33]; and measurement validation changes which metrics are credible [39].
Not supported. That health-services research funding levels determine clinical practice at the margin, or that a specific reduction in AHRQ’s appropriation would cause a specific practice change. No study estimates that, and the causal chain runs through many intermediaries.
Under-appreciated. The instruments themselves. A guideline can be reissued by a professional society; a validated indicator set, a national survey with a consistent series, and a quality-reporting infrastructure cannot be recreated quickly by anyone else. Section 12 takes this up. ## 9. Public-Health Spending and Population Outcomes: The Ecological Literature
9.1 Why this literature is central, and why it is fragile
If any body of work could tell a clinician what happens when public-health funding contracts, it is the spending-and-mortality literature. It is also the body of work most vulnerable to the objection that sicker places spend more, that richer places spend more and are healthier, and that both are driven by things nobody measured. This section reports the effect sizes, then states precisely what each design can support.
9.2 The reference estimate: Mays and Smith
Mays and Smith examined whether changes in spending by local public-health agencies over thirteen years contributed to changes in community mortality from preventable causes, including infant mortality and deaths from cardiovascular disease, diabetes, and cancer. They found that mortality rates fell between 1.1% and 6.9% for each 10% increase in local public-health spending, with the largest effects in low-resource communities, and cautioned that more money alone is unlikely to generate sustained gains without improvements in public-health practice [8]. Public-health spending in their framing accounts for less than 5% of US health spending.
The design is a longitudinal analysis of local agencies with area and time controls. It is stronger than a cross-section because it asks whether changes in spending within an area track changes in mortality, which removes fixed differences between areas. It cannot remove time-varying shocks that affect both — a plant closing that depresses the tax base and raises mortality, for example — and it cannot establish the direction of causation when mortality trends influence budget decisions.
9.3 Program-specific estimates: Florida and Canada
Two later studies improve on the aggregate approach by using programme-specific spending.
Bernet and colleagues used a unique longitudinal dataset from the Florida Department of Health covering all 67 counties from 2001 to 2014, and applied generalised method of moments estimation to address endogeneity and serial correlation. They found that a 10% increase in targeted public-health spending per infant led to a 2.07% decrease in infant mortality, with larger effects among Black infants than white infants [40]. In a companion analysis of maternal mortality, a 10% increase in pregnancy-related public-health expenditures was associated with a 3.9% decline in overall maternal mortality (weakly significant), no significant effect for white mothers, a 13.5% decline among Black mothers, and a 20.0% reduction in the black–white gap [41].
The programme-specificity is the methodological advance: spending on maternal and infant programmes is more plausibly linked to infant mortality than total health spending, and the first-stage relationship is tighter. The remaining threat is that counties increase maternal-infant spending in response to rising infant mortality — which would bias the estimate toward showing harm, making the observed protective association more credible, or that they do so in response to a favourable political cycle that also improves other determinants.
Ammi and colleagues examined the long-run relationship in Canada, the OECD country spending most on public health, using provincial data from 1979 to 2017 and error-correction models. They found evidence of a long-run relationship but no consistent short-run association: a 1% increase in public-health expenditure was associated with a 0.22% decrease in preventable mortality in the long run, larger for males (−0.29%) than females (−0.09%), with no consistent short-run relationship between curative expenditure and treatable mortality except for males [42].
That lag structure is the single most policy-relevant finding in this cluster. If the effect of public-health spending on preventable mortality operates over years rather than months, then a one-year funding interruption may have little measurable effect on mortality — and a multi-year contraction may have a large one that appears after the political cycle in which it occurred has ended.
9.4 The elasticity literature and its meta-analysis
Edney and colleagues estimated the elasticity of health outcomes with respect to public healthcare spending in Australia using instrumental variables. A 1% increase in public health spending was associated with a 2.2% reduction in standardised years of life lost (p<0.05), and instrumental-variable quantile regression indicated that marginal returns were significantly greater in areas with poorer baseline health [43].
An elasticity of −2.2 is far larger than most of the literature, and the meta-analytic estimate explains why. Gallet and Doucouliagos assembled 65 studies published between 1969 and 2014 and used meta-regression to examine how study attributes influence reported elasticities. After correcting for publication selection bias, the spending elasticity for mortality was near −0.13, while the elasticity for life expectancy was near +0.04; the mortality elasticity was particularly sensitive to data aggregation, the specification of the health production function, and the nature of spending, and the life-expectancy elasticity was sensitive to the age at which life expectancy was measured and to whether endogeneity was addressed [9].
An elasticity of −0.13 implies that a 10% cut in health spending would raise mortality by about 1.3% — an enormous effect if taken literally, and a good illustration of why it should not be. The estimate is an average across countries, periods, and outcome definitions, and the meta-regression’s central message is that the number you get depends heavily on design choices. That is a statement about the fragility of the literature, not a refutation of it.
9.5 Austerity as the closest analogue to contraction
The English austerity period is the best-studied instance of deliberate, sustained constraint in a high-income health system. Watkins and colleagues compared actual mortality in 2011–2014 with counterfactual rates expected from pre-2010 trends, and estimated that spending constraints between 2010 and 2014 were associated with 45,368 (95% CI 34,530–56,206) more deaths than expected. Deaths among people aged 60 and over and in care homes accounted for the majority; each £10 per capita decline in real social-care spending was associated with an increase of 5.10 (95% CI 3.65–6.54) care-home deaths per 100,000; associations persisted in lag analyses and after adjustment for macroeconomic factors, and appeared to operate partly through changes in nurse numbers [44].
This is a time-trend analysis, not a controlled experiment, and its central assumption — that pre-2010 trends would have continued — is untestable. Its value is that it is the closest available analogue to what a multi-year contraction in a high-income system looks like, and that the mechanism it identifies (social care, staffing, older adults) is specific enough to be falsifiable.
9.6 Local capacity, not just local spending
Schenck and colleagues examined local health department spending, staffing, and services in 85 North Carolina departments from 2005 to 2010. A 1% increase in full-time-equivalent staffing per 1,000 population was associated with a decrease of 0.01 infant deaths per 1,000 live births, and provision of women’s and children’s services was associated with a reduction of 1 to 2 infant deaths per 1,000 live births [45].
Two further studies complicate the spending story in useful ways. Reszczynski and colleagues found that local public-health department spending increases between 2008 and 2016 did not go to communities with greater need: increases were associated with a higher percentage of adults over 65 and with lower unemployment, and not with the proportion of residents who were Black or Hispanic, in poverty, or uninsured [46]. Vilda and colleagues, examining state and local spending and pregnancy-related mortality, found that transportation spending was associated with 11% lower pregnancy-related mortality (adjusted rate ratio 0.89, 95% CI 0.83–0.96), and that spending on higher education, highways, and parks was likewise associated with lower rates [47].
The second result is a warning. Parks and highways are not plausible biological determinants of maternal survival; their association with lower mortality most likely reflects the fiscal capacity of the jurisdictions that fund them. Any single-domain spending estimate that does not account for correlated spending across domains will attribute to health spending an effect that belongs to local prosperity.
9.7 Cross-national consistency
Park and colleagues examined 36 OECD countries from 2000 to 2019 using fixed effects. A one-percentage-point increase in health spending was associated with a 1.43% decrease in the death rate per 100,000 (95% CI −1.86 to −1.01) and a 0.68% increase in years lived with disability; a one-percentage-point increase in social spending was associated with a 0.29% reduction in disability-adjusted life years and a 0.07% increase in life expectancy; associations persisted with one- and two-year lags [48].
Červený and van Ours used administrative data on heart-attack patients and compared local patients with visitors and stayers with recent movers to address selection into regions, finding that mortality after heart attacks was substantially lower in regions with higher health-care spending [49]. The design is clever and the outcome specific, and it addresses a different question from the public-health literature: not prevention, but acute care intensity.
9.8 Appraisal
Supported. Higher public-health spending is associated with lower preventable mortality across multiple countries and designs, with effect sizes typically in the range of a 1–7% mortality reduction per 10% spending increase [8][40][42]. The association is stronger in low-resource settings [8][43] and operates with a multi-year lag [42].
Contested. The magnitude. Estimates range from an elasticity near −0.13 in meta-regression [9] to −2.2 in one instrumental-variable analysis [43], a sixteen-fold spread that no reader should average.
Not supported. That a one-year interruption in a specific federal programme will produce a detectable change in population mortality. The lag structure, the small share of public-health spending in total health spending, and the coarseness of vital-statistics outcomes all argue against detectability at that timescale.
10. Withdrawing Coverage and Services: Natural Experiments and Their Lessons
10.1 Why these are analogues, not predictions
Nothing in this section is about refugees, AHRQ, or the Office of Minority Health. It is about a general question — what happens when a defined population loses a defined service or coverage — studied in populations where the change was observable and someone measured it. The analogues are informative about mechanisms and timelines. They are not forecasts.
