Modeling the impacts of budget reductions at the U.S. NIH on research and innovation

Abstract

Stable public funding underpins biomedical innovation, yet the U.S. research system remains vulnerable to swings in federal budgets. We developed BRIDGE (Biomedical Research Investment Dynamics and Growth Evaluation), a simulation model linking NIH funding to new drug approvals, workforce dynamics, and scientific output. Using 47 longitudinal data measures from 1995–2024, validated against historical trends and projecting outcomes through 2050, we examined multiple funding trajectories, including stable growth, flat budgets, large reductions, and later restorations. Results show that across scenarios, even temporary disruptions create lasting setbacks for drug development, researcher training, and basic science. Later funding increases cannot repair the damage for decades, with persistent losses in capacity and output that highlight how fragile the system is to short-term shocks and why stable public investment is essential for long-term innovation. For example, if the NIH budget were cut by 40% in 2026, even doubling that reduced budget at the start of the next administration would still require 36 years to recover the cumulative loss in novel drug approvals.

Introduction

Federal funding plays a central role in setting the trajectory of biomedical research in the U.S. and, as the world’s largest funder, shapes scientific progress worldwide. The National Institutes of Health (NIH) has long been the cornerstone of U.S. biomedical investment, funding research, training, and infrastructure while guiding priorities, norms, and capacity across the system^1–3. Its budget, however, has periodically been subject to political and economic pressures. Such pressures raise concerns about the continuity of research programs and the broader infrastructure and workforce that underpin the U.S. biomedical research enterprise^4,5.

A growing body of evidence has shown that these investments support innovative science, help train the next generation of researchers, and catalyze collaborations with industry that lead to therapeutic development^1,6,7. For example, decades of NIH-funded work on the Human Genome Project lay the foundation for cancer immunotherapy, named “breakthrough of the year” in 2013^8,9. Ongoing initiatives, such as the BRAIN Initiative and NIH-supported antibiotic resistance research, could accelerate treatments for Alzheimer’s and help avert projected deaths from resistant infections^10–12. These areas often lack commercial investment, given uncertain returns and long timelines. At the same time, their broad public benefit makes public support essential.

While the value of government research spending is occasionally questioned, studies consistently demonstrate that NIH funding yields significant returns in the form of new knowledge, improved health outcomes, and long-term economic benefits^13,14. In 2024, NIH funding supported about 408,000 jobs and generated $94.6 billion in new economic output nationwide, an estimated return of $2.56 for every dollar awarded¹⁵. Research also shows that even temporary shifts in NIH funding produce magnified downstream effects due to multi-year grant obligations¹⁶. A study estimated that if the agency’s historical funding had been about 40% lower, NIH-funded research linked to more than half of modern drug discoveries (286 of 557, or 51%) approved between 2000 and 2023 would have been at risk¹⁷.

What remains less understood, however, is how different funding paths—such as sustained reductions, stagnation, or later restoration—shape the long-term impacts on the biomedical research system, and what this means for preventing or mitigating disruption to innovation. Understanding the likely impacts of budget reductions raises critical questions: How will the contraction affect the size and composition of the biomedical research workforce? What are the likely consequences for the pace of scientific discovery, translational progress, and therapeutic innovation? And if a large-scale cut is reversed, how quickly and to what extent can the system recover?

To address these questions, this study develops a dynamic simulation model of the biomedical research ecosystem—BRIDGE (Biomedical Research Investment Dynamics and Growth Evaluation)—grounded in historical data and prior empirical insights into funding trajectories, workforce shifts, and innovation outcomes. A systems modeling approach is well-suited for capturing how changes in policies ripple through components of an ecosystem, especially when delays, accumulations, and feedback loops are involved. These mechanisms are common in large-scale adaptive systems. They capture time-lagged and feedback-driven effects that retrospective or regression-based analyses may miss. Rather than extrapolating past trends, the model generates projections from the way elements within the biomedical system interact over time, allowing shifts such as slower therapeutic development or changes in workforce composition to emerge from the system’s structure.

While the model incorporates a range of upstream and midstream processes, including training, workforce dynamics, and research outputs, our primary outcome is novel drug approvals since it is a measurable and policy-relevant proxy for downstream biomedical innovation. This focus does not imply that NIH funding serves only drug development; rather, the model is designed to reflect how public investment in basic and translational research enables innovation, including but not limited to therapeutics. We simulate a period of reduced NIH budget, along with alternative scenarios in which funding returns to pre-cut levels at varying speeds.

The model integrates 47 longitudinal data measures from 1995 to 2024, is validated against historical benchmarks, and projects outcomes through 2050. Full model details and validation procedures are provided in the supplementary materials.

Results

Simulated NIH budget scenarios

We examined several counterfactual scenarios for post-2026, shown in Fig. 1. In the “baseline” scenario, appropriations grow with inflation, keeping purchasing power constant over time. In the “no-increase” scenario, appropriations remain fixed with no further growth. In the “cut” scenario, funding is reduced by 40% in 2026, after which appropriations grow with inflation from the reduced base.

