Variation in Prescription Use and Spending for Lipid-Lowering and Diabetes Medications in the Veterans Affairs Healthcare System

October 15, 2010
Walid F. Gellad, MD, MPH

,
Chester B. Good, MD, MPH

,
John C. Lowe, RPh, MBA

,
Julie M. Donohue, PhD

Volume 16, Issue 10

Substantial variation in prescription spending and use of brand-name drugs exists across the VA healthcare system, with no apparent relationship to quality of care.

Objectives:

To examine variation in outpatient prescription use and spending for hyperlipidemia and diabetes mellitus in the Veterans Affairs Healthcare System (VA) and its association with quality measures for these conditions.

Study Design:

Cross-sectional.

Methods:

We compared outpatient prescription use, spending, and quality of care across 135 VA medical centers (VAMCs) in fiscal year 2008, including 2.3 million patients dispensed lipid-lowering medications and 981,031 patients dispensed diabetes medications. At each facility, we calculated VAMC-level cost per patient for these medications, the proportion of patients taking brand-name drugs, and Healthcare Effectiveness Data and Information Set (HEDIS) scores for hyperlipidemia (low-density lipoprotein cholesterol level <100 mg/dL) and for diabetes (glycosylated hemoglobin level >9% or not measured).

Results:

The median cost per patient for lipidlowering agents in fiscal year 2008 was $49.60 and varied from $39.68 in the least expensive quartile of VAMCs to $69.57 in the most expensive quartile (P < .001). For diabetes agents, the median cost per patient was $158.34 and varied from $123.34 in the least expensive quartile to $198.31 in the most expensive quartile (P < .001). The proportion of patients dispensed brand-name oral drugs among these classes in the most expensive quartile of VAMCs was twice that in the least expensive quartile (P < .001). There was no correlation between VAMC-level prescription spending and performance on HEDIS measures for lipid-lowering drugs (r = 0.12 and r = 0.07) or for diabetes agents (r = -0.10).

Conclusions:

Despite the existence of a closely managed formulary, significant variation in prescription spending and use of brand-name drugs exists in the VA. Although we could not explicitly risk-adjust, there appears to be no relationship between prescription spending and quality of care.

(Am J Manag Care. 2010;16(10):741-750)

Despite the existence of a closely managed formulary, significant variation in prescription spending and use of brand-name drugs exists across the Veterans Affairs Healthcare System (VA). Although we could not explicitly risk-adjust, there was no apparent relationship between prescription spending and quality of care.

  • Although practice pattern variation is well-documented in Medicare and other settings, few studies have examined variation in prescribing.

  • Identifying and eliminating unnecessary variation in prescribing should be part of all quality and efficiency improvement programs.

  • This study has important implications for drug spending outside the VA, where the use of closed formularies is less common, rates of generic drug utilization are lower, and overall cost implications are even more profound.

Geographic variation in healthcare utilization and spending in the United States is well-documented.1,2 After accounting for differences in prices and disease prevalence, researchers reported 2-fold variation in Medicare spending in 2006 across hospital referral regions, largely owing to differences in the volume of healthcare services delivered.3 Studies4,5 to date have shown that areas with higher Medicare spending perform no better than low-spending areas on process quality-of-care and health outcome measures and may, in fact, perform worse. Areas of healthcare that exhibit a high degree of regional variation in use and spending tend to be those in which clinical evidence is weak and there is a lack of consensus-based treatment guidelines.6

One area in which spending has increased dramatically in the past 15 years is prescription drugs.7 Despite the debate over the value of this increased spending,8 few studies9-16 have examined regional variation in prescription drug use. Most of these studies were conducted with data from the late 1990s, were focused on variation within limited geographic regions, or did not examine the association between prescription drug spending and health outcomes.

We chose the Veterans Affairs Healthcare System (VA) as the setting in which to study variation in outpatient prescription drug use and spending, and we focus herein on facility-level variation. In the VA, we can isolate differences in prescription use independent of health system factors known to affect use such as insurance coverage for prescription drugs and the existence of formularies. In addition, variation in drug spending will not be due to differences in price, as acquisition costs for medications generally do not vary between regions in the VA. Our objectives for this research were (1) to assess the level of variation in outpatient prescription use and spending in the VA for 2 widely used therapeutic categories (lipid-lowering drugs and diabetes medications) and (2) to determine if VA medical centers (VAMCs) that spent more per patient for these medications performed better on routinely collected quality measures for hyperlipidemia and diabetes.