10.2 Medicaid expansion: the best-studied positive shock
Sommers, Baicker, and Epstein compared three states that substantially expanded adult Medicaid eligibility after 2000 (New York, Maine, and Arizona) with neighbouring states that did not, using county-level mortality data from 1997 to 2007. Expansion was associated with a reduction in adjusted all-cause mortality of 19.6 deaths per 100,000 adults, a 6.1% relative reduction (P=0.001), with the largest reductions among older adults, non-white residents, and residents of poorer counties; coverage rose 2.2 percentage points, uninsurance fell 3.2 points, and delayed care because of cost fell 2.9 points [10].
Lee and colleagues extended the analysis nationally, using death certificate data for adults aged 25–64 from 2010 to 2018 across 32 expansion and 17 non-expansion states. Medicaid expansion was associated with a reduction of 11.8 deaths per 100,000 adults (95% CI −21.3 to −2.2), but state-level effects ranged from −63.8 (95% CI −134.1 to −42.9) in Delaware to +30.4 (95% CI −39.8 to 51.4) in New Mexico [11].
That range is the most important number in this section. A national average of −11.8 with a state range spanning both directions means the average is not a prediction for any state, and that the modifiers — the authors identified the state-level proportion of women and of non-Hispanic Black residents among them — matter more than the headline.
10.3 The infant-mortality literature is genuinely contested
Three analyses of the same policy reach different conclusions, and the disagreement is instructive rather than embarrassing.
Bhatt and Beck-Sagué compared infant mortality in expansion and non-expansion states from 2010 to 2016 and reported that the mean rate declined in expansion states (5.9 to 5.6) while rising slightly in non-expansion states (6.4 to 6.5); the decline among African American infants in expansion states was 14.5%, more than twice that in non-expansion states (6.6%) [50].
Cook and Stype, using linked birth/death files from 2011 to 2017 and a difference-in-differences estimator with fixed effects, found no statistically significant change in the full sample; reductions appeared only when the sample was restricted to states using the 2003 birth certificate form and excluding states with a Medicaid waiver, and the reduction was seen for babies born to white mothers but not Black mothers [51].
Wiggins, Karaye, and Horney, using state-level aggregate data from 2010 to 2017, found no association overall (difference-in-differences coefficient 0.04; 95% CI −0.39 to 0.46) but a reduction among Hispanics (−0.53; 95% CI −1.02 to −0.03) [52].
The lesson is methodological: infant mortality is a rare outcome measured at state-year level, so the effective sample is small, and estimates are sensitive to which states are included, which comparison window is used, and which subpopulation is examined. A clinician reading a headline about Medicaid and infant mortality should ask which of these three designs produced it.
10.4 Coverage loss: the unwinding as a prospective natural experiment
The 2023–2024 unwinding of continuous Medicaid coverage is the closest available analogue to an abrupt withdrawal of eligibility for a defined population, and the results are consistent across data sources.
A multimodal survey of 2,210 low-income adults in four Southern states found that 12.5% of adults who had Medicaid were no longer enrolled at the time of interview, with state estimates from 7.0% in Kentucky to 16.2% in Arkansas; among adults who left Medicaid and reported coverage status, 47.8% were uninsured [53].
Administrative data tell a similar story with more precision. In a multistate network of community health centres, 8.7% of 450,146 paediatric patients were disenrolled from Medicaid to uninsured status during the first year of unwinding; American Indian or Alaska Native patients had disenrollment of 17.1% versus 9.4% for white patients (adjusted odds ratio 1.95, 95% CI 1.81–2.09) [54].
Prescription data show a clinical consequence. In a difference-in-differences analysis of the IQVIA Longitudinal Prescription Database covering 92% of US pharmacies, residence in a state with high Medicaid enrolment loss was associated with a greater decrease in days of active prescriptions for some chronic medication classes among children, and with a greater increase in the risk of having no prescriptions and of paying cash for all five classes among young adults [55].
And the transition out of childhood coverage is a structural cliff independent of policy. Among 9.4 million individuals in 47 state Medicaid programmes, the conditional probability of disenrollment at age 19 was 13.4% (95% CI 13.2–13.6) among young adults with complex medical conditions and 35.6% among those without, with state probabilities for the complex group ranging from 2.6% to 37.0% [56].
10.5 What coverage does over a lifetime
Thompson estimated the effect of cumulative Medicaid and CHIP eligibility from birth to age 18 on adult health, exploiting variation in programme generosity across states and over time. An additional year of public health insurance eligibility during childhood improved a summary index of adult health by 0.079 standard deviations, and substantially reduced health limitations, chronic conditions, and asthma prevalence while improving self-rated health [57].
That is a small annual effect that compounds over eighteen years, and it is the strongest available evidence that coverage in childhood has consequences that appear decades later — which is also why short-run evaluations will systematically understate them.
10.6 Service withdrawal without coverage change
Three service-level analogues are worth recording because they isolate delivery from insurance.
Community health centres. After Oregon’s randomised Medicaid expansion, gaining coverage significantly increased primary care visits and laboratory use among patients at 108 community health centres, with intent-to-treat differences also evident in behavioural-health visits, referrals, and imaging [58]. The design — a randomised lottery — is the strongest in this literature, and it measures utilisation rather than health.
Family planning. Texas reduced funding for family planning services by 67%, resulting in more than 80 clinic closures. Using a difference-in-differences design comparing Texas counties that lost funding with counties outside Texas that had publicly funded clinics, Packham found that the cuts increased teen birth rates by approximately 3.4% over four years, with effects concentrated two to three years after the initial cuts [59].
HIV services after an abrupt funding withdrawal. When the US government froze foreign aid and halted PEPFAR in January 2025, four non-governmental organisations in Uganda and Zimbabwe that received stop-work orders contributed monthly and quarterly service data. In an interrupted time-series analysis with February 2025 as the interruption point, prevention fell by 36.8 to 62.7 index points, PrEP by 43.0 to 71.6, treatment by 22.0 to 31.5, and testing by 58.7 to 63.8, where 100 represents the mean 2024 amount; services remained substantially below baseline through March 2026 [12].
This is the most directly relevant analogue in the entire review, because the exposure — an abrupt, politically initiated stop-work order affecting organisations that deliver services to a defined population — matches the structure of the September 2026 event more closely than any coverage expansion study. Two cautions apply. The organisations were NGOs serving key populations, not government grantees, and the health system context was low-income, so the magnitudes may not transfer. But the shape of the response — immediate, large, and persistent — is the best empirical answer available to the question “what does a stop-work order do to service volume?”
10.7 Capacity loss without any policy change
Two studies document what happens when delivery capacity disappears for reasons other than funding policy.
Rural hospital closures have been studied with staggered difference-in-differences designs. Across 32 closures, counties experienced an increase in quarterly hospitalisations of 111.6 per 10,000 older adults (95% CI 53.4–170.9) and an increase in average length of stay of 0.34 days; average distance to the hospital used did not change, and measures of post-acute function and population mortality did not worsen, which the authors attributed to care-delivery adjustments [60]. The lesson is that systems adapt, and that adaptation can mask harm in the outcomes researchers happen to measure.
The 2008 recession produced a natural experiment in public-health capacity. In three nationally representative surveys of local health departments, 53% were experiencing cuts to core funding by early 2010, and more than 23,000 local health department jobs were lost in 2008–2009; all programme areas were affected and more than half of departments reduced or eliminated at least one programme area [13]. Follow-up surveillance through 2013 found that 26.9% of departments still reported a reduced budget in early 2013, that the percentage making programme cuts never fell below 40%, and that maternal and child health services were among those most often cut [61].
10.8 Appraisal
Supported. Losing coverage or services produces rapid, measurable changes in utilisation and in process outcomes, and in some settings in mortality [10][53][12]. Effects are heterogeneous across states and subgroups, often substantially so [11][56].
Contested. Mortality effects of coverage expansion, especially for infant mortality, where three credible analyses disagree [50][51][52].
Not transferable. Nothing here licenses a quantitative prediction about the September 2026 rescissions. The analogues differ in population, financing, delivery, and duration, and the affected accounts fund services that are complements to, rather than substitutes for, the coverage studied above.
11. Refugee and Unaccompanied-Minor Health: The Clinical Evidence on Need, Screening, and Services
11.1 The need is documented and specific
A scoping review of long-term physical health outcomes among resettled refugees in the United States identified 33 studies meeting inclusion criteria from five databases (2008–2019). Refugee adults had higher odds of chronic disease than non-refugee immigrant adults and increased risk of diabetes compared with US-born controls; diabetes and hypertension were the most commonly reported chronic conditions among Iraqi, Somali, and Bhutanese adults, and the authors recommended considering screening and evaluation for chronic conditions in the early resettlement period [62].
Among refugee children arriving in high-income countries, a scoping review of 43 studies (38 cross-sectional, three longitudinal, two cohort) grouped findings into seven themes: nutritional deficiencies including stunting, wasting, and vitamin deficiency; infectious diseases including latent tuberculosis and hepatitis; neglected tropical diseases such as scabies and schistosomiasis; dental caries; elevated blood lead levels; inadequate immunisation status; and mental health problems [63].
