Technical Report 49: Allocating VA Research Funds: Can Allocations Reflect Veteran Burden?
Todd H. Wagner, PhD, Kristina Cordasco, MD, MPH, MSHS
Suggested CitationWagner TH, Cordasco K. Allocating VA Research Funds: Can Allocations Reflect Veteran Burden? Technical Report 49. Health Economics Resource Center, VA Palo Alto Health Care System, U.S. Department of Veterans Affairs. June 2026.
1. Introduction
The US Department of Veterans Affairs (VA) runs a large intramural research program, with annual funding approved by Congress, which has directed VA’s research to focus on service-related conditions as well a broad range of biomedical, mental health, rehabilitative, and prosthetic topics that affect Veterans. This raises the question of whether VA could use its own data to identify how well its research allocations are aligned with Veterans’ burden of disease. Therefore, we asked whether cost-of-illness (COI) methods applied to VA administrative data can reliably identify misalignments between disease burden and research investments.
COI methods were developed in the 1960s, motivated by a desire to help policy makers allocate health care resources in proportion to disease burden.1 The method uses population burden information, such as disease prevalence, in combination with medical and non-medical expenditures to estimate the overall cost of illness.1 There have been a number of methodological advances over time, and contemporary studies frequently use regression models to estimate the incremental costs associated with a disease.2 This approach has attracted considerable methodological criticism, while others have raised theoretical concerns, noting that COIs are descriptive and are not suited to guiding funding allocation decisions.3
The objective of this study was to examine whether analyzing VA administrative data could help guide resource allocations for VA research. Using VA administrative data from Fiscal Year (FY) 2024, we replicated a prior study that estimated the incremental VA spending associated with common chronic illnesses.4 We then estimated disease prevalence and subsequently multiplied the incremental costs by the condition prevalence to estimate disease burden.
Any method that guides resource allocation must be robust. We replicated this approach using different analytical methods to test for robustness. We used three different regression specifications: ordinary least squares (OLS), a generalized linear model (GLM) with a gamma distribution and log link, and a square root transformation OLS retransformed using a smearing estimator. We also replicated the approach using different measures of morbidity. We present the results across these different specifications. Finally, we conclude with a discussion on the inherent challenges of allocating resources for research and why analyses of past spending may not lead to more optimal research allocations.
2. Identifying Diseases and Conditions
Using diagnostic codes, we measured the FY 2024 prevalence of 86 conditions and diseases as defined in version 24 of CMS’s Hierarchical Condition Categories along with the 63 conditions in the Psychiatric Categories of Care among VA patients. As these 149 conditions include some with substantial clinical overlap, we grouped them into 23 discrete sets of condition groups based on their clinical domains. Table 1 shows the prevalence of these 23 condition groups among Veterans, ranked by prevalence. We used information on each individual condition group, using dummy variables in the regression model.
Table 1. Prevalence of 23 Condition Groups in FY24
Another way to measure burden of disease is to count condition groups. Table 2 shows the number of Veterans with each of these 23 condition groups; 21% of Veterans did not have any of these conditions, but among those who did, the median number of condition groups was 3.
Table 2. Frequency Count of VA Patients by Number of Chronic Condition Groups in FY24
Finally, we measured prevalence of conditions using medication data. VA has a generous and comprehensive pharmacy benefits package, and we analyzed medications as reported in the Managerial Cost Accounting pharmacy dataset. To link medications to condition groups, we measured the number of medications a person was taking across the 28 different VA drug classes as classified by PBM (see Appendix 1). Hence, a person categorized as 3 indicates that they take medications in three different drug classes. The distribution of Veterans by the number of drug classes is shown in Table 3.
Table 3. Frequency Count of VA Patients with Medications by Number of Drug Classes in FY24
3. Marginal Costs Per Condition Type
We computed the total VA spending per Veteran in FY 2024. This included VA-provided and VA-paid care. Details including the average and median spending per person are shown in Table 4. The average spending per Veteran was $20,594 (SD $54,194) and the median spending was $6,684, indicative of a right-skewed distribution.
