Relapse of patients with schizophrenia is defined as a clinical deterioration or an exacerbation of symptoms based on a predefined threshold or on rehospitalization (
1,
2). Relapse negatively affects patient functioning and quality of life (
3) and increases costs of care (
4,
5). In 2010, approximately 22% of patients with schizophrenia who were hospitalized in the United States were readmitted within 30 days (
6), although relapse rates may depend on the patient population and clinical setting (
7,
8). Using antipsychotic medication to treat patients with schizophrenia reduces the frequency and duration of hospitalizations (
2); thus continuous treatment is recommended to avoid relapse (
9,
10).
Poor adherence to treatment is a strong predictor of relapse in schizophrenia (
7,
11,
12). Long-acting injectable antipsychotics (LAIs) offer a convenient alternative to daily oral medications. LAIs can be a valuable treatment option for patients with schizophrenia who have problems with adherence (
13), and LAIs may reduce relapse rates. However, a meta-analysis of 21 randomized controlled trials (RCTs) reported no difference between LAIs and oral antipsychotics in preventing relapse (
1). In contrast, a meta-analysis of 25 mirror-image studies, which compared periods of LAI versus oral antipsychotic treatment among the same patients, showed a strong benefit of LAIs over oral antipsychotics in reducing the risk and number of hospitalizations (
14). A recent mirror-image study examined treatment compliance and health care costs of 1,992 patients with schizophrenia or schizoaffective disorder after their first prescription of an LAI. During the following year, medication compliance improved significantly, and the number of hospitalization days was reduced (
15).
Although RCTs are considered the gold standard for scientific comparisons, they may not be suitable for comparing LAIs and oral medications, in part because the adherence benefit of LAIs versus oral medications in clinical practice may not be well captured in highly structured RCTs (
16). Patients with a history of or tendency toward nonadherence may be less likely than others to participate in RCTs, and adherence behavior may change during the course of the study because adherence is monitored. The tendency for higher treatment adherence in RCTs may affect results more for patients who are taking oral medications than for those taking LAIs (
16,
17). Although the population and study design in mirror-image studies better reflect the population receiving LAIs in clinical practice versus RCTs, such a study design also introduces selection bias that may affect the assessment of effectiveness of LAI and oral antipsychotics (
17).
As an alternative approach, this study analyzed medical claims of Medicaid beneficiaries, a population with a greater prevalence of schizophrenia (8.5%) (
18) than is estimated in the overall U.S. population (1.1%) (
19). The objective was to compare hospital readmission rates between patients with schizophrenia who were first treated with LAIs versus oral antipsychotics after an index hospitalization discharge and followed for one year.
Methods
Data Sources and Study Population
This was a retrospective cohort analysis of patients with schizophrenia (ICD-9-CM code 295.xx) included in the Truven Health MarketScan Multi-State Medicaid Database from October 1, 2007, through September 30, 2012. The database contains inpatient and outpatient medical claims and outpatient prescription administrative claims data for >28 million Medicaid enrollees and those enrolled in both Medicaid and Medicare from multiple geographically diverse states in the United States.
Patients with a schizophrenia diagnosis were identified if they had at least one inpatient claim in one year with schizophrenia listed as the primary or secondary diagnosis code or had two or more outpatient claims—separated by ≥30 days—with schizophrenia listed as the diagnosis. Patients codiagnosed as having bipolar disorder or major depressive disorder were included in the overall analysis but excluded from the analysis of patients with a sole diagnosis of schizophrenia. Included patients were 18 to 64 years of age at six months before the date of the index hospitalization and had at least one prescription filled for a first- or second-generation antipsychotic (oral or LAI) within a 12-month continuous Medicaid enrollment period after the index admission. Index hospitalization was defined as the first hospital admission within the analyzed period with the primary or secondary diagnosis of serious mental illness: schizophrenia, bipolar disorder, or major depressive disorder. Index hospitalizations preceding any other (nonserious mental illness) hospitalization within a period of less than three months were excluded. Antipsychotic medication use was identified in pharmacy claims based on the National Drug Code or J-codes on inpatient or outpatient medical claims. Patients with missing age data and patients with less than one year of follow-up or continuous enrollment following the index admission were excluded from the analyses.