A statewide linkage of Nebraska Medicaid immigration data with hospital discharge records for 2011–2015 found that refugee patients were at increased risk for pregnancy complications, abdominal pain, upper respiratory infections, viral infections, mood disorders, dental disorders, deficiency and anaemia, urinary disorders, headache, nausea and vomiting, limb fractures, spondylosis, hypertension, and uncomplicated diabetes [64]. The design is ecological at the patient level and cannot separate the effects of displacement from those of poverty, but the diagnostic profile is consistent with the scoping reviews.
11.2 Mental health in unaccompanied minors: high prevalence, wide range
The most precise prevalence estimates come from a systematic review of 23 studies in nine countries covering 80,651 unaccompanied refugee minors. Prevalence ranged from 4.6% to 43% for post-traumatic stress disorder, 2.9% to 61.6% for depression, 32.6% to 38.2% for anxiety, and 4% to 14.3% for behavioural problems; two studies examining suicide attempts and deaths observed higher rates than in the host population of the same age; and a meta-analysis of four studies found lower PTSD prevalence among accompanied refugee minors (standardised difference −1.14, 95% CI −1.56 to −0.72) [65].
The breadth of those ranges is itself a finding. Prevalence estimates differ by instrument, by country of origin, by placement type, and by time since arrival, to a degree that makes a single pooled estimate misleading. A rapid review of 14 studies published after 2018 reached the same conclusion: consistently high levels of PTSD and post-traumatic stress symptoms across contexts, with depression and anxiety associated with discrimination, limited language attainment, and daily hassles, but evidence quality variable and assessment heterogeneous [66].
11.3 What happens over time
Three longitudinal studies answer a question cross-sectional data cannot.
Vervliet and colleagues followed 103 unaccompanied minors from arrival in Belgium to 18 months later. Scores on anxiety, depression, and PTSD were generally high and did not change significantly over time, suggesting persistence; the number of traumatic experiences and the number of daily stressors were each associated with higher symptom levels [67].
Jensen and colleagues followed a Norwegian cohort at six months, two years, and five years after arrival. Mean depression scores decreased significantly by five years, but mean anxiety, post-traumatic stress, and externalising symptoms did not; females and the severely trauma-exposed had higher symptom levels; and five years after arrival many still experienced clinical levels of problems, with daily hassles an important predictor [68].
Nilsen and colleagues compared 81 unaccompanied minors settled in Norway with an age- and sex-matched sample of 324 Norwegian young people. Unaccompanied minors reported more emotional problems, more often reported being alone and being bullied, and also reported being more helpful and sharing; the Strengths and Difficulties Questionnaire showed weak factor loadings and low internal consistency in this group, indicating that the instrument’s standard five-factor structure fits poorly [69].
That last point is methodological and important: the instruments used to measure refugee child mental health are often not validated in the populations to which they are applied, which inflates uncertainty around every prevalence estimate in this literature.
11.4 Post-migration factors are modifiable
The clearest signal in this literature is that post-migration conditions matter, and that they are the part a receiving country controls.
Salhi and colleagues surveyed 383 Somali refugees across five cities in the United States and Canada at two waves. Fifty-six percent reported exposure to violence after resettlement by wave two, and at wave one the associations of pre- and post-resettlement violence with symptoms were comparable in magnitude for depression (β 0.39 vs 0.36), anxiety (0.33 vs 0.38), and PTSD (0.55 vs 0.47) [70]. Violence after arrival is not a residual category; it carries weight similar to violence before it.
Hilado and colleagues used data from a Midwest refugee resettlement programme to examine predictors of mental health care enrolment and engagement, finding that gender, nationality, and presenting problems significantly predicted enrolment and length of engagement, and drawing implications for resettlement agencies [71]. This is one of the few studies retrieved here that examines the service pathway rather than the need.
11.5 What services change: the thin part of the evidence
The literature on service effectiveness is much weaker than the literature on need, and it is important to say so.
A scoping review of mental health screening for resettling refugees identified 25,862 citations, of which 70 met eligibility criteria; among 45 publications describing screening programmes, 25 were screening-tool validation studies, and the review characterised 85 different screening tools [72]. Eighty-five tools is not a sign of a mature evidence base; it is a sign of a field that has not converged. The review’s own framing — that delayed diagnosis and care can reduce the impact of resettlement services and lead to poor long-term outcomes [72] — is a plausible mechanism rather than a demonstrated effect.
For unaccompanied children, one US programme report examined trauma-focused cognitive behavioural therapy in 138 referred children, 122 of whom began treatment and 41.8% of whom completed it. Youth reported an average of 3.80 traumatic events, and there were significant improvements on most self-report scales, with within-subject Cohen’s d ranging from 0.07 to 1.03 and d=0.44 on a brief progress-monitoring measure [73]. This is a single-group pre–post design with substantial attrition; it supports feasibility and probable benefit, not efficacy.
A meta-analysis of vaccination in forcibly displaced paediatric populations pooled 39 studies from 24 countries and estimated full vaccination coverage at 21% (95% CI 11–36%), with extreme heterogeneity (I²=98.90%) [74]. The point estimate is alarming and the heterogeneity means it should not be applied to any particular setting; what it establishes is that incomplete vaccination is the norm rather than the exception in this population, which is precisely what a medical-screening and catch-up system exists to address.
Growth data provide a cleaner longitudinal signal. Among 512 refugee children and 1,175 matched non-refugee low-income children, the prevalence of overweight or obesity among refugees aged 2 to 16 rose from 8.9% at baseline to 20% at 10–24 months, and refugee children had a steeper increase in BMI z-score per year than non-refugees (0.18 vs 0.03) [75]. The transition from undernutrition to overnutrition within a year or two of arrival is one of the more robust findings in this literature, and it implies that resettlement services need a nutritional component that extends well beyond the initial screening visit.
11.6 Appraisal
Supported. Refugees and unaccompanied minors arrive with a high and specific burden of chronic disease, latent infection, incomplete immunisation, and mental-health symptoms [62][63][65][74]. Symptoms in unaccompanied minors persist over years and are strongly associated with post-migration stressors [67][68][70].
Weakly supported. That specific resettlement services improve health outcomes. The evidence consists mainly of single-group pre–post studies, programme descriptions, and screening-tool validations; this review found no randomised or well-controlled quasi-experimental evaluation of a resettlement service package on health outcomes.
Not supported. Any estimate of the health consequence of reducing the Refugee and Entrant Assistance account. The need is documented; the service effect is plausible; the link between a specific appropriation and a specific outcome is not established.
12. Measurement Infrastructure: What Becomes Unknowable
12.1 Measurement is a product, not a by-product
The least intuitive claim in this review is that some of the most valuable things AHRQ and similar agencies produce are not findings but instruments. The claim is testable by asking what happens when an instrument stops.
12.2 How the instruments are built
MEPS is a nationally representative survey of health-care use, expenditure, and insurance, designed to provide annual population estimates. Its design innovations included incorporating CAHPS content to assess patient experiences nationally, adding SF-12 and EQ-5D items to measure health status, adding condition-specific questions for diabetes, asthma, hypertension, and heart disease to determine whether care was consistent with practice guidelines, and increasing sample size to improve precision [76].
Every element of that description is a design decision with a cost. Sample size determines the precision of subgroup estimates; questionnaire content determines which conditions can be studied; the overlap with CAHPS determines whether patient experience can be linked to expenditure. A survey is not a camera pointed at the world; it is a set of choices, and the choices constrain what can be known.
12.3 What happens when a survey is interrupted
The COVID-19 pandemic provided an unplanned test. The National Health Interview Survey responded to disrupted operations with several changes to continue production in 2020, but data files from the 2020 NHIS were not expected to be publicly available until autumn 2021; to fill the gap, the National Center for Health Statistics turned to two online platforms, the Census Bureau’s Household Pulse Survey and the NCHS Research and Development Survey [77].
The substitution was better than nothing and worse than the original. The authors describe the trade-offs explicitly in terms of a data-quality framework — accuracy, relevance, timeliness — and note that large national surveys have always balanced these dimensions [77]. A rapid online survey can tell you whether people lost coverage last month; it cannot tell you whether their blood pressure is controlled, because it does not measure it.
Declining response rates are the slower-moving version of the same problem. The 2019 NHIS questionnaire redesign — intended to reduce respondent burden and increase participation — was associated with improvements in response rate, declines in burden, and increases in data quality and relevancy; the authors note that declining response rates may lead to biased estimates and weaken users’ ability to draw valid conclusions [78]. That is a straightforward statement with a large implication: a survey that is defunded does not merely produce fewer data points; it produces biased ones, because non-response is not random.
12.4 What becomes unknowable
Four categories of knowledge depend on continuity, and each has a different failure mode.
Trends. A time series with a gap is not a shorter time series; it is a different object. Interrupted series can sometimes be bridged with modelling, but the bridge is an assumption, and it is usually untestable.
Subgroup estimates. Precision for small groups — unaccompanied minors, specific refugee nationalities, rural residents — depends on oversampling and sample size. When sample size falls, the first thing lost is the ability to say anything about the groups with the greatest need, which is exactly the pattern documented in the local public-health spending literature [46].
