HYPOGLYCEMIC DRUG UTILIZATION AND ADHERENCE TO PRESCRIBED REGIMENS: A PHARMACOEPIDEMIOLOGIC STUDY USING RETAIL PHARMACY DATA

Diabetes mellitus is a highly prevalent condition, afflicting an estimated 6% of the United States adult population . It is also a complex condition to manage. Dietary, exercise, and drug therapies are essential for reducing the risk of various neurologic and vascular diseases related to disease progression. Tight control of blood glucose as achieved through intensive pharmacologic therapy has been shown to decrease the risk of developing several types of diabetic complications. Success in achieving tight blood glucose control is contingent upon adherence to the prescribed hypoglycemic drug regimen , a behavior known to often be sub-optimal. The objectives of this study were a) to describe hypoglycemic drug utilization, and compare drug regimens prescribed and costs among age groups, insurance plans, and by gender; b) to assess adherence to prescribed hypoglycemic therapies, and to explore the association between nonadherence and change in the strength or type of hypoglycemic medication prescribed; and c) to identify the frequency of nonadherence among patients who are prescribed monotherapy with a sulfonylurea or metformin, as compared to the frequency of nonadherence in patients who are prescribed dual therapy with both medications. Analyses were performed using retail pharmacy data. The data provided included over one-quarter-million dispensings to 5056 diabetic patients. From this population , 2901 patients that received at least 2 dispensings for a hypoglycemic medication during a 12-month period were selected for study. Sulfonylureas were the mainstay of treatment for those receiving oral therapy: 82% of patients received a sulfonylurea as monotherapy or in combination with another hypoglycemic medication. The most frequently observed drug regimen was sulfonylurea monotherapy (40.3%); followed by insulin use only (24.9%); dual therapy with sulfonylurea plus metformin (13.9%); and metformin monotherapy (6.96%). Differences in the drug regimen utilized were found among age groups and between genders. Most notably, sulfonylurea monotherapy was prescribed most frequently for patients 65 years of age and older (age 65 years or older: 48.4%; age 50-64: 43.6%; age under 50: 30.8%, p < 0.0001 ). The 12-month cost of hypoglycemic medication dispensed was lowest among patients 65 years of age or older. The medication possession ratio (MPR) was used as an estimate of adherence. Possession of medication was found to be associated with a change in the strength or type of hypoglycemic medication dispensed. Sulfonylurea users who failed to possess medication for at least 80% of days during a four-month period were 41.7% more likely to receive a dispensing for a different strength of medication in subsequent months, as compared with those possessing medication for at least 80% of days (OR 1.42, 95% Cl 1.02 1.96). Additionally, among patients receiving either monotherapy with a sulfonylurea or metformin, those possessing medication for at least 80% of days were 36.4% more likely than those possessing at least enough medication for 80% of days to receive a dispensing for a different strength of medication (OR 1.36, 95% Cl 1.019 1.83), or for a different strength or type of hypoglycemic medication (OR 1.39, 95% Cl 1.03 1.87). This finding was not statistically significant in the smaller sample of patients receiving monotherapy with metformin. Medication possession was also found to be associated with the number of hypoglycemic drugs prescribed. Patients who were prescribed a regimen of dual therapy with a sulfonylurea plus metformin were found to be less likely to possess medication for 80% or 90% of days, as compared with those prescribed monotherapy with either a sulfonylurea or metformin. In multivariate analyses controlling for age and the total number of dispensings, patients receiving dual therapy were more than 3 times more likely to fail to possess medication for at least eighty percent of days (OR 3.14, 95% Cl 2.42 4.08), or 90% of days (OR 3.20, 95% Cl 2.49 4.11 ). The findings of this pharmacoepidemiologic research provide insight into the drug utilization patterns of diabetic patients. Among patients in this study, the type of drug regimen prescribed differed in frequency among age groups and between genders. The strength and type of hypoglycemic medication utilized was found to change frequently, particularly among patients that were classified as nonadherent. Overall, a substantial percentage of patients were found to be nonadherent with hypoglycemic drug therapy. Patients least frequently adherent to drug therapy included those under 65 years of age and those prescribed dual therapy with a sulfonylurea plus metformin. Presuming that lack of medication possession results in poor glucose control, patients who do not possess medication are at increased risk for diabetic complications.

Analyses were performed using retail pharmacy data . The data provided included over one-quarter-million dispensings to 5056 diabetic patients. From this population , 2901 patients that received at least 2 dispensings for a hypoglycemic medication during a 12-month period were selected for study.
Sulfonylureas were the mainstay of treatment for those receiving oral therapy: 82% of patients received a sulfonylurea as monotherapy or in combination with another hypoglycemic medication . The most frequently observed drug regimen was sulfonylurea monotherapy (40.3%); followed by insulin use only (24.9%); dual therapy with sulfonylurea plus metformin (13.9%); and metformin monotherapy (6.96%). Differences in the drug regimen utilized were found among age groups and between genders. Most notably, sulfonylurea monotherapy was prescribed most frequently for patients 65 years of age and older (age 65 years or older: 48.4%; age 50-64: 43.6%; age under 50: 30.8% , p < 0.0001 ). The 12-month cost of hypoglycemic medication dispensed was lowest among patients 65 years of age or older.
The medication possession ratio (MPR) was used as an estimate of adherence. Possession of medication was found to be associated with a change in the strength or type of hypoglycemic medication dispensed.
Sulfonylurea users who failed to possess medication for at least 80% of days during a four-month period were 41.7% more likely to receive a dispensing for a different strength of medication in subsequent months, as compared with those possessing medication for at least 80% of days (OR 1.42, 95% Cl 1.02 - 1.96). Additionally, among patients receiving either monotherapy with a sulfonylurea or metformin, those possessing medication for at least 80% of days were 36.4% more likely than those possessing at least enough medication for 80 % of days to receive a dispensing for a different strength of medication (OR 1.36, 95% Cl 1.019 - 1.83), or for a different strength or type of hypoglycemic medication (OR 1.39, 95% Cl 1.03 - 1.87). This finding was not statistically significant in the smaller sample of patients receiving monotherapy with metformin.
Medication possession was also found to be associated with the number of hypoglycemic drugs prescribed . Patients who were prescribed a regimen of dual therapy with a sulfonylurea plus metformin were found to be less likely to possess medication for 80% or 90% of days, as compared with those prescribed monotherapy with either a sulfonylurea or metformin . In multivariate analyses controlling for age and the total number of dispensings, patients receiving dual therapy were more than 3 times more likely to fail to possess medication for at least eighty percent of days (OR 3.14, 95% Cl 2.42 -4.08), or 90% of days (OR 3.20, 95% Cl 2. 49 -4.11 ).
The findings of this pharmacoepidemiologic research provide insight into the drug utilization patterns of diabetic patients. Among patients in this study, the type of drug regimen prescribed differed in frequency among age groups and between genders. The strength and type of hypoglycemic medication utilized was found to change frequently, particularly among patients that were classified as nonadherent. Overall, a substantial percentage of patients were found to be nonadherent with hypoglycemic drug therapy. Patients least frequently adherent to drug therapy included those under 65 years of age and those prescribed dual therapy with a sulfonylurea plus metformin. Presuming

Objective
To describe hypoglycemic drug utilization patterns among diabetic patients ; and to determine if such patterns differ by patient characteristics.

Methods
Cross-sectional descriptive study of hypoglycemic medication dispensings to 2901 diabetic patients. Drug utilization was classified into categories representing 10 various hypoglycemic drug regimens. The frequency of drug regimen utilization was compared by age group, insurance plan , and between genders. The total cost of hypoglycemic dispensings during a 12-month period was also determined , and compared within each age group and insurance plan , and between genders.

Results
The most frequent hypoglycemic drug regimen utilized was monotherapy with a sulfonylurea (n = 1168; 40 .26%). Use of insulin and no oral hypoglycemic medication was the second most frequently utilized regimen (n = 723; 24.92%). Dual therapy with both a sulfonylurea and metformin was utilized by 13 $580. 80, Pr> F less than 0.0001 ).

