James W. Davis, PhD

Diabetes is one of the most common and costly diseases, affecting 25.8 million people, or 8.3% of the US population.

 Diabetes is also the seventhleading cause of death in the United States and can result in serious health complications including heart disease, stroke, kidney failure, blindness, and lower-extremity amputations. In 2007, the estimated cost of diabetes in the United States was $1.7 billion, with direct medical expenditures for people with diabetes averaging nearly 2.4 times that of people without the disease.

Glycated hemoglobin (A1C) is one of the most common measurements used in the assessment of glycemic control; it is thought to reflect the average glycemic control over several months

 and predict the occurrence of diabetes-related complications.

 The most recent guidelines put forth by the American Diabetes Association recommend lowering A1C to below or around 7%.

 This recommendation was based on findings from multiple trials performed in both type 1 and type 2 diabetes patients demonstrating significant decreases in microvascular and neuropathic complications associated with reductions in A1C.

 In addition to the microvascular benefits of intensive glycemic control, both the Diabetes Control and Complications Trial Research Group and UK Prospective Diabetes Study demonstrated the potential of intensive glucose control to lower the risk of cardiovascular events.

The percentage of diabetic patients with poor glycemic control (A1C >9.0%) fell from 21.0% in 1999 to 2000 and 17.8% in 2001 to 2002 to 12.4% in 2003 to 2004.

 Despite these encouraging trends, many patients still struggle to achieve and/or sustain optimal glycemic control over extended periods of time (ie, years).

Some prior evidence suggests that having an A1C level of less than 7% is associated with lower costs of medical care.

 However, many of these studies examined crosssectional relationships that may be confounded by severity of disease. Although all studies required a diagnosis of diabetes, there may have been some false positives (ie, people with a diagnosis and A1C <7% who do not really have diabetes). Also, some people have diabetes of low severity and would not be expected to have high costs. To address these issues, our study started with a sample of patients with poor initial control (A1C >9%) and used propensity score matching to defi ne a comparable cohort.

The purpose of this study was to examine the relationship between sustained glycemic control and total direct healthcare costs among patients with diabetes with an initial A1C level greater than or equal to 9%. Specifically, this study sought to test the hypothesis that lowering A1C from above 9% to below 7%—and sustaining this level&mdash;reduces healthcare costs among patients with diabetes mellitus.

We conducted a retrospective analysis of administrative data from patients with diabetes enrolled in a large health plan in Hawaii. To be included in the study, patients needed to meet the following criteria: (1) be identified as having diabetes, by having either 2 or more claims for type 2 diabetes in medical claims (

International Classification of Diseases, Ninth Revision, Clinical Modification

] code 250.xx) or at least 1 prescription for an oral hypoglycemic agent and/or insulin; (2) be at least 18 but under 75 years old; (3) be enrolled with medical and drug coverage; and (4) have at least 1 A1C value greater than 9% in 2006. This level was chosen because the National Committee on Quality Assurance’s Healthcare Effectiveness Data and Information Set uses a level of 9% to indicate poor A1C control.

Patient information including age, sex, isle of residence, and type of coverage (health maintenance organization, preferred provider organization, Medicare cost contract) was obtained from administrative data. Patient morbidity level was determined by using 

 codes according to the Johns Hopkins Adjusted Clinical Group methodology; levels of 4 or 5 on the 5-point scale were considered high morbidity.

In addition, using disease management algorithms, we created dichotomous variables to identify patients with coronary artery disease and congestive heart failure because these are conditions known to be prevalent in patients with diabetes and to increase costs. Diagnoses of diabetes, coronary artery disease, and congestive heart failure were confi rmed whenever possible through contacting members and their physicians. A physician’s confirmation was required to exclude false positives.

During the baseline year (2006), we identifi ed all patients with diabetes who had a A1C level of greater than 9% (4667 out of 56,921 patients with diabetes). For each subsequent year (2007-2009), we calculated mean A1C levels for these patients. We used propensity scores

 to identify a comparable control cohort for those with A1C of less than 7% in 2007 using demographic and utilization data, including age, sex, isle of residence, type of insurance coverage, comorbid conditions, diabetes duration, number of distinct medications, and morbidity level. We conducted cross-sectional analyses, comparing the average annual direct medical costs of patients whose mean A1C level was less than 7% with the costs for those who did not meet this glycemic level in the matched sample (4093 observations for 1304 individuals).