Fig. 1. Annual NIH congressional appropriations, nominal and adjusted to 2025 USD using Biomedical Research and Development Price Index, historical (1995-2025) and projected (2026-2050).

A Nominal and B adjusted to 2025 USD using Biomedical Research and Development Price Index. Scenarios include: baseline (dashed line); no-increase (purple line); cut (orange line); and restoration to baseline level after one year (green) and four years (blue). The two restore scenarios have a four-year ramp. All values are adjusted using the Biomedical Research and Development Price Index (BRDPI), which accounts for inflation specific to biomedical research; as a result, nominally constant appropriations appear as a decline in real terms over time.

In addition, in the “restore” scenarios, funding is also reduced by 40% in 2026 but then gradually returns to the baseline trajectory through steady nominal annual increases, beginning either one year after the cut (2027, the “1-year restore” case) or four years after the cut (2030, the “4-year restore” case). These timings reflect two plausible policy paths: one in which funding is adjusted within the current administration, and another in which changes occur under a future administration. The pace and scale of these restoration scenarios are comparable to the historical doubling era, where the NIH budget increased in unadjusted dollars from about $14 billion to $27 billion between 1998 and 2003¹⁸.

In the baseline scenario, the nominal NIH appropriations are assumed to continue to rise, as has often been the case historically (see Fig. 1-A), following the annual increase of 2.7% beyond 2025, based on the forecast of the Biomedical Research and Development Price Index (BRDPI). The BRDPI, provided by the Department of Commerce’s Bureau of Economic Analysis, measures annual changes in the cost of supplies and workforce necessary for conducting biomedical research¹⁹. The BRDPI is routinely cited in annual congressional appropriations discussions as a benchmark for justifying budget increases, making it a relevant and policy-aligned measure for modeling future appropriations²⁰. Because research costs typically rise faster than general consumer prices, BRDPI offers more accurate comparisons than the consumer price index, the most widely used measure of inflation. Hence, we used BRDPI to adjust budget projections in constant 2025 USD as key model inputs to estimate the number of research projects the NIH budget could support, and in turn, other downstream outcomes. Figure 1 presents both nominal (panel A) and BRDPI-adjusted (panel B) values reported in constant 2025 USD. The figure shows that while nominal funding increases, the inflation-adjusted (real) value may decline over time if appropriations do not keep pace with rising research costs.

Additionally, as a supplementary test, we examined an extreme case in which, following four years of sustained cuts, the NIH budget doubles in nominal terms in fiscal year 2030, and then grows annually at the rate of BRDPI. This scenario is not intended as a realistic forecast but as a test to examine how quickly the system could recover under extreme conditions, illustrated in the supplementary materials.

Projections of the simulated scenarios

Using historical data on NIH appropriations, workforce composition, research outputs, and innovation indicators, the model closely replicated observed trends from 1995 to 2024 with a mean absolute percentage error of 10.5% across series. Model fit is shown in the supplementary materials. Here, we first report projections for the size and composition of the workforce, followed by research outputs and drug innovation, and conclude with economic implications and robustness checks.

Figure 2 presents the projected outcomes in raw numbers, while percentage differences are plotted in the supplementary materials. Panel (a) shows the number of principal investigators (PIs), including early-stage and established researchers with active NIH grants. The number of principal investigators (PIs) has remained largely stagnant since 2009. Under the baseline scenario (dashed line), it is projected to remain near current levels with only a slight decline, whereas in the no-increase scenario (purple line), the number of PIs declines gradually at first, then more rapidly, reaching 46% below the baseline by 2050. Under the cut scenario (orange line), the number of PIs declines sharply, reaching a low point within 9 years at 50% below the baseline. It then recovers slightly but remains 36% below the baseline by 2050. In the 1-year restoration scenario (green line), the number of PIs also drops sharply but begins a gradual recovery once funding is restored. However, the recovery is not symmetrical: the decline is rapid, while the rebound is slower, returning to baseline in about 20 years. The 4-year restoration scenario (blue line) follows an overall similar trend.

Fig. 2. Projected impacts of NIH funding trajectories on the U.S. biomedical research system (1995-2050).

Principal investigators with active NIH grants (A), postdoctoral researchers (B), doctoral students (C), NIH-supported fundamental research papers (D), annual novel drug approvals (E), and cumulative novel drug approval shortfalls relative to the baseline scenario from 2026 to 2050 (F). Scenarios include: baseline (dashed line); no-increase (purple line); cut (orange line); and restoration to baseline level after 1 year (green) and 4 years (blue). Vertical line indicates the initial year for scenario analysis.

These reductions also translate into substantial losses in research continuity among PIs, figures presented in the supplement. Relative to baseline, the no-increase and cut scenarios resulted in approximately 193,000 and 349,000 cumulative funded PI-years lost by 2050, respectively, while the 1-year and 4-year restoration scenarios resulted in approximately 94,000 and 162,000 lost PI-years. Over time projections further show that average funding duration per PI declined sharply following funding disruptions, falling from nearly 10 years to approximately 5.4 years in the no-increase scenario and 4.3 years under funding cuts. Although restoration scenarios eventually recovered, the rebound was gradual and delayed, indicating that interruptions in funding produced persistent effects on research continuity even after appropriations returned to baseline levels. Among investigators who remained as PIs, average funding-years lost among surviving PIs were 2.1 years in the no-increase scenario, 2.7 years under funding cuts, and less than 0.2 years under the restoration scenarios.

