Associations Between Patient Characteristics and Utilization in a Commercial Schizophrenia Population

Background: Schizophrenia is a prevalent and costly condition in the United States. Many studies have been conducted on the schizophrenia populations receiving government sponsored insurance, but less is known about the 16% of the population that receives commercial insurance. A better understanding of the utilization and outcomes in this population is essential to ensure that outreach programs target the groups most in need, that these programs are aimed at the most important aspects of utilization, and that those factors are tied to meaningful clinical outcomes. Objectives: The purpose of this research has been to better understand the patient characteristics, utilization patterns, and outcomes in patients with schizophrenia that participate in commercial insurance plans. Three studies have been completed to address the following specific aims: 1) To describe the schizophrenia population, 2) To determine if the sociodemographic, clinical, and employment characteristics of these patients are associated with their utilization patterns, and 3) To determine if adherence to therapy is associated with the rate of hospitalization for mental health conditions. Methods: In order to accomplish these goals several studies have been completed utilizing claims data from calendar years 2000 and 2001. The first is a retrospective cohort analysis identifying relationships between utilization of first and second generation antipsychotics, switching between therapies, and combination therapy and patient characteristics; the second study identifies the associations between patient characteristics and adherence; the final study utilizes a retrospective cohort design to determine the association between adherence and hospitalizations. Results: Patient characteristics are a significant predictor of utilization, with individuals living in the North Central region and individuals with comorbid bipolar disorder significantly more likely to use second generation antipsychotics. Adherence was associated with comorbid diabetes and mental health disorders. Adherence as measured by an MPR greater than or equal to 80% was associated with a lower risk of hospitalization due to mental health conditions. Conclusion: This series of studies has identified significant associations between comorbidities and increased likelihood to switch medications, utilize a second generation antipsychotic, or combine therapies. Comorbidities also increase the likelihood that someone will not be adherent to their therapy. Low adherence to therapy in turn increases the likelihood of hospitalization.


Second generation antipsychotic use was associated with living in the North
Central area of the United States (OR 1.3, 95% CI 1.053 -1.603), comorbid bipolar disorder (OR 1.271, 95% CI 1.072 -1.507), and switching (OR 1.937, 95% CI 1.568 -2.393). Individuals who switched therapy were nearly twice as likely to have been taking a first generation antipsychotic before the switch (OR 1.962, 95% CI 1.583 -2.431) Combination therapies were less likely in the North Central (0.727, 95% CI 0.554 -0.955) and Southern (0.534, 95% CI 0.411 -0.694) regions as compared to the North East.

Conclusion
There were significant associations between certain demographic, clinical, and employment characteristics and the antipsychotic therapies received by individuals in the commercial population.

Background:
Prevalence Mental health disorders are prevalent in the United States, and treating these conditions can be difficult and expensive. A study conducted by Kessler et al, found that the lifetime prevalence of schizophrenia for adults in the United States is between 0.3% and 1.6%. 1 A systematic review of the literature found that the period prevalence of schizophrenia is between 0.13% and 0.82%. 2 An evaluation of a commercially insured population found that the prevalence of schizophrenia in that group is near the lower end of the range, at about 0.13%. 3 Despite being less prevalent in the commercial population, people in this group make up a significant portion (16%) of the schizophrenic population. 3

Treatment Decisions
Finding an appropriate treatment for schizophrenia is difficult. It can take up to sixteen weeks to see the positive effects of treatment, 4 and even with effective treatment the negative symptoms of schizophrenia (such as social and emotional withdrawal, poor rapport, and blunted affect) are still common. 5 As a result, roughly 74% of patients are expected to discontinue treatment during the first 18 months. 6 In addition to difficulty finding and adjusting to a treatment, the adverse events associated with both first and second generation antipsychotics are associated with lower adherence. 7 In first generation antipsychotics, the extrapyramidal side effects were closely associated with discontinuation, while metabolic effects lowered adherence in users of Second Generation Antipsychotics, 6,8 A study by Staring, et al, found that high adherence was associated with decreased quality of life, as was low adherence due to the balance of symptoms and adverse events. 9 It has become clear that adequate treatment with antipsychotics will lower the mortality rate in the schizophrenia population, 10 but there are also associations between antipsychotics and increased cardiovascular mortality that have yet to be fully understood. 11 There are several studies that have tried to define the best strategy for selecting an initial antipsychotic. The CATIE and CutLASS studies were prospective open label trials, but the treatments included for study, population characteristics, and study limitations have limited the acceptance of their findings by clinicians. 12 Several studies have provided support for the use of First Generation Antipsychotics. In a Medicare population, Second Generation Antipsychotic users were significantly more likely to be hospitalized as compared to non-users, while First Generation Antipsychotic users had roughly the same hospitalization rate as non-users. 13 CATIE found that users of perphenazine saved roughly $300-600 over users of Second Generation Antipsychotics. 14 Amongst Second Generation Antipsychotics, a review of the literature found no clear benefit to any one choice. 15 Despite the limited support in the literature, and no preference in treatment guidelines, there has been a major shift in use of antipsychotics from the first to second generation agents. 16 Race, age, and comorbidities were all factors closely associated with the decision to use a second generation antipsychotic in a Veteran population. 16 The population utilizing antipsychotics is not limited to individuals with schizophrenia. In an analysis of a Veterans population, 60.2% of the population had no indication of a schizophrenia or bipolar diagnosis. 17 The most common off label use in this population was Post Traumatic Stress Disorder (PTSD) (about 40% of the population had this diagnosis). In a Medicaid population in Oregon, only 15% of patients receiving had a documented diagnosis of schizophrenia. 18 In addition to adverse events, comorbidities complicate the treatment of schizophrenia. It has been found that despite physician knowledge of the problem, schizophrenic patients still do not receive appropriate treatments for their comorbid conditions. 19 Patients with these conditions are often diagnosed with both mental health and general health issues, with one of the most common in schizophrenics being bipolar disorder. 20 In schizophrenia, comorbid depression is closely linked to relapses, 21 and anxiety is diagnosed in roughly 60% of the population. 22 The major general comorbidities driving treatment decisions are diabetes and cardiovascular disease. 23,24 Diabetes occurs in 10.3% of schizophrenics, as compared to 5.6% of the general population. 25 In the schizophrenic population, these increased risks are associated with greater risk of death from comorbidities, 26 and a nursing home admission rate in  year olds that is 3.9 times higher than the general population. 27