For longitudinal analyses, we created a dichotomous variable indicating whether or not the patient had sustained A1C control at target levels (A1C <7%) for all 3 years (2007-2009). Of the 1304 individuals in the matched sample, 518 were enrolled, had A1C values for all 4 years (2006 through 2009), and were included in the longitudinal analyses of cost change.

We calculated average baseline costs of care (2006) for these patients and subsequent average annual costs in years 2007 through 2009 by using medical claims data. Costs included direct medical expenses paid by the health plan. We analyzed total costs and cost broken into cost categories: facility, physician services, and pharmaceutical. All costs in the study were adjusted to constant 2009 dollars using the medical care component of the Consumer Price Index. For samples matched using propensity scores, we compared costs (total and by category) in a given year for patients with A1C levels less than 7% with the costs for patients with higher A1C levels. We also examined differences in cost changes for patients who were able to sustain A1C levels of less than 7% for 3 years compared with patients who had A1C levels higher than 7% for at least 1 year.

Next, to account for year and baseline costs, we used generalized estimating equations with robust standard errors to compare annual average healthcare costs by cost category for patients at recommended levels (A1C <7%) with the costs for patients with A1C levels at or above 7% for the matched groups.

Interaction terms between year and A1C level were included to account for differences in the relationship over time.

Generalized linear models with a negative binomial distribution and log link were used to examine the relationship between sustained control (A1C <7% for 3 years) and direct medical costs according to cost category (total, inpatient facility, physician services, and pharmaceutical) in 2009, adjusting for baseline costs. Inpatient costs included costs of all facilities such as hospitals and long- term care institutions. Physician services included both inpatient and outpatient physician reimbursement. Institutional review board approval was obtained from the University of Hawaii. All analyses were conducted with Stata statistical software, release 11 (StataCorp, College Station, Texas).

Of members with A1C levels greater than or equal to 9% in 2006 (n = 4667), the percentage of members able to achieve mean A1C levels of less than 7% were 10.0% in 2007, 12.7% in 2008, and 13.3% in 2009. Only 3.6% of members with A1C levels greater than 9% in the baseline year were able to reduce their A1C level to goal (<7%) and sustain this level for 3 years.

Although there were signifi cant differences in 2007 between those achieving an A1C level of 7% in the unmatched sample (n = 4667) in terms of age, isle of residence, type of coverage, and duration of diabetes, there were no significant differences in demographic characteristics and utilization for the groups matched using propensity scores (n = 1304, 

For the matched sample in 2007, the mean age of those whose A1C level was less than 7% was 54 ± 0.53 years, and the mean age for those not at recommended levels was 54 ± 0.41 years. Forty-three percent of patients in both groups were female (Table 1). Most patients were in the preferred provider organization (67% of those with A1C <7% and 66% of those with A1C >7%). The average duration of diabetes was 6.9 years for both groups, and patients in both groups were taking 9 distinct medications for a year on average. Approximately 44% of patients at target had high morbidity compared with 41% of those whose A1C level was greater than or equal to 7%. With respect to comorbid conditions, 13% of patients in both groups had congestive heart failure, and 23% of patients whose A1C level was less than 7% had a history of coronary artery disease compared with 24% of patients not at recommended A1C levels. This suggests that the propensity matching was successful in creating groups that had similar observable characteristics.

Difference in Average Costs and Cost Change

In cross-sectional comparisons, the average annual costs for patients whose A1C level was less than 7% was $14,821 compared with $12,108 for the matched sample of patients whose A1C level was greater than or equal to 7%, for a difference of $2713 (95% confi dence interval [CI], $285-$5140) (

). In contrast, when we examined the change in cost from 2006 to 2009 for patients who had sustained levels of A1C at less than 7% for all 3 years, we found that total cost of care for patients with sustained control decreased by $2207 compared with a $3006 increase for patients without sustained control, for a difference of —$5214 (95% CI, –$10,163 to –$264).

) are consistent with fi ndings in Table 2. In the panel data analyses, costs were higher for patients whose mean A1C level was less than 7% compared with patients who had higher mean A1C values. For total costs, the incidence rate ratio (IRR) was estimated at 1.35 (95% CI, 1.28-1.44). These results were consistent across all cost categories (Table 3).