Panel (b) shows that postdoctoral researchers has increased in previous years. Under the baseline scenario, growth is projected to continue but at a slower rate. In the no-increase scenario, the number of postdocs is projected to decline, reaching about 31% below the baseline by 2050. The cut scenario results in a faster initial decline, with the number of postdoctoral researchers falling to about 33% below the baseline at its lowest point in 13 years, before recovering slightly to 31% below baseline by 2050. In both the 1-year and 4-year restoration scenarios, the number of postdoctoral researchers also declines sharply after the initial cut, reaching lows of 12% and 25% below baseline in 5 and 7 years, respectively. However, postdoctoral numbers rebound more quickly than the PI count, returning to near-baseline levels in 12 and 14 years, respectively.

Panel (c) shows that the number of doctoral students in science, engineering, and health has overall risen steadily in recent decades. Under the baseline scenario, enrollment is projected to continue increasing, but at a slower rate after 2036. In all other scenarios, the number of doctoral students is projected to remain above 2025 levels, though the levels vary relative to the baseline. By 2050, the number of doctoral students is projected to be 7% below baseline in the no-increase scenario and 15% below baseline in the cut scenario. In the 1-year and 4-year restoration scenarios, the declines are milder, at 5% and 9% below baseline, respectively, yet recovery still does not return the number of doctoral students to baseline.

Additionally, panel (d) presents NIH-supported fundamental research publications. The number of fundamental research papers has not increased since peaking around 2010, and under the baseline scenario, this stagnation is projected to continue. In the no-increase scenario, a slow decline begins after one year and accelerates after 2030, reaching 61% below baseline by 2050. Under the cut scenario, the number of publications drops sharply and by 2050 reaches levels last observed in 1996, approximately 70% below baseline. In the 1-year and 4-year restoration scenarios, publications fall to lows of 28% and 44% below baseline in 10 and 12 years, respectively, before getting back on a recovery track. However, recovery is slow, and by 2050, counts remain 9% and 18% below baseline, respectively.

Finally, panel (e) shows the annual number of Food and Drug Administration (FDA)-approved novel drugs. Under the baseline scenario, approvals remain near current levels, with a slow decline beginning after about a decade. The no-increase scenario follows a similar trajectory until 2035, after which the decline becomes more pronounced, resulting in 27% fewer annual approvals by 2050. Under the cut scenario, the decline is earlier and sharper, with approvals falling to about 44% below baseline by 2050. In the 1-year and 4-year restoration scenarios, approvals also decline and reach the lows of 13% and 22% below baseline by 2050, respectively.

As panel (f) illustrates, lower approvals lead to a cumulative shortfall in novel drugs between the time of cut, 2026, and 2050. Specifically, the total number of approvals since 2026 falls from 1,122 in the baseline scenario to 1037 in the no-increase scenario, which is about an 8% decline. In the cut scenario, with 920, the cumulative approvals fall 18% short of baseline. In the 1-year restoration scenario, cumulative approvals reach about 1039, representing a 7% drop from the baseline, while in the 4-year restoration scenario, they total around 988, reflecting a 12% decline.

Translating these novel drug shortfalls into approximate foregone economic value, we estimate losses of about $0.20–$0.88 trillion (median: $0.54 trillion) for the no-increase scenario; $0.48–$2.08 trillion (median: $1.28 trillion) for the cut scenario; $0.20–$0.85 trillion (median: $0.52 trillion) for the 1-year restoration scenario; and $0.32–$1.38 trillion (median: $0.85 trillion) for the 4-year restoration scenario. These numbers are expressed in 2025 dollars and should be viewed as rough estimates, intended to illustrate the potential order of magnitude rather than provide precise valuations. Even under an extended scenario where the NIH budget initially follows the cut trajectory but begins to recover after four years—doubling by fiscal year 2030 (an unprecedented increase)—the novel drugs still recover slowly. Cumulative approved novel drugs may not offset the baseline trajectory until around 2062, reflecting a lag of 36 years. Similar inertia is observed across other outcomes, underscoring how delayed or uneven funding recovery can leave lasting gaps in research capacity and outputs. We visualize the results of this extended scenario in the supplementary materials. We report the ranges of estimated parameters and uncertainty intervals for the projected outcomes in the supplementary materials, showing that the overall trends remain similar across a plausible range. We further present intermediate outputs, including the application success rate, principal investigators disaggregated by career stage, total faculty counts, the number of clinical trials, the industry research workforce, and other related measures.

We also estimated elasticity linking funding changes to downstream innovation outcomes (see supplementary materials). Elasticity increased over time across scenarios, particularly higher in restoration scenarios where funding levels partially recovered but innovation deficits persisted. This reflects the delayed and path-dependent nature of the biomedical research system, in which downstream innovation outcomes may continue lagging behind baseline trajectories long after funding levels begin recovering.

