Seniors with Chronic Health Conditions and Prescription Drugs: Benefits, Wealth, and Health

Authors


  • An earlier version of this work was presented at the June 2001 annual meeting of the Academy for Health Services Research and Health Policy.

Barry G. Saver, Department of Family Medicine, University of Washington, Box 354696, Seattle, WA 98195-4696, USA. E-mail: saver@u.washington.edu

ABSTRACT

Objectives:  The objectives of this study were to examine the relationship between prescription benefit status and access to medications among Medicare beneficiaries with hypertension, congestive heart failure, coronary artery disease, and diabetes and to determine how income, wealth, and health status influence this relationship.

Methods:  We analyzed survey and administrative data for 4492 Medicare + Choice enrollees aged 67 and above enrolled in a predominantly group-model health maintenance organization in 2000. Outcome measures included difficulty affording medications, methods of coping with medication costs including obtaining medicines from another country, using free samples, and stretching out medications to make them last longer. Independent variables included prescription benefit status, income, wealth measures, health status, and out-of-pocket prescription drug spending.

Results:  Lacking a prescription benefit was independently associated with difficulty affording medications (25% of those without a benefit vs. 17% with a benefit) and coping methods such as stretching out medications. Lower income, lower assets, and worse health status also independently  predicted  greater  difficulty  as  measured by these outcomes; there was no effect modification between these factors and benefit status. Relative to national figures, out-of-pocket spending in this setting was quite low, with only 0.2 and 13% of those with and without a benefit, respectively, spending over $100 per month. Higher out-of-pocket spending predicted greater difficulty affording medications but not stretching out medications.

Conclusions:  Efforts to improve medication accessibility for older Americans with chronic conditions need to address not only insurance coverage but also barriers related to socioeconomic status and health status.

Introduction

Prescription drug spending continues to soar [1–3]; rising prices for medications, increasing per capita numbers of medications taken, and shifts toward newer, more expensive medications are all implicated in this dramatic escalation of costs [4]. In 1987, persons aged 65 and above accounted for 11.8% of the US population but 34.4% of prescription drug expenditures. In addition, costs per person climbed substantially as the number of chronic conditions rose [5]. Combined with this rise in costs, the increasing therapeutic importance of prescription drugs has made the absence of a prescription benefit in Medicare the focus of a national debate. Seniors in Medicare + Choice programs with common, chronic health conditions would seem to be an obvious group in which to study effects of providing prescription benefits, but we are aware of no published studies doing so.

In 1998, 73% of community-dwelling Medicare beneficiaries were estimated to have some type of prescription drug coverage supplemental to basic Medicare for at least part of the year, leaving more than one-quarter without any coverage [6]. Nevertheless, even those with coverage are at risk financially—only about one-half of Medicare beneficiaries have continuous coverage over a 1-year period [7] and many of those with coverage have substantial copayments and caps [8–10]. When the Medicare + Choice program was established by the Balanced Budget Act of 1997, allowing Medicare beneficiaries to choose options such as health maintenance organizations (HMOs) for their care, one of the major attractions was the availability of prescription coverage in many programs. Although the erosion in offering a prescription benefit in Medicare + Choice plans appears to have stabilized from 2001 to 2002, the value of the prescription benefit continues to erode, with increasing premiums and new cost-sharing requirements that particularly affect the chronically ill [11,12]. Thus, even persons with coverage, particularly sicker persons with greater need for medications, may face substantial financial barriers to obtaining essential medications.

We conducted a study of the associations of prescription benefit status with difficulty affording prescription medications and methods of coping with their costs, such as stretching out the interval between prescription fills, in a sample of Medicare beneficiaries with at least one common, chronic condition enrolled in a Medicare + Choice program in a large HMO in Washington state. We also sought to understand how income, wealth, and health influenced these relationships, hypothesizing that having a prescription benefit would be more beneficial for poorer persons, with lower copayments being more protective than higher copayments.