Switching and Combinations
Switching between antipsychotics is a fairly common practice. The CATIE trial found that in the best case scenario, 36% of patients taking olanzapine remained on treatment at 18 months. Only 17% of quetiapine patients, 26% of risperidone patients, and 20% of ziprasidone patients remained on these second generation antipsychotics at 18 months. Perphenazine represented first generation antipsychotics in this trial, and 25% of patients remained persistent at 18 months. 14 In almost all cases patients switched to a second generation antipsychotic, in only 1.2% of treatment episodes did an individual move from second generation to first generation. An earlier study by Menzin, et al, found that 58% of first generation antipsychotic users switched to another antipsychotic (most often a second generation antipsychotic) while only 33% of second generation antipsychotic users switched to another antipsychotic (also most often a new second generation antipsychotic). 28 The use of combination therapy is highly controversial. The practice has been observed in between 5% and 18% of the population according to a review of the literature. 4 There is scant evidence that polypharmacy results in better outcomes for individuals, 4 although patients taking multiple medications do demonstrate poorer adherence. 29 The most common combinations seen in database analyses involve two second generation antipsychotics, or a first and a second generation medication. 30 The Joint Commission for Quality Improvement has laid out a set of narrow guidelines where combinations of two or more antipsychotics might be reasonable, such as in patients utilizing Clozapine, multiple failed trials of monotherapy, or if discharged from a hospitalization with two or more medications. 31

Study Purpose and Justification
There have been many studies conducted to better understand the populations utilizing antipsychotics. The vast majority of the literature identified above focuses on specific populations with a high incidence of schizophrenia, those individuals utilizing Medicaid and VA programs. The commercial population likely differs significantly from the Medicaid and VA groups where patients are typically older, and (especially in the VA) more likely to be male. 16,28 This population is also different from the CATIE population, in which 85% of the population was unemployed. 6 A better understanding of how antipsychotic medications are utilized within commercial populations is critical. Sixteen percent of schizophrenics receive services through private insurance. 3 Programs designed to better manage these populations will utilize the improved information to better understand their participants. 32,33 The goal of this analysis is to describe the primary medication taking characteristics of commercially insured patients utilizing antipsychotic medications; these characteristics are: the use of first or second generation agents, switching treatments, and combination of treatments. These characteristics will be examined in relationship to the demographic, clinical, and employment characteristics of the patients.

Study Design:
A retrospective cohort study was conducted utilizing the Thompson MarketScan database. Those individuals utilizing antipsychotic medications, and diagnosed with schizophrenia were identified. The primary treatment outcomes studied were choice of first or second generation antipsychotic, switching of treatments, and use of combination treatment. Associations between these outcomes and the demographic, clinical, and employment characteristics of the populations were assessed.

Data:
Data for this analysis was made available through the Thomson Medstat dissertation support program. This program provides access to de-identified data in the MarketScan database from years 2000 and 2001. This database contains enrollment and demographics data, as well as medical and pharmacy claims for nearly 5.9 million individuals, including employees and their dependents working for large companies spread across the United States, and insured by roughly 100 different payers.

Inclusion Criteria:
Continuous enrollment is a key criterion for inclusion in this analysis.
Although this requirement narrows the study population significantly, it is essential to ensure that key events such as prescription dispensings and encounters with medical professionals are recorded in the database, and available for the description of the population. Individuals over 65 years of age during the study period were also excluded from the study population in order to avoid biases resulting from missing data associated with the coordination of benefits between Medicare and commercial insurers. Individuals under the age of 18 were excluded in order to avoid similar benefit coordination issues in a pediatric environment. Patients also had to have at least one diagnosis code indicating schizophrenia during the 2 year study period (Table I-1).

Operational Definitions:
Continuous enrollment was defined as having 366 days of continuous enrollment in 2000, and 365 days of continuous enrollment in 2001 with both medical and pharmacy coverage. This data was abstracted from the enrollment data sets provided.
Diagnoses for specific conditions were identified through the presence of one or more ICD-9 codes in the inpatient or outpatient records between January 1, 2000 and December 31,2001. Anxiety, Bipolar Disorder, Depression, Schizophrenia, Other Mental Health Disorders not included above, and Diabetes were all identified. The specific list of ICD-9 codes used to identify these conditions is provided in table I-1. This methodology is similar to that used in a variety of other studies, 3,[34][35][36] and validated by Rawson, et al, in 1997. 36 A treatment episode with a given therapy was defined as receipt of at least 2 dispensings on different days within the study period. Individuals with no treatment episodes were excluded from the bivariate and multivariate analyses.
Therapies were defined as a "new treatment" if there was no record of that treatment being received during the six months preceding the first fills. The "therapeutic period" for a treatment was defined as the time from the first fill, to the date of the last fill plus the last days supply. "Combination Therapy" was defined as two treatments overlapping by a minimum of 90 consecutive days. A "switch" in therapy is defined as a change from one treatment to another where there is no more than a 90 day gap in therapy, and no more than 90 days of overlap in therapy. A "gap" in treatment is defined as a period of 90 days or more following the end of a therapeutic period. Studies have typically used periods of 30 to 90 days to define gaps in therapy. 34,35,37 This study utilized the 90 day period to match the longest days supply routinely received by patients.
Demographic variables were examined for missing or obviously erroneous data through examination of distributions and outliers. An individual's age was defined as the difference between their birth year, and the year 2000. There were no instances in the data where an individual had more than 1 gender on record. Descriptions of employment type (primary vs. secondary policy holder, full time vs. part time, and hourly vs. salaried) and geographic variables were defined for each individual as the value that turned up most often for that individual (if more than one value was available).