Linked analyses. MEPS can be linked to other data; HCUP supports state-level policy research; the value of both depends on consistent identifiers and definitions over time. A change in questionnaire content or coding breaks linkage silently, and the break is often discovered only when a trend looks implausible.
Validation. The patient-safety indicator study is the clearest example: positive predictive values ranged from 18–49% for several indicators and 90–99% for others [39]. Without periodic validation, users cannot distinguish a real improvement in care from a change in coding. The indicator survives the defunding; the check on the indicator does not.
12.5 The workforce dimension of measurement
Measurement capacity is people. The National Association of County and City Health Officials’ profile survey — described by its authors as the main source of comprehensive data collected from local health departments — estimated a local health department workforce of 182,100 employees or 163,200 full-time equivalents in 2022. Between 2019 and 2022 the total workforce grew 19%, but the largest increase was among contract workers (175%) while the full-time workforce grew about 7%; the authors warned that with the surge in temporary and contract workers, the sustainability of the workforce is in question without continued strategic funding [79].
A workforce that is one-fifth contract labour is a workforce that can be demobilised quickly. The capacity to collect data is embedded in the same people who deliver services, so a service contraction and a measurement contraction tend to arrive together — which is one reason the 2008–2013 recession data are unusually complete and the years after a sudden contraction often are not [13][61].
12.6 Appraisal
Supported. National surveys require deliberate design choices that trade accuracy, relevance, and timeliness [77][76]; interruptions delay public data and force substitution with instruments that measure different things [77]; declining response produces bias, not merely imprecision [78]; and indicator validation changes the interpretation of every downstream comparison [39].
Inference, not evidence. That a reduction in AHRQ’s appropriation would degrade measurement capacity in a way that matters clinically. This is plausible given what MEPS and HCUP are, and it is not demonstrated, because no one has run the experiment of removing them and measuring the consequences.
A note on asymmetry. The cost of losing measurement capacity is paid later and diffusely, while the saving is realised now and specifically. That asymmetry is a general property of public goods and is the strongest theoretical reason to treat measurement cuts differently from programme cuts — but it is an argument from political economy, not from the empirical literature, and is labelled as such. ## 13. Historical Precedents: Sequester, the ARRA Cliff, and Pandemic-Funding Wind-Down
13.1 Why precedents matter more than projections
A reviewer asked to say what will happen cannot answer. A reviewer asked what happened the last several times federal health funding was interrupted can answer, at least in part, and can say how well each episode was measured. This section assembles four episodes with different mechanisms and different quality of evidence.
| Episode | Mechanism | What was measured | Design quality |
|---|---|---|---|
| 2013 sequestration | Across-the-board automatic cuts, mid-year | Grant counts, contract reductions, staff layoffs, institutional responses | Descriptive and journalistic; no causal estimate |
| 2009 ARRA and its 2012–2014 cliff | Large temporary infusion, then expiration | Citation impact per dollar; neighbourhood employment | Quasi-experimental for both |
| 2020–2023 pandemic infusion and 2023–2024 wind-down | Emergency supplemental funding, then expiry | Workforce composition; survey continuity; service volume | Descriptive time series |
| 2025 PEPFAR/USAID stop-work | Abrupt political termination | Service delivery indices | Interrupted time series |
13.2 The 2013 sequester: what the record actually contains
The 2013 sequestration imposed across-the-board cuts in the middle of a fiscal year, which meant agencies had to absorb a full-year reduction in the remaining months. The peer-reviewed record of its effects is thin and mostly descriptive. What exists is specific enough to be useful.
The Framingham Heart Study, a 65-year-old cohort study funded through an NHLBI contract of approximately $9 million per year, had that contract reduced by $4 million on 1 August 2013; the renewal scheduled for 2015 was expected to run two or three years instead of the usual seven; and 19 of 90 staff members were laid off [80]. That is a single institutional anecdote reported in a journal interview, not an evaluation, but it illustrates the arithmetic that matters: a mid-year cut of that magnitude is absorbed by stopping work and releasing staff, because the fixed costs of a cohort study cannot be reduced gradually.
At the system level, the decade’s funding trajectory is visible in aggregate data. Between 2006 and 2016, NIH applications rose 18.7% while funded grants fell 6.7% and R01s fell 17.0% [19]. The sequester sits inside that window and is one contributor among several, including the end of the budget-doubling era; separating its effect from the trend is not possible with the data retrieved here.
13.3 The ARRA cliff: the best-measured research-funding episode
ARRA is the closest thing the US has to a natural experiment in rapid research-funding expansion followed by contraction, and it produced two useful findings.
On scientific output, the NHLBI comparison described in Section 5 found that 165 grants funded only because of ARRA produced normalized citation impact per $1 million statistically indistinguishable from 458 payline grants (2.15 vs 2.03; P=0.61), despite shorter durations and smaller budgets [23]. The contraction that followed — the “ARRA cliff” — was widely predicted to damage labs, but this review did not find a peer-reviewed evaluation that estimates the cliff’s effect on investigator retention or output. The expansion was measured; the contraction was not.
On community effects, ARRA’s $2 billion investment in federally qualified health centres was evaluated with a propensity-score-matched longitudinal design across 50,381 census tracts, 2,223 of which contained at least one funded health centre. Jobs grew at 3.84 per 1,000 population per year on average; in ARRA-funded tracts jobs grew at a higher annual rate of 4.34 per 1,000 (95% CI 2.56–6.12) relative to tracts with similar social vulnerability, and there was no decline in healthcare establishments even as non-healthcare establishments declined [81].
The design is credible for the question it asks, which is about neighbourhood economic recovery rather than health. Its relevance here is that it demonstrates a general point: a temporary infusion can produce durable local capacity, and the size of that capacity is measurable. It says nothing about what happened when the infusion ended, which is the question a contraction review would most want answered.
13.4 The pandemic wind-down: capacity that was built to be temporary
The COVID-19 period produced the largest peacetime expansion of public-health funding in decades, and its wind-down is the most recent completed natural experiment in contraction.
The workforce data are unambiguous. Between 2019 and 2022, the local health department workforce grew 19% to 182,100 employees (163,200 FTE), but the growth was concentrated in contract workers, which rose 175%, while the full-time workforce grew about 7%; the authors concluded that with the surge in temporary and contract workers there are concerns about sustainability, and that without continued strategic and sustained funding the workforce may be in jeopardy [79].
That is a precise description of a specific failure mode: the expansion happened, it was real, and it was structured so that it could be reversed quickly. The clinical consequence is indirect but not speculative — the same workforce performs tuberculosis contact tracing, immunisation, and sexually transmitted infection services — but this review found no study that measured service levels after the wind-down.
The measurement system was itself disrupted. NHIS data files for 2020 were not expected to be publicly available until autumn 2021, and NCHS substituted two online platforms in the interim [77]. The substitution preserved some information and lost other information, and the loss was documented by the agency itself — which is the best-case scenario. In a less well-instrumented system, the loss would have gone unrecorded.
13.5 The 2025 foreign-aid stop-work: the closest structural analogue
The January 2025 freeze and halt of PEPFAR, described in Section 10, is the most recent and most structurally similar episode to a stop-work order on service-delivery organisations. The interrupted time-series analysis found declines of 36.8 to 62.7 index points in prevention, 43.0 to 71.6 in PrEP, 22.0 to 31.5 in treatment, and 58.7 to 63.8 in testing, sustained through March 2026 [12].
Two features make this evidence unusually valuable. First, the exposure was abrupt and political rather than gradual and budgetary, which is the relevant comparison for a rescission. Second, the outcome was measured at the level of the organisation delivering the service, not at the level of a national aggregate, so the estimate is not diluted by unaffected providers. Its limitation is external validity: four NGOs serving key populations in two low-income countries are not a state health department or a university research centre.
13.6 What the precedents jointly show
Four findings recur across episodes:
- Mid-year cuts are absorbed by stopping work and releasing people, because fixed costs cannot be reduced proportionally [80][13].
- Temporary expansions are often built on temporary employment, so the capacity they create is reversible [79].
- Abrupt stop-work orders produce immediate, large, and persistent service declines where service data exist [12][59].
- The expansion is usually measured better than the contraction. ARRA’s output was evaluated; the cliff was not. The pandemic workforce surge was measured; the wind-down largely was not.
The fourth finding is a limitation of the evidence base, not a finding about the world, and it has a practical implication for anyone who wants to answer this question in future: the data have to be collected prospectively, because the organisations that would collect them are the ones being defunded.
14. International Comparators and Stability Mechanisms
14.1 What can and cannot be compared
Cross-national comparison of research-funding stability is tempting and methodologically treacherous. Health systems differ in financing, in the share of research funded publicly, in the separation between research and service budgets, and in the legal durability of multi-year commitments. This review found no peer-reviewed study that directly compares funding-stability mechanisms across countries and estimates their effect on research output or health outcomes. What exists is evidence on spending levels and constraints within systems, which is a different question.
That absence should be stated plainly rather than papered over with institutional description.