Methods

Study Setting and Population

The study took place in Group Health Cooperative (GHC), a mixed-model HMO providing health services to approximately 400,000 adults in western Washington state. Founded in 1947, GHC is the nation's oldest and largest consumer-governed HMO. GHC contracts with the Group Health Permanente medical group to provide care for the 85% of the plan's enrollees receiving care within a group-model delivery system, whereas the remainder are cared for in a network-model system. GHC's enrollees are not representative of the entire nation but are representative of the communities in which it provides care [13]. Group Health offers a comprehensive, coordinated care Medicare + Choice plan and was the first health plan in the United States to provide Medicare services on a prepaid, capitated basis. Approximately 75% of seniors in GHC have a prescription drug benefit as part of their medical coverage. GHC discontinued an outpatient drug benefit for new individual Medicare enrollees in 1994 but seniors newly insured through a group plan may still have a prescription drug benefit as part of their group coverage. Individuals enrolled before 1994 were allowed to retain their drug benefit while still enrolled in GHC. GHC currently charges $115 per month for the individual pharmacy benefit. GHC's prescription benefits had no annual or lifetime caps on expenditures and differed only in copay amounts, which ranged from $0 to $15 for a 1-month supply of medication. Persons with Medicaid comprised 15% of those with prescription coverage and had no copayments or caps.

Patients with hypertension, diabetes, congestive heart failure, and coronary heart disease were chosen for this study owing to the central role played by prescription medications in the management of these conditions. Patients continuously enrolled in GHC for at least 2 years aged 67 and above who had been diagnosed with one or more of these conditions were identified from administrative databases and chronic disease registries maintained by GHC. These data are a key component of GHC's clinical improvement efforts and have been validated in a variety of research and clinical applications [13–16]. Study entry was limited to persons aged 67 and older to ensure that all subjects would have Medicare coverage throughout the retrospective 2-year period of data availability. Persons residing in nursing homes or having dementia or psychosis were excluded because they would be unlikely to be responsible for taking their own medications; persons with cancer not in remission for at least 5 years were also excluded as their treatment for cancer may have taken priority over treatment for the study conditions. Out of approximately 55,000 Medicare + Choice enrollees in GHC, 17,064 met these inclusion criteria.

Given budgetary limitations, we surveyed a stratified, random sample from this population. We stratified by health condition based on power calculations for condition-specific outcomes for another portion of the overall study. For example, the sample size for persons with hypertension was based on the number to detect clinically meaningful differences in blood pressure level. Attempting to get better representation for the overall study of groups with particular national policy importance, we oversampled the following groups: 1) those without a prescription benefit through the plan; 2) persons with Medicaid; and 3) persons likely to have low incomes (estimating income by geocoding addresses to census block groups). Of the 5533 enrollees selected for surveying, 4763 persons responded for an overall response rate of 86%. We subsequently excluded 265 respondents whose prescription benefit status could not be verified as unchanged during the 2-year observation window and an additional 6 respondents who did not provide consent for use of their administrative data.

This study was reviewed and approved by the University of Washington's Human Subjects Review Committee.

Dependent Measures

Respondents were asked, “Are you sometimes unable to afford your medicines?” Five possible responses were given: “All of the time,”“Most of the time,”“Some of the time,”“A little of the time,” and “None of the time.” Responses were dichotomized for analysis as “all of the time” through “some of the time” (which we will refer to as “unable to afford medications”) versus “a little of the time” or “none of the time”; dichotomizing as “none of the time” versus any other response yielded similar results and is not presented. Respondents were also asked how many of their prescription medicines were obtained through the following methods: traveling to another country, using someone else's insurance coverage, participating in drug-company-sponsored medication assistance programs, and receiving free drug samples from clinics or pharmacies. Responses to these questions were on a 5-point Likert scale: “I don’t do this,”“25% or less,”“half,”“around 75%,” and “all or almost all.” Our final outcome was the yes/no response to the question, “Do you ever stretch out your medicines to make them last longer (for example, take less than the prescribed amount)?”