Conclusions
We found sulfonylureas to be the mainstay of treatment for the majority of patients receiving oral therapy. Patients 65 years of age and older were less frequently dispended troglitazone, or metformin in monotherapy, and were most frequently prescribed sulfonylureas as the only hypoglycemic medication . For patients 65 years of age or older, the 12-month cost of hypoglycemic medications dispensed was roughly 20% less than the average hypoglycemic drug utilization cost for younger diabetic patients.
Differences in drug utilization among senior patients may reflect differences in co-morbidities that influence drug selection, differences in insurance coverage and associated out-of-pocket costs, or perhaps reluctance of prescribers to utilize newer therapies in older diabetic patients. Further research is needed to determine if older patients are less frequently prescribed newer therapies, and if diabetes control and progression is impacted.

INTRODUCTION
Diabetes mellitus has been identified as an epidemic throughout the world (1)(2)(3)(4)(5)(6)(7)(8)(9)(10)(11)(12)(13)(14)(15). Approximately 16 million Americans are afflicted with the condition, a prevalence rate that exceeds 6% of the United States population (16). The number of Americans living with diabetes has increased steadily. Based upon data from the US National Health Interview Surveys, the prevalence of diabetes has grown in a near-linear manner, rising from under 1 % in 1958 to approximately 7% in 1993 (17). Recently, Mokdad et al reported that the prevalence rate of diabetes increased 33% from 1990 to1998 (4.9 to 6.5%) (18). The number of Americans living with diabetes is expected to double by 2020, and the disease is expected to become more prevalent in younger age groups (19).
Positive family history (20)(21)(22), older age (23)(24)(25)(26), and obesity (27)(28)(29) are associated with an increased risk of developing diabetes. The rise in diabetes prevalence reflects the growing number of older adults in the US population, and is compounded by an increased rate of the disease in this population (17).
An increasing prevalence of obesity among Americans of all adult ages has also contributed to the rise in diabetes prevalence (30). Diabetes has a major impact on the health of those afflicted with the condition. Diabetes is the seventh leading cause of death among U. S. citizens (31 ), and co ntributes significantly to other leading causes of death such as cardiovascular disease (32 , 33), stroke (34), and kidney failure (35)(36)(37).
Type 2 diabetes, formerly termed adult-onset or non-insulin dependent diabetes mellitus, accounts for roughly 90 % of all cases of diabetes (17).
Patients with type 2 diabetes mellitus generally develop the condition after the age of thirty (20), and are usually managed with oral hypoglycemic medications when dietary therapy and exercise fail to control blood glucose levels. Sulfonylureas, discovered in 1942 by Janborn (38), have long been the mainstay of type 2 diabetes treatment. During the past several years newer oral therapies have become available, which exert a hypoglycemic effect through differing biological mechanisms. These agents have been approved for use in combination with other medications (39)(40)(41)(42)(43)(44)(45) or, for some drugs, for use as mono-therapy (40, 43 -45).
The choice of oral hypoglycemic medication may reflect patient characteristics.
For example, metformin has been shown to be particularly efficacious in patients with obesity (46), and sulfonylureas may cause rash in some patients (47). However, despite the several classes of hypoglycemic medications to choose from , specific guidelines or algorithms for selection of drug therapy in We examined hypoglycemic drug utilization in a large diabetic population , using data obtained from retail pharmacies during a 24-month period. We determined the frequency of dispensing of medications, and the characteristics of patients receiving dispensings for specific hypoglycemic agents. Changes in hypoglycemic drug regimen during a 12-month period were determined, including the number of patients switching to a different drug or combination of drugs. The average cost of hypoglycemic medications dispensed during a 12month period was calculated, and compared between age groups, gender, and insurance types .

Data source and study population
Consumer Value Stores (CVS). Inc provided data from 198 retail pharmacies throughout Pennsylvania . The data included all pharmacy claims for 5056 diabetic patients from April, 1997 -May, 1999. The total number of pharmacy dispensings for these patients during the 2-year period was 288, 171.
All patients were enrolled in a comprehensive diabetes management program through one of two health insurance plans. Pharmacy services for these patients were provided by CVS pharmacies. and patients were not reimbursed for prescriptions filled by other pharmacies. Thus, the data represents patients' utilization of hypoglycemic drugs, though some patients may have filled prescriptions elsewhere using a different third party plan for drug reimbursement, or by paying cash.
To create a standard basis of comparison, we included only patients who were dispensed at least 2 prescriptions over a time period spanning at least 12 months. This was accomplished by determining the number of days between the first and last dispensing of any hypog lycemic medication during the twoyea r period for whi ch data was available. Of the 5056 patients in the population, 2901 patients (57.4%) met this criteria . For these patients. a 8 sample was created that included only the first 12-months of hypoglycemic drug dispensings. This sample was used for all analyses.
Hypoglycemic medications were categorized by therapeutic class, or by generic name if the drug was the only available product of its class. The following categories were created: insulin , sulfonylurea , metformin, alphaglucosidase inhibitor, meglitinide , glimepramide, and troglitazone (Rezulin®), the first available drug from the thiazoladinedionne class. The study period preceded the withdrawal of troglitazone from the U.S. market, and also preceded the introduction of two newer thiazoladinedionne agents to the US market: pioglitazone (Actos®) and rosiglitazone (Avandia®).
Patients were categorized as receiving one of ten hypoglycemic drug regimens, based upon the class(es) of hypoglycemic agent prescribed during the first 90 days of the 12-month period. These drug regimens are presented in Figure 1. For comparison , we also categorized drug regimens based upon hypoglycemic agents dispensed during the last 90 days of the 12-month period.
For all patients, we determined the age , gender, and insurance plan associated with prescription dispensings . Age was categorized into three groups: less than 50; 50-65; and 65 years or older. To enhance the focus on drug regimens used by type 2 diabetics, some analyses were performed restricting the population to those 40

Statistical analyses
Descriptive statistics were used to determine and present gender, age category, and insurance type. Drug regimens utilized and changes in drug regimen were presented as frequencies and percentages. Non-parametric (chi-square) analyses were used to determine if the frequency of drug regimen utilized differed in statistical significance by the gender, age category, and insurance type of subjects. Chi-square tests were also to determine differences in dispensing of brand name products and first generation sulfonylureas among age groups, gender, and insurance types; and to determine differences in the frequency of troglitazone dispensing in the three age groups.
Analysis of variance procedures using Tu key's test were used to determine if differences in 12-month hypoglycemic drug utilization costs among age groups and insurance types were statistically significantly different. We used the student's t-test for independent samples to determine if hypoglycemic drug spending differed by gender. 11 Statistical analyses were performed using SAS for windows version 8.01 . 12

Descriptive statistics
A total of 2901 patients received hypoglycemic drugs spanning at least a 12- receiving mono-therapy with metformin , troglitazone, or glimepramide also did not differ significantly. These findings are summarized in Table 4. We also examined the relation between insurance type and drug regimen utilized.

DISCUSSION
Pharmacy data is a useful and often available information source for pharmacoepidemiologic investigations of drug utilization patterns. Pharmacy data can be used to identify trends in prescribing patterns (53), assess the impact of regulatory changes (54), and to determine drug expenditures (55).
Though the lack of information related to patient diagnoses can be limiting (56), pharmacy data can be superior to the medical record for determining if and when prescriptions are actually filled by patients; an action necessary for adherence to therapy. Pharmacy data is an excellent source for describing drug use in populations, and for comparing patterns of use between populations.
In this study, pharmacy data was used to describe the utilization of hypoglycemic drugs in a population of diabetic patients. We created nine categories of drug regimens utilized, based upon the types of hypoglycemic drugs dispensed to patients. We compared the frequency of regimen utilization among age groups, gender, and payment type. Identification of regimen utilized was based upon hypoglycemic drugs dispensed during the first 90 days of a 12-month period , and was compared with the frequency of regimens utilized in the last 90-days of the 12-month period. We identified which patients changed medication regimen , and which different regimen was utilized.

