Problems Due to Medication Costs Among VA and Non-VA Patients With Chronic Illnesses

,
The American Journal of Managed Care, November 2004 - Part 2, Volume 10, Issue 11 Pt 2

Objective: To compare rates of cost-related medication underuseand other problems due to medication costs amongDepartment of Veterans Affairs (VA) patients with rates amongpatients with Medicaid, private health insurance, Medicare, and noinsurance coverage.

Study Design: Nationwide survey.

Methods: A total of 4055 chronically ill patients completed anonline questionnaire regarding cost-related adherence problemsfor medications used to treat 16 chronic illnesses. Respondentsalso reported whether they cut back on necessities due to medicationcosts, increased their debt burden to pay for prescriptions, andworried about how they would pay for their medications.

P

P

P

P

Results: Rates of cost-related medication underuse were loweramong VA patients (12%) than among patients with Medicaid(25%; =.0004), Medicare (22%; =.001), or no insurance (35%; < .0001). In multivariate analyses, patients with Medicare or noinsurance coverage were more likely than VA patients to foregomedication at least once per month due to cost (adjusted oddsratios of 3.4 and 3.9; both &#8804; .001). Patients with Medicare or noinsurance coverage also were more likely than VA patients toforego basic needs to pay medication costs, borrow money to payfor their treatments, and worry frequently about how they wouldpay for their medication.

Conclusion: The VA's prescription benefits may prevent problemsdue to medication costs. Studies assessing the impact of VAprescription coverage on health outcomes and service use will beneeded to evaluate the cost-effectiveness of VA drug benefit policies.

(Am J Manag Care. 2004;10(part 2):861-868)

Managed care plans and their patients are strugglingwith the burden of skyrocketing prescriptiondrug costs. Patients with chronic illnessesoften have inadequate prescription coverage,1-3 andmany cut back on medication use due to cost pressures.2,4-10 Because medication underuse can lead toincreased morbidity, mortality, and preventable acute-careuse,10-12 managed care organizations are striving todevelop drug coverage programs that are financiallyviable while ensuring that plan members can afford thetreatments they need.

Department of Veterans Affairs (VA) medical centersoffer more comprehensive medication coverage thanalmost any other payer in the United States. Drugs onthe VA formulary are 100% covered for patients withlow incomes or service-connected disabilities, and otherVA patients pay a $7 copayment for a 30-day prescription.VA patients have no cap on the total cost of theircovered drugs, and most patients who incur $840 ormore in medication costs during a given year have allsubsequent copayments waived. Between 1995 and2001, the number of patients receiving VA outpatientcare increased from 2.8 million to 4.1 million, withmany of these new patients entering the system toaccess prescription drug coverage.13-15 In the context ofthe current debate over prescription drug costs, theVA's drug coverage policy and its impact on patients'use of prescribed medications provide an importantmodel for inquiry.

In a prior study,12 we compared rates of cost-relatedmedication underuse among diabetes patients in 3 VAfacilities to rates among patients with private insurance,Medicare, Medicaid, and no health insurance treated ina county and a university health system. We found thatVA patients were less likely than patients with any othertype of coverage to report medication underuse due tocost. We also found that patients reporting cost-relatedadherence problems had poorer glycemic control, moresymptoms, and worse physical and mental health functioning.These findings suggest that the VA's comprehensivedrug benefits may contribute to the bettertreatment outcomes among chronically ill VA patientscompared with Medicare patients.16,17 In our priorstudy,12 however, all patients were drawn from 5 healthcaresystems, 4 or which were located in NorthernCalifornia. Thus, it remains unclear whether the findingsreflect nationwide patterns of adherence among VAand non-VA patients with chronic illnesses.

The problems chronically ill patients face due tomedication costs extend beyond their treatment adherence.For example, one recent study found that 21% ofMedicare beneficiaries spent less on basic needs tocover the cost of their prescription drugs.4 However, weknow little about the extent to which the VA's prescriptiondrug coverage may buffer chronically ill patientsfrom these other consequences of out-of-pocket treatmentcosts.

Accordingly, we surveyed a nationwide panel ofchronically ill older adults about their problems due toout-of-pocket medication costs. We found that rates ofcost-related medication underuse were high amongindividuals with a variety of chronic health problems.9We also found that more than 22% of patients reportedcutting back on spending for basic needs in response tomedication cost pressures, and that 16% reportedincreasing their debt burden to pay medication costs.18In the current study, we compared rates of cost-relatedmedication underuse as well as other burdens frommedication cost pressures among survey respondentsusing VA care with rates among patients with otherforms of health insurance coverage.

