Differential Medication Adherence Among Patients With Schizophrenia and Comorbid Diabetes and Hypertension

Data were obtained from the VA National Psychosis Registry ( 28 ), which includes clinical and health service utilization data for all VA patients with a psychotic disorder. Information in the registry is compiled from annual extracts of administrative databases representing all VA inpatient and outpatient services, as well as all medication refills. The VA Ann Arbor Healthcare System Human Subjects Committee approved analyses such as those in this study with a waiver of informed consent.

Using the registry, we identified patients in fiscal year 2001 (FY 2001; October 1, 2000, through September 30, 2001) who had ICD-9 diagnosis codes indicating schizophrenia or a schizoaffective disorder, as well as both diabetes and hypertension. Only patients with refill information indicating that they were using oral medications for each condition during the year were eligible for inclusion in the analytic sample of 3,397 patients. For these patients, we evaluated medication adherence in FY 2002. Once patients are prescribed medication for schizophrenia, diabetes, and hypertension, they may switch drugs but rarely discontinue medication use altogether. Thus we assumed that all patients should be refilling each medication type throughout the observation period. No patients filled prescriptions for insulin during the study period, suggesting that all should have been using oral hypoglycemic agents continuously. We excluded 1,623 patients who were using more than two medications within a class, because adherence measures for these complex regimens have not been validated. Finally, we excluded patients with fewer than 90 noninstitutionalized days of observation in FY 2002 and those who filled two or more prescriptions as outpatients during an inpatient stay. The final analytic sample included 1,686 patients.

Medication possession ratios (MPRs) provided a validated measure of patients' adherence on the basis of their pattern of medication refills ( 5, 9, 29 ). For patients with only one medication within a therapeutic class, MPRs were calculated by dividing the total number of days of medication supplied by the number of days' supply needed if the patient was taking a full dosage of medication continuously during outpatient periods in FY 2002. For patients taking more than one medication within a therapeutic class, we calculated medication-specific MPRs and then a weighted average to estimate the class-specific MPR. If two medications within a class were prescribed concurrently, the denominator for the second drug's "days' supply needed" extended only until the end of its prescription if the other drug within the class continued to be prescribed. The equation was thus MPR=number of days of medication supplied by outpatient pharmacies)/(number of days' supply needed for continuous outpatient medication use).

The numerator of the MPR was calculated by adding the number of days' supply from each outpatient prescription filled during the year. Medications received at the time of discharge from inpatient stays were included in the outpatient supply. For prescriptions filled in FY 2001, we included the number of days' supply, if any, that extended into FY 2002. The denominator of the MPR was calculated as the number of noninstitutionalized days between the start of the fiscal year and either the end of the year or the date of death.

An MPR of 1.0 indicates that the patient received all the medications needed to take a full dosage continuously throughout the period, whereas an MPR of .5 indicates medication sufficient for only half of needed days. As in prior studies ( 5 ) some patients had MPRs greater than 1.0, most likely because these patients lost prescriptions or providers frequently changed their prescription dosages.

Information on patients' sociodemographic characteristics and health service use was obtained from the National Psychosis Registry. Medication copayments and other limits on coverage are waived for VA patients with military service-connected disabilities, and we captured data on service connection from administrative files. Patients' distance to VA care was calculated on the basis of zip code of residence according to standard procedures ( 30 ). We used indicators of psychiatric and nonpsychiatric hospitalizations in the 12 months before FY 2002 to measure the severity of patients' psychiatric and physical illness. Patients' involvement with outpatient care was measured by the number of mental health and medical outpatient visits during the year. We included a count of all medications filled during FY 2002 as a measure of patients' overall medication burden.

In addition to medication type, we considered three other characteristics of each treatment regimen: whether the patient filled one or two different medications of one type during the year, the average number of days of medication supplied per refill, and the proportion of medication refills that were mailed to the patient versus dispensed at VA outpatient centers.

The primary goal of our analyses was to examine intrapatient variation in medication adherence across treatment types and to estimate the impact on adherence of other medication-specific characteristics and patient-level covariates. We fit multivariate logistic regression models in which MPRs less than .8 were used as the dependent variable to indicate a significant adherence problem. Medication type was the primary predictor variable of interest. MPRs less than .8 have been found in prior studies to be associated with poorer outcomes ( 5, 29 ). All multivariate analyses were multilevel, with patients' adherence represented by three observations, one observation for each therapeutic class. Additional medication-level data were included (such as average refill interval) as well as patient-level covariates (such as race). We controlled for the nesting of adherence data within patients and patients within facilities by using the cluster command in SAS ( 31 ).

Three nested multivariate models were estimated. In model 1, we included medication type indicators as well as indicators of patients' sociodemographic characteristics as controls for confounding. In model 2, we included all of the variables in model 1, as well as measures of patients' health service use. Model 3 included all previous variables as well as additional information about the number of medications in each therapeutic class (one or two), average number of days supplied, and the proportion of refills dispensed by mail versus in outpatient centers.

We conducted extensive auxiliary analyses to test the robustness of the models. Analyses with MPR cutoff points of 50%, 60%, 70%, and 90% showed a similar pattern of effects. We refit models for the subset of 1,318 patients with two full years of refill data—in other words, refills in FY 2002 and FY 2003—in order to expand the opportunities to detect poor adherence for treatments with relatively high numbers of days supplied per refill.

Most patients were male, Caucasian, and unmarried ( Table 1 ). Most had not been hospitalized in the prior 12 months, although patients had relatively high rates of outpatient mental health and physical service use. Single-medication regimens were more common for antipsychotic treatment (78% of patients) than for treatment of diabetes (54%) or hypertension (50%) ( Table 2 ). Patients were more likely to receive a medication supply of 30 days or less when refilling their antipsychotic medications (79%) than when filling prescriptions for their hypoglycemic (16%) or antihypertensive medications (11%).

Bivariate analyses suggested substantial intrapatient variation in MPRs. Figure 1 shows the MPRs for hypoglycemic versus antihypertensive medications. The distributions were similar for hypoglycemic versus antipsychotic medications and for antipsychotic versus antihypertensive medications. [Figures illustrating the MPRs for these other bivariate comparisons are available in an online supplement to this article at ps.psychiatryonline.org.] The bivariate correlation of MPRs was highest for patients' two physical disorders ( Figure 1 ), hypertension and diabetes (r=.53). Statistically significant but weaker associations were observed between the MPRs for each physical condition and patients' antipsychotic medication. Information about antipsychotic medication adherence explained only 13% of the variance in patients' antihypertensive MPRs, and antipsychotic MPRs explained only 16% of the variance in MPRs for hypoglycemic medications.

In unadjusted analyses, patients were less likely to have poor adherence (MPR less than .8) to their hypoglycemic (29%) or antihypertensive medications (26%) than to their antipsychotic medications (35%; p< .001). Multivariate analyses controlling for patients' sociodemographic characteristics and health service use ( Table 3, model 2), showed a similar pattern. However, when other characteristics of each medication regimen, including the average days' supply, were controlled for (model 3), the adjusted odds of poor adherence were significantly higher for hypoglycemic and antihypertensive medications than for antipsychotic medication (both adjusted odds ratios were 1.5, p<.001). Regimens with an average 30 days' supply or less were associated with a 3.3-fold increase in the relative odds of nonadherence as compared with a supply of 60 days or more.