Frequency of regimen category utilized
Sulfonylureas, a staple of therapy for type 2 diabetes for several decades, were the most frequently prescribed oral hypoglycemic agent in this population. Factors contributing to the popularity of these agents include a long history of use (38), relatively low expense (57) Consistent with its approved indications, troglitazone was utilized as monotherapy or in combination with insulin or oral hypoglycemic agents.
Troglitazone was discontinued in 1999 after awareness of an association with hepatic failure and death among patients prescribed this agent (59)(60)(61)(62). The increased risk for liver toxicity in those treated with troglitazone was recognized while the drug was available for prescribing (63), potentially explaining why troglitazone was prescribed less frequently for older patients in this population (Table 7).
Despite the effective treatment of diabetes with sulfonylurea, metformin , or insulin , glycemic control is known to become poorer as diabetes progresses (64)(65)(66) be unstable, and medication needs increase over time (20).
As compared to utilization based upon the first 90 days of the 12-month period, the utilization of sulfonylurea mono-therapy was less, and the utilization of combination therapies was greater during the last 90-days of the 12-month period . The decrease in sulfonylurea mono-therapy and increase in use of oral combination therapies may be related to several factors . First, it is possible that the increase in utilization of combination regimens is partly the resu lt of increased medication needs associated with disease progression over time. Second , the increased use of combination therapy may reflect the availability of newer hypoglycemic agents approved for use in combination therapies during the 12-month period, such as repaglinide, glimepramide, acarbose, and troglitazone. A third possible influencing factor is the increased awareness and popularity of combination therapy as a means to achieve improved glycemic control.
Of patients categorized as users of insulin as their sole hypoglycemic medication during the first 90 days of the 12-month period, 5% added troglitazone to their regimen during the following 9 months. Anoth er 5% of insulin users added an oral agent other than troglitazone to their regimen . Thus, when comparing drug regimens utilized among age groups, differences in the regimen utilized was presumed not to reflect differences in the proportion of type 1 and type 2 diabetics.
We found differences in the frequency of drug regimen utilized by patients 65 years of age or older. First, though there was no statistically significant difference in insulin prescribing among age groups, seniors were less frequently prescribed troglitazone in combination with insulin. Though removed from the US market in 2000 due to risk of life-threatening hepatic injury, troglitazone, the first available thiazoladinedionne, was recommended as a useful agent in the management of diabetes (72)(73)(74). Prescribers may have avoided prescribing troglitazone in older patients due to concern that older patients may be at greater risk for troglitazone-induced hepatic toxicity.
Though prescription drug coverage was available to older patients in this population, it is also possible that prescribers avoided prescribing this product in older patients due to financial considerations. The retail price for a 30-day supply of troglitazone exceeded the average one-month cost of a generic sulfonylurea, a regimen utilized with increased frequency among older subjects. Since the cost of the dispensing to the patient (co-payment) was not available, the influence of cost on drug regimen utilization cannot be directly assessed . However, it is interesting that seniors were most frequentl y prescribed mono-therapy with a sulfonylurea, the oldest and most inexpensive oral hypoglycemic drug regimen available. Additionally, seniors utilized metformin as mono-therapy less frequently, but were not less frequent users of regimens of sulfonylurea plus metformin . Thus metformin was used more frequently as an adjunct to sulfonylurea therapy in seniors. Contrastingly, younger patients were prescribed mono-therapy with metformin more frequently than older patients (p = 0.0107). Metformin has been shown to be particularly useful in obese diabetic patients (65). The less frequent use of metformin mono-therapy among seniors may reflect a lower prevalence of obesity among those 65 years of age and older in this population .
We also examined the frequen cy of drug regimen prescribed for each insurer type. We did not find the frequency of drug regimens prescribed to differ 26 significantly between the two main insurance plans included in this study. One exception was the use of glimepramide as mono-therapy, which was less frequently prescribed for patients insured by plan B. However, only a small percentage of patients were prescribed this drug as mono-therapy (2.6%; 76/2901 ). Of the patients that paid cash for their prescription , one-half were prescribed sulfonylurea mono-therapy.

Additional analyses
We found a low prevalence of first generation sulfonylurea utilization (1 .1 %; 17/1168), and older persons were not more frequently dispensed these agents. The infrequent use of these products reflects an awareness of the increased potential for hypoglycemia associated with chlorpropamide (58), and the lower likelihood for adverse effects such as hyponatremia and disulfiramtype reactions (75)(76)(77). Use of first-generation sulfonylureas did not differ by age group or by gender. We did not expect to find a difference in dispensings for brand-name sulfonylurea products among age groups or by gender.
However, although patients covered by health plan B utilized brand name sulfonylureas less frequently, the average 12-month cost of hypoglycemic drugs dispensed to patients in plan B was the highest of all insurance types.
Thus, it appears that factors other than dispensing of generic products were important determinants of drug utilization costs; and may include factors such as the frequency of use of combination therapies, disease severity, or a more frequent use of newer drugs for which generic products are not available. It is also important to note that drug expenditures were calculated using AWP-10%. We selected this figure mainly to provide a common metric for assessing costs of hypoglycemic drug utilization. The dollar totals do not include the cost of syringes or blood glucose monitoring devices or supplies, or medications used for reasons and conditions other than blood glucose control.
Though various insurance types are represented , a majority of study patients were enrolled in a diabetes management program through one of two health 29 plans. Generalizability to other populations is impaired by a lack of knowledge of race and socio-economic status, and the absence of diagnostic information such as diabetes type and co-morbidities.

CONCLUSION
Retail pharmacy data were used to describe hypoglycemic drug utilization and associated costs in a large population of ambulatory diabetic patients. Perhaps the recent attention directed towards tight glycemic control and enhanced control of post-prandial blood glucose will generate an increase in the use of combination drug therapies that includes newer hypoglycemic agents . However, though glycemic control may be enhanced through the use of some of the newer agents, there is no evidence to suggest that sulfonylureas are less effective than newer agents at preventing death and disability due to diabetes. In this population, sulfonylureas were by far the 31 most frequently used agents, particularly in older patients. Older patients were also less frequently dispensed prescriptions for troglitazone, and metformin in combination with a sulfonylurea. These findings may or may not suggest a difference in the standard of care delivered to seniors, but we believe it indicates a difference in the type of care received .
Effect of intensive blood-glucose control with metformin on complications in overweight patients with type 2 diabetes (UKPDS 34

Objective
To determine if nonadherence, as determined by medication possession , is associated with changes in hypoglycemic drug prescribing; either as a change in medication strength or a change in type of hypoglycemic medication dispensed .

Methods
. A retrospective cohort study examining retail pharmacy dispensings of sulfonylureas and metformin. The medication possession ratio (MPR) was used to assess adherence, with patients that did not receive a sufficient quantity of medication to cover eight of ten days in the period (MPR

Results
For patients receiving dispensings only for a sulfonylurea , those failing to possess medication for eight of ten days (MPR < 8: 10) were 41