METHODS

Study Participants

The study was based on surveys completed duringNovember and December 2002 by a nationwide panel ofadults living in the United States. Details about thepanel's sampling approach and recruitment process areavailable elsewhere.19-21 In brief, panel members wererecruited using random digit dialing; the sampling frameconsisted of the entire US population with an assignedtelephone number. Potential panel members wereoffered WebTV and free Internet access in exchange forcompleting Web-based surveys several times per month.At the time of this study, the panel included over 40 000members.

We used sociodemographic and health status informationabout panel members to identify all 5644 individualsaged 50 years and older who reported takingprescription medication for diabetes, depression, heartproblems, hypertension, or high cholesterol. Thesociodemographic characteristics of these respondentswere similar to those of Americans aged 50 years andolder, as reported in 2000 Census data. For example,the survey and Census populations were similar withregard to the representation of African Americans (8%vs 10%), women (50% vs 54%), married individuals (70%vs 65%), and individuals with household incomes of$20 000 or less (17% vs 23%). Institutional reviewboards approved the study protocol.

Survey Description and Variable Creation

Respondents were asked about their use and cost-relatedunderuse of prescription medication for each of16 chronic health problems: arthritis; asthma; chronicback pain or sciatica; diabetes; high cholesterol; chronicobstructive pulmonary disease ("chronic bronchitis,emphysema, or chronic obstructive pulmonary disease");depression; heartburn, acid reflux, or irritablebowel syndrome; atherosclerosis ("blocked arteries inthe heart, angina, or chest pain from heart disease");heart failure; high blood pressure or hypertension;myocardial infarction ("heart attack"); migraineheadache; osteoporosis; stomach or duodenal ulcers;and stroke. On a condition-by-condition basis, participantswere asked: "In the past 12 months, have youever taken less of this medication than prescribed byyour doctor because of the cost?" For the current analyses,respondents were coded as having cost-relatedadherence problems if they reported any medicationunderuse due to cost for 1 or more of their medicationsin the prior year.

Respondents reporting cost-related underuse wereasked how often they engaged in each of the followingbehaviors because of the medication cost: taking fewerpills or a smaller dose, not filling a prescription at all,putting off or postponing getting a prescription filled,using herbal medicine or vitamins when feeling sickrather than taking their prescription medication, or takingthe medication less frequently than recommendedto "stretch out" the time before getting a refill.Participants were coded as foregoing medication at leastmonthly if they reported any of these behaviors "at leastonce per month" due to the cost.

Respondents also were asked about 4 other types ofproblems associated with out-of-pocket medicationcosts: spending less on basic needs such as food or heatto pay medication costs, borrowing money from familyor friends to pay medication costs, increasing creditcard debt to pay for medications, and worrying abouthow to pay for medications at least once per month. Seethe footnotes to Tables 2, 3, and 4 for the wording ofeach item.

For the current study, respondents were groupedinto mutually exclusive categories using a hierarchybased on their primary source of medical insurance.Patients were assigned to the insurance category thatwe expected would be associated with the most generousform of prescription drug coverage. Specifically, weidentified the following patient groups: VA patients,Medicaid patients who did not use the VA, privatelyinsured patients (without VA use or Medicaid),Medicare patients with no other form of health insurance,and uninsured patients. These categories identifysubgroups of patients of particular interest to policymakers(eg, patients with Medicare coverage only) andrecognize that individuals often are unable to identify indetail the provisions of theirprescription drug coverage.22,23The sociodemographic variableswe examined as possible predictorsof problems due to out-of-pocketmedication costsincluded respondents' race, age,sex, educational attainment,and annual household income.We also examined indices ofpatients' medication cost pressures,including their total numberof prescription medications,total monthly out-of-pocketmedication costs, and numberof chronic health problems.