Bivariate Statistical Analysis:
Bivariate analyses were conducted in order to determine the associations between the three outcomes (antipsychotic generation choice, switching, and combination use) and the demographic, clinical, and employment variables. The students t-test was utilized for age, the only normally distributed continuous variable. Chi-square analysis was conducted for the binomial and categorical variables.

Multivariate Statistical Analysis:
Because of the potential for strong relationships between many of the explanatory variables explored in the bivariate statistical analysis, logistic regression was also used in each of the populations to examine the following: 1. Factors associated with the decision to use a second generation antipsychotic or a first generation antipsychotic 2. Factors associated with switching 3. Factors associated with combination use The dependent variables tested were: type of antipsychotic, switch, and combination. The independent variables included in each model were: age, gender, region, rural / urban, full time employment, employee or dependent, industry, and comorbidies (schizophrenia, bipolar disorder, depression, anxiety, other mental health, and diabetes). Variables were included in the model if they were found to be associated with the outcome variable in bivariate testing (p < 0.2). The exceptions are age and gender, which were included regardless due to their importance in understanding the make-up of the population. Interaction terms including the various combinations of age group, gender, and comorbidities were also included if they were associated with the dependent variable as measured by the chi-square test with a significant p-value (<0.2).
Chi-square analysis was also used to assess several potential associations amongst independent variables to determine if they were independent. If the variables were not independent (chi-square value was less than 0.05), the less important variable was dropped from consideration for the model. Stepwise backwards elimination was used to optimize the model, using the -2 log likelihood to test the significance of changes. Multicollinearity was assessed based on the variation inflation factor (VIF), and eignevalues. These were calculated utilizing a separate model with the proc reg function in SAS with the VIF, TOL, and collin options. Hosmer-Lemeshow was used to assess goodness of fit.
. Switching between one or more antipsychotics occurred prior to 21.7% (N=702) of the treatment episodes (in 13.6% of the population), while combination therapy was observed in 421 (13.0%) treatment episodes (in 12.9% of the population). The most common switches were between second generation antipsychotics. Table I-4 demonstrates the switches identified in the antipsychotic users. The majority of combinations were with first and second generation antipsychotics (N=262), with combinations of two second generation antipsychotics also common (N=136). Table I-5 demonstrates the combinations utilized by this population.

Bivariate Analysis
First generation antipsychotic users were on average 3.1 years older than those using second generation antipsychotics in this population (table I-6), and 3.6% more individuals in the North East received a first generation antipsychotic than individuals living elsewhere; 9.2% more individual patients diagnosed with comorbid bipolar disorder received a second generation antipsychotic than those who were not diagnosed. After switching medications, the frequency of second generation antipsychotic use was 14.5% higher than if they were treatment naïve or starting on a therapy after a long gap.
Patients switching medications were on average, younger than those who did not switch, although there was no difference in age for those using combination treatments. Table I-7 describes associations with switching and combination use in this population. Patients with bipolar disorder in addition to schizophrenia were more likely to switch medications (p<0.0001). Those with anxiety were more likely to switch (p<0.001), but less likely to use combinations (p= 0.004). Type of employment was not associated with the rate of switching.

Multivariate Analysis:
In this population antipsychotic type was modeled against age, gender, region, comorbid diagnoses (bipolar disorder, depression, anxiety, and diabetes), switching, combination use, and employment variables for the full model. Salary, full time employment, combinations, and anxiety and depression diagnoses did not contribute significantly to the model (see table I Use of combinations of treatments in the schizophrenia population was modeled against age, gender, rural location, region, diagnoses, use of a second generation antipsychotic, switching, and employment variables. Rural location, use of a second generation antipsychotic, and diagnoses besides anxiety were removed from the model because they did not make a significant contribution to the model (table I-

Discussion:
The results of this study draw some significant distinctions between the commercially insured and government insured populations receiving antipsychotic therapy for schizophrenia. Understanding these differences will improve the ability of the commercial managed care organizations to direct resources and focus on ensuring appropriate care for those individuals most likely to be switching or combining therapy, and help to ensure that those prescribing decisions are appropriate. The ability to do this could lead to improved quality of care and lower overall costs for both the payors and patients.