Independent Variables and Analytic Plan

We adapted the Andersen-Newman model of access to care [17] for our analyses. Predisposing factors were age, race (collapsed owing to sample size considerations to non-Hispanic white/other in multivariate analyses), sex, education, and household configuration. Enabling factors were prescription benefit status, copayment amount, source of Medicare + Choice enrollment (classified as individually purchased with choice of purchasing a prescription benefit, individually purchased with no option to purchase a prescription benefit, private employer-sponsored, government employer-sponsored, or Medicaid), group- versus network-model clinic, family income, estimated wealth, home ownership status (classified after preliminary analyses as own without a mortgage, own with a mortgage, and rent/other), and out-of-pocket spending for prescription drugs for the past 2 years. After initial analyses indicated that source of employer-sponsored enrollment, employer-sponsored versus individual enrollment, and copayment amount were not associated with our outcomes, these were dropped from our multivariate models and a 5-level variable combining benefit status with whether an individual enrollee had an opportunity to purchase a benefit was created. Need factors were health status as represented by the Physical Component Scale (PCS) and Mental Component Scale (MCS) of the Medical Outcomes Study Short Form-12 (SF-12), potentially ranging from 0 to 100 with higher scores representing better health status [18], indicator variables for the four study health conditions, and out-of-pocket pharmacy spending. Complete pharmacy utilization data for persons without a prescription benefit could be obtained only for those using Group Health pharmacies exclusively for their medications, so out-of-pocket prescription expenditures were missing for the 1251 respondents who reported using non-Group Health pharmacies at least some of the time. This was much more common among subjects in the network-model than the group-model part of Group Health (90% vs. 16%, respectively). After initial evaluation and to facilitate comparison with other reports, out-of-pocket prescription medication spending was categorized as < $50/month, $50 to $100/month, or > $100/month.

Chi-square tests were used to measure association between categorical variables; missing re-sponses for categorical independent variables (e.g., income, assets, and categorized out-of-pocket drug expenditures) were assigned to a “missing” category to retain as much data as possible for the analyses. Student's t tests were used to compare means of continuous variables. To further explore possible factors associated with reporting being unable to afford medications, stretching out medicines, and receipt of free samples, we carried out multivariate modeling using logistic regression. Missing data for categorical variables were represented by a missing category to retain observations in the analyses. We evaluated interaction terms between prescription benefit status and other factors for significance at the P < .05 level. All multivariate tests of significance for categorical variables and interactions were for the entire constructs. Persons with Medicaid were dropped from models when testing for an interaction with: 1) out-of-pocket prescription medication expenditures owing to collinearity between Medicaid status and $0 out-of-pocket expenditures and 2) clinic type owing to small numbers of subjects with Medicaid attending network-model clinics. Because we had stratified our sampling on benefit status, health condition, and census block group income, we used SUDAAN software (Version 8.01, 2002) [19] to adjust our estimates for the complex survey sampling.

Results

As shown in Table 1, the 4492 eligible survey respondents were predominantly non-Hispanic white, reflecting the population of western Washington. Overall, 88% of respondents received care in group-model, owned clinics of GHC. Having a prescription benefit was associated with being older, more educated, and more affluent. The finding for age resulted from Group Health's dropping the availability of a prescription benefit for new, individual Medicare + Choice enrollees in 1994. Additional findings, not shown in Table 1, were that respondents reported taking a median of four chronic medications and had median out-of-pocket expenditures for prescription medications during the 2-year study period of $453 for persons with a private benefit and $706 for those with no benefit. Survey responses revealed that no more than 10 respondents had MediGap prescription coverage so this category is not represented in our tables and analyses.

Table 1.  Characteristics of study population (weighted) and prescription coverage
 % with no prescription coverage% with private prescription coverage (unweighted, n = 1525)% with Medicaid (unweighted, n = 198)
Unweighted (n = 2769)SE
  • Note: aP < .05, bP < .01, and cP < .001 for comparison of those with a private prescription benefit versus no prescription benefit.