Analysis

P

P

P

In initial analyses, we examinedbivariate differences in thesociodemographic characteristicsand medication cost pressuresacross groups of patientsdefined by their primary sourceof health insurance. We thenexamined bivariate associationsbetween insurance type andeach outcome (medicationunderuse, foregoing necessities,borrowing from family orfriends, increasing credit carddebt to pay medication costs,and medication cost worry).Finally, we constructed multivariatelogistic regression modelsto determine the independenteffect of insurance type on eachoutcome, controlling for patients'sociodemographic characteristics(race, sex, age,educational attainment, andincome), number of chronichealth problems, and number ofprescription medications. SeeTable 1 for the categorization ofordinal and categorical covariates. We included all ofthe covariates in each regression model although somecovariates were not associated with all the outcomevariables, both for consistency and because each of thevariables has been found to be related to medicationadherence problems in prior studies. In all bivariate andmultivariate analyses, we used a standard adjustment tointerpret the values associated with the 4 pairwisecomparisons between VA patients and patients in the 4other insurance categories. Thus, a 2-sided value of.0125 (.05 &#247; 4) was considered equivalent to a nominal value of .05 in defining statistically significant differences.

Because our hypothesis was that VA prescription benefitswould lead to lower rates of medication cost problemsby decreasing patients' out-of-pocket costs,out-of-pocket costs were not included as a covariate inour primary multivariate models. Confirmatory analysesincluded models with out-of-pocket costs as acovariate and fitting models limited to men only(because of the high number of male VA patients). In allanalyses, we used sampling weights to adjust the distributionof respondents to match the distribution of theUS population on age, sex, race/ethnicity, education,region, and metropolitan residence, thereby correctingfor oversampling and survey nonresponse.24,25 TheBureau of Labor Statistics Current Population Surveyfor October 2002 provided data on the distributionof the US population.26 All analyses were done usingStata 8.1.27

RESULTS

Respondent Characteristics

P

P

P

P

P

P

P

A total of 4264 people (76% of the 5644 patients sampled)completed the online informed consent and questionnairein the fall of 2002. We subsequently excluded185 participants because they reported that they wereno longer taking medication for any of the 5 index conditionsand 24 additional patients because they weremissing data on income. The final sample size was 4055.Compared with eligible nonrespondents, respondentsto the survey were more likely to be white (88% vs 81%; <.001), to be older (mean age of 65 years vs 63 years; <.001), and to have some college education (66% vs60%; < .001). Respondents and nonrespondents weresimilar with regard to sex ( = .29), home ownership( = .44), marital status ( = .16), and income ( = .41).

P

P

P

P

P

P

VA patients were more likely to be Caucasian andmale than patients with other insurance types (Table 1).As expected, Medicare patients were older than otherpatients, and most patients with Medicaid coverage hadannual household incomes of less than $20 000 per year.VA patients reported using more prescription medicationsthan patients with private insurance ( <.0001), Medicare( = .003), or no health coverage ( <.0001). For example,a greater proportion of VA patients reported usingcholesterol-lowering drugs (63%) than patients withMedicaid (48%), private insurance (50%), Medicare(46%), or no health insurance (44%; overall = .007).Nevertheless, VA patients reported lower monthly out-of-pocket medication costs than patients with either privatehealth insurance ( = .0002) or no insurancecoverage ( <.0001).

Bivariate Relationships Between Insurance Typeand Medication Cost Problems

P

P

P

P

In bivariate analyses (Table 2), less than half asmany VA patients reported foregoing medication atleast once per month due to cost (7%) than patients withMedicaid (20%; = .0005), Medicare (19%; = .004), orno health insurance (28%; <.0001). VA patients alsowere substantially less likely than patients withMedicaid, Medicare, or no health insurance to cut backon necessities to pay medication costs, borrow moneyfrom family or friends to pay for their prescriptions, orworry about their medication costs at least once permonth ( &#8804;.0002 for each pairwise comparison).

Multivariate Analyses

P

P

P

Controlling for patients' disease burden, income,number of prescription medications, and sociodemographiccharacteristics, patients with Medicare coveragewere 2.6 times as likely to report cost-related underuseas VA patients ( = .001), and patients without anyhealth insurance coverage were 3.2 times as likely to doso ( <.001; Table 3). The differences between VApatients and those with other forms of health coveragewere even more pronounced when only frequent cost-relatedadherence problems (at least once per month)were considered. Compared with their VA counterparts,Medicare patients were 3.4 times as likely, andpatients with no insurance coverage were nearly 4times as likely, to report frequent medication underuse(both <.001).