Utilization of First vs. Second Generation Antipsychotics
Age was a significant predictor of medication choice, with older individuals more likely to receive first generation antipsychotics than younger patients. One other study was identified that has used claims data to identify factors associated with utilization of antipsychotics. The association between older age and lower frequency of second generation antipsychotic use was similar to that identified in a Texas Veteran's population by Yang, et al. 16 In agreement with our findings

Switching Rates
Individuals can switch between antipsychotics for a variety of reasons, including adverse events, lack of effectiveness, or concerns about cost.
Although patients are commonly more adherent to treatment after a switch, there are potential adverse events associated with switching, especially if patients do not titrate propperly. 39 The results of this study are useful in establishing a baseline switch rate for individuals in a commercial population.
The likelihood of having switched therapies was twice as high among those individuals currently using second generation antipsychotics, and roughly 40% higher in those with bipolar disorder or depression. Age and gender were not significant influences on medication switching. Although having a comorbid diagnosis of diabetes was associated with increased use of first generation antipsychotics, those with diabetes were also 34.5% more likely to switch medications. The increased rate of switching makes sense in those individuals with a more complex clinical situation due to the increased likelihood of adverse events and poorer adherence leading to inadequate outcomes, 28 and these trends are also supported by literature evaluating a Medicaid population. 28,39 The rate of switching from first generation antipsychotics to second generation antipsychotics was also similar to that seen in the CATIE trial, which found that 18% of patients switched medications during an 18 month time period. 14

Use of combination therapy
The frequency of combinations between first generation antipsychotics and second generation antipsychotics, as well as multiple second generation antipsychotics seen in our population was similar to that seen in the literature. 30 The overall combination rate of 12.9% was in the range identified in a review by Stahl, et al, which was 5-18% of all users of these medications within outpatient populations. 4 Combination therapy was significantly less common in individuals working full time, or currently responsible for providing insurance coverage.
These factors are likely to be closely associated with disease severity, although given the short time period studied, it cannot be determined whether they simply are better responders to monotherapy, or if this group has a less severe underlying condition overall. Generally, the existing literature does not support the utilization of polypharmacy, due to a lack of improved outcomes, increased adverse events, and higher costs, 4 although there are narrow circumstances in which combining multiple medications may be necessary and acceptable, such as after failure to respond in 3 or more trials of monotherapy, individuals using Clozapine, and those released from inpatient treatment with combination therapy. 31 Combination therapy was also less common in individuals living in the North Central and Southern regions as compared to the North East. Further study is needed to understand why these regional differences exist, and no other studies conducted in the United States have been identified that address this discrepancy in therapy across regions. A Danish study identified regional differences in the understanding of clinical guidelines as one possible reason for differences in polypharmacy. 40

Limitations
Although the goal of this analysis was to ensure generalizability to the commercially insured US population, the requirement of 2 years continuous data may have limited the inclusion of some of the more severe patients that were not enrolled for the entire study period. However, the proportion of schizophrenic individuals excluded from the study population due to our continuous enrollment requirement was smaller than the proportion of the overall population. Because the time available in the data set is relatively short, it was not possible to reliably establish a temporal sequence, limiting the ability to understand the relationships between the observed factors and medication choice, switching, or combination use. The age of the data also limits generalizability due to the addition of new treatments and guidelines that may have subsequently changed practice over time.
This study is reliant on claims data submitted by physicians and pharmacies for the purposes of billing, therefore there are some limitations seen across retrospective database analyses. Pharmacy data indicating a medication is

Conclusions
This analysis demonstrates the significant differences between a commercially insured schizophrenia population and the more commonly evaluated populations in federally insured programs. Comorbidities with diabetes and bipolar disorder were key drivers of increased therapy switches, while age and gender played a smaller role than that observed in other populations.
Additional study is necessary to determine if these factors also impact an individual's ability to adhere to therapy.

Background
Adherence and persistence to medications for schizophrenia is typically less than that seen in other classes of medication. Additionally, studies of adherence to antipsychotics are often completed in populations with a high prevalence of schizophrenia, which may not be generalizable to the commercially insured population. The purpose of this analysis was to determine the factors associated with adherence to therapy in a commercially insured population.

Methods
A retrospective cohort study was completed in order to determine if demographic, clinical, or employment characteristics described in administrative claims data are associated with adherence and persistence in this population. Adherence was calculated based on a medication possession ratio of greater than 80%, and persistence was based on a gap of 90 days or more in therapy.

Results
There were 1,086 individuals identified with an ICD-9 code indicating schizophrenia and at least 2 dispensings of the same medication.

Conclusion
Several factors drive an individual's adherence and persistence to medication regimens. Adherence and persistence were both decreased in individuals with comorbid mental health conditions such as bipolar disorder or anxiety.
Demographic and employment characteristics were not significant predictors of adherence or persistence.

Prevalence of Schizophrenia
Treatment of Mental health disorders can be difficult and expensive, and these conditions occur fairly commonly in the United States. The lifetime prevalence of schizophrenia is between 0.3% and 1.6%, 1 with the period prevalence of schizophrenia between 0.13% and 0.82%. 2 Although less common in commercially insured populations, the prevalence of schizophrenia in that group is still not negligible at about 0.13%. 3 Despite being less prevalent in the commercial population overall, people in this group make up a significant portion (16%) of the schizophrenic population. 3

Antipsychotic Treatment
Adherence to treatment is a key stumbling block for many patients receiving antipsychotics. A review of the literature by Llorca found that several studies demonstrate a significant relationship between low adherence and relapses and hospitalization. 4 The difficult side effects of antipsychotics, which include extrapyramidal effects in first generation antipsychotics and metabolic effects in second generation antipsychotics, alone are a barrier to adherence. A study by Bulloch found that 35% of individuals were non-adherent to their antipsychotics, 5 and Second Generation Antipsychotics were associated with a better likelihood of adherence than First Generation Antipsychotics in several studies. 6,7 Less complicated regimens 8 , levels of social support, 9 and patient attitude towards treatment are additional areas associated with patient adherence. 10 It has also been demonstrated that treatment naïve patients are less likely than those who have switched from another medication to be adherent and persistent, 11 demonstrating the difficulty in selecting a first medication.
In the schizophrenia population, it is clear that decreased adherence leads to a decreased treatment response, 12 a higher rate of relapse, 13,14 and greater risk of hospitalization. [15][16][17] The Law study in particular found that within 10 days of a missed refill, the risk of hospitalization increases significantly. 17 In schizophrenic patients, adherence was better on second generation antipsychotics, 18 23 Barriers such as lacking social support also affect adherence in this population. 24