  • *

    PCS score, Physical Component Scale of the Medical Outcomes Study Short Form-12.

  • MCS score, Mental Component Scale of the Medical Outcomes Study Short Form-12.

  • Out-of-pocket expenditures are missing for persons reporting obtaining at leasty some of their medications from non-GHC pharmacies.

Predisposing factors
 Age (years)c
  67–7029  19 15
  71–8461  72 70
  85 and older10  10 15
 Sex
  Male48  46 18
  Female52  54 82
 Race/ethnicitya
  African American 2   3  5
  Asian/Pacific Islander 4   4  7
  White92  91 81
  Other 2   7 
  Missing 1   1  1
 Educational attainmentb
  Less than high school17  13 25
  High school or more81  86 74
  Missing 2   2  1
Enabling factors
 Medicare + Choice coverage sourcec
  Individual enrollee, option to purchase prescription benefit301.8 42NA
  Individual enrollee, no option to purchase prescription benefit54  0NA
  Private employer-sponsored161.2 18NA
  Government employer-sponsored 0  40NA
  MedicaidNA0.0NA100
 Copayment for those with private coverage
  $0–5 for a 30-day supplyNA  38NA
  >$5 for a 30-day supplyNA  62NA
 Clinic typec
  Group model82  97 96
  Network model18  3  4
 Annual Income from all sourcesc
  Less than $20,000411.2 32 84
  From $20,000 to 50,000371.3 38  6
  $50,000 or more 7  10  1
  Missing152.2 20  9
 Assetsc
  Less than $10,000191.6 14 63
  $10,000 or more48  50  7
  Missing331.0 36 30
 Tenancya
  Own, no mortgage490.9 53 19
  Own, with a mortgage181.7 17 10
  Rent or other241.7 19 68
  Missing 9  11  4
Need factors
 Out-of-pocket medication expendituresc
  <$50/month39  79100
  $50–$100/month17   8  0
  >$100/month 8   0  0
  Missing36  13  0
 Mean self-reported health status
  PCS score*36.9  38.0 28.5
  MCS score50.9  51.8 47.5
 “In general, would you say your health is”a
  Excellent 2   3  0
  Very good14  14  5
  Good45  45 27
  Fair31  32 43
  Poor 8   6 25
 Hypertension
  Yes53  40 59
  No47  60 41
 Congestive heart failure   
  Yes23  19 30
  No77  81 70
 Coronary artery disease
  Yes23  17 29
  No77  83 71
 Diabetes
  Yes46  53 56
  No54  47 44

Table 2 presents the frequencies of reporting being unable to afford prescription medications and using various methods for coping with the costs of prescription medications. All of the outcomes were strongly associated with prescription benefit status, with a private prescription benefit decreasing the likelihood of all outcomes. Nevertheless, half of persons with Medicaid reported difficulty affording their medications despite having no copayments or benefit caps and all of the outcomes except travel to another country were more common among persons with Medicaid than those with a private benefit. “Stretching out medications to make them last longer” was the most commonly reported coping mechanism, cited by 12% of respondents overall, whereas only 3% of respondents reported obtaining medications in another country. Analyzing outcomes shown in Table 2 by type of clinic attended (group model vs. network model), we found substantial variation in receipt of free samples—25% of persons attending network-model clinics reported this compared to 4% of persons attending group-model clinics. Similarly, use of drug-company-sponsored medication assistance programs was more than three times more common among persons attending the network-model versus group-model clinics (2.8% vs. 0.8%). These differences are larger than those associated with benefit status shown in Table 2 for these outcomes.

Table 2.  Inability to afford medications and use of strategies to cope with prescription costs
 % of all respondents% of population without prescription benefit% of population with private prescription benefit% of population with Medicaid
  1. Note: All comparisons between persons with a private prescription benefit and those without a prescription benefit are significant at the P < .001 level.