Table 4. Annual Person Level Costs for VA-Provided and VA-Paid Care in FY24 (n=6,833,735)
Table 5 shows the marginal cost, prevalence and total cost associated with each condition group. This was estimated with an OLS model with a square root transformation. We selected the square root transformation based on the modified Park test, though results remained sensitive to model specification (see limitations).5 Table 6 shows the estimated burden of chronic conditions across the three regression models. The estimated total costs changed considerably (see Table 6), whereas the ordinal rankings provided more stability.
Table 5. Marginal Costs Associated with Each Condition Group
Table 6. How Overall Spending by Condition Group varied by Statistical Model
4. Different Measures of Disease and Condition
How do you allocate funding when chronic conditions cluster together? To show the correlation between morbidity and spending, we cross-tabulated the count of condition groups and count of drug classes. For each cell, we estimated the median spending per person by count of condition groups and count of drugs used across different drug classes. We then multiplied the median spending by the prevalence rate for that cell. Of the 6,833,735 Veterans treated in FY 2024, 168,316 had diagnoses in 1 condition group and took medications in one drug class, while 92,390 had diagnoses in 2 condition groups and took drugs in two drug classes. So, while the count of condition groups and count of drug classes are correlated, they are different.
We created a contour map of prevalence-weighted spending (see Figure 1). The Y axis shows the number of drug classes and the X axis shows the number of conditions. Cooler colors reflect areas that have the combination of fewer Veterans and lower spending. Warmer colors represent areas of more Veterans and more spending. The contour map would suggest that VA should focus its resources on studies in the warmer colors. However, substantial effort has tried to target the red group—the top 5% most expensive Veterans.6,7 Those studies have failed to bend the cost curve, in part because this group changes from year to year and many in that group die.6,7 This approach assumes current prescribing patterns reflect disease burden rather than prescribing culture or formulary constraints.
Figure 1. Contour map of prevalence-weighted spending
5. Discussion
It is feasible to use statistical models to estimate the total cost of conditions among VA users. However, we found these models would be unreliable for allocating resources for 5 reasons: (1) different statistical models produced varying results, (2) the reference group for these analyses is not well defined, (3) coding is not always accurate, (4) historical investment in select conditions will impact their costs and perceived value, and (5) COI analyses measures spending rather than health. Alternate approaches evaluating the prevalence of chronic conditions or number of chronic conditions and drug classes are more reliable ways to understand morbidity. Additional details on these analyses are provided in the sections, “Analyses Estimating the Costs of Chronic Conditions” and “Alternative Approaches.”
The second problem is that these marginal estimates assume that there is a defined reference group and that the condition is exogenous. However, the reference group is hard to define. For example, when we estimate the marginal effect of neurologic conditions, it is compared to people who do not have neurologic conditions, holding other conditions constant. Yet there are many other conditions we are not measuring that are correlated with the error term and this is biasing the estimate of neurologic conditions. Put simply, we did not randomize a patient to have a neurologic problem, so the statistical estimate of neurologic conditions is endogenous and biased.
Third, these analyses depend on coding diagnostic coding during VA medical encounters and diagnostic coding is not always accurate or specific to a single condition. For example, there are diagnoses for tobacco use and homelessness, but these are used less frequently. Thus, these estimates underestimate the cost of these conditions. Note also that the prevalence for dementia is 3.27%, using this method. That is 3x lower than the current best estimates in the published literature.8 Similarly, the prevalence of pain appears to be very low. One explanation is that pain is subsumed in other categories, such as neurology or oncology, but again it raises concerns about this approach. In summary, the magnitude of coding discrepancies suggests these data may be fundamentally unsuitable for allocation decisions; conditions with systematic undercoding would be deprioritized not due to low burden or costs but due to documentation practices.
Fourth, we tend to find problems today where we invested a lot of money yesterday. We fully expect that cardiovascular problems will be high on this list because VA (and the US) heavily invests in cardiology. Sarcopenia is not on this list though it is quite common (prevalence 15-30% of older adults). Thus, not being on this list should not mean that frailty is not worthy of research. For instance, direct-acting antiviral treatments transformed infectious disease care, and retrospective burden analyses would have missed this prospective opportunity to help implement treatment for Veterans with hepatitis C.