INTRODUCTION
Patients with diabetes are at risk for numerous adverse health outcomes .
Beyond consequences immediately related to blood glucose regulation, such as hypoglycemia or ketoacidosis (1 ), diabetes increases risk for 'macro' vascular diseases such as myocardial infarction and stroke (2). Diabetics are also more likely to develop hypertension and dyslipidemia (3), which are risk factors for developing these outcomes. In addition to macrovascular diseases, diabetes also increases the risk of developing diseases resulting from damage to the smaller blood vessels, such as retinopathy, neuropathy, and nephropathy (4). The quality of life in patients with diabetes often diminishes over time as diabetic complications become disabling (5).
Appropriate management of diabetes can reduce and delay the sequelae of this disease. Therapeutic interventions have been proven effective in controlling blood glucose levels (6) and in the longer-term, preventing some types of diabetes complications (7-9). Dietary changes and routine exercise are fundamental interventions. Often , however, normalization of blood glucose control can only be achieved through drug treatment (10).
In addition to resolving blood glucose instability in many patients , drug therapy has been demonstrated to confer the added benefit of preventing some types of diabetes complications. Results from two large trials have provided evidence that drug therapy can reduce the incidence of micro-vascular disease in patients that are aggressively managed .
Researchers in the Diabetes Control and Complications Trial (DCCT) (9) compared standard care with intensive drug treatment and monitoring regimens in 1,441 type 1 diabetics during a ten-year period . In this study, the risk of developing retinopathy, nephropathy, and neuropathy was reduced by 76%, 50%, and 60% respectively among diabetics receiving the intense treatment. This finding led researchers to conclude that intensive drug therapy "delays the onset and slows the progression" of these diseases .
The United Kingdom Prospective Diabetes Study (UKPDS) (7) investigated whether intensive drug therapy reduced the incidence of diabetic complications in type 2 diabetes . In this study, 3867 patients with newly diagnosed disease were followed during a 10-year period . As in the DCCT, researchers attempted to determine if intense treatment and monitoring would reduce the incidence of diabetic complications. The results of the UKPDS provided compelling evidence that drug therapy can reduce the incidence of certain types of complications in patients with type 2 disease. Compared with patients treated with conventional care, the risk for developing microvascular complications in patients receiving intensive therapy was reduced by 25%.
Further, the risk of developing retinopathy was reduced by 21 %, while the risk of developing microalbuminurea was reduced by 34 %. The incidence of myocardial infarction was also lesser in the group treated intensively, though the 16% difference observed between groups failed to achieve statistical significance (p = 0.052).
These two trials provided strong evidence that tight control of blood glucose can prevent the microvascular complications in patients with diabetes, and may reduce the risk for macrovascular complications such as myocardial infarction. Since the publication of these trials, standards for the treatment of diabetes have incorporated tight blood glucose control as an objective of therapy. Thus, drug treatment is aimed at maintaining normal blood glucose throughout the day to obtain the benefits described in these trials.
Patient adherence to prescribed drug regimens is implicit to a strategy of tight blood glucose control. When medications are not taken according to instructions, optimal control of blood glucose will clearly be compromised .
This may be the case for many diabetic patients. Patient adherence with prescribed medication regimens for chronic diseases is known to be suboptimal (11-13) (14), and patients with diabetes in particular are known to have difficulty adhering to prescribed dietary, exercise, and drug therapies (15,16).
Further complicating matters is the progressive nature of diabetes, as patients generally experience a worsening of glycemic contro l over time. Therefore, not only is adherence important in reducing the risk of diabetic complications, it is a fundamental component of active participation in therapy that continually requires assessment of medication needs for sustaining glucose control.
Thus, both disease progression and poor adherence can cause inadequate blood glucose control, and it is important to distinguish between the two. The patient who is poorly controlled because his disease has progressed potentially requires a different intervention than the patient who is poorly controlled due to poor adherence. Increasing the dose of medication for the patient in the latter scenario may result in a dangerous hypoglycemic reaction .
Our working hypothesis was that patients that do not adhere to prescribed drug regimens are more likely to be poorly controlled; and such poor control will be manifested as a change in medication dose or a change in medication prescribed . To investigate this hypothesis, we assessed adherence with sulfonylurea or metformin therapy using pharmacy claims data. We determined whether patients that were nonadherent to prescribed therapy were more likely to receive a change in dose or change in hypoglycemic drug dispensed during subsequent months.

Identification of the study samples
We conducted a retrospective cohort study using prescription dispensing information. Consumer Value Stores, Inc (CVS) provided pharmacy utilization data used in this study. The data included over one-quarter million dispensings to 5056 diabetic patients between April 1997 and May 1999.
Most patients included in this population were enrollees of one of two area health plans, and participants in a special managed care program in diabetes through these plans. Patients were restricted to designated CVS pharmacies for pharmacy services. The data provided represented pharmacy utilization by patients enrolled in these two health plans from the designated pharmacies .
The restriction to designated pharmacies was an advantageous feature, potentially resulting in a more complete representation of pharmacy utilization .
Of the patients included in the total population, we identified those that received any hypoglycemic medication during a 12-month observation period.
From this population, we identified patients receiving mono-therapy with a sulfonylurea or mono-therapy with metformin during the first four months of this 12-month period . Patients who received two or more different types of hypoglycemic medication during this four-month period were excluded. Thus, two samples were created: patients receiving monotherapy with sulfonylurea and patients receiving monotherapy with metformin .
Patients who did not receive the same strength of medication and same number of tablets per day for all dispensings during the four-month period were excluded from study. Thus, the two samples represented patients receiving dispensings for either only a sulfonylurea or only metformin at the same dose during the four-month period, and receiving at least one dispensing of a hypoglycemic medication during the following eight months.

Determining change in dose or medication regimen
We assessed the influence of adherence for two main outcomes: change in dose of sulfonylurea or metformin ; or change in hypoglycemic drug dispensed .
To determine change in dose, we compared the initial strength of medication with additional subsequent dispensings during the following eight months.
Also , we identified patients receiving an increased quantity of the same strength of medication (e .g. two tablets daily instead of one). To identify such patients, we compared the number of tablets in the daily dose for all dispensings. By dividing the quantity supplied by the days supply of medication received, we were able to determine the number of tables prescribed per day. Patients who received a different strength of a same medication, or who were prescribed a different number of tablets per day during the eight months following the four-month period were classified as havi ng a change in dose of sulfonylurea or metformin .
In addition to change in medication dose, we identified patients who received a dispensing for a different type of hypoglycemic drug in the months following the initial four-month period . The two outcomes were also used to create a combined outcome of receiving a dispensing for either a different strength or type of hypoglycemic medication.

Calculating the medication possession ratio
We and depression (21 ). Patients who received only one dispensing during the four-month period but who received medication in following months were classified as non-adherent.
A limitation of the MPR as used to assess adherence is the potential for misclassification , since a patient may possess but not consume a medication .
Possession of medication has however, been considered a useful 'first-order' measure of adherence (22), since patients must first possess medication before they can adhere to therapy

Potential confounding variables
We also examined the effect of four other factors on change in dose of hypoglycemic medication or change in type of hypoglycemic medication prescribed . One such factor, the number of medications prescribed , was thought to potentially be an important influence on medication possession and change in hypoglycemic therapy. To approximate the number of medications prescribed , we identified the total number of dispensings for any class of medication during the four-month period; including medications for conditions other than diabetes. Three categories of this variable were created: less than 5 dispensings, 5-15 dispensings, or more than 15 dispensings. We also assessed the effect of patient age, and categorized this variable using categories of less than 50 years of age, 50-65 years, and greater than 65 years of age. Stratifications for the total number of prescriptions dispensed and patient age were based upon assessment of parametric form , as described in the following section .
The type of insurance used for prescription reimbursement was also included as a potential confounder. We categorized insurance type as health plan A, health plan B, or a third category that included all other insurance types and cash. We also included gender as an additional potential confounding variable.
Based on the methodology for variable identification described above and as presented in figure 2, we determined the relative risk of a change in medication dose, change in class of hypoglycemic medication prescribed, or a combined outcome of either for patients having a medication possession ratio less than 8:10, as compared to those having an MPR ~8: 10. Additionally, we assessed the influence of four other variables : the total number of prescriptions dispensed, age category, insurance type, and gender; attempting to control for the effect of these variables where necessary.

Statistical methods
Univariate statistics were used to describe the frequency and percent of monotherapy with a sulfonylurea or metformin. For each of the two samples, we categorized patients by gender, age, insurance type, and the total number of prescriptions dispensed. These characteristics were also presented as frequencies and percentages. We also determined the frequency and percent of subjects having an MPR greater than 8:10 and 9: 10  Bivariate analyses were used to assess the relationship between the MPR and other independent variables, and between independent and dependent variables. Ch i square analyses were used to assess the relation between the medication possession ratio and the potential confounding variables. We also used chi-square analyses to examine the relation between all independent variables and the outcomes of change in dose and change in medication dispensed . For each independent variable, bivariate logistic regression was used to determine the relation between variable and change in dose and change in medication dispensed. These results were presented as an odds ratio with 95% confidence intervals.
Multivariate statistics were used to identify the presence of collinearity between independent variables, to determine the presence of interactions between independent variables, and to assess the influence of the MPR on outcome variables when other potential confounding variables were included.
Collinearity diagnostics were performed using the PROC REG procedure for multiple regression as suggested by Allison (23). Collinearity was assessed for all independent variables in separate models with dependent variables change in dose, change in medication dispensed , and the combined outcome of change in either. Presence of collinearity was determined using thresholds for condition index and proportion of variance shared as described by Tabachnick and Fidell (24).
Test for interaction between variables was performed using the chunk test for multivariate logistic models as described by Kleinbaum (25). For this procedure, we calculated the difference in the -21og statistic between full and reduced models. Full models included all possible interaction and single terms: Reduced models included single terms (variables) only. The difference in -21og statistic between full and reduced models was tested for significance using the Chi-square distribution , with degrees of freedom equal to the difference in terms between the two models. A difference in -21og value that was less than the X 2 statistic indicated that an interaction was not present. The age stratifications presented reflect the parametric form of this variable.
Nearly half of all patients in both samples were between 50-65 years of age.
A notable disparity was the higher percentage of patients in youngest age strata among metformin users, as compared with subjects in the sulfonylurea sample. Conversely, the sulfonylurea sample included a greater percentage of patients 65 years of age or older as compared with the metformin sample.
In both the sulfonylurea and metformin samples , nearly 90% of patients were covered by one of the two predominant insurance plans. The remainder of patients used other insurance plans for prescription payment, or paid cash.
Percentages of categories of insurance type were similar for the sulfonylurea and metformin samples.
More than half of patients were categorized as receiving a total of 5-15 prescriptions during the four-month period, a strata that represented the second and third quartiles of the distribution of this variable . We chose to combine these quartiles due to a similar relationship between each quartile and the outcome variables. However, the first and fourth quartile of the frequency distribution of this variable differed from the second and third quartile in relation to the dependent variables, necessitating the three levels created .