Are you sometimes unable to afford your medicines?19.731.314.349.4
Do you ever stretch out your medicines to make them last longer (for example, take less than the prescribed amount?)12.219.4 9.319.4
Reports obtaining at least some of medications via
 Free drug samples from clinic or pharmacy 5.710.8 3.9 5.4
 Travel to other countries (i.e., Canada, Mexico) 3.0 6.2 1.9 0.6
 Drug-company sponsored medication assistance programs 0.9 1.8 0.5 4.7
 Using someone else's insurance coverage 0.3 0.9 0.1 1.2

Table 3 presents results of multivariate logistic models for having difficulty affording medications and stretching out medications to make them last longer. Income had the strongest association with difficulty affording medications and stretching out medications. Having limited financial assets was also significantly associated with both outcomes. Having a private (non-Medicaid) prescription benefit, whether self-purchased or employment-related, was protective in both cases, although individual enrollees who had not had the option of purchasing a prescription benefit were not significantly more likely to report stretching out medications than those with a benefit. After adjustment, persons with Medicaid were no more likely than those with other prescription benefits to report stretching out their medications; adjusting for income and assets in the models produced this change. Nevertheless, persons with Medicaid remained as likely as those without a benefit to report trouble affording their medications. Higher out-of-pocket payments for medications over the 2 years of the study were associated with greater likelihood of reporting difficulty affording medications but not with reporting stretching out medications.

Table 3.  Multivariate models of being unable to afford and stretching out medications
 Sometimes unable to afford medicationsStretching out medications to make them last longer
OR (95% CI)P valueOR (95% CI)P value
  • *

    PCS, Physical Component Summary of the SF-12, here divided by 10 for scaling.

  • MCS, Mental Component Summary of the SF-12, here divided by 10 for scaling.

Predisposing factors
 Age (years) 0.00 0.09
  67–741 1 
  75–840.57 (0.40–0.81) 0.71 (0.52–0.97) 
  85 and older0.36 (0.22–0.61) 0.42 (0.17–1.06) 
 Female sex1.27 (0.96–1.69)0.090.94 (0.67–1.31)0.70
 Racial/ethnic background 0.05 0.38
  Non-Hispanic white1 1 
  Other1.49 (1.02–2.17) 1.27 (0.77–2.09) 
  Missing1.07 (0.43–2.67) 1.44 (0.70–2.98) 
 Marital status 0.13 0.17
  Married/partnered1 1 
  Widowed0.73 (0.50–1.07) 1.16 (0.77–1.75) 
  Divorced/separated0.57 (0.38–0.85) 0.80 (0.54–1.21) 
  Never married0.61 (0.26–1.41) 0.42 (0.24–0.72) 
  Missing0.99 (0.34–2.82) 0.90 (0.52–1.54) 
 Educational attainment 0.21 0.58
  Less than high school0.94 (0.69–1.27) 0.84 (0.50–1.40) 
  High school or more1 1 
  Missing2.35 (1.06–5.15) 1.24 (0.49–2.14) 
Enabling factors
 Prescription coverage status 0.002 0.01
  No benefit, no choice to purchase2.06 (1.26–3.38) 1.40 (0.80–2.43) 
  No benefit, had choice to purchase2.12 (1.33–3.41) 2.31 (1.32–4.02) 
  Benefit, employer-sponsored1.01 (0.58–1.77) 0.81 (0.44–1.50) 
  Benefit, chose to purchase1 1 
  Medicaid1.90 (1.11–3.26) 1.06 (0.57–1.99) 
 Clinic type 0.43 0.05
  Group model1 1 
  Network model1.16 (0.80–1.66) 1.60 (1.01–2.54) 
 Annual income from all sources <0.0001 0.02
  Less than $20,00019.4 (10.5–35.7) 4.18 (2.24–7.82) 
  From $20,000–50,0007.50 (4.2–13.3) 2.97 (1.69–5.21) 
  $50,000 or more1 1 
  Missing7.72 (4.20–14.2) 4.21 (1.92–9.20) 
 Estimated asset value 0.00 0.04
  Less than $10,0002.24 (1.48–3.37) 1.69 (1.09–2.60) 
  $10,000 or more1 1 
  Missing1.11 (0.80–1.54) 0.98 (0.61–1.59) 
 Tenancy 0.007 0.009
  Own, no mortgage1 1 
  Own, with a mortgage1.90 (1.31–2.75) 1.72 (1.14–2.59) 
  Rent or other1.27 (0.84–1.92) 1.41 (0.90–2.23) 
  Missing1.18 (0.79–1.75) 0.57 (0.32–1.00) 
Need factors
 Out-of-pocket payment for prescriptions 0.04 0.60
  <$50/month1 1 
  $50–100/month1.50 (1.02–2.20) 1.26 (0.79–1.99) 
  >$100/month1.79 (1.35–2.37) 1.15 (0.88–1.50) 
  Missing1.23 (0.88–1.73) 1.11 (0.74–1.65) 
 PCS/10*0.77 (0.68–0.87)<0.00010.80 (0.69–0.92)0.002
 MCS/100.74 (0.65–0.84)<0.00010.73 (0.62–0.87)0.0004
 Hypertension1.26 (0.98–1.61)0.070.79 (0.56–1.11)0.17
 Congestive heart failure0.86 (0.64–1.17)0.330.84 (0.53–1.33)0.45
 Coronary artery disease1.23 (0.92–1.67)0.170.86 (0.60–1.25)0.43
 Diabetes1.33 (0.95–1.85)0.090.93 (0.65–1.34)0.71