In contrast, when examining the impact of duration of A1C control on cost, we found that patients who had 3 years of sustained A1C control (A1C <7%) had smaller cost increases than patients without sustained control (IRR = 0.71; 95% CI, 0.59-0.86; Table 3). Cost ratios of patients with sustained control were signifi cantly less than 1 for all cost categories except for pharmaceutical costs, which did not differ between groups.

Fewer than 4% of the study population who had poor glycemic control in 2006 (A1C >9%) were able to reduce their A1C level to less than 7% and sustain this level for 3 years (2007-2009); however, achievement and maintenance of glycemic control were associated with lower total healthcare costs in 2009. To our knowledge, our study is the first to examine the impact of sustained A1C control on costs and to use propensity score matching to minimize potential confounding.

Although cost increases in pharmaceutical treatment were similar in both groups, cost reductions for both physician and inpatient services were found for the group with sustained A1C control. Prior evidence has shown that complications are key drivers of the direct medical costs in patients with diabetes.

 Hence, some of the cost reductions seen in our study with sustained glycemic control may be due to a reduced incidence of diabetes complications and morbidity.

Our analyses of panel data reveal, however, that these cost savings are not immediate. In our study, during any given year that patients achieved A1C levels of less than 7%, their costs were signifi cantly higher than those for patients whose A1C levels were greater than or equal to 7%. We acknowledge that even though the group that sustained good glycemic control had cost savings of more than $5000 from 2009 to 2006, some of these savings could be counterbalanced by cost increases in the intervening years, as additional physician visits and medications might be needed to achieve this control. The costs are immediate, but the benefits may take years to be realized. Even so, it is likely that costs will continue to be lower in subsequent years if patients maintain glycemic control.

These findings contrast with those of several prior studies that have found short-term cost savings. Shetty and colleagues11 divided a cohort of patients with diabetes in a managed care setting into 2 groups, those at the target A1C level (

7%) and those above the target A1C level (>7%), and examined cost savings after 1 year of follow-up. They found substantial cost savings of 32% for patients at target A1C level. In another study, grouping patients with type 2 diabetes by glycemic control (good [A1C

9%], and poor [>9%]), Oglesby and colleagues

 found diabetes-related costs to be 16% lower for patients with good control than for patients with fair control and 20% lower than for patients with poor control.

 examined medical charges related to A1C and found that after controlling for demographics and cardiovascular disease, costs increased by 30% as A1C increased from 6% to 10%.

Our findings of increased costs for the patients whose A1C level was less than 7% within a year differed from the results of these prior studies. We believe the main difference was that we required patients to have elevated A1C levels (>9%) at baseline, while these other studies did not. Hence, those researchers may have had in their at target group patients with fairly “normal” A1C levels even without treatment. It may also be that initial interventions for patients with poor control that drive up immediate costs were not evaluated in these cross-sectional cohort studies.

A subsequent study by Gilmer and colleagues

 reported that in a large Minnesota health plan, higher A1C in patients with either type 1 or type 2 diabetes predicted higher 3-year total healthcare costs for patients with A1C greater than 7.5%. This study differed from ours in that it did not examine the impact of changes in A1C levels. Both studies by Gilmer and colleagues emphasized that cardiovascular disease was a stronger predictor of costs than glycemic control. We did not examine the association between cardiovascular disease and costs in our study.

 examined a cohort of patients with diabetes in a managed care setting. The investigators found the odds of having at least 1 diabetesrelated hospitalization were not significantly associated with higher mean A1C levels except for patients with a mean A1C level of at least 10%. They did find that for hospitalized patients, mean costs were higher among patients with higher mean A1C levels. However, because they did not examine total costs of care or changes in costs related to changes in A1C levels, we cannot directly compare their findings with ours.

was the most similar to ours in that it focused on changes in cost related to changes in A1C levels. They examined data for diabetic patients enrolled in a staff model health maintenance organization in the mid-1990s. The investigators defined improvement as having at least a 1% decrease in A1C levels. They found costs for the improved group were lower each subsequent year. but differences were statistically significant only for those with baseline A1C levels of greater than or equal to 10%. Our study design differed from theirs in that we examined costs for patients who dropped their A1C levels from greater than 9% to less than 7%. Hence, we required a higher level of improvement (2 percentage point reduction) as well as sustained control, and found that patients whose initial A1C levels were greater than 9% experienced significant cost reductions.

Our study may underestimate potential savings from A1C reduction to target levels. Tissue damage from inadequate glycemic control may not be promptly overcome or mitigated. Clinical trials have demonstrated that approximately 8 years are needed to realize fully all of the microvascular benefi ts of glycemic control.