Multivariate results
We assessed the influence of the MPR on three outcomes: change in dose, change in drug dispensed , and the combined outcome of change in dose or drug dispensed. We examined the effect of MPR on these outcomes for three populations: patients dispensed sulfonylureas , patients dispensed metformin , and a combined sample of patients dispensed either monotherapy with a sulfonylurea or monotherapy with metformin . In sum, nine multivariate models were created to assess these relationships .
For each of these nine models , we assessed collinearity and interaction between variables. Collinearity between independent variables was not detected at condition indices above the threshold of 30. However, collinearity diagnostics indicated a high degree of shared variance between the number of prescriptions dispensed and age. Interaction between the MPR and other independent variables was assessed using the chunk test described by Kleinbaum. We did not detect a significant interaction between any combination of independent variables, as determined by comparing the difference in -21og statistic between full and reduced models with a chi-square value. These results are presented in table 7.
Tables 8a-c, 9a-c, and 1 Oa-c present the multivariate results of the nine models . In these tables we have presented the resu lts of models that include all independent variables in the logistic regression , followed by a final model containing only significant terms. For models having no significant terms, only the model containing all independent variables is presented .
Tables 8a , 8b, and 8c present the multivariate results from the sulfonylurea sample for outcomes of change in dose, change in medication dispensed , and the combined outcome of change in dose, respectively .

DISCUSSION
Adherence to prescribed drug regimens is a common problem in patients with chronic diseases. Adherence to prescribed drug regimens in patients with diabetes is particularly important, since tight control of blood glucose has been demonstrated to reduce the incidence of many types of diabetes complications. Nevertheless, non-adherence to hypoglycemic therapy has been identified as a major barrier to effective management of diabetes. Lack of adherence with prescribed hypoglycemic medications has been correlated with diminished blood glucose control (26)(27)(28), which in turn can result in adverse health outcomes, both in the long and short-term.
Much research has been conducted investigating the cause for lack of adherence to drug therapy. Immediate barriers to adherence include the cost of medication, access to medication , and the complexity of the drug regimen .
Beyond these barriers, determinants of adherence are more complex, and include physiologic, cognitive, behavioral , and environmental factors (29). For example, the health belief model has been reported to predict adherence to drug therapy (30).
According to the health belief model, adherence will be improved when patients perceive their disease to pose a threat to their health, and believe that prescribed medications will be effective in decreasing the risk of morbidity. In In this study, we attempted to demonstrate that lack of adherence to prescribed hypoglycemic medication increases the likelihood that prescribed therapies will change. We assumed that a change in medication dosage or a change drug type was an action in response to poor blood glucose control.
Appropriate interventions for patients who are non-adherent to therapy would logically involve efforts intended to improve adherence. Increasing the amount of medication prescribed to a non-adherent diabetic patient is a potentially inappropriate intervention. Such a response may also be dangerous, resulting in increased risk of hypoglycemia (38) and longer-term complications resulting from continued poor glucose control.
In this study, the medication possession ratio was used to assess adherence.
Also included in analyses were four other factors that were considered to be possibly associated with the MPR or related to the outcomes of change in medication dose or change in type of hypoglycemic medication dispensed .
The age and gender of the patient were two of these factors. According to various studies, the relation between adherence and these two factors is unclear. For example, some studies have shown age and gender to be influences on adherence (39)(40), though others have not (41)(42). In one study of adherence with sulfonylurea therapy, age and gender were found to be statistically significant influences on drug regimen adherence (43) . In our study, we also included insurance type as a potential factor of influence.
Though we had no knowledge of details regarding co-payment structures or formulary systems, we thought that attributes of the benefit design may have had an impact on adherence or change in dose or type of drug dispensed . The 77 total number of medications dispensed was the fourth of the other independent variables included in our study. We included this factor based upon research suggesting that adherence may be associated with the number of medications prescribed.
As determined by the medication possession ratio, adherence to drug therapy was poor among patients in both the sulfonylurea and metformin samples.
Only greater concern among older patients for the protection of health, with a corresponding adherence with prescribed medications.
It was interesting that a greater number of prescriptions dispensed was more frequently associated with possession of medication for at least 80% of days.
This finding may also be related to concerns for maintaining heath . Patients receiving more medications are presumably less healthy, and perhaps more likely to take medications as prescribed .

Multivariate analyses : Change in dose of hypoglycemic medication
A change in dose of sulfonylurea or metformin was more than twice as frequently observed than a change in type of hypoglycemic medication dispensed . Diabetes is a progressive disorder, and even patients that adhere to therapy will require increased amounts of medication over time. In both the sulfonylurea and metformin samples, the youngest of the three age groups most frequently changed dose or type of drug dispensed , and the oldest age group least frequently experienced a change dose or type of drug dispensed.
A possible explanation for the less frequent changes in drug dose among in older patients is that perhaps these patients are receiving a maximum dose of medication, a result of having diabetes for a longer duration . Betz-Brown et al.
found that prescribers may tend to continue sulfonylurea therapies despite evidence that these agents are failing (45 We did not find gender, insurance type , or the total number of prescriptions dispensed to be associated with change in dose, change in drug dispensed , or a combined outcome of either change in dose or change in drug dispensed.
Thus, we did not include these factors in final multivariate logistic models.
This study was conducted solely through the use of pharmacy claims data.
Such data is useful for assessing adherence with drug therapy (46) in determining blood glucose control and its correlation with adherence to drug therapy. Such information could have been used to validate the conceptual model presented in Figure 1. Last, information regarding prescription copayment would have been of use, given evidence that acquisition costs affect medication purchase.
Despite these limitations, we feel that this study provides evidence to support the hypothesis that poor adherence leads to greater instability; in this case manifested as a change in medication dosage, or in the combined outcome of change in medication dose or type of medication dispensed.
Lack of adherence to prescribed drug therapy is an acknowledged problem.
This problem can be complicated by a patient's lack of awareness of an adherence problem, or a lack of truthfulness in describing medication-taking behavior to the physician (49). The result of poor adherence with therapies for controlling blood glucose can have adverse consequences. Prescribers that increase the amount of medication prescribed to a poorly adherent diabetic patient may cause great risk of inducing dangerous hypoglycemic reactions.
Further, the incidence of microvascular, and perhaps macrovascular complications will be increased in patients with adherence problems and corresponding poor glucose control.

CONCLUSION
Non-adherence with prescribed sulfonylurea regimens , as measured by possession of medication less than 80 percent of days, increases the likelihood that prescribed hypoglycemic therapy will be changed by 43 to 50%.
Such change in therapy, presumably a response to poor blood glucose control , fails to address the underlying cause of poor glucose control, and may potentially cause dangerous hypoglycemic reactions. Increased awareness of non-adherence with prescribed medications as a cause of inadequate blood glucose control is necessary. Further, interventions aimed at improving adherence should be increasingly examined as a potential effective means to improve diabetes care.