P

In multivariate models, we also observed significantlyhigher rates of other problems due to medicationcosts among non-VA patients compared with VApatients (Table 4). Specifically, patients with Medicareor no insurance were both 2.9 times as likely to reportcutting back on spending for basic needs due to medicationcost pressures, 6.4 and 4.9 times as likely to borrowmoney from family or friends to pay medicationcosts, and 3.2 and 2.8 times as likely to report worryingabout their medication costs at least once per month(all <.001). Rates of other problems due to medicationcosts were similar between VA patients and patientswith Medicaid or private health insurance.

P

P

P

Although the overall sample in this study was large,we had limited statistical power to examine differencesin rates of cost-related medication underuse associatedwith specific treatment types. Nevertheless, we observedhigher rates of cost-related adherence problems amongnon-VA patients who used several types of clinicallyimportant treatments, especially when comparing VApatients with individuals who had Medicare coverage orno health insurance. For example, in multivariate modelscontrolling for the same covariates as in Table 3,we found that, among patients using prescription cardiovascularmedication after a myocardial infarction(n = 841), uninsured patients were 11.5 times as likelyas their VA counterparts to report underuse of thosetreatments due to cost (95% confidence interval [CI]:2.0, 65.2; = .006). Among patients using antihypertensivemedications (n = 2838), the rate of cost-relatedunderuse was 5.7 times as high among patients withoutany insurance coverage than it was among VA patients(95% CI: 1.9, 16.9; = .002). Among patients usingback pain medication (n = 523), uninsured patientswere 7.8 times as likely as their VA counterparts toforego those treatments due to cost (95% CI: 1.9, 31.9; = .005).

Confirmatory Analyses

P

P

P

P

P

Although general studies of medication adherence28,29as well as cost-related adherence problemsshow few consistent differences between men andwomen, we re-fit each of the regression models shownin Tables 3 and 4 after limiting the samples of VA andnon-VA patients to men only. Although the overall patternof effects was similar, some differences betweenVA and non-VA patients were larger in the male subsetof respondents. Controlling for covariates, male patientswith Medicaid were more likely than male VA patientsto report any cost-related medication underuse (adjustedodds ratio [AOR] = 2.9; = .009), to report cost-relatedunderuse at least once per month (AOR = 3.3; = .009), to forego basic needs due to medication costs(AOR = 3.1; = .004), to borrow money to pay medicationcosts (AOR = 5.5; = .006), and to worry abouttheir medication costs at least once per month (AOR =3.0; = .002).

P

P

As noted in the Methods section, patients' out-of-pocketmedication costs were not included in our multivariatemodels because we hypothesized thatreductions in these costs were a primary mechanismlinking VA system use to lower rates of medication costproblems. To test this conceptual framework, we re-fiteach of the multivariate models shown in Tables 3 and4, including an additional covariate characterizingpatients' monthly out-of-pocket medication costs. In allinstances, out-of-pocket costs were a strong predictor ofpatients' likelihood of medication cost problems; and inmost cases, statistically significant differences betweenVA patients and patients with other insurance typeswere substantially reduced. For example, the differencebetween Medicare patients' and VA patients' likelihoodof foregoing medication due to cost was no longer statisticallysignificant after including out-of-pocket costs(AOR = 1.7; = .08), and Medicare patients were nolonger more likely than VA patients to forego basicneeds due to cost pressures (AOR = 1.7; = .07).Nevertheless, some statistically significant differencesbetween VA and non-VA patients' likelihood of experiencingmedication cost problems persisted even whencontrolling for total out-of-pocket drug costs. We concludethat out-of-pocket cost differences are a majordriver of the effects shown in Tables 3 and 4.

DISCUSSION

Compared with other patients, VA participants inthis study reported using more prescription drugs whileexperiencing lower out-of-pocket medication costs. VApatients also were less likely than patients withMedicare or no health insurance coverage to report cuttingback on medication use due to cost pressures. Theresults of this study confirm our prior study's findings12that fewer VA diabetes patients reported cutting backon medication use due to cost than patients with otherforms of coverage. That prior study, however, was limitedto patients with diabetes drawn from only 5 healthsystems, with all but 1 located in Northern California. Inaddition, VA patients in the current study were significantlyless likely to experience other burdens from prescriptionmedication costs than patients who hadMedicare or no coverage, such as cutting back on spendingfor basic needs, borrowing money to pay for theirmedication, or worrying frequently about how theywould pay for their prescription drugs. These otherproblems due to medication cost pressures, although farless studied than medication underuse, also mayadversely affect patients' health and well-being. Becausethe current study is based on a nationwide survey ofolder adults with a variety of chronic health problemsrecruited through a community-based sampling approach,the pattern of effects we observed is likely to reflect differencesbetween VA and non-VA patients with chronicillnesses nationwide.

