Assertive Community Treatment in Veterans Affairs Settings: Impact on Adherence to Antipsychotic Medication

Study data were obtained from the VA National Psychosis Registry (NPR), which is developed and maintained by the VA Serious Mental Illness Treatment Resource and Evaluation Center in Ann Arbor, Michigan. The NPR includes national VA health system records for all VA patients with diagnoses of schizophrenia, bipolar disorder, and other psychoses.

Using the VA ACT programs’ enrollment registry and data from the VA NPR, we identified all 763 VA patients newly enrolling in ACT between fiscal years 2001 and 2004 (October 1, 2000–September 30, 2004) who had a diagnosis of schizophrenia in the 12 months before enrollment, were alive for 12 months after enrollment, met VA-defined criteria for high hospital use (three or more mental health inpatient admissions or 30 or more mental health inpatient days in the prior 12 months) (20), lived within 60 miles of a VA ACT team (the usual operating area of VA ACT teams), and had 90 or more outpatient days in each of five sequential six-month periods of interest (zero to six months before the date of ACT entry or index date for the control group and zero to six, seven to 12, 13 to 18, and 19 to 24 months after ACT entry date or the index date). To have MPRs calculated during these study periods, patients needed to have had an antipsychotic fill six to 12 months before ACT entry or the index date and also to have had 90 or more outpatient days in each six-month period.

Propensity scores for the probability of enrolling in ACT were used to match ACT enrollees 1:1 with patients who met all of the same inclusion criteria but did not enter an ACT program. Propensity scores were calculated on the basis of adherence at baseline, age, sex, race, marital status, homelessness, substance use disorder diagnosis, Charlson comorbidity score, number of inpatient psychiatric stays in the past year, number of inpatient psychiatric days in the past year, service connection, number of domiciliary days in the past year, number of residential rehabilitation days in the past year, number of mental health visits in the past year, number of substance abuse visits in the past year, and the Veterans Integrated Service Network (VISN) in which the patient received care. Propensity score matching is often used to balance between-group differences in the distribution of the key covariates, allowing matching on multiple factors (21).

As in prior studies, data on the days’ supply of antipsychotic medication dispensed during each six-month study period were used to calculate MPRs. The MPR is defined as the number of days’ supply of antipsychotic received from an outpatient pharmacy divided by the number of days’ supply needed for continuous outpatient antipsychotic use.

For each period, MPRs were calculated by adding the number of days’ supply of antipsychotic medication dispensed by the outpatient pharmacy during the six-month period plus any remaining days’ supply from fills in the prior six months that would have covered days during the period of interest. Medications received at the time of discharge from inpatient settings were included in outpatient pharmacy supplies. Days that patients spent in institutional settings (VA hospitals or nursing homes) were subtracted from the outpatient days’ supply needed.

MPRs calculated from VA pharmacy data have been shown to correlate with important intermediate patient outcomes, including psychiatric admission of patients with schizophrenia to VA facilities and other health care settings (2,22).

Data for all covariates were obtained from administrative data in the VA NPR. Measures included patients’ demographic data (age, sex, race, and marital status). Other covariates were constructed from data from the 12 months before the index date, including an indicator for homelessness (ICD-9 code V60.0 or services utilization in specific clinics or “bed sections”) (23), number of psychiatric inpatient days, number of psychiatric inpatient admissions, number of antipsychotics prescribed, and a substance use disorder diagnosis (ICD-9 codes 291.X, 292.X, 303.0, 303.9–305.0, and 305.2–305.9 without a fifth digit of 3 to signify remission). A modified version of the Charlson Comorbidity Index indicating the presence of 19 medical diagnoses in the year before the index date was included as a measure of medical comorbidity (24).

For an exploratory analysis of the level of ACT participation among enrollees and medication adherence, we constructed a variable for full and partial ACT participation. Level of participation was measured by the frequency of contact between the ACT team and the client. Patients who completed at least 42 ACT visits in a 12-month period were considered to be fully participating in ACT treatment, whereas patients who completed 41 visits or fewer in a 12-month period were considered to be partially participating in ACT care. This represents a relatively conservative estimate for full ACT involvement. VA criteria for ACT programs indicate that ACT usually involves two or three contacts per week, with lower levels of contact only after a year of treatment. VA health system regulations also specify that VA ACT programs must have at least 41 contacts per client per year to qualify for a higher-than-normal budgetary allowance for these patients (25).

Univariate analyses of patient characteristics and good adherence (MPR ≥.8) at various points were described by means and frequencies.

Bivariate analyses assessing the relationship between good adherence and ACT enrollment were completed by using chi square statistics. To assess differences in rates of good antipsychotic adherence among patients who entered or did not enter ACT, we used a generalized linear model with logit link and a generalized estimating equation (GEE) method. These analyses adjusted for patient demographic variables, baseline adherence, homelessness, substance use diagnosis, medical comorbidity score, number of antipsychotics prescribed, and prior inpatient utilization. GEE analyses more appropriately estimate regression coefficients and variance when correlated data are used (26). We used a GEE approach because adherence observations were nested within patients over the two-and-one-half years of the study.

In exploratory analyses, we used chi square statistics to explore whether patients who enrolled in ACT showed significant differences in medication adherence by level of participation in ACT. We used GEEs to assess whether adherence (MPR) in the first year after the index date predicted inpatient admissions in the second year after the index date. A mixed-effects model was used to account for correlation of inpatient days within patient.

Statistical analyses were completed by using SAS software, version 9.2.

Table 1 shows the characteristics of ACT enrollees and ACT-eligible nonenrollees after propensity score matching. After propensity matching, there were no significant group differences across any covariates that might be associated with both adherence and with ACT enrollment.

The study sample (N=1,526) was predominantly male (89%) and older, with a mean±SD age of 50±10 years. Most patients were white (57%), but black patients constituted a substantial minority (35%). Approximately 22% of patients had an indicator for homelessness in the year before the index date, and 43% had a concurrent substance use diagnosis.

In the six months before the study index date, there were no differences in adherence (MPR ≥.8) among patients who enrolled (72%) or did not enroll (70%) in ACT. However, in all six-month periods that followed the index date, the percentage of patients with MPRs ≥.8 was higher among ACT enrollees than among nonenrollees, both in unadjusted (Table 2) and adjusted analyses (Table 3). In multivariate logistic regression analyses, the odds of good adherence among ACT enrollees were greatest in the six-month period immediately following enrollment. However, the odds of good adherence were twice as high or higher among ACT enrollees versus nonenrollees across all periods following enrollment.

After adjustment for demographic and clinical variables, GEE analyses that examined the impact of ACT enrollment over time indicated that ACT enrollees were 2.3 times more likely than nonenrollees to have good antipsychotic adherence (Table 4).

As outlined in Table 5, ACT enrollees who received higher levels of ACT services were more likely to have MPRs ≥.8. Between seven and 24 months postenrollment, approximately 80% of patients who averaged 42 or more ACT visits in a year had MPRs ≥.8, compared with 55% to 70% of patients with fewer ACT visits. Only 31% to 42% of patients who discontinued ACT involvement entirely in the second year had good antipsychotic adherence in that year.

An exploratory analysis of the study sample indicated that adherence in the first year was not significantly related to the number of inpatient admissions in the second year after enrollment. Only the number of inpatient stays, substance use, patient age, and the number of antipsychotics prescribed in the first year were significantly associated with the number of inpatient stays in the second year after enrollment. [A table presenting the results of this mixed model is available online as a data supplement to this article.]