Study Purpose and Justification
There have been many studies conducted to better understand factors in prospective open-label trials such as CATIE, in which 85% of the population was unemployed. 26 It is critical to gain a better understanding of medication utilization patterns the commercial population. Sixteen percent of schizophrenic patients receive services through private insurance, and represent a fairly significant cost in this population. 3 Programs designed to better manage these populations will utilize the improved information to better understand their participants, leading to more targeted interventions. 27,28 In addition to better understanding this population, there are also a wide variety of methods utilized to measure adherence to antipsychotics described in the literature, although the MPR method is most common. 17,29,30 MPR and proportion of days covered (PDC) have both been shown to be associated with poor outcomes such as hospitalization. 29 Additionally, evaluating large gaps in therapy as a proxy for discontinuation provides insight into longer drug free periods and may be associated with different factors than MPR and PDC. 17 The goal of this analysis was to evaluate factors that may be associated Diabetes was also identified due to the association between second generation antipsychotics and an increased risk of metabolic outcomes. The specific list of ICD-9 codes used to identify these conditions is provided in Therapy with a given treatment was defined as receipt of at least 2 dispensings of that medication within the study period. The "therapeutic period" for a treatment is the time from the first fill, to the date of the last fill plus the last days supply, and any days spent in the hospital. "Combination Therapy" is defined as two treatments overlapping by a minimum of 90 days. A "switch" in therapy was defined as a change from one treatment to another where there is no more than a 90 day gap in therapy, and no more than 90 days of overlap in therapy. A discontinuation in treatment is defined as a period of 90 days or more without therapy following the end of a therapeutic period, excluding any days in the hospital. Studies have typically used periods of 30 to 90 days to define gaps in therapy. 18

Bivariate analysis
Because of the typical left skewed, truncated nature of the distribution of MPR, each individual treatment episode was categorized as adherent (MPR ≥80%) or non-adherent (MPR <80%). The 80% cut-off has been utilized as the threshold in several publications describing adherence rates and outcomes due to poor adherence in populations treated for psychiatric conditions including schizophrenia. 18,21,38 Treatment gap of 90 days or more was treated as a dichotomous variable. The relationship between the outcomes of interest (adherence and discontinuation) and the clinical and demographic variables (first or second generation antipsychotic utilization, treatment combination or switch, demographic, and employment variables) was tested using the chi-square analysis. The association between age, the only continuous explanatory variable, and adherence and discontinuation was also tested utilizing the student's T-test.

Multivariate Analysis:
Two separate Logistic regression models were used in order to adjust for multiple factors influencing MPR and discontinuation. Adherence to therapy was modeled against first or second generation medication choice, switching, combination use, demographic, clinical, and employment factors. Interaction terms including the various combinations of age group, gender, and comorbidities were also included if they were associated with the dependent variable as measured by the chi-square test with a significant p-value (<0.2).
Chi-square analysis was also used to assess several potential associations amongst independent variables to determine if they were independent. If the variables were not independent (chi-square value was less than 0.05), the less important variable was dropped from consideration for the model. Stepwise backwards elimination was used to optimize the model, using the -2 log likelihood to test the significance of changes. Multicollinearity was assessed based on the variation inflation factor (VIF), and eignevalues. These were calculated utilizing a separate model with the proc reg function in SAS with the VIF, TOL, and collin options. If the condition number was greater than 30, or the VIF was greater than 5, increased scrutiny was given to the affected variables. Hosmer-Lemeshow was used to assess goodness of fit.

Multivariate results:
The multivariate analysis found several associations between adherence and clinical, demographic, and employment related factors. Table II- both associated with less likelihood to be persistent. Table II-9 provides the odds ratios from the final model.

Discussion:
This analysis demonstrates the importance of comorbidities in the likelihood that a patient will be adherent to and persistent on therapy. Observed decreases in adherence associated with a diabetes or other mental health diagnoses may lead to treatment being ineffective, with resultant increases in the need for medical care, as well as increases in the indirect costs of schizophrenia.
This analysis gives caregivers, providers, and insurers new insight into the factors that increase the risk of non-adherence and non-persistence in the commercial population, and these can be utilized to identify the patients who would be the best targets for outreach attempting to improve adherence.

Factors Associated with MPR
Adherence measured by MPR is the most commonly utilized methodology in literature addressing claims based studies. The adherence rate found in our study (61.3% of the population having an MPR above 80%) is higher than that seen in other studies, for example Ascher-Svanum found that 58% of participants in the US Schizophrenia Care and Assessment Program (US-SCAP) had an MPR above 80%, Valenstein found that 60% of the VA population studied had an MPR greater than 80%, and Gilmer found that only 41% of the Medicaid population they studied was adherent using the same criteria. 20,38,39 There are a number of potential explanations for this finding. First, studies have shown that social support is a key driver of adherence, 9 in this analysis dependents tended to be more adherent than primary policy holders, thus dependent status may be an indicator of the social support available to these individuals. In addition to differences in the population, there may also be differences in the methodology used to calculate MPR, where some studies do not exclude days in the hospital from the denominator, or use the entire study period as the denominator, rather than just the treatment period.
The multivariate results show that certain clinical characteristics are key drivers associated with adherence. There are differing reports on superior adherence rates in first vs. second generation antipsychotic users, with a VA population having lower adherence to second generation antipsychotics (62.2% and 58.8% with an MPR >80% in first vs. second generation antipsychotics respectively), 20 and a study based on patients with a first episode of psychosis in the community found that second generation antipsychotics resulted in better adherence (MPR = 59.4% vs. 34.5% in first generation users); 6,9,20 in the study described here, we found that there was no significant difference in adherence for second generation antipsychotic users as compared to first generation antipsychotic users.
Individuals with other mental health disorders also had lower adherence rates, which correlate with the existing literature. 40 A study by Lang et al found that combination users were more likely to be adherent than those using single treatments (71% in combination users vs. 70% in first generation users and 64% in second generation users), agreeing with our study that found that combination users trended towards higher adherence than single medication users (OR 1.361, 95% CI 0.939 -1.972). 40 There was also lower adherence among patients having diabetes in our study, which may be indicative of concern about metabolic adverse events, or associated with increased complexity of care in general. 41 Other studies have also found limited associations between demographic, economic / work, and clinical variables and adherence. 19