 The UK Prospective Diabetes Study demonstrated that maintaining a lower A1C (7%) over 10 years is associated with a substantially reduced risk of microvascular complications.7 Hence, further risk reduction and cost savings may be observed within an extended study period (8-10 years).

This study had several limitations. First, it was conducted with enrollees of a single health plan in Hawaii. The percentage of patients in poor control (A1C >9%) in 2006 (8%) was lower than the national average, so our results may not be generalizable to other areas or uninsured populations. Second, we only estimated direct medical costs covered by the health plan. Free drug samples distributed by physicians, use of aspirin, and other noncovered costs would not be included in our analyses. Adding indirect measures such as productivity gains and decreased absenteeism would most likely have increased the cost savings for the group with sustained glucose control.

 We were also not able to separate out diabetes-specific costs.

Third, we relied on administrative data from a health plan to identify patients with diabetes and to estimate costs. The possibility of false positives arising from identifi cation based on claims data was diminished in our study as we required a diagnosis of diabetes and an A1C value greater than 9% at baseline. Fourth, this study also did not examine the quality-of-life implications, which could be substantial. Fifth, only 3% of patients were able to achieve target level A1C for 3 years. This group may have been different in other ways. For instance, they may have eaten healthier diets, smoked less, and exercised more than those not able to sustain control. These factors may have been as important to sustained glycemic control as medical interventions, but the costs would have primarily been absorbed by the patients and they were unmeasured in this study. Future research is needed that takes a more comprehensive approach to measuring factors that might contribute to cost changes.

In our study of 1304 managed care enrollees with poor initial A1C control, we found that patients who were able to achieve target goals had lower costs after 3 years compared with patients not able to sustain these target levels.

Quality indicators for health plans and physicians often include A1C screening and reaching target levels. Our study suggests that while achieving these target goals may not immediately result in cost reductions, sustained A1C control is associated with lower costs in a 3-year time frame, suggesting that efforts to support and reward physicians and health plans for achieving these target goals may be a good investment in the intermediate term.

Articles

Total cost of care for patients whose glycated hemoglobin levels were 

Association Between Sustained Glycated Hemoglobin Control and Healthcare  Costs

Hypertension, dyslipidemia, and diabetes mediate cardiovascular disease in many patients, and patients with 2 or 3 of these conditions face the greatest risks.

 All 3 conditions are potentially controllable with medications.

A number of studies have demonstrated associations between good adherence and lower risks in routine care settings. One study, for example, reported that patients taking medications for hypertension for 80% or more of the days of supply attained substantially improved blood pressure control compared with patients who adhered 50% or less of the time.

 A second study reported that appropriate drug treatment for blood pressure reduced the incidence of stroke by 30% to 40% and the incidence of myocardial infarction by 20% to 25%.

 A report on lipid-lowering medications estimated that patients with good adherence reduced their serum levels more than patients with poor adherence (30% vs 11%).

 Ho et al observed that patients with diabetes who had good medication adherence had fewer all-cause hospitalizations and reduced mortality.

Despite the accepted importance of high medication adherence as expressed in national guidelines, patients often take medications irregularly, or start and subsequently stop their medications.

 The chronic nature of hypertension, dyslipidemia, and diabetes coupled with the absence of overt symptoms and the costs of medications may contribute to the low adherence. One study of statins for cholesterol control found that only about 60% of patients had high adherence 12 months after starting treatment.

Only 26% of the patients remained adherent after 5 years. A study of hypertension medications in 13 managed care organizations reported that a fourth of patients taking drugs for hypertension did not maintain high adherence.13 A review of studies of patients with diabetes reported drug adherence ranging from 36% to 93% depending on the population and the type of medication.

A still unresolved issue is whether patients taking a greater number of medications have increased or decreased adherence. Some studies found patients taking more medications were less adherent,

 whereas other studies reported that patients taking more medications had the highest adherence.

 Patients with multiple conditions that require long-term management face the greatest challenge in maintaining adherence.

In this study we investigated patients taking medications for hypertension, dyslipidemia, and diabetes. The study population consisted of patients who took drugs for 1 or more of the conditions and who were enrolled with the largest insurer in Hawaii between 2004 and 2008. The objectives were to investigate by the combinations of medications patients were taking (1) trends in adherence over time, (2) relative medication adherence, and (3) the consistency of individual patient adherence from year to year.