Weighing the costs and benefits of VA's more generouscoverage from a health system's perspective requiresboth a careful assessment of the indications for pharmacotherapyand a comprehensive enumeration of theimpact of this coverage on nonpharmaceutical resourceuse. Some payers are concerned that expanded pharmacy benefits may lead to unnecessary prescribing oftreatments with limited clinical value. More generousbenefits indeed have been found to increase the numberof prescriptions for problems that are generally consideredless serious, such as colds, allergies, and skin problems.30 Although VA patients in the current study didreport more prescriptions on average than otherpatients, they also reported significantly more seriouscomorbidities for which medications often are indicated.

There is growing evidence that the short-term costsof more generous pharmacy benefits may be offset byreductions in future health service use. One seminalstudy demonstrated that restrictions in Medicaid prescriptioncoverage for schizophrenic patients led to adramatic increase in acute psychiatric hospitalizations,negating any possible cost savings associated with thereduction in drug benefits.10 More comprehensive coverageof other medication classes, such as gemfibrozilfor coronary heart disease patients with low high-densitylipoprotein levels, also may provide cost-savings,31and some experts have recommended restructuringcopayment levels to reflect the expected benefits ofspecific drugs for targeted patient subgroups.32 Otherstudies have elucidated the dramatic effects of changesin health plan formulary administration (eg, adoption ofvarious incentive-based copayment structures) onenrollees' out-of-pocket spending and continued use ofprescribed medications.33 Unfortunately, many healthplans are moving forward with increased cost-sharingfor prescription drugs without empirical evidenceregarding the possible implications of those changes formembers' adherence, health status, or use of acute care.

Even though pharmaceutical use may improvepatients' health and decrease healthcare costs in thelong term, the immediate cost of providing comprehensiveprescription drug coverage is of significant concernto VA policymakers and other insurers considering thescope of their benefits. VA pharmacy costs increased160% between 1996 and 2002, due to the growing numberof VA system users, increases in the number of prescriptionsper patient, and (to a lesser extent)increasing costs per unit prescribed.15 To maintain reasonablecoverage while containing costs, pharmacybenefits must be based on effective negotiations withwholesalers to keep costs down—negotiations in whichthe VA has been very successful but which plannedMedicare reforms specifically prohibit.34 VA healthcaresystems also strictly control their formulary, requiringprescribers to use approved drugs according to explicitclinical guidelines before more expensive classes ofmedications (eg, cyclooxygenase-2 inhibitors) can beprescribed. Although it is challenging to develop andmanage a cost-effective formulary in a large health system,most VA healthcare providers report that theirpatients receive the drugs they need.35 Of course, manysmaller health plans cannot avail themselves of the collectivenegotiating strength, economies of scale, andother benefits of a large, centralized system such as theVA. However, by adopting these management tools (egclosed or incentive-based formularies, high-volumenegotiations with suppliers, and more rationally structuredcopayment systems), non-VA health plans maybe able to maintain a fiscally viable program of prescriptiondrug coverage that maximizes cost savingswhile also enabling patients to afford their prescribedtreatments.

Although this is the first nationwide study to compareproblems due to medication costs among VA andnon-VA patients, it has several limitations. Because theinterview was Internet-based, individuals who are lesscomputer literate may have been underrepresented,and these may be the same patients who are more likelyto experience medication cost problems. Althoughthese sampling biases may affect the rates presented inTable 2, they are less likely to influence the relativerates estimated via multiple regression models inTables 3 and 4.

Perhaps the most significant limitation of this studyis that patients were identified according to their self-reportedhealth insurance, rather than according totheir type of prescription drug coverage. VA prescriptionbenefits are consistent nationally. Medicaid andprivate insurance coverage for prescription drugs, however,varies considerably across plans, and manyMedicare patients have supplemental coverage througha managed care plan. Thus, the findings reported hererepresent average differences between patients with VAcoverage and patients with other forms of coverage.The comparisons may overestimate or underestimatedisparities in coverage and medication cost problemsbetween VA patients and patients with more or lessextensive pharmaceutical benefits.