METHODS

Setting and Data Sources

The VA is one of the largest integrated healthcare systems in the United States, with more than 8 million veterans enrolled and more than 5 million receiving care.17,18 In fiscal year 2008 (FY2008), the VA spent $3.1 billion on outpatient pharmaceuticals. Between 1996 and 2009, the VA transitioned from a decentralized system of more than 170 individual drug formularies to a single VA National Formulary.19,20 Healthcare providers have access to all US Food and Drug Administration— approved medications but must request and cite justification for nonformulary drugs before their use.

The VA Pharmacy Benefits Management Services database contains all individual VA outpatient pharmacy transactions in the United States. We used aggregate outpatient pharmacy data for FY2008 to study prescription use and spending across 135 VAMCs nationwide. The database includes aggregate VAMC-level data on the cost for each dispensed product (acquisition cost to the VA), the number of unique patients receiving prescriptions, and the number of prescriptions and dosage units dispensed. We also obtained VAMC-level quality measures from the VA Office of Quality and Performance.

Population and Drugs Studied

We aggregated total prescription spending for lipid-lowering and diabetes medications (including insulin) at the VAMC level. Each VAMC may include several hospitals and community-based clinics. For example, the VA Pittsburgh (Pennsylvania) Healthcare System Veterans Integrated Service Network (VISN) 4 includes 3 different hospitals or divisions. VISN 15 (Missouri, Kansas, and parts of Illinois) aggregates the prescription spending of its 6 VAMCs into 2 groups (rather than at the facility level, as measured in other regions of the country). We maintained the designation of these 2 groups as VAMCs for the purposes of the analysis of variation in spending.

Table 1

We chose to study lipid-lowering and diabetes medications, which are commonly used drugs available in brand-name and generic forms and treat conditions with objectively measured quality outcomes (). For lipid-lowering medications, we excluded bile acid sequestrants because of the likelihood that these agents are used for conditions other than hyperlipidemia (eg, cholestyramine resin for chronic diarrhea or pruritus). We also excluded niacin products because the VA recommends the use of brand-name Niaspan® (long-acting niacin) over “nutritional” generics, and these agents are clinically used primarily for high-density lipoprotein cholesterol and triglyceride control rather than for low-density lipoprotein cholesterol (LDL-C) control.

Outcome Variables

Drug Spending. Our main drug spending variable was cost per patient in FY2008 for lipid-lowering and diabetes medications. We divided the total amount spent by each VAMC

for these medication classes (ie, acquisition cost for all dispensed units or pills in the class) by the number of unique patients who were dispensed these medications. For example, if a VAMC dispensed 100,000 pills of simvastatin at an acquisition cost of $0.08 per pill, then the total amount spent by that VAMC would be $8000. Because VA acquisition cost varies little across VAMCs and because VA Pharmacy Benefits Management Services data include only drug costs and not pharmacy labor, supply, or overhead costs, differences in spending among VAMCs should reflect differences in prescribing practices (eg, type of medicine and intensity of use) rather than differences in drug price or labor costs.

Our primary interest was estimating the extent of variation in spending due to prescribing differences as opposed to differences in adherence, which would affect the number of prescriptions filled and thus total cost per patient. Therefore, we also calculated a second VAMC-level spending variable, namely, the mean cost per 30-day prescription. The mean cost per prescription at a VAMC depends on what drugs are filled (ie, medication choice) rather than on the total days’ supply dispensed during the year. The VAMCs using more expensive brand-name agents will have a higher mean cost per prescription than facilities using fewer brand-name agents, regardless of whether patients miss certain doses during the year or are taking 1 or 2 agents to control their disease. We compare our 2 spending measures and assume that if they are highly correlated, VAMC-level spending differences are not driven by differences in adherence.