The study population consisted of patients enrolled with the largest insurer in Hawaii who had hypertension, dyslipidemia, and/or diabetes and who filled a prescription for a medication for 1 or more of the 3 conditions between 2004 and 2008. The insurer covers approximately 650,000 lives, or about half of the state’s nonmilitary population. The study used existing data without patient identifiers and was granted an exemption from review by the University of Hawaii Institutional Review Board.

As a primary outcome, we calculated medication possession ratios, which were defined as the proportion of days enrolled in a year that were covered by a filled prescription. Medical possession ratios were calculated separately for medications for hypertension, dyslipidemia, and diabetes. If more than 1 prescription for a condition was filled on the same day, the day was counted only once toward the numerator. Patients with a medical possession ratio of 80% or higher were considered adherent. This definition has often been used in studies of drug adherence and has been shown to measure adherence with medications that require long-term use.

 With administrative data, however, filled prescriptions do not necessarily mean continuous use, as patients may not take all of their purchased medications. Patients were included in analyses for calendar years in which they had no more than a 45-day gap in enrollment. This choice was made to measure adherence among patients enrolled long enough to reflect their usual adherence during the year.

The analyses included regression models adjusted for multiple variables that might explain difference in adherence by the combinations of medications taken. Age, sex, island of residence, and morbidity were included to adjust for patient characteristics. Morbidity was also measured, based on claims submitted during a calendar year using the Johns Hopkins Adjusted Clinical Groups Case-Mix System (www.acg.jhsph.edu). Calendar year was included to adjust for possible temporal trends. Health plan was adjusted for, so that members within the same health plan would be compared.

For analyses, the frequency of adherence was summarized by year and by the combinations of the medications taken. Analyses also summarized the proportion of patients adherent in the current year by whether or not they had been adherent in the past year. Adherence adjusting for patient characteristics was modeled using negative binomial regression to provide relative risks. Interaction terms between age and morbidity with medication cohorts were included in some regression models. Regression models were fit using generalized estimating equations to correct for correlations among repeated measurements within people. Most patients had multiple years of adherence data. The same methodology was used to model adherence in the current year as a function of adherence in the previous year.

A total of 148,119 people qualified for the study. At the start of their first year in the study the population was primarily 35 years and older. Only 2.5% were aged 20 to 34 years, whereas 20.1% were aged 35 to 49 years, 38.7% were aged 50 to 64 years, and 38.7% were 65 years and older. Slightly over half (51.7%) were female, and the majority (73.3%) lived on Oahu, the most populous of the Hawaiian Islands. About one-third had a high morbidity index (32.6%), and 21.6% had heart disease. In their first year in the study, 48.2% of patients were taking medications for a single illness (32.4% antihypertensive medications, 19.7% lipid-lowering medications, and 6.1% medications for diabetes). Of the other participants, 21.8% were taking antihypertensive and lipid-lowering medications, 5.7% were taking antihypertensive and diabetes medications, 3.3% were taking lipid-lowering and diabetes medications, and 11.1% were taking antihypertensive, lipid-lowering, and diabetes medications.

Between 2004 and 2008 the percentage of patients who adhered with medications for hypertension, hyperlipidemia, and diabetes remained nearly constant (

). Adherence with the individual medications, however, varied from less than 65% to nearly 80% by the combinations of other medications taken. More patients were adherent to diabetes medications than to other medications, and the fewest patients were adherent to lipid-lowering medications. Adherence with antihypertensive medications was intermediate. Among patients taking a particular medication such as an antihypertensive medication, however, adherence varied by what other medications they were taking. Among patients using medications to lower lipid levels, for example, those also taking medications for both hypertension and diabetes had the highest adherence, followed by patients taking hypertensive and lipid-lowering medications. The next highest adherence rate was among patients taking diabetes and lipid-lowering medications, followed by patients taking only medications to control their lipid levels.

As a general observation, patients taking 1 of the 3 types of medication had higher adherence if they also were taking a second type of medication, and the highest adherence if they were taking all 3 types. A group with especially low adherence was patients taking diabetes medications alone; their adherence averaged less than 60%.