Monotherapy with a Sulfonylurea or Metformin, and Dual Therapy With
Both Agents

ABSTRACT Background
Adherence to prescribed drug therapies for chronic conditions is known to be poor. Lack of optimal adherence to drug therapies for diabetes is particularly problematic, since poor glucose control has been associated with an increased risk of certain types of diabetic complications.

Objective
To determine if diabetic patients prescribed dual therapy with both a sulfonylurea and metformin as separate prescriptions are more likely to be nonadherent than patients receiving monotherapy with either agent. for at least 80% of days, and 43% of such patients possessing medication for at least 90% of days. Using multivariate logistic models controlling for the effects of age and the number of dispensings, we determined that patients receiving dual therapy with a sulfonylurea and metformin were more than 3

Conclusions
As determined using the medication possession ratio , adherence to oral hypoglycemic drug therapies was frequently sub-optimal.
Patients that possessed medication for less than 80% of days were likely not achieving tight control of blood glucose. Adherence was poorest among those receiving the combination of a sulfonylurea plus metformin.
The success of intensive therapy in tightly controlling the blood glucose level is contingent upon patient behavior. Though perhaps difficult to predict, adherence can be assessed. Diabetes management strategies should include the assessment of adherence with prescribed hypoglycemic drug regimens, and the application of interventions designed to improve adherence.
Additionally, the complexity of the drug reg imen should be considered as a potential barrier to optimal adherence.

INTRODUCTION
Non-adherence to prescribed drug regimens is a primary cause of sub-optimal health outcomes. Despite the efforts of prescribers and pharmacists, as many as one-half of patients or more will not achieve the full benefit of prescribed medications due to problems with adhering to therapy (1, 2).
Various types of interventions intended to improve adherence rates have been studied. However, research has not uncovered any specific intervention that effectively improves adherence to the degree where the full benefit of therapy is realized (3)(4)(5). The utility of such interventions ultimately depends upon the ability of health care providers to identify non-adherence with treatment regimens and , most importantly, depends upon characteristics of individual patients (6).
Further complicating the picture, despite greater than two decades of study regarding adherence with prescribed therapies, researchers have been mostly unable to consistently demonstrate associations between the behavior of adherence and various potential predictive factors (7). For example , some researchers have reported an association between adherence with prescribed therapies and age (8,9), gender (9, 10), and race (11,12), though others have not (13 , 14). Beyond such patient-specific factors , researchers have attempted to understand and describe the underlying psychological determinants of adherence. Models borrowed from the psychological sciences have been used to characterize and predict the behavior of adherence with some success. For example, the health belief (15), self-efficacy (16), and stages-of-change models (17) have all been shown to be useful in explaining adherence to drug therapy.
Factors specific to the drug regimen prescribed have also been examined as potential influences of adherence to therapy. Regimens having increased complexity and greater behavioral demand have been shown to be associated with decreased adherence (18,19). The number of drugs prescribed and number of required daily dosages have been shown to be associated with adherence, with the probability of non-adherence increasing in patients prescribed multiple medications and receiving divided daily doses (20,21 ).
This phenomenon has been perhaps best demonstrated in studies of patients treated for hypertension (22)(23)(24) and human immunodeficiency virus infection (25). The cost of and access to medications may also be important barriers to adherence for some patients.
Adherence to drug therapies prescribed for diabetes is particularly important.
Hypoglycemic therapy is essential for preventing acute complications resulting from elevated blood glucose, such as the nonketotic hyperosmolar state.
Additionally, tight control of blood glucose has been demonstrated to reduce the incidence of several types of chronic diabetic complications (26)(27)(28) In this study, we sought to determine if the complexity of the drug regimen was an influence of adherence to therapy. Specifically, we hypothesized that diabetic patients would be less likely to adhere to regimens that utilized two drugs as compared with regimens consisting of a single hypoglycemic agent.
To test this hypothesis, we used pharmacy data to assess hypoglycemic drug utilization among patients receiving dispensings for a sulfonylurea, metformin, or both of these agents together but as separate tablets. 120

Population
We conducted a cross-sectional study using data provided by Consumer Thus, the final sample included only patients that received dispensings for a sulfonylurea and/or metformin during a four-month period, and received some type of hypoglycemic medication throughout a 12-month period. Patients were classified by type of hypoglycemic drug regimen prescribed : either monotherapy with sulfonylurea or metformin, or dual therapy using both drugs.

Calculating the medication possession ratio
We We assessed adherence using MPR thresholds of 8:10 and 9:10. Nonadherence was defined as failing to possess medication for at least 80% or 90 % of days in the period . For example, the patient that received 90 tablets in 113 days would be classified as adherent using the 8:10 threshold , but classified as non-adherent using the 9:10 threshold . Results based upon these two MPR thresholds were presented separately. For patients dispensed both sulfonylurea and metformin as dual therapy, non-adherence was defined as failing to possess either medication for a number of days sufficient to cover eight of ten or nine of ten days in the period . Thus, patients prescribed both drugs must have adhered to both medications to be classified as adherent.
The MPR threshold of 8: 10 has been used in previous studies. Some researchers have used a ratio of 8:10 , or 80% of days covered, in studies of adherence to other chronic therapies such as hormone-replacement therapy (31 ); hypertension (24,32 ); and depression (33). Additionally, we also used second MPR threshold of 9:10 to define adherence based on estimates of what might be required to achieve tight glycemic control.

Potential confounding variables
In addition to the type of hypoglycemic drug regimen prescribed , we assessed the influence of five other potential confounding factors that were derivable from the pharmacy data provided. Three of these variables , age, gender, and insurance type, were obtainable from the patient profile. We stratified age into three categories: < 50, 50-64, and 65 years of age or older; representing younger, middle-aged , and senior patients. Insurance type was also categorized into three types. Of all insurances associated with prescription refills, two area health plans accounted for nearly 90 % of all dispensings.
These two insurance types were classified as health plan A and health plan B.
A third category of insurance type included patients whose prescription coverage was an insurance type other than health plan A or B, or patients that paid cash for all dispensings. In cases where more than one insurance type (or cash) was identified , the insurance type associated with the first non-cash dispensing was used .
We also sought to examine the effect of the number of medications prescribed , hypothesizing that a greater number of prescriptions would be associated with a greater likelihood of non-adherence to therapy. For this variable, we determined the total number of dispensings during the four-month period , including medications for conditions other than diabetes.
DiMatteo et al found depression to be a risk factor for non-adherence with prescribed drug therapies (34). Thus, we also decided to include a fifth potential confounding variable representing patients who had received an antidepressant medication. We decided not to include patients receiving an tricyclic anti-depressant in our categorization method, since the popularity of these agents for use in depression is declining (35,36) and these agents are used for conditions other than depression (37)(38)(39)(40)(41). Patients were considered to have depression if they received at least one dispensing for a medication presented in table1.

Statistical methods
Univariate statistics were used to determine the frequency and percentage of patients that received dispensings for a sulfonylurea, metformin , or both. The Chi-square analyses were used to assess bivariate relationships between proportions for all independent variables and the medication possession ratio.
Bivariate logistic models were also constructed, with separate models assessed for each independent variable and the medication possession ratio.
For the dependent variable of medication possession , ratio thresholds of 8:10 and 9:10 were assessed . Multivariate statistics were used to identify the presence of collinearity between independent variables, to determine the presence of interactions between variables, and to assess the association between type of reg imen prescribed and medication possession, controlling for other potential confounding variables. Collinearity diagnostics were performed using the PROC REG procedure for multiple regression as suggested by Allison (42). Collinearity between independent variables was assessed separately for MPR thresholds of 8:10 and 9:10. The presence of collinearity was determined using thresholds for condition index and proportion of shared variance as described by Tabachnick and Fidell (43).
The likelihood ratio test was used to test for interaction between variables in multivariate logistic models , using the chunk test as described by Kleinbaum (44 ). For this procedure, we calculated the difference in the -21og statistic between full and reduced models. Full models included all possible interaction and single terms; reduced models included single terms only. The difference in -21og statistic between full and reduced models was tested for significance using the chi-square distribution with alpha .05, and with degrees of freedom equal to the difference in terms between the two models. A difference in -21og value that was less than the chi-square statistic was considered sufficient evidence that the model could not be better fit by including interaction between variables.
Various multivariate models were assessed. Initial models contained all independent variables: non-significant terms were removed to create several other models. When non-significant terms were removed from models, we examined the magnitude of change in the parameter estimate beta and confidence interval for the variable regimen type (dual or monotherapy). The association between regimen type and medication possession is presented in a summary table that presents the parameter estimate, standard error, odds ratio, and 95% confidence interval obtained for each model.