Worse self-rated physical and mental health status were both associated with higher risks of both outcomes in the models. To aid the interpretation of these odds ratios for PCS and MCS scores, in Fig. 1 we graphically present multivariate-adjusted proportions of persons reporting being unable to afford medications according to global self-rated health status. After adjustment, 24% of subjects with fair or poor health status reported difficulty affording their medications, compared to 4% of those in excellent health.

Figure 1.

Adjusted* proportions of persons reporting being unable to afford medications by health status (*adjusted for age, sex, race/ethnicity, marital status, education, prescription benefit status, income, wealth, out-of-pocket prescription drug payments, and presence of study health conditions). Error bars indicate 95% CIs.

We also sought to better understand the receipt of free sample medications using multivariate modeling. We evaluated separate models for persons attending group-model and network-model clinics owing to their very different rates of receipt of free samples. We found no significant predictors of free sample receipt among persons attending group-model clinics. With multivariate adjustment, 14% (95% CI, 5%−24%) of network-model clinic attendees with a prescription benefit reported receiving free samples versus 38% (95% CI, 35%−42%) of those without a benefit. Other significant predictors of free sample receipt were lower income, owning a home with a mortgage, worse physical health status, and being female; out-of-pocket costs could not be modeled owing to the large amount of missing data (data not presented in the tables).

We also tested whether there was significant effect modification by prescription benefit status for the observed associations with other factors. There were no significant findings for our income, wealth, and health measures. The only significant effect modification was with clinic type for difficulty affording medications. After adjustment, excluding persons with Medicaid because there were almost none in the network-model clinics, and using a benefits status indicator that, because of small cell sizes, did not include whether or not a subject had a choice about purchasing prescription coverage, 16 and 23% of persons with and without a benefit, respectively, in the group-model clinics reported difficulty affording medications, whereas in network-model clinics, the comparable figures were 6 and 29% (P = .02; data not shown in the tables). Including this interaction term did not appreciably change other findings.

Discussion

In this group of Medicare recipients with significant chronic health conditions enrolled in a primarily group-model HMO in Washington state, we found a moderate incidence of problems affording medications and of using strategies to deal with these costs. These problems were more common among poorer, sicker persons and those without prescription coverage. Nevertheless, income, wealth, and health status did not influence the relationship between prescription benefit status and the dependent measures. Approximately one-fifth of the respondents reported difficulty affording their medications some of the time to all of the time. “Stretching out” medications to make them last longer was the most commonly reported method of coping with the high cost of prescription medications. This is consistent with findings from two other recent surveys of Medicare beneficiaries [9,20] and a survey of low-income African Americans residing in rural Georgia [21].