Quality of Care. The VA collects Healthcare Effectiveness Data and Information Set (HEDIS) quality measures at each facility using independent reviewers from the External Peer Review Program, who examine cross-sectional random samples of medical records.21 We used facility-level HEDIS measures obtained from the VA Office of Quality and Performance to measure quality of care for hyperlipidemia and for diabetes.22 Lipid-lowering and diabetes therapy is associated with improvements in these intermediate measures.23-25 Evidence about the association between surrogate measures and outcomes such as mortality is mixed26; however, these measures are accepted quality metrics and are used to evaluate the performance of physicians, institutions, and health plans.

HEDIS scores were obtained for LDL-C levels less than 100 mg/dL. Data were abstracted on the percentage of patients aged 18 to 75 years with a diagnosis of diabetes or with ischemic heart disease who had undergone a full lipid panel during the past year and whose most recent LDL-C level was less than 100 mg/dL.

HEDIS scores were also obtained for glycosylated hemoglobin (A1C) levels exceeding 9% or not measured (ie, poor control) in the past year. The External Peer Review Program abstracts data on the percentage of patients aged 18 to 75 years with a diagnosis of diabetes having an A1C level exceeding 9% or for whom A1C testing was not performed. This measure is the most up-to-date clinical measure of diabetes control used and reported by the VA Office of Quality and Performance. For ease of comparison across the 2 conditions, we reverse-scored the A1C measurement (1 minus the HEDIS score), so that a higher score represented better quality of care.

Independent Variables

We expected drug spending to be associated with the level of brand-name drug use. For lipid-lowering agents, we measured the percentage of patients on lipid-lowering therapy at each facility who were receiving brand-name drugs. We classified rosuvastatin calcium (Crestor®), atorvastatin calcium (Lipitor®), fluvastatin sodium (Lescol®), ezetimibe (Zetia®), and ezetimibe-simvastatin (Vytorin®), which are nonformulary in the VA, as brand-name agents and examined variation in prescribing of these agents. For diabetes medications, we measured the percentage of patients taking diabetes drugs who received thiazolidinediones (ie, pioglitazone hydrochloride [Actos®] or rosiglitazone maleate [Avandia®]), which are a brand-name—only class of oral hypoglycemics.

To examine whether patient-level and facility-level differences might explain variation in outpatient prescription use and spending, we obtained facility-level data on organization characteristics from VA administrative sources such as the percentage of patients older than 65 years, the total number of outpatient visits, and the percentage of uninsured patients seen as outpatients at each VAMC in FY2008. In addition, we stratified facilities by “complexity” rating (1a or 1b on the VA Allocation Resource Center Complexity Scale). The VA rates each VAMC on this 5-level scale, with high-complexity facilities seeing the largest volume of patients and having the highest patient risk. Patient risk is based on all VA patient diagnoses and uses the same diagnostic cost group risk scores as Medicare. In addition, high-complexity facilities employ a greater relative number and breadth of specialists, offer the highest level of intensive care unit healthcare (based on the type of services provided and the availability of subspecialty services), train more medical residents, and engage in more research activities. These variables are combined in a weighted mean using established methods to create these ratings.27 More complex VAMCs have been shown to be more likely to have year-round emergency department access, academic physicians, formal residency training in primary care, local clinical champions, and designated nurses for quality improvement.28 Facility characteristics included in the VA complexity ratings have previously been used to measure facility complexity in VA research.29

Statistical Analysis

We categorized VAMCs into spending quartiles and used Kruskal-Wallis tests to assess differences in the median cost per patient across each quartile for lipid-lowering and diabetes agents. We also used Kruskal-Wallis tests and c2 tests to assess differences in VAMC characteristics across quartiles. We used analysis of variance to test for differences in the mean proportion of patients using brand-name nonformulary drugs across quartiles. We used Spearman rank correlation coefficients and scatterplots to examine the relationship between cost per patient for lipid-lowering and diabetes agents at each VAMC. We also used Spearman rank correlation coefficients and scatterplots to compare spending at each VAMC for lipidlowering and diabetes agents with HEDIS quality scores for hyperlipidemia and diabetes.