Overall, these projections suggest that even temporary reductions in the NIH budget can generate long-lasting effects across the complex system that underpins U.S. biomedical research. We developed an online simulator, available at mj-lab.mgh.harvard.edu/nih-budget-simulator. This tool allows users to interact with the model to examine additional scenarios and explore a wider range of outcomes beyond what we can report here.

Discussion

This study examines how changes in NIH funding may reshape the U.S. biomedical research system under different scenarios. It is the first data-driven, system-level model calibrated to historical data from 1995 to 2024, and projecting through 2050, linking changes in appropriations to long-term shifts in research workforce development and capacity, and the innovation pipeline. More broadly, the results illustrate how different trajectories of NIH funding—growth with inflation, stagnation, sudden reductions, or restoration—can lead to markedly different outcomes over both short- and long-term horizons. These findings are intended not only to highlight potential risks but also to provide a framework for anticipating outcomes and a decision-support tool to buffer the research ecosystem against volatility.

Overall, our results suggest that NIH funding cuts can have compounding, multi-layered effects, not only reducing the immediate number of grants or workforce participants but also weakening the interdependencies that sustain long-term innovation. For example, fewer research grants reduce mentoring opportunities, which slows workforce development; fewer basic science studies delay clinical trials, which then bottleneck downstream drug development. These effects are not isolated but reinforcing, revealing how the NIH functions as a keystone actor in the broader U.S. innovation system.

Although the NIH nominal budget historically rose, once adjusted for biomedical research inflation, the effective budget has been stagnating. As a result, several measures, including the number of researchers, fundamental research papers, and annual FDA approvals of novel drugs, have either already peaked or are projected to peak soon. Beyond those inflection points, our projections show a gradual long-term decline even if the real budget (adjusted for inflation) remains constant. Sudden, large-scale budget reductions add another layer of challenge, with the potential to trigger cascading disruptions across the biomedical research ecosystem, reducing the number of doctoral researchers, limiting institutional research capacity, and suppressing downstream innovation.

For instance, a funding cut of 40% in 2026 would reduce the NIH budget to approximately 2015 levels in nominal terms (i.e., without adjusting for inflation), which is equivalent to levels seen nearly three decades ago (1996-1997) when adjusted for inflation. The scale of the proposed 40% cut is so substantial that, everything else being equal, it would not even be sufficient to cover existing multi-year grant commitments, let alone fund any new awards. This raises a policy challenge: when budgets shrink to merely cover existing commitments, the system effectively stalls, leaving little to no room for new ideas, investigators, or approaches to enter the biomedical system in the U.S.

After such a large cut, even when funding is restored to pre-cut levels, the recovery is slow and often incomplete (i.e., key measures like new fundamental research papers and annual FDA approvals of novel drugs remain below baseline trajectories). This reflects the time it takes to rebuild training infrastructure, rehire and sustain faculty, and re-establish capacity to produce research outcomes. Once lost, many of these do not simply spring back. The lag between workforce loss and reconstitution is particularly important, as it means a large number of researchers may be diverted to other sectors. Delays in early-stage research also create ripple effects that slow the development of clinical advances and reduce the long-term flow of commercial innovation, specifically FDA-approved novel drugs. Importantly, these effects extend beyond reductions in the number of funded investigators. Funding disruptions also shortened research continuity among remaining PIs, reducing average funding duration and generating substantial cumulative losses in funded PI-years. Even under restoration scenarios, recovery in funding continuity was delayed. Consistent with this pattern, scenario-specific elasticity estimates generally increased over time, particularly in recovery scenarios where funding levels partially rebounded but innovation outputs continued to lag behind baseline projections. These results align with earlier findings that even modest and temporary cuts led to clinical trial delays, lab closures, and disrupted career trajectories for trainees and staff^16,21.

Even when modeling an optimistic, unprecedented rebound, where the NIH budget remains at the reduced level for four years and is then doubled in 2030, the system shows signs of prolonged disruption. This highlights the inertia and fragility of the biomedical research pipeline: once disinvestment has occurred, even dramatic reinvestment may not fully restore lost ground. Notably, cumulative FDA drug approvals still lag the baseline trajectory by about 36 years. Delays in workforce replenishment, research program ramp-up, and long-cycle innovation processes mean that missed opportunities in the early years compound over time.

Additionally, debates around NIH funding occasionally include concerns about inefficiency, redundancy, or workforce saturation. One may question whether all NIH-supported research or training is equally impactful, or whether some degree of reallocation or streamlining could improve the overall efficiency of the system. Similarly, concerns about an oversupply of publications or the limited number of long-term research or faculty positions for doctoral students are not new²². But such concerns reflect questions of strategic alignment, how talent and funding are matched to long-term national goals, rather than a simple problem of too much research activity. Addressing these challenges calls for targeted adjustments in funding priorities, workforce development, and institutional incentives, not broad reductions in overall support. Uniform funding reductions do not distinguish between high-impact and low-impact projects, or between fields with saturated vs. underserved talent pipelines. Even modest but sustained cuts can erode the workforce across academia and industry, reduce institutional flexibility, and limit the system’s ability to respond to emerging needs. In that sense, the issue is not just how much is cut, but how indiscriminately.