The observed differences in adherence might have resulted from demographic factors such as older patients having more prescriptions and being more prone to take their medications. To address this possibility, we analyzed associations between the medications taken and adherence with adjustment for patient characteristics (

). The order of adherence by combinations of medications after adjustment remained the same as that observed in the unadjusted frequencies. Of the adjusted results, the highest relative risks occurred with diabetes medications as a consequence of the especially low adherence of the

reference group, patients who were taking medications only for diabetes.

Other analyses examined possible differences in the strengths of associations by comparing members within medication cohorts by age group (65 years and older versus younger ages) and by high compared with low morbidity. The older patients within a medication cohort tended to be adherent less frequently. As an example of one of the stronger differences, patients with hypertension, hyperlipidemia, and diabetes who were younger than 65 years had 11% higher adherence than older patients (95% confidence interval [CI] 8%, 19%). Patients with low morbidity tended to be adherent less frequently than those with high morbidity. For example, patients with hypertension, hyperlipidemia, and diabetes and low morbidity had 8% lower adherence than patients with high morbidity (95% CI 6%, 10%). More than half of the differences by age and morbidity, however, were less than 5%, and not all were statistically significant.

Adherence was examined further by measuring the consistency of adherence of individual patients from year to year (

). Results mirrored the results for adherence by calendar year in several respects. Patients taking medicines for more conditions were less likely to stop adherence in the next year if they had been adherent in the past year, and more likely to start adherence if they had not been adherent in the past year. The most favorable pattern was evident among patients taking medications for hypertension, hyperlipidemia, and diabetes. The percentage of patients remaining adherent if they had been adherent in the past year by the various medications taken ranged from 64% to 88%. The percentages for becoming adherent if they had been nonadherent in the past year ranged from 17% to 47%.

Changes in adherence from year to year were examined further in regression models adjusted for patient characteristics (

). Patients taking lipid-lowering, antihypertensive, and diabetes medications together were the least likely to stop adhering with any of the medications. Patients taking medications for a single illness were the most likely to stop. Among patients nonadherent in the past year, the largest percentage who did adhere in the current year were those taking all 3 medications and the smallest percentage were those taking medications for a single illness. As with adherence, the strength of associations within medication cohorts was examined by comparing younger and older ages and high versus low morbidity. The only consistent pattern observed was for patients taking medications for diabetes. Healthier members were less likely to become adherent if they had been nonadherent in the past year and more likely to become nonadherent if they had previously been adherent. The magnitude of differences ranged from 10% to 19%.

Our results show first that patients had a lower average adherence with lipid-lowing medications than with antihypertensive medications, and had the highest average adherence with medications for diabetes. Trends over time were relatively flat, exhibiting only minor fluctuations regardless of the combinations of medications taken. Among patients taking medications for the same illness, adherence was better when they were also taking medications for 1 or 2 of the other illnesses. Patients taking medications for hypertension, dyslipidemia, and diabetes had the highest adherence. Especially for hypertension and diabetes, patients may often be managing more than 1 medication for the condition simultaneously, although adherence was defined as 1 or more filled prescriptions on a given day. Adherence by combinations of the medications being taken remained in an order that persisted across the study years. The relative rates of adherence by the combinations of medications prescribed remained in the same order after adjusting for patient characteristics.

Patients were more likely to be adherent in the current year if they had adhered in the past year, as one might expect. The extent to which patients stayed adherent from year to year, however, varied by the combinations of prescribed medications, as did the percentage of patients who were nonadherent in the past year and became adherent in the current year. The more illnesses for which patients were taking medications, the more likely they were both to become adherent if nonadherent in the past and to stay adherent if adherent in the past.

A number of other studies have also reported that patients taking more medications have higher adherence. These studies typically investigated adherence with single medications and examined adherence by the total number of other drugs that patients were taking. A study of new users of statins to control lipid levels, for example, found that the more medications patients were taking at the initiation of statin therapy, the better their daily medication adherence.

 A second study of new statin users observed that patients with a greater medication burden had higher medical possession ratios.

A study of newly diagnosed hypertensive patients found those taking multiple drugs for other conditions had better adherence with their antihypertensive medications.

Contrary to these examples, other studies have observed that patients with a greater medication burden have lower adherence. One study of lipid-lowering drugs among patients 65 years and older reported patients with the highest number of medications in the past year had the lowest adherence.