14.2 Constraints in a tax-financed system: England
The English austerity period is the best-documented instance of sustained constraint in a high-income, tax-financed system. As described in Section 9, time-trend analysis estimated 45,368 (95% CI 34,530–56,206) more deaths than expected between 2011 and 2014, with the majority among people aged 60 and over and in care homes; each £10 per capita decline in real social-care spending was associated with 5.10 (95% CI 3.65–6.54) additional care-home deaths per 100,000, and the association appeared to operate partly through nurse numbers [44].
The finding that social-care spending was more strongly linked to care-home mortality than health-care spending is the most transferable insight in this section. It implies that the health consequences of a funding contraction may appear in the part of the system that is least visible to clinicians — social care, home support, community services — rather than in hospitals.
14.3 Long-run returns in a federal system: Canada
Canada’s provincial public-health data produce the clearest estimate of the time profile of spending effects: a 1% increase in public-health expenditure associated with a 0.22% decrease in preventable mortality in the long run, with no consistent short-run association, and larger effects for males than females [42]. The authors’ conclusion — that public-health expenditure reaps benefits primarily in the long run — is a direct challenge to annual budget cycles, in any country.
A Delphi study of 25 senior Canadian public-health decision-makers (from 86 invited) reached consensus on six of nine propositions about financing, organisation, workforce, and interventions, with propositions rated important relating to targeted public-health budgets [82]. This is expert opinion, not evidence of effect, and its value is in showing what practitioners prioritise when asked — a form of evidence that should be labelled as such.
14.4 Cross-national spending associations
The OECD fixed-effects analysis described earlier found that a one-percentage-point increase in health spending was associated with a 1.43% decrease in the death rate (95% CI −1.86 to −1.01), while social spending was associated with a 0.29% reduction in DALYs and a 0.07% increase in life expectancy [48]. The finding that health spending predominantly influenced mortality while social spending influenced quality-of-life measures is consistent with the English result and suggests that the two streams do different work.
Mexico’s Seguro Popular expansion provides a lower-middle-income comparator with a stronger design: using hospital-level affiliation data mapping geographical expansion and double- and triple-difference estimation, the programme was associated with a reduction in late neonatal mortality of 0.139 deaths per 1,000 live births (95% CI 0.032–0.246), or 7% relative, and with reductions in infant mortality from covered conditions [83]. The design exploits the staggered rollout of insurance, which is the closest analogue available to a controlled expansion.
14.5 Stability mechanisms: what is asserted and what is known
Policy discussion of funding stability typically proposes instruments: ring-fenced budgets, multi-year settlements, standing research levies, and independent funding bodies insulated from annual appropriations. Each has an intuitive rationale. This review found no peer-reviewed evaluation of whether any of these instruments reduces the volatility of research output or improves health outcomes relative to annual appropriation.
That is a genuine gap, and it is worth being precise about its size. The literature can tell you that volatility is costly in principle (Section 4), that interruptions reduce service volume (Section 10), and that spending effects operate with long lags (Section 14.3). It cannot tell you that a specific institutional design solves the problem, because no one has run the comparison.
The honest conclusion for this section is therefore negative: the international evidence supports the claim that sustained spending matters and that lags are long, and it does not support any specific prescription for stabilising funding.
15. What a Clinician or Researcher Can Do Now
15.1 The scope of this section, stated honestly
Everything in this section is inference from the evidence reviewed above plus ordinary clinical and administrative reasoning. None of it is a tested intervention, and nothing here is a prediction that any specific programme will be cut. The section is organised by role, and each recommendation is labelled with the evidence it rests on and the confidence it warrants.
15.2 For clinicians whose patients depend on the affected services
Verify coverage before assuming it. The unwinding evidence shows that coverage loss is often invisible until the point of care: 12.5% of adults with Medicaid in four Southern states were no longer enrolled at interview, and among those who left, 47.8% were uninsured [53]. For patients in refugee and immigrant families, eligibility rules are complex and change frequently, and the disenrollment data show that adolescents and young adults with complex conditions face the sharpest cliffs [56]. A eligibility check at every visit is low-cost and supported by the mechanism, if not by a trial.
Anticipate medication disruption. The prescription data show that disenrollment is followed by fewer days of active prescriptions and more cash payment, particularly among young adults [55]. For patients with epilepsy, diabetes, asthma, or serious mental illness, a refill-gap question at the next visit is a reasonable screen. This is clinical inference from an observational association.
Know what the safety-net providers are absorbing. Community health centres that gained coverage through the Oregon experiment saw increased primary care, laboratory, and behavioural-health use [58]; the corollary is that centres serving populations that lose coverage absorb uncompensated demand. Whether that is happening locally is an empirical question a clinician can answer by asking their own institution’s finance office, not by inference from national data.
Use screening and navigation where the yield is high. Refugee health screening identifies latent tuberculosis, incomplete immunisation, anaemia, lead exposure, and mental-health symptoms at high prevalence [63][62]; full vaccination coverage in forcibly displaced children was pooled at 21% [74]. Screening is only useful if a treatment pathway exists, and the treatment-pathway evidence is the weakest link in this literature [72].
15.3 For researchers and research staff
Treat funding continuity as a portfolio problem, and expect the portfolio to be correlated. The grant-funding literature suggests that marginal NIH dollars partly substitute for other dollars [5], but that substitution operates at the level of the investigator, not the project: a lab with two grants from the same programme is not diversified. Foundation and specialty-society awards are small but have measurable conversion to federal funding — 22% overall and 46% for career-development awards in one orthopaedic cohort [25] — and their value in a contraction is continuity rather than amount.
Protect the things that cannot be restarted. The Framingham precedent is the clearest illustration: a mid-year contract reduction of $4 million on a $9 million contract was absorbed by laying off 19 of 90 staff [80]. Cohorts, biobanks, and longitudinal series are the assets most vulnerable to a funding gap and the least recoverable afterwards. If a programme is at risk, the priority ordering is: preserve the identifiable data and the follow-up contact information first, the specimens second, the staff third — because staff can be rehired and data cannot be recreated.
Do not over-interpret your own grant outcomes. Peer review explains little of subsequent impact: the area under the ROC curve for percentile rank predicting top-10% articles was 0.52 [7], and reviewers asked to rate the same applications showed no agreement [21]. A funded application is not proof of quality and a rejected one is not proof of its absence. That is worth saying to trainees, who are the group most likely to internalise the signal.
For physician-scientists, the pipeline data argue for earlier independence, not more perseverance. The documented problems are time from degree to first R01 and retention beyond the first award [26], and the consensus recommendations converge on shortening time to independence and building career-development infrastructure rather than exhorting individuals to persist [27][28]. Those are institutional actions; an individual can advocate for them but cannot implement them alone.
15.4 For institutions
Model the exposure, not the headline. A rescission proposal of $27.7 million against a $345 million appropriation [1] does not translate into a proportional institutional cut. What matters for a university or health system is the share of its portfolio in the affected accounts, the unobligated balance at risk, and the fraction of staff whose salaries are charged to those awards. Institutions that can compute those three numbers in a week are in a different position from those that cannot.
Keep the measurement functions funded internally if the federal source is interrupted. This is a direct inference from Section 12: a survey or indicator set that stops produces biased as well as missing data [78], and the cost of the gap is paid by whoever needs the trend later. For a health system, that means maintaining internal quality and utilisation series even when national benchmarks are disrupted.
15.5 What not to do
Three responses are not supported by the evidence reviewed here.
Do not claim that a specific cut will cause a specific number of deaths. No study supports that mapping, the lags are long [42], and the heterogeneity across states is larger than most reported averages [11]. Advocates who make such claims expose the underlying evidence to deserved criticism.
Do not claim that funding does not matter because marginal productivity is low. The marginal-productivity estimates concern additional grants to investigators near a threshold [5]; they say nothing about the loss of an entire measurement system or service line, which has no close substitute.
Do not treat peer review as a quality guarantee in either direction. The evidence that review is weakly predictive [7][21] cuts against the meritocratic defence of any particular allocation, and it is not a licence to conclude that allocation is arbitrary — it is a reason to prefer mechanisms, such as lotteries among meritorious proposals, that are transparent about the randomness [22].
Limitations
The literature is observational, and the strongest designs answer narrow questions. The studies with credible identification — the Oregon lottery [58], the NHLBI payline comparison [23], the Texas family-planning cut [59], the PEPFAR interruption [12] — each answer a specific question in a specific setting. None of them estimates the effect of a small, temporary, legally contested rescission on a national programme.
Publication and advocacy bias run in both directions. Studies finding that spending improves health are more likely to be published than null results, which is why the meta-regression correcting for publication selection is important and why its corrected elasticity (−0.13) is much smaller than several individual estimates [9]. Conversely, several sources in this review are documents with institutional positions: the physician-scientist workforce reports are consensus statements by professional bodies [27][28], and the local-health-department survey data are collected by a membership association [13][61]. They are cited for description, not for causal claims.
The evidence is asymmetric between expansion and contraction. Almost every well-identified study examines the addition of funding or coverage, not its removal. The exceptions — the unwinding analyses [53][54], the family-planning cuts [59], the PEPFAR withdrawal [12] — are recent and few. Extrapolating from expansion to contraction assumes symmetry, and Section 4 explains why that assumption is questionable.