To better characterize the independent effect of facility-level cost per patient on quality outcomes, we used logistic regression analysis to model the probability of being in the highest quartile of quality for lipid-lowering and diabetes medications separately. Our primary independent variable was cost per patient at the VAMC level, used as a continuous measure, and we included all facility characteristics aforedescribed as additional covariates in the analysis. We used commercially available statistical software (SAS, version 9.2; SAS Institute, Cary, NC) for all analyses.

RESULTS

Overall, 135 VAMCs were included in the analysis, representing 2.3 million veterans who were dispensed lipid-lowering medications and 981,031 veterans who were dispensed diabetes medications in FY2008. Table 1 lists each of the medications studied, along with the number of unique pharmacy users, the mean unit acquisition cost of the medications, and the brand-name versus generic classification of the medications. On average, generic agents were available at 10% to 20% of the acquisition cost of brandname drugs.

Table 2

lists additional characteristics of VAMCs overall and by quartile of drug spending. For lipid-lowering agents prescribed in FY2008, VAMCs in the most expensive quartile had slightly fewer outpatient visits and veterans taking the medications (P > .05), but the proportion of patients 65 years and older was the same as in the least expensive quartile. For diabetes agents, VAMCs in the most expensive quartile had fewer patients using insulin compared with VAMCs in the least expensive quartile (P = .049) and had a lower proportion of patients overall receiving diabetes medications (20% vs 21%), although the difference was not statistically significant (P = .38).

For patients taking lipid-lowering agents in FY2008, the median cost per patient per year was $49.60 (interquartile range, $42.80-$61.00) and ranged across VAMCs from $23.50 to $125.00 (Table 2). The VAMCs in the most expensive quartile had a median cost per patient of $69.57, which was 75% higher than that of VAMCs in the least expensive quartile ($39.68). The median cost per patient per year for diabetes agents was $158.34 (interquartile range, $136.23-$182.13) and ranged across VAMCs from $106.95 to $306.58. The VAMCs in the most expensive quartile had a median cost per patient of $198.31, which was 60% higher than that of VAMCs in the least expensive quartile ($123.34). Calculations using cost per 30-day prescription showed similar variation, and the 2 measures were highly correlated (r >0.95). The percentage of patients taking brandname oral medications was greatest among VAMCs with the highest prescription spending for lipid-lowering agents and for diabetes medications (P <.001), with percentages in the highest quartile more than twice the percentage in the lowest quartile for lipid-lowering drugs and almost twice the percentage in the lowest quartile for oral diabetes medications.

Figure 1

There was a moderate correlation between VAMC-level cost per patient for lipid-lowering drugs and cost per patient for diabetes agents (r = 0.41, P <.001) (). Almost half (47%) of VAMCs in the highest quartile for lipid-lowering drugs were also in the highest quartile for diabetes drugs (data not shown).

Figure 2

There was no statistically significant correlation between the mean prescription spending and performance on disease-specific HEDIS measures across VAMCs for lipid-lowering drugs or for diabetes agents (). The correlation coefficient for patients with diabetes having LDL-C levels less than 100 mg/dL was r = 0.12, P = .16. The coefficient for patients with heart disease having LDL-C levels less than 100 mg/dL was r = 0.07, P = .42. The coefficient for patients with diabetes having A1C levels exceeding 9% was r = −0.10, P = .27.

Figure 3

When we limited the analysis to VAMCs rated as high-complexity facilities, the correlation between prescription spending and quality approached statistical significance for lipid-lowering drugs, although it remained weak (r =0.28, P = .06 for patients with diabetes having LDL-C levels <100 mg/dL; r = 0.12, P = .40 for patients with heart disease having LDL-C levels <100 mg/dL). There remained no relationship between prescription spending and performance for diabetes measures (r = 0.10, P = .49 for A1C levels >9%) (). Similarly, no correlation was found between cost per 30-day prescription and quality of care (data not shown).