RESULTS
A total of 1537 patients were identified as receiving dispensings for either a sulfonylurea, metformin, or both. Of these three types of hypoglycem ic drug regimens , a majority of patients were classified as users of sulfonylurea only The frequency and percent of patients possessing medication for at least 80% and 90 % of days is presented in Table 3. Patients receiving dispensings for only a sulfonylurea were most frequently determined to have medication possession ratios greater then 8: 10 and 9: 10 (77% and 69% respectively) .
Patients receiving dispensings for only metformin were less frequently identified as having an MPR above 0. 80  Likewise, type of insurance did not differ significantly between groups for either MPR threshold.
Patients receiving greater than 15 total dispensings were least frequently categorized as having an MPR less than 8:10 or 9:10, and those dispensed less than 5 prescriptions were most frequently categorized as not possessing a sufficient quantity of medication using either MPR threshold. These differences were statistically significant overall , and among the group of patients receiving dispensings for sulfonylurea only (p < 0.01 for each). The lesser number of patients in the metformin only and in the metformin plus sulfonylurea regimen categories likely affected the ability to detect a statically significant difference for these groups.
Among patients dispensed sulfonylureas, those who received a dispensing for a medication for depression more frequently had an MPR less than 8:10 or 9:10 (p < 0.05). The percentage of patients who were prescribed an antidepressant did not differ in statistical significance for other regimen categories or overall.
To determine the appropriate form for inclusion into logistic models , the parametric form of the continuous variables age and number of dispensings was assessed. Age was relatively linearly related to the MPR, with the likelihood of being categorized as having an MPR less than 8:10 or 9:10 decreasing with advancing age. Age categories used for analyses were based upon the frequency distribution of ages and for ease in presentation and interpretation. We did not find the presence of a strong linear trend between the total number of dispensings and the MPR. However, patients receiving 12 or more dispensings were less likely to be classified as having an MPR less than 8:10 or 9:10, while a total of less than 12 dispensings did not prove to be associated with the MPR. Thus, for use in logistic models, we dichotomized the total number of dispensings at 12 prescriptions.
Tables 5a and 5b present the results of bivariate logistic models for each independent variable using MPR thresholds of 8:10 and 9: 10  These results provide evidence that the self-management of diabetes is far from ideal for many patients. Clearly, patients who are not in possession of medication cannot be achieving the 'tight control ' of blood glucose that has been demonstrated to reduce the incidence of many types of diabetes complications. Unfortunately, the finding that many diabetic patients do not strictly adhere to prescribed hypoglycemic drug regimens is not unexpected .
Other researchers have provided evidence that adherence to hypoglycemic therapy is often sub-optimal. Venturini et al (45) reported a mean medication possession percentage of 83 % of days among HMO patients prescribed a sulfonylurea . In a study using Medicaid claims data, Sclar et al (46) found that only 39.4% of newly treated diabetics receiving a second-generation sulfonylurea obtained at least 6 months supply of medication during a 12month period . In an assessment of hypoglycemic drug possession among nearly 3,000 Scottish diabetics, Morris et al (47) found that roughly one-third of patients prescribed sulfonylureas received enough medication to cover 90% of days. Poor adherence to prescribed therapies among diabetic patients has been described by several researchers (48)(49)(50)(51)(52)(53).
The primary aim of this study was to determine if adherence to prescribed drug regimens was poorer among patients prescribed dual therapy with a sulfonylurea plus metformin as compared with patients receiving only one medication as monotherapy. Indeed , patients prescribed the two hypoglycemic drugs together were less likely to possess medication . In addition , the difference in adherence was pronounced. As compared with patients receiving dispensings for only a sulfonylurea or metformin , the percentage of patients having a medication possession ratio less than 8:10 or An influence of the number of medications prescribed on adherence has also been described. Sellars and Hayes (54) note that in general the probability of non-ad herence can be expected to increase with the number of medications prescribed. Treatment complexity has been shown to be a barrier to adherence in diabetic patients (55). In other health conditions, the number of medications prescribed has been shown to be an influence of drug utilization .
For example, persistence with a single-pill combination of two antihypertensive agents was found to be superior to persistence rates when the same agents were prescribed as separate pills (56). Additionally, treatment complexity is a recognized feature of adherence to drug therapies for human immunodeficiency virus (57,58).
The age of the patient was also found to be associated with medication possession. In th is study, bivariate analyses revealed that patients 50 years of age or older were less likely to be categorized as non-adherent. This finding is consistent with research by Sclar, who also found older diabetic patients to be more likely to obtain oral hypoglycemic medication regularly (46). This finding is explainable using the health beliefs model as described by Becker (59).
According to this model , patients that perceive their disease to pose a threat to health will be more likely to adhere to therapies that they perceive will lessen the risk of illness. Thus, in this study, perhaps younger diabetic patients were poorly adherent to drug therapy due to a lessened concern about poor longerterm outcomes resulting from persistent uncontrolled blood glucose.
Conversely, older patients may be more concerned about impending ill health, and thus be more likely to adhere to therapy.
Patients receiving 12 or more dispensings were 55% percent less likely to be non-adherent for MPR thresholds of 8:10 and 9:10 days. However, our calculations of total dispensings do not equate to the total number of medications prescribed, since non-adherence and differing days supply of medications may be responsible for some portion of the difference in the number of dispensings. Nevertheless, patients that received less than 12 dispensings were more likely to be identified as having an MPR below 8:10 and 9:10 days. A possible explanation for this finding is that patients receiving less than 12 dispensings during the four-month period were less ill, and 139 perhaps less adherent for reasons related to the health beliefs model as previously described. Also, it is likely that patients receiving 12 or more dispensings were afflicted with other co-morbid conditions and thus more vigilant in their adherence to therapy in an effort to stave off ill health .
Removing the term for the total number of prescriptions from multivariate models resulted in a decrease in the odds of non-adherence among patients dispensed both a sulfonylurea and metformin. Thus, controlling for the number of prescriptions dispensed resulted in an increase in the odds for nonadherence among those prescribed both drugs. Such effect was lacking for all other independent variables except age. Gender and insurance type were not found to be significant influences of medication possession .
Depression was identified as a risk factor for non-adherence by Di Matteo (34 ).
In an attempt to control for the presence of depression , we identified patients that received a dispensing for one of several anti-depressant medications.
For such patients, the risk of nonadherence to drug therapy was greater.
However, the 95% confidence interval did not exclude the possibility that no Additionally, patients receiving dual therapy may have initially failed monotherapy due to poor adherence . Thus, the group of patients using two hypoglycemic medications may inherently be more likely to be nonadherent.
Importantly, this study did not include many important factors known to be associated with adherence to prescribed drug regimens . Though we assessed the influence of gender and age, other patient-related factors may have been important to include. For example, race may have been an influence on adherence with sulfonylurea treatment, as found by Sclar et al (46). More notably, other complex behavioral factors may have been responsible for the differences in medication possession observed . For example, a patient's self-efficacy in adhering to prescribed regimens has been associated with superior adherence (29). Add itionally, a patient's stage of readiness to adhere to a prescribed regimen may prove explanatory (62 , 63).
In this study, the identification and incorporation of factors related to perception and beliefs may have influenced the association between drug regimen type and medication possession .
Additionally, important factors related to the characteristics of the disease were not included in this analysis. It is possible , for example, that some 142 patients classified as non-adherent in fact no longer required medication.
perhaps from adherence to recommended diet or exercise programs. Also , a percentage of patients classified as non-adherent may have been hospitalized during the assessment period. Additionally, the severity of disease may have also been important to assess. as highly symptomatic patients may have been more likely to adhere to therapy. Severity of disease may partly explain the finding of increased medication possession among older patients.
Other potentially important confounding factors relate to obtaining medication.
Information related to prescription co-payment was not provided , thus the influence of out-of-pocket expenditure on medication possession cannot be assessed . Additionally, though patients were required to obtain prescriptions at designated pharmacies for reimbursement. it is possible that some patients filled prescriptions at other locations by using other insurances or by paying cash. A last potentially important factor related to the drug regimen not included in this analysis relates to the type of drug therapies studied. Though sulfonylureas and metformin remain popular therapies in the treatment of type 2 diabetes. newer hypoglycemic agents have become increasingly utilized during the past several years. Though dual therapy with a sulfonylurea plus metformin was found to be associated with an increased likelihood of nonadherence. the association between other hypoglycemic drug combinations and medication possession is uncertain . One must not assume that adherence to therapy would also be lesser for combination therapies that include, for example, a thiazoladinedionne, meglitinde , or alpha glucosidase inhibitor.
Finally, it is important to note that dose frequency was not included as an influence on medication possession . Both metformin and sulfonylureas may be prescribed as one or two daily doses; depending upon the agent prescribed for the latter. (64 ). A sustained release preparation of metformin was not available during the period of study.
Despite these limitations , we found that patients receiving dispensings for both a sulfonylurea and metformin possessed medication for a lesser number of days as compared with patients who were dispensed only a sulfonylurea or metformin . This effect was strongest when controlling for the age of the patients and the total number of prescription dispensings. In this model , patients who received dispensings for both drugs were 3.14 times more likely to be non-adherent as defined as possessing medication for less than 8 of 10 days in the measurement period . Such patients were also 3.20 times more likely to be non-adherent using an MPR threshold of 9 in 10 days . The association between dual therapy with sulfonylurea plus metformin and an MPR of less than 8:1 O or 9: 1 O was also significant in bivariate analyses.
Adherence with drug therapy was found to be sub-optimal among all patients, regardless of regimen type. Greater than one in four of all patients did not 144 possess a sufficient quantify of medication to adhere to prescribed therapy at least 80% of days; and roughly two in five patients did not possess enough medication to adhere to therapy at least 90 % of days. Additionally, these estimates likely underestimate the true rate of adherence, since possessing medication is but one step in the process of adherence .
Despite the limitations described , this research should add to evidence demonstrating that adherence to hypoglycemic drug therapy is a fundamental problem , and potentially the greatest influence on the development of diabetic complications. Accordingly, adherence to drug therapy for patients with diabetes deserves at least as much attention as other components of diabetes management, such as preventative exams and monitoring of blood pressure and lipid levels. 145