Factors Associated with Persistence
Persistence indicates the length of time an individual is able to remain on treatment. Due to the relatively short timeframe available for this study, most individuals were already on a treatment at the beginning of the study, and still taking it at the end. Therefore we were unable to calculate a true estimation of treatment duration in this population, but with nearly 50% of the population receiving therapy without more than a 90 day gap at the beginning and end of the study, the commercially insured population performs significantly better than the CATIE population, where 74% of new users discontinued therapy by 18 months, 26 as well as a VA population where the median time to discontinuation was 120 days. 33 Being persistent on therapy is crucial as the risk for poor outcomes such as hospitalization can begin to increase in as little as 10 days after discontinuing therapy. 17 Our study demonstrated an increased risk for discontinuation in populations with comorbid bipolar disorder or anxiety. These conditions likely complicate therapeutic regimens due to their episodic nature, leading to discontinuations due to changes in overall disease states, but also due to confusion on the part of the individual. 8 Overall, there were very few variables associated with discontinuation following adjustment with logistic regression. This is in agreement with the literature where Gianfrancesco et al found that there was no difference in persistence between first and second generation antipsychotics in a commercial population. 18 In an evaluation of the VA population, there was less likelihood of discontinuation in second generation antipsychotic users as compared to first generation antipsychotic users. Several issues may be causing this difference in outcomes. First, the VA study did not control for comorbid conditions in their analysis. Additionally there are some substantial differences in the populations, with the VA consisting largely of older male patients at the time, and the Medicaid population likely having significant differences in sociodemographic characteristics such as income and social support that could not be measured in either study.

Limitations
Studies utilizing claims records to measure adherence are prone to several well documented limitations. For instance, without observing an individual taking a medication one cannot be sure that it is being taken.
Pharmacy claims can also be incomplete if patients receive samples from their physician, or pay cash rather than their copay for inexpensive generics, resulting in underestimates of adherence.
The short time period available in our data also limited the ability to follow patients over a long period of time based on trade-offs with population size.
Future studies using a broader population, and longer timeframe may also be able to identify more significant drivers of adherence and persistence than those seen in our population. The age of this data may also limit generalizability due to the addition of new treatments and changes in guidelines over time.
Generalizability is also limited to the commercially insured population represented by the data.

Formatted for submission to the Journal of Managed Care Pharmacy (JMCP)
, not yet submitted.

Background
Adherence to treatments for schizophrenia are associated with significant adverse events that often lead to poor adherence and discontinuation of therapy.
The lack of adherence to therapy may limit its effectiveness and be associated with poor outcomes. The purpose of this analysis is to determine if there is a significant difference in mental health associated hospitalization rates for those individuals with high adherence as compared to those with lower rates of adherence.

Results
Matching was successful in selecting a non-adherent comparison group for 76.5% of the adherent population, with significant support across the full range of propensity scores. The matching process was successful in limiting the differences in demographic and clinical characteristics between the groups. The risk of hospitalization was higher in the non-adherent population (RR 1.55, 95% CI 1.07 -2.25) than the adherent population after matching. This is slightly lower than the relative risk seen in the unmatched group (RR 1.74, 95% CI 1. 22 -2.49 in the adherent population vs. the non-adherent population).

Conclusion
Adherence to antipsychotic treatment has a significant association with a lower likelihood of hospitalization in the schizophrenia population. Additional analysis is necessary to understand the impact of persistence, and other medication taking behaviors such as switching or polytherapy.

Treatments and Adherence
Adherence to treatment is a key stumbling block for many patients receiving antipsychotics. It can take up to 16 weeks for a treatment regimen to become effective for relieving symptoms, but adverse events can occur far earlier. 4 A review of the literature by Llorca found that several studies demonstrate a significant relationship between low adherence and poor outcomes. 5 The side effects of antipsychotics alone are a barrier to adherence.
These adverse events are largely metabolic for second generation antipsychotics, while first generation antipsychotics are associated with extrapyramidal side effects. A study by Bulloch found that 35% of individuals were non-adherent to their antipsychotics, 6 and second generation antipsychotics were associated with a better likelihood of adherence than first generation antipsychotics in several studies. 7,8 Less complicated regimens, 9 levels of social support, 10 and patient attitude towards treatment are additional areas associated with improved patient adherence. 11 It has also been demonstrated that treatment naïve patients are less likely than those who have switched from another medication to be adherent and persistent, 12 demonstrating the difficulty in selecting a first medication.
Adherence rates in the literature vary substantially depending on the population, and methods used to measure it. The medication possession ratio (MPR) is based on the sum of the days supply divided by the duration of therapy, whereas the proportion of days covered (PDC) is the number of days with therapy available divided by the length of the study. 13 Individuals with a diagnosis of schizophrenia and a medication possession ratio (MPR) greater than 80% range between 35% and 60%, 6, 14-17 coincidentally, the average MPR or PDC also ranged between 35% and 60% in different studies. 8,10 This demonstrates the importance of ensuring that results are interpreted with population differences in mind, as results from one group may not be generalizable to another.