Other sources of research support, including industry and philanthropy, play important roles but are insufficient to fully replace the functions of NIH funding. While some substitution is possible, it would require fundamentally different priorities and incentives to close such a gap. NIH plays a central role in supporting early-stage, long-horizon research that often lacks clear commercial incentives (e.g., mechanistic studies of disease pathways or development of novel research methods). In contrast, industry tends to invest in late-stage development with shorter time-to-market and clearer returns^23,24. Also, philanthropic funding, while important, is comparatively small and typically focused on specific conditions or populations. Even if private investment increased in response to public disinvestment, the result could be narrower therapeutic targeting, reduced emphasis on basic science, or cost shifts to consumers and payers. These dynamics are important but beyond the scope of this model.

Our findings align with the growing body of research indicating that reductions in NIH funding have system-wide impacts. In 2013, the NIH faced a $1.55 billion (5%) budget cut, leading to fewer research grants and widespread disruption across all research programs²⁵. The immediate disruptions signaled broader effects: many researchers were forced to halt clinical trials, lay off staff, and abandon projects, resulting in lost opportunities for treatment development and delays in potentially life-saving therapies²⁶. Additionally, researchers have estimated that a 33% cut in NIH funding would be associated with a 15.3% decline in patents associated with new drugs, suggesting a proportional reduction in the pipeline of novel therapies^6,27. A preliminary macroeconomic analysis has similarly projected that a 50% reduction in NIH research and development (R&D) spending could reduce U.S. GDP by approximately 3.7% over the long run²⁸.

Our findings have implications for policy decisions beyond the current funding cycle. First, budget cuts can generate increased workforce attrition and underused infrastructure, such as laboratory facilities, shared research cores, or national data platforms, that are expensive to maintain and vulnerable to obsolescence as global standards evolve. Such dynamics may slow the pace of discovery, reduce long-term health gains, and shift U.S. positioning in global science, as countries with more stable public investment may attract top talent and accelerate innovation. Indeed, if early-career researchers lose trust in the stability of federal science funding or opt for more stable, non-research careers, the long-term talent pipeline may narrow, with downstream effects that take decades to correct. This may also deter prospective students from pursuing research careers altogether, shrinking the pipeline before talent even enters it. Second, abrupt disruptions to public funding may compromise the system’s ability to respond to future national needs. For example, reduced investment in early-stage research today may limit the pool of expertise available to respond to future crises, including pandemics, antimicrobial resistance, or climate-related health threats.

Third, the effects of reductions are not symmetric: the damage from reductions occurs quickly, while recovery is slow and incomplete. Periods of funding instability may shift how institutions plan and prioritize long-term research investments, favoring short-term, lower-risk projects that can yield quicker returns. This strategic conservatism could further suppress breakthrough innovation, particularly in foundational science, where payoffs are inherently long-term and less certain. In this way, short-term austerity may lead to long-term stagnation, compounding the very economic and health burdens that scientific research is intended to alleviate. Fourth, our results caution against assuming that later surges in funding can readily compensate for earlier cuts. The model illustrates how scientific output is not immediately responsive to financial inputs: it must be cultivated, retained, and protected over time.

Finally, the impacts of large-scale budget cuts on novel drug approvals translate into potential present-value economic losses in the range of several hundred billion to over a trillion dollars (2025 USD) by 2050, depending on the scenario. These losses are substantial relative to the fiscal savings from the budget reduction, reflecting how disinvestment in biomedical research can generate long-lasting and disproportionate reductions in innovation output.

We provide a decision-support tool (through a user-friendly, online dashboard) for policymakers considering tradeoffs in fiscal planning, enabling exploration of both short-term budget savings and long-term innovation impacts.

This study has several limitations. The model is designed to capture the key mechanisms linking NIH funding to workforce capacity, research activity, and novel drug development, but it does not represent every factor that could influence the biomedical landscape over the next 25 years. For example, shifts toward medical devices or diagnostics, advances in artificial intelligence-enabled care, regulatory changes, or global shifts in research priorities could alter drug approvals independently of NIH investment. Advances in artificial intelligence and related research tools could also affect research productivity over time, although the magnitude and timing of such effects remain uncertain and were not explicitly modeled in this study. However, because the analysis focuses on comparisons across funding scenarios, broad productivity improvements that affect all scenarios similarly may shift overall research outputs without necessarily eliminating the relative differences between scenarios. These dynamics are outside the present scope.