CONCLUSION
As compared with monotherapy with a sulfonylurea or metformin, dual therapy with a sulfonylurea plus metformin is associated with a greater than three fold increase in the likelihood of non-adherence, controlling for the age of the patient and total number of medications dispensed. Patients receiving monotherapy with sulfonylurea were most frequently adherent, with 77% of patients possessing medication for at least 8of10 days in the study period , and 67% receiving at least enough medication to cover 9 of 10 days. Of those receiving monotherapy with metformin , 77% possessed enough medication to adhere 8 of 10 days , and 57% of patients possessed enough medication to adhere 9 of 10 days. The lowest rate of adherence was among patients receiving dispensings for a sulfonylurea and metformin. Among patients receiving both drugs, 57% received enough medication to adhere 8 of 10 days , while only 43% of patients possessed at least enough medication for 9 of 10 days.
In this population many patients did not adhere to drug therapy to the extent considered necessary to achieve tight glycemic control. Such patients can be expected to be at increased risk for several types of diabetic complications.
Thus, components of diabetes management must include the assessment of adherence with prescribed hypoglycemic drug regimens , and the application of interventions known to be effective in improving adherence . Additionally, the complexity of the drug regimen should be considered as a potential barrier to optimal adherence.      Intensive therapy has not been shown to significantly reduce the incidence of macrovascular diseases such as myocardial infarction or stoke. Though a trend for a reduction in these endpoints was observed among those receiving intensive therapy in the UKPDS, the frequency of such events was substantially greater than the frequency of microvascular complications (13).
Thus, though the UKPDS trial was sufficiently powered to detect a statistically significant reduction in macrovascular complications, no such reduction was observed . Additional information regarding the association between regimens of tight control and macrovascular disease will be provided from the ongoing Veteran 's Affairs Diabetes Trial (14), which is assessing the effect of intensive blood glucose control on cardiovascular compl ications in patients with type 2 disease. Meanwhile, current evidence suggests that control of blood glucose alone is not sufficient to significantly reduce the incidence of heart attack, stroke, or diabetes-related death . Aggressive management of other cardiovascular risk factors is essential , including reduction of LDL-C cholesterol and blood pressure to below target levels (15).

Creation of drug regimen categories
The frequency of dispensing of hypoglycemic medications was assessed using the PROC FREQ procedure. Regimen categories were defined based on dispensing frequencies and with a priori awareness of regimens of interest; such as users of su lfonylurea only and users of sulfonylurea plus metformin.
To minimize misclassification resulting from changes in regimen during the 12month period, we based drug regimen classification on the first three months of the 12-month period . Using the regimen categories presented in Figure 1, 98.3% of patients were classified as receiving one of nine types of hypoglycemic drugs regimen , with 1.72% of patients classified as receiving an 'other regimen' . The variable REGVAR was created , representing regimen categories 1-10. To assess changes in drug regi men during the 12-month period, we also categorized patients based upon the last th ree months of the 12-month period, using the same regimen classification strategy. The ten levels of variable EN DVAR represented the regimen category utilized during the last three months of the 12-month period .

Identifying changes in reg imen category and dose
Changes in drug regimen utilized between the first and last three months of the 12-month period were identified by comparing val ues for variables REGVAR and ENDVAR. We also determined which regimens patients were switching to, presenting both the frequency and percentage of change.
We also identified changes in the dose of medication utilized.

Determining the medication possession ratio
The assessment of medication possession was based upon the first four months of the 12-month period . We calculated the total days supply of medication received for all dispensings preceding the last dispensing during the four-month period ; and determined the total number of days between the first and last dispensing. The medication possession ratio (MPR) was defined as the total days supply of medication received compared with the total number of days between the first and last dispensing. Patients were considered to be nonadherent if the MPR was less than 8:10, or less than 9:10 for some analyses. Patients who received only one dispensing during the four-month period but who received medication in subsequent months were classified as non-adherent.
Among patients dispensed either sulfonylureas or metformin , those dispensed a greater number of prescriptions for any health condition were less frequently categorized as failing to possess medication for less than 80% of days. The outcomes of change in dose or change in class of medication dispensed were not found to be significantly influenced by differences in the medication possession ratio.
The aims of the analyses described in manuscript 3 were to assess adherence with hypoglycemic medication via the medication possession ratio and to compare medication possession among patients dispensed either monotherapy with sulfonylurea or metformin with patients receiving dual therapy with both medications. Additionally, we examined the association between medication possession and age, gender, insurance plan, the total number of dispensings, and the dispensing of a medication typically used to treat depression.
Based upon medication possession, adherence to any hypoglycemic drug regimen was frequently sub-optimal. Greater than one in four of all patients studied (28.8%) did not receive a quantity of medication sufficient to cover 8 of medication for at least eight of ten or nine of ten days in the assessment period , respectively. The least adherent sub-population was patients receiving dispensings for both metformin and sulfonylureas . Less than half of such patients possessed medication for at least nine of ten days (43 .1 %), while 56.9% of patients received at least enough medication to cover eight of ten days.
In both bivariate and multivariate logistic models, dual therapy with a sulfonylurea and metformin was associated with an increased likelihood of failure to possess medication for at least eight of ten and nine of ten days. In bivariate logistic analyses, patients were nearly 2.5 times more likely to be classified as nonadherent if they received dispensings for both drugs (MPR < were found to be a significant influence on medication possession.
In sum, the hypoglycemic drug utilization patterns of diabetic patients and various sub-populations can be characterized as frequently changing, and differing substantially between age categories. Older patients in this population more frequently received dispensings for sulfonylureas only, less frequently received dispensings for troglitazone or metformin as monotherapy, and more likely to refill prescriptions when due. Additionally, the 12-month total cost of all hypoglycemic medication dispensed was lowest among