Impact of Poor Adherence
Non-adherence to medications leads increasingly poor outcomes over time. 5 Poor adherence is a key driver of relapses in schizophrenia, leading to costs that are 3 times higher than non-relapsers. 18 One key driver of this cost is hospitalization. A Canadian study in 2006 found that individuals who had a medication possession ratio (MPR) of greater than 80% had 63% fewer hospitalizations. 19 A Medicaid population in Wisconsin had twice as many hospitalizations in the non-adherent group, and costs were 4 times higher. 20 The risk of hospitalization increases quickly following discontinuation, with a study by Law et.al. observing the risk increase in as little as 10 days. 21 Medication therapy choice regarding switching, combination therapy, and use of first vs. second generation antipsychotics might also impact hospitalization rates. In a Medicare population, second generation antipsychotic users and those on combination therapy were more likely to be hospitalized as compared with those utilizing first generation antipsychotics without combining, although the impact of adherence was not addressed in this analysis. 22

Study Purpose and Justification
It has been estimated that hospitalizations could be decreased by 12.3% in the Medicaid population, saving $103 million per year in the United States if gaps in therapy of longer than 15 days could be eliminated. 23 Interventions have been successful in improving adherence rates, but additional information regarding the causes of adherence, and the populations where non-adherence is most likely could make these programs more effective. 24 A key to making this information useful is ensuring that it is generalizable to the population where these programs are implemented.
There have been many studies conducted to better understand the factors leading to hospitalizations in the schizophrenia population. The vast majority of the literature identified above focuses on specific populations with a high incidence of schizophrenia, those individuals utilizing Medicaid and VA programs.
The commercial population likely differs significantly from the Medicaid and VA groups where patients are typically older, and (especially in the VA) more likely to be male. 22,23,25 This population is also different from that identified in Clinical Antipsychotic Trials of Intervention Effectiveness (CATIE), a prospective comparative effectiveness trial, in which 85% of the population was unemployed. 26 A better understanding of medication use within the commercial population is critical. Sixteen percent of schizophrenics receive services through private insurance, and represent a fairly significant cost in this population. 3 The purpose of this study is to determine the association between adherence to antipsychotics and decreased rates of hospitalization, while controlling for other medication taking behaviors and patient characteristics that could be associated with increased hospitalization rates in the commercially insured population receiving antipsychotics.

Study Design
A matched retrospective cohort study was conducted. Individuals who were adherent to antipsychotics were matched to non-adherent individuals based on propensity scores measuring the likelihood of adherence to medication. codes to identify schizophrenia and comorbidities matches the methodology used in a variety of other studies, 3,27,28 and is similar to methodology utilizing hospitalization billing records validated by Rawson, et al, in 1997. 29 Therapy with a given treatment was defined as receipt of at least 2 fills within the study period. The "therapeutic period" for a treatment is the time from the first fill, to the date of the last fill plus the last days supply. "Combination Therapy" is defined as two treatments overlapping by a minimum of 90 days. A "switch" in therapy is defined as a change from one treatment to another where there is no more than a 90 day gap in therapy, and no more than 90 days of overlap in therapy. A "gap" in treatment is defined as a period of 90 days or more following the end of a therapeutic period, plus any days in the hospital (for Propensity scores were developed for the likelihood that a patient will be adherent given their demographic, clinical, and employment characteristics (see table III-2 for a list of specific variables). Chi-square analysis was conducted in order to identify key characteristics based on a p-value greater than 0.2. The full logistic regression model was then optimized using backwards stepwise regression. The log-likelihood test was used to determine if there were significant differences in the models. Correlations were utilized to identify risk of collinearity, and the Hosmer-Lemeshow test was used to assess goodness-of-fit.

Matching and Statistical Analysis
Individuals with an MPR greater than 80% were matched 1:1 to nonadherent individuals based on the propensity scores. A nearest neighbor matching algorithm was used, utilizing calipers set to ½ of a standard deviation of the propensity scores. Standardized differences were used to assess balance in variables before and after matching. This methodology is the same at that used by Bangelore et. al. 35 in their analysis of the impact of beta-blocker use on clinical outcomes. The difference in hospitalization risk between adherent and non-adherent individuals was also assessed using the Chi-square test. Logistic regression was also utilized to control for factors differing between the hospitalized and non-hospitalized populations following matching. This model was optimized and tested in the same manner as the propensity score models described above.

Population:
This population examined in this study is described in table III-2. The mean age of the population was higher in the pre-match population than the post-match population (46.7 vs. 46.3) prior to matching, but the difference was

Propensity Scores:
The propensity score model was successfully optimized (see table III-3) based on changes in -2 log likelihood. Goodness-of-fit was adequate based on a c-statistic = 0.782, and Hosmer-Lemeshow p-value = 0.482. The key drivers of adherence based on this model were age, gender, region, anxiety, other mental health diagnoses, diabetes, full time employment, and salary vs. hourly pay.
There was support for matching adherent to non-adherent individuals across nearly the full range of assigned propensity scores (see figure III-2).