Beyond these scope boundaries, the model has several specific limitations. While the model integrates historical data and estimates, it includes simplifications. It does not explicitly represent how universities might shift internal priorities, restructure grant portfolios, or adapt staffing models in response to external shocks. The model also does not capture institutional financial dynamics such as indirect cost recovery, infrastructure support, or broader university budget adjustments that may create additional pressures on research capacity during funding reductions. The model uses novel drug approvals as a primary outcome to represent downstream innovation, but it does not capture other important public health impacts of NIH-funded research, such as diagnostics or behavioral interventions. The model also does not distinguish workforce diversity or equity impacts, nor does it account for regional or institutional disparities, both of which may worsen under funding contractions and compound disparities in training, career advancement, and research opportunities across communities. In addition, the current framework does not explicitly represent career-stage-specific NIH funding allocation mechanisms. Moreover, all projections assume that other key institutions, such as the FDA, continue operating as usual; however, these entities may also be impacted by budget cuts, and such ripple effects were not modeled in this analysis.

Despite these limitations, the model offers a transparent, focused foundation for assessing long-term impacts of NIH budget cuts. No single model can address every question in a complex system, and overly comprehensive models often sacrifice clarity and usefulness²⁹. By concentrating on core structural relationships, this model provides a flexible base for future extensions and more targeted policy analysis.

Taken together, our findings underscore that the biomedical research ecosystem is sensitive not only to how much funding it receives, but also to how predictably and consistently that funding is delivered. When support is withdrawn suddenly, the effects can extend across generations of training, discovery, and application. Recovery is possible, but it could take multiple decades, and some opportunities might never return. Finally, while focused on NIH, these findings may generalize to other large-scale public research institutions, highlighting the broader risks of destabilizing national science infrastructures.

Methods

Model development

We developed a simulation model to capture the interdependent processes through which NIH research funding influences workforce composition and capacity, and downstream innovation. We chose a system dynamics modeling approach due to its strength in representing feedback, delays, and cumulative effects in complex, interdependent systems like research funding and innovation pipelines²⁹.

The model is conceptually grounded in theories of public sector retrenchment and draws on frameworks from innovation economics and science and technology policy, as discussed in our prior research⁵. This foundation reflects a body of empirical research documenting the relationship between NIH funding and scientific productivity, translational activity, and long-term innovation outcomes^1,2,6,7. It also incorporates lags between funding changes and measurable system responses, consistent with prior studies of biomedical R&D trajectories^14,30.

Figure 3 presents a schematic overview of the model, organized into four sections: NIH-funded research grants, the academic research workforce, the industry workforce, and the biomedical innovation pipeline. The model tracks how publicly funded research investments move through the biomedical ecosystem, supporting academic researchers, some of whom later enter industry, and shaping flows of people across training, faculty appointments, and exits. It focuses specifically on R&D roles, excluding those not engaged in research. Innovation emerges from fundamental research (basic research that reveals mechanisms of biology, disease, or behavior³¹) supported by NIH, which is often long-term, infrastructure-focused, and non-commercial, as well as from applied research led by industry, resulting in outputs such as publications, clinical trials, and FDA-approved novel drugs²⁴.

Fig. 3. Overview of the model capturing the U.S. biomedical research ecosystem, encompassing NIH-funded research grants, academic and industry workforce, and biomedical innovation pipeline.

Stocks, represented by rectangular boxes, indicate accumulations. Transitions, or flows, to the stocks areshown with double-lined arrows. Single-lined arrows represent relationships between variables, witharrowheads indicating the direction of the effect. A plus sign denotes that two variables move in thesame direction (e.g., an increase in one leads to an increase in the other and vice versa for the case of adecrease), while a minus sign denotes an inverse relationship (e.g., an increase in one leads to adecrease in the other). These relationships form feedback loops that drive the behavior of the model.

Arrows in the figure are labeled as positive or negative to reflect whether connected variables move in the same or opposite direction. The model includes several feedback loops that reflect real-world system behavior. For example, when NIH funding declines, success rate per application fall, leading to more grant submissions to remain funded (since funding is a requirement to cover salaries and conduct research), which in turn increases total application and further reduces grant application success rates—closing a reinforcing, vicious loop^16,32. Consequently, as more time is spent writing grants, researchers have less time available for conducting research, reducing overall knowledge production^33,34.

The model also includes balancing feedback loops, such as when increased academic position availability draws more researchers into the academic workforce, which then reduces the number of open positions and slows further inflow into the workforce. In addition, when perceived biomedical job prospects in academia and industry decline, fewer individuals enroll in doctoral programs³⁵. Another balancing feedback loop captures how the accumulation of commercial innovations contributes to an overall innovation development burden, reflecting how each new novel drug tends to be harder to develop due to increasing scientific, regulatory, and translational complexity. Although learning and efficiency gains can partially offset these challenges over time, historical patterns suggest the growing development burden has been the dominant force. As therapeutic frontiers advance, subsequent innovations face diminishing returns, higher efficacy standards, and greater biological uncertainty, making development slower and more complex^36,37. The model also links faculty availability to doctoral enrollment, recognizing that without sufficient faculty to teach and supervise, programs may reduce admissions. Additional feedback effects and loops are also presented in the figure. Together, the model mechanisms illustrate how shifts in funding and capacity can cascade across sectors and influence long-term innovation.

Projected outcomes

The primary outcome is the commercial innovations, measured through the annual number of FDA-approved novel drugs. These approvals refer specifically to new molecular entities and original biologics that contain active ingredients not previously approved in the U.S., in any form or combination, and that receive marketing authorization for the first time. Using the drug approval as a proxy for biomedical innovations, consistent with prior research on biomedical innovation^14,30, we report the cumulative shortfall in approved novel drugs by 2050 compared to the baseline.