Legal sources were partly inaccessible. The transmittal letter was retrieved in its printed form and its headline figures verified directly [1]; the enclosure pages of the printed document are scanned images, so programme-level figures were taken from the verified seed record rather than re-keyed from the PDF. The GAO letter of 29 September 2026 could not be retrieved from the agency’s website, which blocked automated access; GAO’s conclusion is reported as described in the verified seed record and in contemporaneous wire reporting [2][14]. The decision number is therefore not cited, because it could not be confirmed.
Several official descriptions are self-descriptions. ORR, AHRQ, and OMH pages describe their own programmes [3][4][16]. They are reliable for what a programme says it funds and unreliable for how well it performs. Performance claims in this review are drawn from peer-reviewed evaluations, not from agency pages.
Search coverage was PubMed-centred. Ninety-five distinct queries were run against PubMed, with targeted retrieval of named landmark studies and title screening of roughly 1,300 unique records; 90 abstracts were read in full. Grey literature was retrieved selectively rather than systematically. Advocacy organisations’ reports were not used as evidence of effect, and no claim in this review rests on a source that was not retrieved and read.
Heterogeneity was reported, not pooled. Where studies disagreed — infant mortality and Medicaid [50][51][52], spending elasticities [9][43], state-level mortality effects [11] — the disagreement is reported as the finding. No meta-analysis was performed here, and the ranges given are ranges across studies, not confidence intervals on a pooled estimate.
And the central limitation. The specific consequences of the September 2026 rescissions are not observable at the time of writing. They depend on unobligated balances, on how agencies implement the hold, on litigation, and on whether Congress acts before 9 November 2026 [2]. Anyone who states those consequences as fact is going beyond the evidence, including in the direction of reassurance.
Bottom Line
The event. Eleven proposed rescissions totalling $810 million were transmitted on 25 September 2026, of which roughly $600 million fell on three HHS accounts: refugee resettlement and medical assistance, AHRQ, and a small Office of Minority Health line [1]. GAO concluded four days later that the affected balances expire at the end of the fiscal year, that the President may not force their expiration, and that Congress’s review would run to 9 November at the earliest, while stating that this was not a view on the programs’ merits [2]. The White House described the mechanism as a pocket rescission, the second such use and the first in 50 years by its own account [1]. The legal dispute is unresolved and this review takes no position on it.
What the evidence supports. Three propositions survive appraisal. First, additional research funding produces real but modest and decelerating gains in output, and peer review predicts later impact poorly, so the quality cost of marginal cuts is likely smaller than advocates assume and the meritocratic case for any specific allocation is weaker than its defenders claim [5][6][7]. Second, public-health and coverage spending is associated with better population outcomes across many designs, with effects that are heterogeneous, context-dependent, and slow — measured in years, not months [8][10][42]. Third, abrupt withdrawal of services produces rapid and large declines in service volume where anyone is measuring, as the PEPFAR and family-planning episodes show [12][59].
What the evidence does not support. It does not support a quantitative prediction about this rescission, a claim that a specific number of patients will be harmed, or a claim that nothing will happen. The relevant literature is observational, asymmetric between expansion and contraction, and — for the exact question of a small, contested, mid-cycle rescission — simply absent.
What follows for a clinician. The defensible response is operational rather than rhetorical: verify coverage and anticipate medication gaps for patients in affected populations [53][55]; recognise that refugee and unaccompanied-minor patients arrive with a high and specific burden of disease that screening can find [63][65]; protect longitudinal data and cohorts in research portfolios, because they are the assets a funding gap destroys permanently [80]; and treat both optimistic and pessimistic claims about this episode with the scepticism that the underlying evidence base requires.
The larger point. The September 2026 package is small in dollars and instructive in structure. American health funding is time-limited, project-based, and competitively allocated, which makes the continuity of money a separate variable from its amount — and makes episodes like this one a recurring feature rather than an anomaly [18]. The evidence base for what that volatility does is strongest where the discontinuity was abrupt and someone kept counting, and weakest exactly where the stakes are highest: the slow erosion of measurement capacity and of the people who produce it.
References
- The White House. Letter from the President to the Speaker of the House of Representatives transmitting 11 proposed rescissions of budget authority, 25 September 2026. Printed as H. Doc. 119-196, 119th Congress, 2d Session, “Proposed Rescission of Budget Authority,” referred to the Committee on Appropriations and ordered to be printed 28 September 2026. https://www.govinfo.gov/content/pkg/CDOC-119hdoc196/pdf/CDOC-119hdoc196.pdf
- U.S. Government Accountability Office. Letter of the General Counsel regarding the proposed rescissions of budget authority transmitted 25 September 2026, dated 29 September 2026 (GAO appropriations law decision; decision number not retrievable from the agency website at the time of preparation). https://www.gao.gov/legal/appropriations-law-decisions
- Administration for Children and Families, Office of Refugee Resettlement. Refugee Program Bureau: programs and services, eligibility, and legislative authority (Refugee Act of 1980; 45 CFR Parts 400-401). Washington, DC: U.S. Department of Health and Human Services. https://www.acf.hhs.gov/orr/programs/refugees
- Agency for Healthcare Research and Quality. About AHRQ: mission, statutory authority (42 U.S.C. § 299 et seq.), and FY2026 research priorities. Rockville, MD: AHRQ. https://www.ahrq.gov/cpi/about/index.html
- Jacob BA, Lefgren L. The Impact of Research Grant Funding on Scientific Productivity. J Public Econ. 2011;95(9-10):1168-1177. doi:10.1016/j.jpubeco.2011.05.005. PMID: 21857758.
- Fortin JM, Currie DJ. Big Science vs. Little Science: How Scientific Impact Scales with Funding. PLoS One. 2013;8(6):e65263. doi:10.1371/journal.pone.0065263. PMID: 23840323.
- Lauer MS, Danthi NS, Kaltman J, Wu C. Predicting Productivity Returns on Investment: Thirty Years of Peer Review, Grant Funding, and Publication of Highly Cited Papers at the National Heart, Lung, and Blood Institute. Circ Res. 2015;117(3):239-43. doi:10.1161/CIRCRESAHA.115.306830. PMID: 26089369.
- Mays GP, Smith SA. Evidence links increases in public health spending to declines in preventable deaths. Health Aff (Millwood). 2011;30(8):1585-93. doi:10.1377/hlthaff.2011.0196. PMID: 21778174.
- Gallet CA, Doucouliagos H. The impact of healthcare spending on health outcomes: A meta-regression analysis. Soc Sci Med. 2017;179:9-17. doi:10.1016/j.socscimed.2017.02.024. PMID: 28237460.
- Sommers BD, Baicker K, Epstein AM. Mortality and access to care among adults after state Medicaid expansions. N Engl J Med. 2012;367(11):1025-34. doi:10.1056/NEJMsa1202099. PMID: 22830435.
- Lee BP, Dodge JL, Terrault NA. Medicaid expansion and variability in mortality in the USA: a national, observational cohort study. Lancet Public Health. 2022;7(1):e48-e55. doi:10.1016/S2468-2667(21)00252-8. PMID: 34863364.
- Cust H, Chabata ST. HIV service delivery after the PEPFAR and USAID HIV funding withdrawal in Uganda and Zimbabwe: an interrupted time-series study. Lancet Glob Health. 2027:104061. doi:10.1016/j.langlo.2026.104061. PMID: 42830089.
- Willard R, Shah GH, Leep C, Ku L. Impact of the 2008-2010 economic recession on local health departments. J Public Health Manag Pract. 2012;18(2):106-14. doi:10.1097/PHH.0b013e3182461cf2. PMID: 22217535.
- Associated Press. Trump administration uses rare authority to claw back nearly $1B in spending approved by Congress. STAT News, 25 September 2026. https://www.statnews.com/2026/09/25/trump-administration-cancel-spending-hhs-programs/
- Administration for Children and Families, Office of Refugee Resettlement. Who are unaccompanied alien children? Unaccompanied Alien Children Bureau. Washington, DC: U.S. Department of Health and Human Services. https://www.acf.hhs.gov/orr/programs/ucs/about
- U.S. Department of Health and Human Services, Office of Minority Health. Office of Minority Health: leadership, data, programs, and grant resources. Washington, DC: HHS. https://www.minorityhealth.hhs.gov/
- Congressional Budget and Impoundment Control Act of 1974, Pub. L. 93-344, title X, § 1012, codified at 2 U.S.C. § 683 (rescission of budget authority), as amended by Pub. L. 100-119, title II, § 207, 29 September 1987. https://www.law.cornell.edu/uscode/text/2/683
- Alberts B, Kirschner MW, Tilghman S, Varmus H. Rescuing US biomedical research from its systemic flaws. Proc Natl Acad Sci U S A. 2014;111(16):5773-7. doi:10.1073/pnas.1404402111. PMID: 24733905.
- Narahari AK, Mehaffey JH, Hawkins RB, Charles EJ, Baderdinni PK, Chandrabhatla AS, et al. Surgeon Scientists Are Disproportionately Affected by Declining NIH Funding Rates. J Am Coll Surg. 2018;226(4):474-481. doi:10.1016/j.jamcollsurg.2017.12.047. PMID: 29482999.