In logistic regression models controlling for facility characteristics (summarized in Table 2), there remained no relationship between prescription spending and highest-quality quartile status for diabetes drugs (odds ratio [OR], 1.00; 95% confidence interval [CI], 0.98-1.01; P = .61) or for lipid-lowering drugs (OR, 1.02; 95% CI, 0.99-1.04; P = .19 for patients with diabetes having LDL-C levels <100 mg/dL; OR, 1.01; 95% CI, 0.99-1.04; P = .27 for patients with heart disease having LDL-C levels <100 mg/dL). Complete results of the multivariate regression analysis are included in an eAppendix (available at www.ajmc.com).

DISCUSSION

To our knowledge, this is the first national study of variation in outpatient prescription spending among adults in the VA and is the first study to assess variation of this kind in a large nationwide sample. We found widespread variation in yearly drug spending for 2 commonly used categories of prescription drugs. This variation in outpatient prescription spending and use exists in the VA despite the existence of a closely managed formulary with uniform drug prices, a commitment to a uniform prescription benefit, and clinical guidance for the appropriate use of nonformulary medications.20 Moreover, VAMCs with higher spending per patient for these medication classes performed no better on quality measures.

Our findings about variation in VA prescribing are consistent with prior investigations in the VA that addressed variation in outpatient prescribing, although previous studies12,14,15

were limited to small segments of the VA or examined variation only across VISNs. Gao and Campbell13 used VA data from 2003 to look at trends in prescription costs and touched briefly on regional variation in prescription spending, although they did not separate inpatient and outpatient spendning and did not assess the relationship between spending and quality. In the only other study we are aware of that examines facility-level variation in outpatient prescription use in the entire VA, Aspinall et al9 reported regional variation in rates of antibiotic prescribing for veterans with upper respiratory tract infections. Health outcomes associated with medication treatment were not examined.

Variation in prescription drug spending within the VA cannot be attributed to differences in prices paid for drugs. Prices (ie, unit costs) are generally negotiated at the national level and vary only slightly across VAMCs, unlike prescription prices outside the VA, which vary markedly across institutions and across classes of payer (eg, uninsured, Medicaid, and private insurer).30-32 Prescription prices can also vary over short intervals (even days) in the private sector, and true prices are difficult to obtain because of undisclosed manufacturer rebates, which would affect the measurement of prescription spending variation outside the VA.

Nor can these differences in drug use and spending be attributed to differences in VAMC formularies. We found significant differences across VAMCs in the use of brand-name nonformulary agents despite the existence of a national formulary and guidance for the use of nonformulary drugs. One explanation for this variation may be that, although the formulary is the same across the VA, the procedures for adjudicating nonformulary requests (the process by which providers ask to use nonformulary agents) may vary by VAMC. We found that VA facilities that spent more per patient for one drug class were more likely to spend more on the other drug class as well. The VA should consider disseminating throughout the system best practices in formulary management from VAMCs with low pharmacy costs that perform well on quality-of-care measures.

Patient-level factors could explain variation in prescribing if patients at certain facilities are more or less likely to require nonformulary medications because of more severe comorbid illness or because of differences in risk of adverse effects. When we repeated our analysis using only VAMCs with the highest-complexity rating, our results were similar; however, our use of aggregate data means that we cannot control for individual-level factors that might affect the need for more expensive agents. Differences in facility characteristics across quartiles of spending do not seem large enough to explain the level of variation that we observed. In fact, the lower percentage of patients using insulin in the most expensive quartile of VAMCs argues against the idea that the presence of sicker patients is leading to higher costs at these VAMCs. In addition, regression analyses adjusting for facility-level factors did not change the relationship (or lack thereof) between spending and quality.

Provider prescribing patterns differ by specialty, practice setting, level of training, age, and academic affiliation, and these factors could explain some of the documented variation. Studies show that much of the variation in medication choice and choice of generic versus brand-name drugs is related to unobserved physician factors.33 For example, it is unclear to what extent pharmaceutical promotion to physicians influences the prescribing behavior of VA physicians. The VA ethics rules prohibit acceptance of significant gifts from the pharmaceutical industry, and generally VAs do not accept free samples, although there is variation across VAMCs in the level of access to pharmaceutical sales representatives. Of note, the VA has no authority to limit interactions with industry representatives for providers who have dual affiliations with non-VA organizations.