Matching:
Matching was successful in limiting the differences in key variables associated with adherence between the adherent and non-adherent populations.
A match was identified for 76.5% of the adherent individuals, utilizing 91.6% of the non-adherent individuals. Figure III-3 describes the standardized differences in key variables before and after matching.
Prior to matching, age was significantly higher in the adherent population than the non-adherent population (47.7 vs. 45.5, p-0.002), but after matching the difference was no longer statistically significant (45.8 vs 46.9, p-0.169 in the adherent and non-adherent groups respectively). Differences between the adherent and non-adherent populations based on gender remained insignificant at the alpha = 0.05 level, as was the case for location and most of the comorbidities evaluated. There was a statistically significant difference before matching in other mental health disorders, which did not exist following matching.
The variables that were significant at the alpha = 0.2 level prior to matching were also no longer significant after matching. Table III-2 provides the details of these results. The details of the model fitting procedure are provided in table III-3.

Impact of Adherence on Hospitalization:
The impact of adherence on hospitalization rates was apparent in the unmatched population, as well as the matched only and matched and statistically controlled groups. The relative risk of hospitalization in the matched nonadherent group was 55% higher than in the adherent population (RR 1.55, 95% CI 1.07 -2.25). In the unmatched population, the relative risk of hospitalization in the non-adherent group was 74% higher than in the adherent population (RR There were also significant differences in a variety of clinical, demographic, and employment characteristics between hospitalized and nonhospitalized individuals both before and after matching. Table III-5 describes these characteristics in the hospitalized and non-hospitalized populations before and after matching. The strongest drivers of hospitalization in addition to adherence in both cases were comorbid bipolar disorder, anxiety, depression, and other mental health conditions, and therapy characteristics including choice of first vs. second generation antipsychotic, switching, and combination use. After controlling for confounders using logistic regression, adherence was still a significant predictor of hospitalization avoidance (OR 0.627, 95% CI 0.394 -0.999). Individuals with bipolar disorder, depression, and other mental health conditions also had significantly higher rates of hospitalization. See table III-6 for   model diagnostics, and table III-7 for odds ratios generated by the logistic model.
The difference in hospitalization rates between users of first and second generation antipsychotics were not significant, but those who used combinations or switched therapies were significantly more likely to be hospitalized. Table III-8 describes the differences in hospitalization rates between the matched and unmatched populations, and table III-9 describes the differences in odds ratios between methods.
Poor adherence is a significant predictor of increased hospitalization risk for individuals with schizophrenia. This analysis clarifies the risk of nonadherence to therapy in a commercially insured population, and provides a clear incentive for patients, providers, caregivers, and insurers to ensure that those patients receiving antipsychotics remain true to their regimen. Identifying those individuals at risk for non-adherence and ensuring they are receiving support to overcome their barriers to adherence could lead to significantly lower numbers of hospitalizations, which in turn could lead to lower costs and essentially pay for the support programs.

Impact of adherence on hospitalizations
A variety of studies have looked at the association between adherence and hospitalizations due to mental health disorders, and many of them have found that hospitalization rates were nearly reduced by half between those who were adherent and those who were not. 19,20,36 This study found a more modest difference in hospitalization rates, with a relative risk of hospitalization of 1.55 (95% CI 1.07 -2.25) in the non-adherent population, as compared to adherent populations. Although adherence to medication is an important factor in avoiding hospitalizations, several other key factors emerged as well, with comorbidities, polypharmacy, and treatment switches increasing the likelihood of hospitalization.
The hospitalization rates seen in this population are below those seen in Medicaid populations 20 (42% in the Medicaid non-adherent population vs. 14.5% in the commercial non-adherent population). This reinforces the need for population specific studies, as the severity of disease in this population is significantly lower as measured by hospitalizations.

Meaning of different methodologies
The importance of properly controlling comorbidities was also evident after examining the results of this analysis. Although the same conclusion holds throughout the unmatched, matched, and matched with logistic regression analyses, the magnitude of the impact decreases significantly with stronger controls for selection bias and potential confounders. Many of the studies performed in the literature did not control as carefully for these biases, and may be over-estimating the impact of adherence on hospitalization rates. Although adherence is important in avoiding hospitalizations, inflating the impact may divert limited funds from managed care outreach programs that address polypharmacy and other therapeutic issues to those focused more directly on adherence. In the end, it would be ideal to address all of a patient's medication taking behaviors, but the statistical analysis in this study shows that getting people settled on a suitable monotherapeutic regimen might result in a lower rate of hospitalizations.

Limitations
This study is limited to inferences that can be made from claims databases. A matched cohort study cannot be used to assess causality, although every attempt was made to address potential confounders, issues known to impact adherence and risk of hospitalization such as social support and disease severity could not be observed based on claims data. The limited timeframe of the study also made it difficult to assess the impact of persistence on outcomes. Users of depot medications were excluded from the study, potentially eliminating individuals with more severe, or long-term illness. The age of the data utilized in this study may also lead to limited generalizability due to changes in available therapies and differences in guidelines. The results are also limited in generalizability to the commercially insured population, and differ significantly for similar studies of Medicare, Medicaid, and Veterans populations.
Studies utilizing claims records to measure adherence are prone to several well documented limitations. For instance, without observing an individual taking a medication one cannot be sure that it is being taken.
Pharmacy claims can also be incomplete if patients receive samples from their physician, or pay cash rather than their copay for inexpensive generics, resulting in underestimates of adherence. Severity of illness is also a confounder that is difficult to control for based on claims data, but may have biased the results if individuals with more severe disease were also less likely to be adherent to therapy.

Conclusions
Regardless of methodology, there risk of hospitalization increases substantially if individuals are not adherent to therapy as prescribed. Although adherence is a significant issue, polypharmacy and switching between treatments are also closely tied to hospitalization, and future studies should assess their impact.     Table III-5