To explain the projections of novel drug approval, we report several other outcomes, including early-career and established PIs funded by NIH, total doctoral students and postdoctoral researchers in science, engineering, and health, and NIH-supported fundamental knowledge output measured in published papers. In addition, we evaluated intensive-margin effects on PI funding continuity beyond changes in PI headcount. Specifically, we estimated cumulative NIH-funded PI-years lost, changes in the average funding duration per PI, and average funding-years lost between 2026 and 2050 among surviving PIs compared to baseline under each scenario.

To provide an illustrative sense of the potential magnitude, we translated the projected shortfall in each scenario into an approximate estimate of foregone economic value. We used an inflation-adjusted per-drug net social value, with an upward adjustment to account for the higher average value of novel approvals³⁸, yielding a range of roughly $5.1-$21.9 billion (2025 USD) per novel drug (see supplement for details). Future losses were discounted to present value using a 3% annual rate.

Additionally, we calculated scenario-specific cumulative elasticity measures linking changes in novel drug approvals to changes in NIH funding relative to baseline. These measures were intended as summary indicators of the proportional responsiveness of innovation outcomes to funding changes. Additional details and equations are provided in the supplementary materials.

Data inputs and model calibration

As presented in Table 1, the model includes 47 longitudinal datasets from 1995 to 2024, where available. Of these, 35 were used as calibration targets, such that unknown parameters were estimated by optimizing the model to closely replicate these historical patterns. The remaining 12 time series were used as external historical data inputs, meaning they were directly fed into the model (e.g., NIH congressional appropriations) as they reflect outside conditions or policy decisions that the model takes as given rather than trying to predict. See more details in the supplementary materials.

Table 1.

Time series data used in the model

Time series data	Sources	Years	Unit
A) Calibration data targets (35 time series)
NIH budget and Research Project Grants
(1) Annual competing research project grant budget	NIH RePORT43.	1995–2023	USD/year
(2) Annual noncompeting research project grant budget	NIH RePORT43.	1995–2023	USD/year
(3) Annual new grant approvals	NIH RePORT43.	1995–2024	Grant/year
(4) Active research project grants	NIH RePORT43.	1999–2024	Grant
(5) Annual research project grant applications	NIH RePORT43.	1995–2023	Grant/year
(6) Annual number of competing investigators	NIH RePORT43.	1995–2023	People/year
(7) Research project grant success rate	NIH RePORT43.	1995–2023	%
R&D Workforce in Academia and Industry
(8) Annual doctoral enrollment	NSF GSS39.	1995–2023	People/year
(9) Doctoral students in science, engineering, and health	NSF GSS39.	1995–2023	People
(10) Annual new doctoral attainment	NSF SED40.	1997–2023*	People/year
(11) New doctorate recipients with commitments to a postdoc position	NSF SED40.	1997–2023*	People/year
(12) New doctorate recipients with commitments to an academic position	NSF SED40.	1997–2023*	People/year
(13) New doctorate recipients with commitments to an industry position	NSF SED40.	1997–2023*	People/year
(14) New doctorate recipients with no definite commitments to academia or industry positions	NSF SED40.	1998-2023*	People/year
(15) Postdoctorates	NSF GSS39.	1997–2023*	People
(16) Total doctorate holders in R&D in academia	NSF SDR41.	1999–2023*	People
(17) Total doctorate holders in R&D in industry	NSF SDR41.	1999–2023*	People
(18) First-time PIs with active grants	NIH ExPORTER44.	1995–2024	People
(19) Established PIs with active grants	NIH ExPORTER44	1995–2024	People
(20) Prior PIs without active grants	NIH ExPORTER44.	1995–2024	People
(21) Annual new PIs awarded their first project grants	NIH ExPORTER44.	1995–2024	People/year
(22) Annual new PIs awarded multiple project grants	NIH ExPORTER44.	1995–2024	People/year
(23) Annual first-time PIs receiving another grant before the first one ends	NIH ExPORTER44.	1995–2024	People/year
(24) Annual first-time PI who has a funding gap after the first grant ends	NIH ExPORTER44.	1995–2024	People/year
(25) Annual established PI who has a funding gap after the grant ends	NIH ExPORTER44.	1995–2024	People/year
(26) Annual prior PIs without active grants receiving a new award	NIH ExPORTER44.	1995–2024	People/year
Biomedical Research and Innovation Pipeline
(27) Publications per NIH grant	NIH ExPORTER44.	1995–2021	Paper/grant
(28) Annual new NIH-funded fundamental research papers	Extracted from PubMed based on45.	1995–2024	Paper/year
(29) Annual new NIH-funded clinical trial registrations	ClinicalTrials.gov46.	2000–2024	Trial/year
(30) Active NIH clinical trials	ClinicalTrials.gov46.	2000–2024	Trial
(31) Annual new industry-funded clinical trial registrations	ClinicalTrials.gov46.	2000–2024	Trial/year
(32) Active industry clinical trials	ClinicalTrials.gov46.	2000–2024	Trial
(33) Annual novel molecular entity and biologics approvals	FDA CDER47,48.	1995–2024	Approval/year
(34) Cumulative novel molecular entity and biologics approvals	FDA CDER47,48.	1995–2024	Approvals
(35) Annual pharmaceutical industry R&D expenditure	PhRMA surveys49.	1995–2023	USD/year
B) External Data Inputs (12 time series)
NIH budget and research project grants
(1) NIH congressional appropriations	NIH Budget Office50.	1995–2025	USD/year
(2) Biomedical R&D Price Index	NIH Budget Office19.	1995–2030	%
(3) Percent of applications from established contacts PIs	NIH RePORT43.	1998–2024	%
(4) Percent first-time contact PI among annual competing PIs	NIH RePORT43.	1995–2024	%
R&D Workforce in Academia and Industry
(5) Population and projections for the 18-year-old cohort	Census51–55.	1995–2050	People
(6) Median time to doctoral degree	NSF SED40.	1995–2023*	Year
(7) Percent of primarily NIH-supported doctoral students	NSF GSS39.	1995–2023	%
(8) Percent of primarily NIH-supported postdoctoral appointees	NSF GSS39.	2000–2023	%
(9) Percent of established active PIs with single grants	NIH ExPORTER44.	1995–2024	%
Research and Innovation Pipeline
(10) Percent of NIH and industry trials to all clinical trials	ClinicalTrials.gov46.	1995–2023	%
(11) Percent NIH grant recipients in fundamental research	NIH Special Reports43.	2009–2022	%
(12) Percent NIH grants for fundamental research	NIH Special Reports43.	2009–2022	%