- Agarwal R, Tu W. NIH Funding, Research Productivity, and Scientific Impact: a 20-Year Study. J Gen Intern Med. 2022;37(1):104-109. doi:10.1007/s11606-021-06659-y. PMID: 33655385.
- Pier EL, Brauer M, Filut A, Kaatz A, Raclaw J, Nathan MJ, et al. Low agreement among reviewers evaluating the same NIH grant applications. Proc Natl Acad Sci U S A. 2018;115(12):2952-2957. doi:10.1073/pnas.1714379115. PMID: 29507248.
- Fang FC, Casadevall A. Research Funding: the Case for a Modified Lottery. mBio. 2016;7(2):e00422-16. doi:10.1128/mBio.00422-16. PMID: 27073093.
- Danthi NS, Wu CO, DiMichele DM, Hoots WK, Lauer MS. Citation impact of NHLBI R01 grants funded through the American Recovery and Reinvestment Act as compared to R01 grants funded through a standard payline. Circ Res. 2015;116(5):784-8. doi:10.1161/CIRCRESAHA.116.305894. PMID: 25722441.
- Svider PF, Mauro KM, Sanghvi S, Setzen M, Baredes S, Eloy JA. Is NIH funding predictive of greater research productivity and impact among academic otolaryngologists?. Laryngoscope. 2013;123(1):118-22. doi:10.1002/lary.23659. PMID: 22991270.
- Hegde V, Johansen D, Park HY, Zoller SD, Hamad C, Bernthal NM. The Relationship Between OREF Grants and Future NIH Funding Success. J Bone Joint Surg Am. 2017;99(16):e87. doi:10.2106/JBJS.16.01278. PMID: 28816904.
- Antman MD, Gorelik R, Kennedy A, Liou GF, Billingslea EN, Corrigan JG, et al. Changes in the National Cancer Institute’s R01 workforce: growth, aging, retention, and policy implications. J Clin Invest. 2021;131(7). doi:10.1172/JCI146925. PMID: 33792559.
- Milewicz DM, Lorenz RG, Dermody TS, Brass LF. Rescuing the physician-scientist workforce: the time for action is now. J Clin Invest. 2015;125(10):3742-7. doi:10.1172/JCI84170. PMID: 26426074.
- Salata RA, Geraci MW, Rockey DC, Blanchard M, Brown NJ, Cardinal LJ, et al. U.S. Physician-Scientist Workforce in the 21st Century: Recommendations to Attract and Sustain the Pipeline. Acad Med. 2018;93(4):565-573. doi:10.1097/ACM.0000000000001950. PMID: 28991849.
- Lambert WM, Wells MT, Cipriano MF, Sneva JN, Morris JA, Golightly LM. Career choices of underrepresented and female postdocs in the biomedical sciences. Elife. 2020;9. doi:10.7554/eLife.48774. PMID: 31898935.
- Galkina Cleary E, Beierlein JM, Khanuja NS, McNamee LM, Ledley FD. Contribution of NIH funding to new drug approvals 2010-2016. Proc Natl Acad Sci U S A. 2018;115(10):2329-2334. doi:10.1073/pnas.1715368115. PMID: 29440428.
- Sampat BN, Pincus HA. Citations in Life Science Patents to Publicly Funded Research at Academic Medical Centers. Clin Transl Sci. 2015;8(6):759-63. doi:10.1111/cts.12361. PMID: 26728947.
- Siddique S, Wang R, Yasin F, Gaddy JJ, Zhang L, Gross CP, et al. USPSTF Colorectal Cancer Screening Recommendation and Uptake for Individuals Aged 45 to 49 Years. JAMA Netw Open. 2024;7(10):e2436358. doi:10.1001/jamanetworkopen.2024.36358. PMID: 39361285.
- Kearns JT, Holt SK, Wright JL, Lin DW, Lange PH, Gore JL. PSA screening, prostate biopsy, and treatment of prostate cancer in the years surrounding the USPSTF recommendation against prostate cancer screening. Cancer. 2018;124(13):2733-2739. doi:10.1002/cncr.31337. PMID: 29781117.
- Song S, Kucik JE. Trends in the Utilization of Recommended Clinical Preventive Services, 2011-2019. Am J Prev Med. 2021;61(2):149-157. doi:10.1016/j.amepre.2021.03.010. PMID: 34039495.
- Berenholtz SM, Lubomski LH, Weeks K, Goeschel CA, Marsteller JA, Pham JC, et al. Eliminating central line-associated bloodstream infections: a national patient safety imperative. Infect Control Hosp Epidemiol. 2014;35(1):56-62. doi:10.1086/674384. PMID: 24334799.
- Brilli RJ, McClead RE Jr, Crandall WV, Stoverock L, Berry JC, Wheeler TA, et al. A comprehensive patient safety program can significantly reduce preventable harm, associated costs, and hospital mortality. J Pediatr. 2013;163(6):1638-45. doi:10.1016/j.jpeds.2013.06.031. PMID: 23910978.
- Rangachari P, Rissing P, Rethemeyer K. Awareness of evidence-based practices alone does not translate to implementation: insights from implementation research. Qual Manag Health Care. 2013;22(2):117-25. doi:10.1097/QMH.0b013e31828bc21d. PMID: 23542366.
- Fleming K, George JL, Bazelak SJ, Roeske JA, Biggs AD, Landry CM, et al. Optimizing Respiratory Therapy Resources by De-Implementing Low-Value Care. Respir Care. 2023;68(5):559-564. doi:10.4187/respcare.10712. PMID: 37015815.
- Havranek MM, Rüter F, Bilger S, Dahlem Y, Oliveira L, Ehbrecht D, et al. Validity of 16 AHRQ Patient Safety Indicators to identify in-hospital complications: a medical record review across nine Swiss hospitals. Int J Qual Health Care. 2023;35(4):0. doi:10.1093/intqhc/mzad092. PMID: 37949115.
- Bernet PM, Gumus G, Vishwasrao S. Effectiveness of public health spending on infant mortality in Florida, 2001-2014. Soc Sci Med. 2018;211:31-38. doi:10.1016/j.socscimed.2018.05.044. PMID: 29885571.
- Bernet P, Gumus G, Vishwasrao S. Maternal Mortality and Public Health Programs: Evidence from Florida. Milbank Q. 2020;98(1):150-171. doi:10.1111/1468-0009.12442. PMID: 31943403.
- Ammi M, Arpin E, Dedewanou FA, Allin S. Do expenditures on public health reduce preventable mortality in the long run? Evidence from the Canadian provinces. Soc Sci Med. 2024;345:116696. doi:10.1016/j.socscimed.2024.116696. PMID: 38377835.
- Edney LC, Haji Ali Afzali H, Cheng TC, Karnon J. Mortality reductions from marginal increases in public spending on health. Health Policy. 2018;122(8):892-899. doi:10.1016/j.healthpol.2018.04.011. PMID: 29759682.
- Watkins J, Wulaningsih W, Da Zhou C, Marshall DC, Sylianteng GDC, Dela Rosa PG, et al. Effects of health and social care spending constraints on mortality in England: a time trend analysis. BMJ Open. 2017;7(11):e017722. doi:10.1136/bmjopen-2017-017722. PMID: 29141897.
- Schenck AP, Meyer AM, Kuo TM, Cilenti D. Building the evidence for decision-making: the relationship between local public health capacity and community mortality. Am J Public Health. 2015;105 Suppl 2(Suppl 2):S211-6. doi:10.2105/AJPH.2014.302500. PMID: 25689215.
- Reszczynski O, Connolly J, Shultz K, Kelly S, Mitra N, Hom J, et al. US local public health department spending between 2008 and 2016 did not increase for communities in need. BMC Health Serv Res. 2022;22(1):237. doi:10.1186/s12913-022-07613-2. PMID: 35189868.
- Vilda D, Walker BC, Hardeman RR, Wallace ME. Associations Between State and Local Government Spending and Pregnancy-Related Mortality in the U.S. Am J Prev Med. 2023;64(4):459-467. doi:10.1016/j.amepre.2022.10.022. PMID: 36658021.
- Park S, Dieleman JL, Kim R, Subramanian SV. Association of Health and Social Spending With Health Outcomes in OECD Countries. Health Serv Res. 2025;60(6):e14660. doi:10.1111/1475-6773.14660. PMID: 40705475.
- Červený J, Ours JCV. Long-term returns to local health-care spending. Eur J Health Econ. 2025;26(2):153-182. doi:10.1007/s10198-024-01695-x. PMID: 38762706.
- Bhatt CB, Beck-Sagué CM. Medicaid Expansion and Infant Mortality in the United States. Am J Public Health. 2018;108(4):565-567. doi:10.2105/AJPH.2017.304218. PMID: 29346003.
- Cook A, Stype A. Medicaid expansion and infant mortality: the (questionable) impact of the Affordable Care Act. J Epidemiol Community Health. 2021;75(1):10-15. doi:10.1136/jech-2019-213666. PMID: 32913131.