In addition, because of the fluid manner in which VA providers cross in and out of the VA (especially in academic centers), local area practice patterns (the same ones that influence private practice physicians) may affect VA physicians.34 VA physician prescribing patterns may in fact be similar to those of non-VA physicians in the same region. In addition, some VA patients bring prescriptions from private physicians to the VA and ask VA providers to “re-prescribe” them, suggesting that the effect of local area practice patterns on patients may be as important to the VA as the effect on providers.

Our work has important limitations. First, our data are aggregated at the facility level, and as already discussed, we cannot risk adjust each facility’s prescription use to understand how differences in patient or provider factors may affect facility-level variation. In addition, there can be legitimate reasons why patients require brand-name drugs in the VA, and our data do not allow us to focus on the appropriateness of use. Second, we cannot account for non-VA prescription use in our analysis, although we believe that veterans who fill rescriptions in the VA (and thus are included in our sample) often do so because of the lower copayment for these medications and would be unlikely to fill the same prescriptions outside the VA. Some veterans fill prescriptions outside the VA using the $4 generic programs from large pharmacy retailers; however, the VA does not have data on the use of these programs by veterans. We believe that the overall effect of these low-cost generics for veterans who are filling prescriptions in the VA was likely small in FY2008. Third, variation in spending across VAMCs might be related to differences in patient refill adherence. Given that the degree of variation in cost per patient was similar to that in cost per prescription, we are confident that differences in adherence are not driving our findings. Fourth, we believe that HEDIS measures for diabetes and hyperlipidemia are important facility-level measures of quality of care and thus use them in our analysis; however, there are factors aside from prescription use that would affect these quality scores, and the use of 2 quality measures may not adequately assess overall quality of care for patients with diabetes or hyperlipidemia.

In conclusion, the VA has been successful in providing high-quality care to veterans35-37 while maintaining low rates of growth in prescription spending.38 However, substantial variation remains in prescription drug spending across VAMCs. Understanding the causal mechanisms underlying this variation can help target prescription quality improvement programs. Our findings also have important implications for prescription drug spending outside the VA, where the use of closed formularies is much less common, rates of generic drug utilization are lower, and variation may be even greater in magnitude and overall cost implications even more profound.

Acknowledgments

We thank Andrew Henriksen, RPh, pharmacoeconomic data manager for the VA Pharmacy Benefits Management Services, for his help with compiling the prescription data. We also thank Michael Valentino, RPh, MHSA, and Sherrie Aspinall, PharmD, MSc, for their helpful comments on early versions of the manuscript.

Author Affiliations: From the Center for Health Equity Research and Promotion (WFG) and Center for Medication Safety (CBG), VA Pittsburgh Healthcare System, Pittsburgh, PA; Department of Medicine (WFG, CBG), Department of Pharmacy (JCL), and Department of Health Policy and Management (JMD), University of Pittsburgh, Pittsburgh, PA; RAND Corporation (WFG), Pittsburgh, PA; and Pharmacy Benefits Management Services (JCL), Veterans Health Administration, Washington, DC.

Funding Source: This study is the result of work supported by resources from the VA Pittsburgh Healthcare System. Dr Donohue was supported by grant KL2 RR024154 from the National Center for Research Resources, a component of the National Institutes of Health Roadmap for Medical Research.

Author Disclosures: The authors (WFG, CBG, JCL, JMD) report no relationship or financial interest with any entity that would pose a conflict of interest with the subject matter of this article. The views expressed in this article are those of the authors and do not necessarily reflect the position or policy of the Department of Veterans Affairs or the United States Government.

Authorship Information: Concept and design (WFG, CBG, JCL, JMD); acquisition of data (WFG, CBG, JCL); analysis and interpretation of data (WFG, CBG, JMD); drafting of the manuscript (WFG, CBG, JCL, JMD); critical revision of the manuscript for important intellectual content (WFG, CBG, JCL, JMD); and statistical analysis (WFG).

Address correspondence to: Walid F. Gellad, MD, MPH, Center for Health Equity Research and Promotion, Department of Veterans Affairs Medical Center, 7180 Highland Dr, Pittsburgh, PA 15206. E-mail: walid.gellad@va.gov.