NIH National Institutes of Health, RePORT Research Portfolio Online Reporting Tools, NSF National Science Foundation, GSS Survey of Graduate Students and Postdoctorates in Science and Engineering, SDR Survey of Doctorate Recipients, SED Survey of Earned Doctorates, FDA Food and Drug Administration, CDER Center for Drug Evaluation and Research, PhRMA Pharmaceutical Research and Manufacturers of America, R&D research and development, PI Principal investigator.

*Data were not available every year in the given interval.

We primarily drew from the Research Portfolio Online Reporting Tools of the NIH to inform model parameters of the NIH budget and grant awards characteristics. Then, we derived workforce statistics (i.e., graduate students, postdoctoral appointees, and doctorate-holding individuals in R&D) using the surveys conducted by the National Science Foundation’s National Center for Science and Engineering Statistics. These include the Survey of Graduate Students and Postdoctorates in Science and Engineering (GSS), the Survey of Earned Doctorates (SED), and the Survey of Doctorate Recipients (SDR). The GSS is a comprehensive survey of U.S. academic institutions that grant research-based graduate degrees; institutions provide data with consistent, near-complete coverage on graduate students and postdoctoral researchers in science, engineering, and health fields³⁹. The SED is a census of all individuals earning research doctorates from U.S. institutions each year, with response rates typically exceeding 90%⁴⁰. The SDR surveys a representative sample of U.S.-trained doctorate recipients exceeding a hundred thousand individuals to track career trajectories and workforce outcomes, with response rates generally above 70%⁴¹. Together, these surveys offer nationally representative data on the U.S. scientific workforce with direct relevance to biomedical research fields.

We also used population estimates and projections by the U.S. Census Bureau to generate age-cohort-based estimates of enrollment potential in the academic pipeline. In addition, we used NIH ExPORTER to extract detailed project-level data, including PI identifiers, which allowed us to construct cohorts of PIs over time. We linked NIH grant identifiers to associated PubMed publications to analyze research outputs over time. Finally, we used various data sources to estimate parameters for the biomedical innovation pipeline, including active clinical trials registered on ClinicalTrials.gov and new drug approvals from the FDA. Further details are presented in the supplementary materials.

Model assessment and reproducibility

We conducted several tests, including extreme conditions testing, model boundary analysis, and comparison of simulated trends to historical data using mean absolute percentage error⁴². For sensitivity analysis, using Markov Chain Monte Carlo, we ran 2 million simulations, discarding the first 1 million as burn-in, and then retained 1 million samples from the joint posterior distribution of the model parameters. From these, we randomly drew 10,000 samples to estimate model parameters and report 95% projection intervals for each outcome and scenario.

The model was created in Vensim version 10. Model assumptions and details are documented in the supplementary materials. As this study relied exclusively on publicly available, aggregated data, approval from the Institutional Review Board was not needed.

Author contributions

Conceptualization: Z.H. and M.S.J. Methodology: Z.H., H.L., and M.S.J. Investigation: Z.H., H.L., and M.S.J. Visualization: Z.H. Supervision: M.S.J. Writing – original draft: Z.H. and M.S.J. Writing – review & editing: Z.H., H.L., and M.S.J.

Data availability

All data are available in the main text or the supplementary materials.

Code availability

The repository containing the code and model files is provided in the Supplementary Materials.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s44401-026-00120-0.