The multivariate models showed that similar factors were related both to reporting being unable to afford medications and to stretching out medications. Having a private prescription benefit appeared protective for both of these outcomes. As expected, lower income and lower assets were associated with both of these outcomes. Nevertheless, we did not expect such a large income effect among persons with a prescription benefit, given that the Group Health benefit has no cap and includes relatively modest copays. This raises the question of whether even modest copayments are a significant barrier to obtaining needed medications for older Americans with limited income and/or assets and chronic health conditions. Within the relatively narrow range of copayments in this HMO, we found no relationship between size of copayment and our outcomes; too few of our respondents had full, non-Medicaid coverage with no copayment to let us assess whether income was significant for this group and whether outcomes were better with no copayments. Modest copayment increases have been shown in some studies to decrease drug utilization among HMO enrollees [22,23], although this does not appear to be uniformly true [24].

It is also worrisome that persons reporting worse physical and mental health status were more likely to report being unable to afford and stretching out their medications, even after adjusting for benefits, income, assets, and out-of-pocket costs. These findings suggest that the sickest persons, who are probably at greatest risk of complications from poor adherence to their medication regimens, are also at greatest risk of having poor adherence to their medications. Although we do not understand why this should be the case, this finding is consistent with other studies [9,25,26] and suggests that, if a prescription benefit program for Medicare beneficiaries is created that is not structured as a universal entitlement, it should be targeted not only at low-income seniors but also at those in poorer health, regardless of income and assets. Nevertheless, creating an equitable system that would target seniors based on health condition/health status as well as income (and, perhaps, assets) would be a daunting task.

We had anticipated that other coping strategies would be used more frequently by our respondents. Given how much attention has been paid to purchasing medications outside of the United States and the residence of most of our study population within 150 miles of the Canadian border, we had expected that obtaining prescription medications in Canada would be reasonably common. Ten percent of the Arizona Medicare HMO members in the report of Cox et al. [9] reported buying some of their medications in Mexico and 38% reported using free samples. It may be that purchasing medications in Canada has increased in frequency in our population since our survey because the potential cost savings have recently received substantial publicity; for example, at least two commercial services taking people to Canada on buses to obtain their prescriptions have started since the time of our survey [27]. It may also be that enrollees in this HMO were less likely than other Medicare beneficiaries living near a border to seek medications outside the United States.

Compared to persons receiving care in other settings, these GHC enrollees do appear to have substantially lower than expected drug costs—only 0.2 and 13% of respondents with and without a prescription benefit for whom we had complete prescription data, all of whom had one or more chronic conditions, had over $100/month in out-of-pocket medication costs versus 17 and 43% of randomly selected seniors in a survey of Medicare beneficiaries in eight states [20] and 12% (with partial coverage) and 26% (no coverage) in another study [25]. This lower drug spending could be a consequence of the factors in GHC tending to lower overall drug costs, including a restricted formulary emphasizing generic medications, active detailing of providers about cost-effective prescribing, and the HMO's tradition of selling “essential” drugs, such as those for high blood pressure and diabetes, at cost in on-site pharmacies to persons without a prescription benefit. Thus, our subjects may have felt less financial pressure to reduce their prescription drug spending than other Medicare beneficiaries.

Use of some coping mechanisms was strongly associated with receiving care in group-model versus network-model clinics. This presumably reflects different styles of practice in owned, group-model practices versus independent, community practices. Employed physicians working in owned clinics may be more likely to assume that patients can obtain medications from on-site pharmacies, whereas community physicians seeing patients with different types of insurance or no insurance may be more aware that patients may have difficulty obtaining prescribed medications. This could also represent greater access by drug detail representatives to community than salaried HMO physicians. We cannot determine from our data which medications were given as samples and whether the higher use of free samples in network-model clinics was beneficial to patients or not. As pharmaceutical companies supply samples of expensive, proprietary medications and free samples tend to consist of small quantities supplied at irregular intervals, providing such samples to the vulnerable seniors in our study seems more likely to contribute to confusion and poor adherence than to improved control of chronic conditions [28–30]. As drug company-sponsored medication assistance programs yield substantial supplies of medications, they may be more likely to have benefited their recipients than free samples [30,31], but we suspect that the substantial administrative barriers to participating in many of these programs limited their use.