1. Wennberg JE, Cooper MM. The Dartmouth Atlas of Health Care in the United States. Chicago, IL: Center for Evaluative Clinical Sciences, Dartmouth Medical School, AHA Press; 1996.

2. Gawande A. The cost conundrum: what a Texas town can teach us about health care. New Yorker. June 1, 2009:36-44.

3. Fisher E, Goodman D, Skinner J, Bronner K; Dartmouth Institute for Health Policy and Clinical Practice. Health care spending, quality, and outcomes: more isn't always better. February 27, 2009. http://www. dartmouthatlas.org/downloads/reports/Spending_Brief_022709.pdf. Accessed September 23, 2009.

4. Fisher ES, Wennberg DE, Stukel TA, Gottlieb DJ, Lucas FL, Pinder EL. The implications of regional variations in Medicare spending, part 2: health outcomes and satisfaction with care. Ann Intern Med. 2003;138(4):288-298.

5. Yasaitis L, Fisher ES, Skinner JS, Chandra A. Hospital quality and intensity of spending: is there an association? Health Aff (Millwood). 2009;28(4):w566-w572.

6. Sirovich B, Gallagher PM, Wennberg DE, Fisher ES. Discretionary decision making by primary care physicians and the cost of U.S. health care. Health Aff (Millwood). 2008;27(3):813-823.

7. Hartman M, Martin A, McDonnell P, Catlin A; National Health Expenditure Accounts Team. National health spending in 2007: slower drug spending contributes to lowest rate of overall growth since 1998 [published correction appears in Health Aff (Millwood). 2009;28(1) doi: 10.1377/hlthaff.28.1.246]. Health Aff (Millwood). 2009;28(1):246-261.

8. Aitken M, Berndt ER, Cutler DM. Prescription drug spending trends in the United States: looking beyond the turning point. Health Aff (Millwood). 2009;28(1):w151-w160.

9. Aspinall SL, Berlin JA, Zhang Y, Metlay JP. Facility-level variation in antibiotic prescriptions for veterans with upper respiratory infections. Clin Ther. 2005;27(2):258-262.

10. Cox ER, Motheral BR, Henderson RR, Mager D. Geographic variation in the prevalence of stimulant medication use among children 5 to 14 years old: results from a commercially insured US sample. Pediatrics. 2003;111(2):237-243.

11. Dubois RW, Batchlor E, Wade S. Geographic variation in the use of medications: is uniformity good news or bad? Health Aff (Millwood). 2002;21(1):240-250.

12. Fultz SL, Good CB, Fine M. Increased costs of diabetes therapy not related to glycemic control at three Veterans Affairs medical outpatient clinics. P & T: a peer-reviewed journal for formulary management. 2004;29(8):500-506.

13. Gao J, Campbell J. Trend and variation of prescription drug cost in the veterans health-care system. Health Serv Manage Res. 2008; 21(1):14-22.

14. Krein SL, Hofer TP, Kerr EA, Hayward RA. Whom should we profile? Examining diabetes care practice variation among primary care providers, provider groups, and health care facilities. Health Serv Res. 2002;37(5): 1159-1180.

15. Lopez J, Meier J, Cunningham F, Siegel D. Antihypertensive medication use in the Department of Veterans Affairs: a national analysis of prescribing patterns from 2000 to 2002. Am J Hypertens. 2004;17(12 pt 1):1095-1099.

16. Steinman MA, Yang KY, Byron SC, Maselli JH, Gonzales R. Variation in outpatient antibiotic prescribing in the United States. Am J Manag Care. 2009;15(12):861-868.

17. Miller DR, Safford MM, Pogach LM. Who has diabetes? Best estimates of diabetes prevalence in the Department of Veterans Affairs based on computerized patient data. Diabetes Care. 2004;27(suppl 2):B10-B21.

18. Office of the Assistant Deputy Under Secretary for Health for Policy and Planning. VHA Enrollment, Patients and Expenditures. Washington, DC: US Dept of Veterans Affairs Intranet Site; 2008. Accessed September 1, 2009.