In summary, we found that chronically ill patientswith access to the VA's prescription drug coverage hadlower rates of cost-related adherence problems thanpatients with Medicare or no insurance coverage, andthat rates of problems due to medication costs were atleast as low among VA patients as among patients withMedicaid or private insurance coverage. VA patientswere also less likely than some non-VA patients toreport other detrimental consequences of medicationcost pressures, such as foregoing necessities to pay fortheir medication or worrying frequently about howthey could pay for their treatments.

Further studies (with more reliable informationabout patients' actual benefits and cost pressures)will be needed to discern whether financial benefitsalone explain differences between VA and non-VApatients, or whether other factors influence whetherpatients experience medication cost problems in theface of equivalent financial demands. Research also isneeded to elucidate the costs and benefits of the VA'sgenerous prescription drug coverage and to furtherassess components of effective non-VA coverage programs.Clearly, however, raising medication copaymentsto levels that prevent patients from taking theirprescribed treatments is one of the poorest strategieshealth plans can adopt to contain prescription drugcosts.

From the Department of Veterans Affairs Center for Practice Management andOutcomes Research, and the Department of Internal Medicine and Michigan DiabetesResearch and Training Center, University of Michigan, Ann Arbor, Mich.

This study was funded by the Department of Veterans Affairs Health Services Researchand Development Service. The views expressed in this article are those of the authors anddo not necessarily represent the views of the Department of Veterans Affairs.

Address correspondence to: John D. Piette, PhD, Center for Practice Management andOutcomes Research, VA Ann Arbor Health Care System, PO Box 130170; Ann Arbor, MI48113-0170. E-mail: jpiette@umich.edu.

Annu Rev Public Health.

1. Adams AS, Soumerai SB, Ross-Degnan D. The case for a Medicare drug coveragebenefit: a critical review of the empirical evidence. 2001;22:49-61.

Med Care.

2. Martin BC, McMillan JA. The impact of implementing a more restrictive prescriptionlimit on Medicaid recipients: effects on cost, therapy, and out-of-pocketexpenditures. 1996;34:686-701.

N Engl JMed.

3. Soumerai SB, Avorn J, Ross-Degnan D, Gortmaker S. Payment restrictions forprescription drugs under Medicaid: effects on therapy, cost, and equity. 1987;317:550-556.

Health Aff (Milwood)

4. Safran DG, Neuman P, Schoen C, et al. Prescription drug coverage and seniors:how well are states closing the gap? 2002;Supp Web exclusives:W253-W268.

Soc Sci Med.

5. Leibowitz A, Manning WG, Newhouse JP. The demand for prescription drugs asa function of cost-sharing. 1985;21:1063-1069.

Med Care.

6. Harris BL, Stergachis A, Reid D. The effect of drug co-payment on utilizationand cost of pharmaceuticals in a health maintenance organization. 1990;28:907-917.

Pharmacoeconomics.

7. Levy RA. Prescription cost sharing: economic and health impacts, and implicationsfor health policy. 1992;2:219-237.

J Gen Intern Med.

8. Steinman MA, Sands LP, Covinsky KE. Self-restriction of medications due to costin seniors without prescription coverage. 2001;16:793-799.

Am J Pub Health.

9. Piette JD, Heisler M, Wagner TH. Cost-related medication underuse amongchronically-ill adults: the treatments people forego, how often, and who is at risk. 2004;94(10):1782-1787.

N Engl J Med.

10. Soumerai SB, McLaughlin TJ, Ross-Degnan D, Casteris CS, Bollini P. Effects oflimiting Medicaid drug-reimbursement benefits on the use of psychotropic agentsand acute mental health services by patients with schizophrenia. 1994;331:650-655.

NEngl J Med.

11. Soumerai SB, Ross-Degnan D, Avorn J, McLaughlin TJ, Choodnovsky I. Effectsof Medicaid drug-payment limits on admission to hospitals and nursing homes. 1991;325:1072-1077.

Med Care.

12. Piette JD, Wagner TH, Potter MB, Schillinger D. Health insurance status, cost-relatedmedication underuse, and outcomes among diabetes patients in three systemsof care. 2004;42:102-109.

13. Lederle FA, Parenti CM. Prescription drug costs as a reason for changing physicians.J Gen Intern Med. 1994;9:162-163.