 A study of patients initiating therapy with both antihypertensive and lipid-lowering medications found that patients taking 6 or more medications were about half as adherent as patients taking 1 or no medication. Adherence in this study was higher with antihypertensive medications than with lipid-lowering medications,

 a finding similar to ours. A second study of patients initiating concomitant antihypertensive and lipid-lowering therapy reported that adherence among patients taking 0, 1, and 2 other medications was 41%, 35%, and 30%, respectively.

The studies that examined adherence by the number of medications taken differ from one another in a number of ways. Some, for example, included all patients taking the study medications, whereas others only enrolled new users. The demographic characteristics of the study populations also varied, as did the medications examined for adherence. As a further issue, the other medications being taken by patients in the studies may have differed, and typically were not characterized. Any combination of these factors may have contributed to the discordant results.

We took the approach of studying adherence specifically among patients for the 3 most common conditions leading to cardiovascular disease. The results suggest patients taking medications for more of the conditions maintained higher adherence. Patients with more chronic illnesses may have a greater sense of vulnerability and a stronger belief that taking medications may protect them from serious adverse events.

 Patients in poor health especially may feel at risk compared with patients with risk factors but without overt disease. One study reported that elderly patients with a history of stroke, congestive heart failure, or hypertension had high statin adherence.

 The issue of adherence by patient morbidity, however, is likely complex; for example, having a coronary event has been associated with subsequent discontinuation of preventive medications.

 As a further issue, patients taking medications for multiple illnesses may experience drug interactions between drugs in the same category or between drugs in different categories. The relationship of polypharmacy, adherence, and the outcome of polypharmacy requires further study. Patients with more illnesses or greater severity of a single illness may see physicians more often, which in turn may affect medication adherence. Some patients may cut their pills in half to reduce expenses, and this practice might vary by the number of illnesses.

This article has a number of limitations that affect interpretation of the results. The study population only included members of a single insurer in Hawaii, so the results may not generalize to other populations. Patients included in analyses had filled at least 1 prescription during the calendar year; patients who stopped their medication use before the start of a year were excluded. Given the chronic nature of the conditions, however, most patients included in the analyses would be long-term users. The definition of adherence as 80% of days with prescriptions is arbitrary, although a choice frequently used in studies of medication adherence.

It is likely some patients obtained free samples from their physicians; if so, these medications would not be included in the insurer’s pharmacy data. If patients taking medications for more illnesses were more likely to obtain free samples, a bias might result. The severity of illnesses among patients with similar illnesses might affect adherence. If any patients had additional drug insurance with a different insurer, prescriptions billed to the other insurer would not have been known. A further issue is that patients taking multiple medications for the same illness but who failed to fill a prescription for 1 of their medications would be counted as being adherent in this study, although not necessarily following their doctor’s orders. Finally, adherence was based on filled prescriptions; the extent to which patients actually took their medications was unknown.

Our primary conclusion is that patients taking medications for more of the 3 conditions studied (hypertension, dyslipidemia, diabetes) have greater adherence than patients taking medications for fewer of the conditions. With medications for more illnesses, the proportion of patients who adhered in a calendar year was higher, the proportion who became nonadherent if adherent in the previous year was lower, and the proportion who become adherent if they had been nonadherent in the previous year was higher. The results suggest that physicians should not be concerned about negatively affecting adherence by prescribing medications for 2 or all 3 of these conditions. Confirmation of the findings in other populations would strengthen this conclusion.

Of concern is the lower adherence of patients taking drugs for 2 of the 3 conditions, and the especially low adherence among patients with single illnesses. Patients taking medications for single illnesses may require education and support to encourage their medication adherence. Patients with a low frequency of adherence to multiple medications also warrant attention. They face the greatest immediate risks of cardiovascular and other adverse events. The constancy of the trends in adherence across the 5 years of the study emphasizes, however, how difficult changing adherence behavior may be, as well as the importance of developing new intervention strategies.

Adherence was examined among patients taking medications, individually and in combinations, to treat hypertension, hyperlipidemia, and/or diabetes.

Adherence With Lipid-Lowering, Antihypertensive, and Diabetes Medications

As healthcare expenditures continue to rise, insurers seek to slow the rate of growth through the implementation of cost-containment measures. A common cost-containment measure is the incentive formulary, which attempts to disincentiviz

that cost-containment measures including tiered formularies

succeed in reducing overall consumption of prescription

drugs by health plan members, reducing health plan

spending on pharmaceuticals, and encouraging the choice

of less expensive over more expensive medications.

effects of cost sharing include switching to lower-cost

drugs, treatment interruptions, and taking medications less

frequently than prescribed to extend the period of use for

 The majority of studies to date have focused

on only a few therapeutic classes of drugs, most notably

angiotensin-converting enzyme inhibitors, proton pump

Diabetes mellitus, with an estimated total prevalence of 7% and medical expenditures totaling $174 billion,

 is a serious public health problem that disproportionately affects ethnic subgroups such as non-Hispanic blacks, American Indians, Alaska Natives, Native Hawaiians, and Filipinos.