19. Department of Veterans Affairs. Pharmacy Benefits Management Services. July 2009. Accessed August 1, 2009.

20. Sales MM, Cunningham FE, Glassman PA, Valentino MA, Good CB. Pharmacy benefits management in the Veterans Health Administration: 1995 to 2003. Am J Manag Care. 2005;11(2):104-112.

21. National Committee for Quality Assurance. HEDIS & quality measurement. 2009. http://www.ncqa.org/tabid/59/Default.aspx. Accessed April 10, 2009.

22. Jha AK, Perlin JB, Steinman MA, Peabody JW, Ayanian JZ. Quality of ambulatory care for women and men in the Veterans Affairs Health Care System. J Gen Intern Med. 2005;20(8):762-765.

23. Bolen S, Feldman L, Vassy J, et al. Systematic review: comparative effectiveness and safety of oral medications for type 2 diabetes mellitus. Ann Intern Med. 2007;147(6):386-399.

24. Baigent C, Keech A, Kearney PM, et al; Cholesterol Treatment Trialists'(CTT) Collaborators. Efficacy and safety of cholesterol-lowering treatment: prospective meta-analysis of data from 90,056 participants in 14 randomised trials of statins [published corrections appear in Lancet. 2008;371(9630):2084 and 2005;366(9494):1358]. Lancet. 2005;366(9493):1267-1278.

25. Law MR, Wald NJ, Rudnicka AR. Quantifying effect of statins on low density lipoprotein cholesterol, ischaemic heart disease, and stroke: systematic review and meta-analysis. BMJ. 2003;326(7404):e1423.

26. Gellad WF, Detsky AS, Choudhry NK. Implications of recent clinical trials on pay-for-performance [published correction appears in Am J Health Syst Pharm. 2009;66(11):973]. Am J Health Syst Pharm. 2009;66(9):864-867.

27. Stefos T, LaVallee N, Holden F. Fairness in prospective payment: a clustering approach. Health Serv Res. 1992;27(2):239-261.

28. Yano EM, Wright S, Cuerdon T. Organizational complexity: does it reflect differences in structure and management process at individual VAMCs? In: HSR&D 2009 National Meeting Abstracts; February 11-13, 2009; Baltimore, MD.

29. Huang PY, Yano EM, Lee ML, Chang BL, Rubenstein LV. Variations in nurse practitioner use in Veterans Affairs primary care practices. Health Serv Res. 2004;39(4 pt 1):887-904.

30. Frank RG. Prescription drug prices: why do some pay more than others do? Health Aff (Millwood). 2001;20(2):115-128.

31. Gellad WF, Choudhry NK, Friedberg MW, Brookhart MA, Haas JS, Shrank WH. Variation in drug prices at pharmacies: are prices higher in poorer areas? Health Serv Res. 2009;44(2 pt 1):606-617.

32. Redelmeier DA, Bell CM, Detsky AS, Pansegrau GK. Charges for medical care at different hospitals. Arch Intern Med. 2000;160(10): 1417-1422.

33. Hellerstein JK. The importance of the physician in the generic versus trade-name prescription decision. RAND J Econ. 1998;29(1):108-136.

34. Ashton CM, Petersen NJ, Souchek J, et al. Geographic variations in utilization rates in Veterans Affairs hospitals and clinics. N Engl J Med. 1999;340(1):32-39.

35. Asch SM, McGlynn EA, Hogan MM, et al. Comparison of quality of care for patients in the Veterans Health Administration and patients in a national sample. Ann Intern Med. 2004;141(12):938-945.

36. Greenfield S, Kaplan SH. Creating a culture of quality: the remarkable transformation of the Department of Veterans Affairs Health Care System. Ann Intern Med. 2004;141(4):316-318.

37. Kerr EA, Gerzoff RB, Krein SL, et al. Diabetes care quality in the Veterans Affairs Health Care System and commercial managed care: the TRIAD study. Ann Intern Med. 2004;141(4):272-281.

38. Good CB, Valentino M. Access to affordable medications: the Department of Veterans Affairs pharmacy plan as a national model. Am J Public Health. 2007;97(12):2129-2131.