Fifth, COI analyses fundamentally measure spending rather than health. Health care, as an industry, is focused on producing health and well-being. Yet we do not, as a science or industry, adequately measure health outcomes. Cutler and others have argued that while it may be helpful to identify high-cost, complex patients, resource allocation decisions should ultimately be guided by where we can improve health, not simply where we spend the most money.9 A COI-based allocation approach risks perpetuating investment in areas with high historical spending rather than directing resources toward conditions where research could yield the greatest health improvements. For example, a relatively inexpensive condition with poor treatment options and substantial impact on quality of life might warrant more research investment than an expensive condition with well-established, effective treatments. By focusing on costs rather than potential health gains, COI methods may systematically undervalue research opportunities with the greatest potential to improve Veteran health and well-being.
Finally, VA's research mission differs from population-wide health research. Veterans experience unique conditions from military service (blast injuries, burn pits, Agent Orange exposure, military sexual trauma) that would not appear as 'high burden' in these analyses due to relatively small numbers, but represent areas where VA has both a special obligation and comparative advantage. A burden-based allocation would systematically defund research most relevant to Veterans' service-connected conditions.
6. Implications for VA Research
This work has several implications for VA research. First, burden data should be used as a screening tool. Grant applicants could be required to provide prevalence and cost context for their proposed research, but funding decisions should ultimately be based on scientific merit and expected impact. Prevalence or cost-of-illness estimates should not determine funding levels.
Second, the VA should prioritize research that intersections with multiple co-morbidities. As shown in Figure 1, more disease complexity is associated with higher costs. The VA should also discourage studies from excluding Veterans with complex needs, even if those studies are more expensive to conduct.
Third, VA could consider commissioning a prospective value-of-information (VOI) analysis rather than relying on retrospective burden estimates. The VOI approach would ask where research investments have the highest expected value given current uncertainty. Artificial intelligence could potentially help develop and maintain this effort.
Fourth, the VA should continue to monitor its research portfolio. There is value in maintaining a diverse portfolio that includes high-risk/high-reward projects, VA-specific conditions, and understudied populations.
Finally, regardless of the disease area selected, VA should invest in rigorous methods, and it should not shy away from replication studies. Well-conducted replication studies can be highly valuable and should not be viewed as waste or inefficiency.
7. References
- Rice DP. Estimating the Cost of Illness. U.S. Department of Health, Education and Welfare, Rockville, MD.; 1966.
- Onukwugha E, McRae J, Kravetz A, Varga S, Khairnar R, Mullins CD. Cost-of-Illness Studies: An Updated Review of Current Methods. PharmacoEconomics. 2016;34(1):43-58. doi:10.1007/s40273-015-0325-4.
- Tarricone R. Cost-of-illness analysis. What room in health economics? Health Policy. 2006;77(1):51-63. doi:10.1016/j.healthpol.2005.07.016.
- Yoon J, Scott JY, Phibbs CS, Wagner TH. Recent trends in Veterans Affairs chronic condition spending. Popul Health Manag. 2011;14(6):293-298. doi:10.1089/pop.2010.0079.
- Manning WG, Mullahy J. Estimating log models: to transform or not to transform? J Health Econ. 2001;20(4):461-94.
- Zulman DM, Chee CP, Ezeji-Okoye SC, et al. Effect of an intensive outpatient program to augment primary care for high-need Veterans Affairs patients: a randomized clinical trial. JAMA Intern Med. 2017;177(2):166-175.
- Yoon J, Chee CP, Su P, Almenoff P, Zulman DM, Wagner TH. Persistence of High Health Care Costs among VA Patients. Health Serv Res. 2018;53(5):3898-3916. doi:10.1111/1475-6773.12989.
- Williamson V, Stevelink SA, Greenberg K, Greenberg N. Prevalence of mental health disorders in elderly US military veterans: a meta-analysis and systematic review. Am J Geriatr Psychiatry. 2018;26(5):534-545.
- Cutler DM, Richardson E, Keeler TE, Staiger D. Measuring the health of the US population. Brook Pap Econ Act Microecon. 1997;1997:217-282.
AcknowledgementsThis work was supported by Merit Review Award Numbers RCS-17-154 and EBP-22-108 from the United States (U.S.) Department of Veterans Affairs HSR and QUERI Service. The funders of the study had no role in the study design, data collection, data analysis, data interpretation, or writing of the report. The views expressed in this article are those of the authors and do not represent the views of the U.S. Department of Veterans Affairs or the U.S. Government.
Last Updated Date: June 8, 2026