 In Hawaii, the setting of this study, the prevalence of diabetes among Native Hawaiians is 11.5% and 10.4% among Filipinos.

 Although previous studies have examined the relationship between cost sharing and medication use among diabetic patients,

 to our knowledge, no studies to date have looked at the interaction between drug switching and cost sharing, an issue with potential implications for diabetes management.

To address this gap, we examined the relationship between copayment level and drug switching among patients who newly initiated a medication regime for the treatment of type 2 diabetes. The conceptual model guiding this investigation was 

the economic framework for healthcare demand proposed

 According to this framework, raising the

cost-sharing level for prescription drugs is expected to result

in (1) reduced consumption of drugs overall; (2) substitution

of less expensive, equivalent drugs; and (3) a decrease

in consumption of “nonessential” drugs (eg, discretionary

would increase the rate of switches to drugs with

lower copayments, and also that copayment levels might

have a differential effect on drug switching depending on

The design was a retrospective observational study. The study population was drawn from members of a single health plan in the state of Hawaii that provides coverage to approximately 650,000 members, or about 50% of the state’s total insured population.

 Inclusion in the study required patients to have (1) a diagnosis of type 2 diabetes and (2) a first filled prescription for a diabetes-related medication after at least 1 full year of prior enrollment. Patients were excluded if (1) they had type 1 diabetes or (2) they filled 2 different prescription diabetes medications on their first or index date. A small number of patients (n = 779) began on polytherapy and were excluded from the final study sample. Data were extracted for eligible members during the study period from January 2002 through August 2007. No universal changes in copayments occurred for this insured population during the study period. However, the study population was limited to patients who newly initiated pharmacologic treatment of their diabetes, in order to isolate “new” diabetes patients using antidiabetic medications from chronic antidiabetic medication users.

As previous studies have documented a lack of communication between patients and providers about the out-of-pocket costs for medications, we did not assume that discussions about cost occurred prior to the time the first prescription was filled.

 In addition, other studies have shown that clinicians often are unaware of the cost of different medications and therefore do not take cost into account when prescribing.

 A patient’s data were included from the date of the first prescription until the end of the study period or the end of continuous enrollment, whichever was sooner. This study was approved by the University of Hawaii Committee on Human Studies.

The medical claims database provided a limited data set on diagnoses, patient utilization, type of insurance plan (fee-for-service or HMO plan), demographic information, and morbidity levels (Johns Hopkins Adjusted Clinical Groups Case-Mix System; see www.acg.jhsph.edu) for the 3-month period preceding the index prescription.

 The pharmacy claims database provided data on medications including prescription fill dates, days of supply, copayment (out-of-pocket cost to the patient), and formulary tier. Medications were categorized by therapeutic class (all classes of oral hypoglycemics as well as insulins) within formulary tier (generic, preferred brand, and nonpreferred brand). Glycosylated hemoglobin (A1C) levels were obtained from laboratory reports from the various diagnostic laboratories. The laboratory value that most closely followed the index prescription was included in analyses. Ethnicity and education were obtained from member satisfaction surveys that are mailed annually to a random sample of the insurer’s members with response rates of 40% to 50%.

Patients were considered to have switched drugs if they discontinued use of the starting drug 

filled a prescription either (1) for a drug in a different diabetes therapeutic class or (2) for a drug in the same therapeutic class but in a different tier. Discontinuation was defined as having a gap between prescriptions greater than twice the length of the first prescription. Similarly, the study window to switch to a second drug was twice the length of the first prescription. Patients who discontinued therapy completely (ie, stopped filling their current prescription drug without adding another drug) were treated as censored from the end date of their last prescription. Gaps were calculated as the difference between the end date of the first prescription (based on the days of supply) and the fill date for the second prescription.Similar definitions of drug switching have been used inprevious studies.

Among insured patients with type 2 diabetes, copayments and glycosylated hemoglobin values were associated with increased rates of drug switching.