- Wiggins A, Karaye IM, Horney JA. Medicaid expansion and infant mortality, revisited: A difference-in-differences analysis. Health Serv Res. 2020;55(3):393-398. doi:10.1111/1475-6773.13286. PMID: 32196658.
- McIntyre A, Sommers BD, Aboulafia G, Phelan J, Orav EJ, Epstein AM, et al. Coverage and Access Changes During Medicaid Unwinding. JAMA Health Forum. 2024;5(6):e242193. doi:10.1001/jamahealthforum.2024.2193. PMID: 38943683.
- Bensken WP, Dankovchik J, Fein HL, Duhon G, Sills MR. Unwinding of Continuous Medicaid Coverage Among Pediatric Community Health Center Patients. JAMA Netw Open. 2025;8(2):e2458155. doi:10.1001/jamanetworkopen.2024.58155. PMID: 39913136.
- Chua KP, Constantin J, Kenney GM, Conti RM, Simon K. Changes in Chronic Medication Dispensing to Children and Young Adults During Medicaid Unwinding. Pediatrics. 2025;155(6). doi:10.1542/peds.2024-070380. PMID: 40314398.
- Cliff BQ, Sinha S, Cross DA, Hickey E, Kirschner K, Caskey R. Medicaid Disenrollment Among Young Adults With and Without Complex Medical Conditions. JAMA Pediatr. 2026;180(5):518-526. doi:10.1001/jamapediatrics.2025.6326. PMID: 41697678.
- Thompson O. The long-term health impacts of Medicaid and CHIP. J Health Econ. 2017;51:26-40. doi:10.1016/j.jhealeco.2016.12.003. PMID: 28040620.
- DeVoe JE, Marino M, Gold R, Hoopes MJ, Cowburn S, O’Malley JP, et al. Community Health Center Use After Oregon’s Randomized Medicaid Experiment. Ann Fam Med. 2015;13(4):312-20. doi:10.1370/afm.1812. PMID: 26195674.
- Packham A. Family planning funding cuts and teen childbearing. J Health Econ. 2017;55:168-185. doi:10.1016/j.jhealeco.2017.07.002. PMID: 28811119.
- Hoffman GJ, Ha J, Fan Z, Li J. Associations between rural hospital closures and acute and post-acute care access and outcomes. Health Serv Res. 2025;60(3):e14426. doi:10.1111/1475-6773.14426. PMID: 39739398.
- Ye J, Leep C, Newman S. Reductions of budgets, staffing, and programs among local health departments: results from NACCHO’s economic surveillance surveys, 2009-2013. J Public Health Manag Pract. 2015;21(2):126-33. doi:10.1097/PHH.0000000000000074. PMID: 24691428.
- Kumar GS, Beeler JA, Seagle EE, Jentes ES. Long-Term Physical Health Outcomes of Resettled Refugee Populations in the United States: A Scoping Review. J Immigr Minor Health. 2021;23(4):813-823. doi:10.1007/s10903-021-01146-2. PMID: 33515162.
- Birhane BM, Dawson A, Hayen A. Health of refugee children upon arrival in high-income countries: A scoping review. J Migr Health. 2025;12:100373. doi:10.1016/j.jmh.2025.100373. PMID: 41328262.
- Xu K, Watanabe-Galloway S, Qu M, Grimm B, Kim J. Common Diagnoses among Refugee Populations: Linked Results with Statewide Hospital Discharge Database. Ann Glob Health. 2018;84(3):541-550. doi:10.29024/aogh.2354. PMID: 30835394.
- Daniel-Calveras A, Baldaquí N, Baeza I. Mental health of unaccompanied refugee minors in Europe: A systematic review. Child Abuse Negl. 2022;133:105865. doi:10.1016/j.chiabu.2022.105865. PMID: 36095862.
- Bamford J, Fletcher M, Leavey G. Mental Health Outcomes of Unaccompanied Refugee Minors: a Rapid Review of Recent Research. Curr Psychiatry Rep. 2021;23(8):46. doi:10.1007/s11920-021-01262-8. PMID: 34196826.
- Vervliet M, Lammertyn J, Broekaert E, Derluyn I. Longitudinal follow-up of the mental health of unaccompanied refugee minors. Eur Child Adolesc Psychiatry. 2014;23(5):337-46. doi:10.1007/s00787-013-0463-1. PMID: 23979476.
- Jensen TK, Skar AS, Andersson ES, Birkeland MS. Long-term mental health in unaccompanied refugee minors: pre- and post-flight predictors. Eur Child Adolesc Psychiatry. 2019;28(12):1671-1682. doi:10.1007/s00787-019-01340-6. PMID: 31004294.
- Nilsen SA, Kvestad I, Randal SB, Hysing M, Sayyad N, Bøe T. Mental health among unaccompanied refugee minors after settling in Norway: A matched cross-sectional study. Scand J Public Health. 2023;51(3):430-441. doi:10.1177/14034948221100103. PMID: 35684945.
- Salhi C, Scoglio AAJ, Ellis H, Issa O, Lincoln A. The relationship of pre- and post-resettlement violence exposure to mental health among refugees: a multi-site panel survey of somalis in the US and Canada. Soc Psychiatry Psychiatr Epidemiol. 2021;56(6):1015-1023. doi:10.1007/s00127-020-02010-8. PMID: 33398495.
- Hilado A, Bond M, Sanchez E. Predicting Mental Health Care Enrollment and Treatment Uptake among Newly Arrived Refugees in U.S. Resettlement Programs. J Health Soc Behav. 2025;66(1):124-138. doi:10.1177/00221465241276778. PMID: 39277775.
- Magwood O, Kassam A, Mavedatnia D, Mendonca O, Saad A, Hasan H, et al. Mental Health Screening Approaches for Resettling Refugees and Asylum Seekers: A Scoping Review. Int J Environ Res Public Health. 2022;19(6). doi:10.3390/ijerph19063549. PMID: 35329237.
- Patel ZS, Casline EP, Vera C, Ramirez V, Jensen-Doss A. Unaccompanied migrant children in the United States: Implementation and effectiveness of trauma-focused cognitive behavioral therapy. Psychol Trauma. 2024;16(Suppl 2):S389-S399. doi:10.1037/tra0001361. PMID: 36107712.
- Virk S, Milewski A, Khan N, Cheung C, Wright DN, Kaur G. Vaccination in Forcibly Displaced, Pediatric Populations: A Systematic Review and Meta-Analysis. JAMA Netw Open. 2025;8(6):e2516237. doi:10.1001/jamanetworkopen.2025.16237. PMID: 40522657.
- Dawson-Hahn E, Pak-Gorstein S, Matheson J, Zhou C, Yun K, Scott K, et al. Growth Trajectories of Refugee and Nonrefugee Children in the United States. Pediatrics. 2016;138(6). doi:10.1542/peds.2016-0953. PMID: 27940678.
- Cohen SB. Design strategies and innovations in the medical expenditure panel survey. Med Care. 2003;41(7 Suppl):III5-III12. doi:10.1097/01.MLR.0000076048.11549.71. PMID: 12865722.
- Blumberg SJ, Parker JD, Moyer BC. National Health Interview Survey, COVID-19, and Online Data Collection Platforms: Adaptations, Tradeoffs, and New Directions. Am J Public Health. 2021;111(12):2167-2175. doi:10.2105/AJPH.2021.306516. PMID: 34878857.
- Zablotsky B, Lessem SE, Gindi RM, Maitland AK, Dahlhamer JM, Blumberg SJ. Overview of the 2019 National Health Interview Survey Questionnaire Redesign. Am J Public Health. 2023;113(4):408-415. doi:10.2105/AJPH.2022.307197. PMID: 36758202.
- Patel K, McCall TC, Cunningham M, Garofalini C, Lee J, Alford AA. Sustainability of the Growth of the Local Public Health Workforce During the COVID-19 Pandemic, 2019-2022. Am J Public Health. 2025;115(8):1271-1277. doi:10.2105/AJPH.2025.308096. PMID: 40505063.
- Levy D. Straight talk with…Daniel Levy. Interviewed by Elie Dolgin. Nat Med. 2013;19(10):1197. doi:10.1038/nm1013-1197. PMID: 24100971.
- Tung EL, Asfour N, Bolton JD, Huang ES, Zhang C, Anselin L. Impact of 2009 American Recovery and Reinvestment Act (ARRA) health center investments on disadvantaged neighborhoods after recession. Health Econ Rev. 2024;14(1):9. doi:10.1186/s13561-024-00482-x. PMID: 38294643.
- Dedewanou FA, Allin S, Guyon A, Pawa J, Ammi M. Prioritization of public health financing, organization, and workforce transformation: a Delphi study in Canada. BMC Public Health. 2023;23(1):544. doi:10.1186/s12889-023-15373-9. PMID: 36949440.
- Celhay P, Martinez S, Muñoz M, Perez M, Perez-Cuevas R. Long-term effects of public health insurance on the health of children in Mexico: a retrospective study. Lancet Glob Health. 2019;7(10):e1448-e1457. doi:10.1016/S2214-109X(19)30326-2. PMID: 31405780.