West J Med.

14. Jain S, Avins AL, Mendelson T. Preventive health services and access to care formale veterans compared to their spouses. 1998;168:499-503.

TheNew York Times.

15. Freudenheim M. V. A. Health Care Strained by Big Wave of Enrollees. April 6, 2002;sect A1.

N Engl J Med.

16. Jha AK, Perlin JB, Kizer KW, Dudley RA. Effect of the transformation of theVeterans Affairs Health Care System on the quality of care. 2003;348:2218-2227.

Ann Intern Med.

17. Kerr EA, Gerzoff RB, Krein SL, et al. A comparison of diabetes care quality inthe Veterans Affairs healthcare system and commercial managed care: results fromthe TRIAD study. 2004;141:272-281.

J Behav Med.

18. Heisler M, Wagner T, Piette JD. Patient strategies to cope with high prescriptionmedication costs: who is cutting back on necessities, increasing debt, orunderusing medications? In press.

19. Knowledge Networks. Weighting procedures for the VA PrescriptionMedication Study. Available at: http://www.knowledgenetworks.com/ganp/docs/061604_weightingdescription.pdf. Accessed June 16, 2004.

20. Krosnick JA, Chang L. A comparison of random digit dialing telephone surveymethodology with Internet survey methodology as implemented by KnowledgeNetworks and Harris Interactive. April 2001. Ohio State University. Available at:http://www.knowledgenetworks.com/ganp/reviewer-info.html. Accessed June 16,2004.

21. Baker LC, Bundorf MK, Singer S, Wagner TH. Validity of the survey of healthand the Internet, and Knowledge Network's panel and sampling. Palo Alto, Calif:Stanford University; 2003. Available at: http://www.knowledgenetworks.com/ganp/reviewer-info.html. Accessed June 16, 2004.

Health Aff (Millwood).

22. Garnick DW, Hendricks AM, Thorpe KE, Newhouse JP, Donelan K, BlendonRJ. How well do Americans understand their health coverage? 1993;12:204-212.

Health Care Financ Rev.

23. Marquis MS. Consumers' knowledge about their health insurance coverage. 1983;5:65-80.

Stata Technical Bulletin.

Stata Technical Bulletin Reprints.

24. Eltinge JL, Sribnew WM. Some basic concepts for design-based analysis ofcomplex survey data. 1996;10:387-393. Reprinted in 1996;6:208-213.

Analysis of Health Surveys.

25. Korn EL, Graubard BI. New York: John Wiley andSons; 1999.

26. Bureau of Labor Statistics and the Bureau of the Census. CPS (CurrentPopulation Survey) basic monthly survey. Available at:http://www.bls.census.gov/cps/bovrvw1.htm. Accessed April 3, 2003.

27. Stata Corporation. Stata statistical software. Release 8.1. College Station, Tex:Stata Corporation; 2003.

Ann Pharmacother.

28. Vik SA, Maxwell CJ, Hogan DB. Measurement, correlates, and health outcomesof medication adherence among seniors. 2004;38:303-312.

Int J Nurs Stud.

29. Ryan AA. Medication compliance and older people: a review of the literature. 1999;36:153-162.

Med Care.

30. Stuart B, Grana J. Ability to pay and the decision to medicate. 1998;36:202-211.

Arch Intern Med.

31. Nyman JA, Martinson MS, Nelson D, et al. Cost-effectiveness of gemfibrozilfor coronary heart disease patients with low levels of high-density lipoprotein cholesterol. 2002;162:177-182.

Am J Manag Care.

32. Fendrick AM, Smith DG, Chernew ME, Shah SN. A benefit-based copay forprescription drugs: patient contribution based on total benefits, not drug acquisitioncost. 2001;7:861-867.

N Engl J Med.

33. Huskamp HA, Deverka PA, Epstein AM, Epstein RS, McGuigan KA, Frank RG.The effect of incentive-based formularies on prescription-drug utilization andspending. 2003;349:2224-2232.

The New York Times.

34. Pear R, Bogdanich W. Some successful models ignored as Congress works ondrug bill. September 4, 2003;sect A1.

VA Drug Formulary: Better Oversight IsRequired But Veterans Are Getting Needed Drugs.

35. US General Accounting Office. January 2001. Available at:http://www.gao.gov/new.items/d01183.pdf. Accessed June 16, 2004.