We had anticipated that dually eligible persons with Medicaid would report few problems obtaining their medications, given that, in Washington state, Medicaid provided first dollar coverage for medications with no caps on number or cost of medications. Adjustment for income and assets eliminated the difference between Medicaid recipients and other persons with a prescription benefit for stretching out medications, but not for being unable to afford medications. While unexpected, this latter finding is consistent with a recent report that Medicaid recipients aged 18 to 64 reported substantial problems affording their medications [32]. Whether Medicaid recipients in our study responded to questions about medication affordability based on problems before obtaining Medicaid, problems affording other basic needs, difficulty obtaining medications owing to regulations limiting fills to a 1-month supply, with early refills requiring full payment, or other factors is not known and deserves further investigation.

This study has a number of limitations. First, the study population consisted of persons who chose to enroll in a predominantly group-model Medicare HMO for at least 2 years and who had at least one of the selected chronic health conditions—it cannot be assumed to be representative of the Medicare population in general. Nevertheless, there is no reason to believe that, other than having less ethnic diversity than the overall US population, our study population is markedly different from other US seniors with the same chronic conditions enrolled in Medicare + Choice plans. Second, prescription drug costs faced by survey respondents are lower than for many Medicare respondents with the selected medical conditions for several reasons, as discussed above. These lower costs should tend to minimize differences between those with and without a prescription benefit, making our estimates conservative. Third, respondents might have minimized their reports of some strategies, such as using someone else's medications. Nevertheless, this should not bias our other findings. Fourth, very few of our respondents indicated they had Medigap prescription coverage so we cannot make any assessments of the associations of this type of prescription benefit with our outcome measures. Fifth, as an observational study, our findings cannot prove causality and could be biased if having or lacking coverage is strongly driven by endogenous factors. Nevertheless, for individual enrollees difficulty affording medications was associated with prescription benefit status and not whether or not they had a choice about purchasing a benefit; the greater likelihood of reporting stretching out medications by persons who had the option of purchasing prescription coverage and did not do so than those who did not have the option suggests that this purchase decision may be motivated more by financial than attitudinal factors. Additionally, for those with employment-related Medicare + Choice coverage, type of employer (government vs. private industry) was associated with likelihood of having prescription coverage but not with our outcomes. Sixth, some of our variables have significant amounts of missing data. The association of attending network-model clinics with missing status for out-of-pocket spending and our findings in Table 3 indicate that the data are not missing at random and, thus, multiple imputation techniques to address this could introduce bias that could be greater than our approach of including categories for missing status in our analyses.

Our findings indicate that a variety of factors are important for access to prescription medications by seniors with common, chronic health conditions in addition to prescription benefit status. Reducing financial barriers through Medicaid coverage or a generous Medicare + Choice prescription benefit and low drug costs did not eliminate strong associations of lower income, lower wealth, and worse health status with more access difficulties. To provide the greatest overall public health benefit, efforts to improve access of Medicare beneficiaries to prescription drugs equitably should address not only financial access via prescription benefits but also other access barriers, particularly for indigent seniors and those in poor health.

We are indebted to Ella Thompson for assistance with surveying study participants.

This work was supported by Grant HS10318 from the Agency for Healthcare Research and Quality and a grant from the Robert Wood Johnson Foundation's Changes in Health Care Financing and Organization (HCFO) Initiative. Barry Saver and Mark Doescher were partially supported by Advanced Research Training Grants from the American Academy of Family Physicians.

Ancillary