Chronic and comorbid conditions were identified from the database with ICD-9-CM diagnostic codes. [Details are provided in the online supplement to this article.]
Outcome
The study outcome was rehospitalization for any cause at 30 or 60 days after the index hospitalization discharge date, stratified by the first antipsychotic medication prescribed after the index admission [see figure in
online supplement]. All-cause rehospitalization was chosen as the reported outcome because quality metrics of interest proposed by the Healthcare Effectiveness Data and Information Set/National Committee for Quality Assurance focused on all-cause rehospitalization (
20), in part because any readmission, regardless of diagnosis code, can reflect the quality of the care received during and after the index admission, including the effectiveness and tolerability of prescribed medication. This outcome also aligns with the goals of the IMPACT (Improving Medicare Post-Acute Care Transformation) act of 2014 (
21), to foster cross-indication comparison. Rehospitalization was defined as a subsequent admission for any cause after the index hospitalization discharge date. The link between first antipsychotic use and rehospitalization becomes weaker over time because other factors, such as treatment adherence and medication switching, might influence treatment outcomes. Therefore, analyses were reported for data up to 60 days after index admission, with a one-year follow-up.
Rehospitalization rates were compared between LAIs on the basis of generic molecule names (second-generation antipsychotics aripiprazole, olanzapine, paliperidone, and risperidone; first-generation antipsychotics haloperidol and fluphenazine) and oral antipsychotics (second-generation medications aripiprazole, asenapine, clozapine, iloperidone, lurasidone, olanzapine, paliperidone, quetiapine, risperidone, and ziprasidone; first-generation medications chlorpromazine, fluphenazine, haloperidol, and perphenazine). Clozapine carries the risk of life-threatening adverse events and is generally prescribed to patients with treatment-resistant schizophrenia as a last resort. Therefore, statistical analyses controlled for clozapine use to avoid systematic and otherwise unobserved differences in disease severity. Statistical analyses also controlled for use of short-acting injectable antipsychotics (SAIs), often limited to emergency situations for acute agitation.
Statistical Analysis
Multivariate logistic regression and propensity score matching (PSM) methods were used to compare the effect of LAIs with oral second-generation antipsychotics on the probability of rehospitalization after discharge from the index admission. Patients prescribed LAIs may differ from those prescribed oral antipsychotics in parameters that ultimately affect patient outcomes, and logistic regression may not sufficiently mitigate such selection bias. Methods that match patients from different treatment groups on the basis of observable characteristics can reduce the imbalance and help isolate the effect of the treatment on outcomes. Therefore, a PSM model using kernel-weighted matching was implemented. Using a probit model, we estimated in stage 1 the probability that a patient was first prescribed an LAI as a function of age, race and ethnicity, and a set of comorbid conditions. In stage 2, using the Heckman difference-in-difference matching estimator, we estimated the effect of being prescribed an LAI on the probability of rehospitalization (
22). The Heckman model allows LAI-treated patients to be matched to LAI-untreated patients, who are weighted, with the use of kernel weights, by their similarity to the LAI-treated patient. The estimated model was
where y
i equals the probability of readmission to the hospital at 30 or 60 days, π captures the effect of using an LAI rather than an oral antipsychotic first after the index admission, and X′
i is a set of individual characteristics including age, race and ethnicity, comorbid conditions, and use of antipsychotics before index admission.
The PSM model generated a control group on the basis of patient characteristics, such as age, use of LAI or SAI during the preindex hospitalization period, and selected comorbid illnesses. [The online supplement provides details.]
Results
Data were analyzed for 1,450 patients with a sole diagnosis of schizophrenia and for 15,556 overall patients with schizophrenia, a sample which includes patients with a codiagnosis of bipolar disorder or major depressive disorder
(Table 1). In the overall sample, there were slightly more women than men, patients were well distributed across the four age categories, and a majority were white. The readmission rate was 30.8% at 60 days. After the index hospitalization, 6.3% of patients were prescribed an LAI as their first postindex antipsychotic, and 84.4% of patients were prescribed an oral antipsychotic; the remaining 9.3% of patients were first prescribed an SAI. Most initial treatments were prescribed within 30 days of the index hospitalization. In the population for the PSM model, none of the patients who used an LAI first after the index hospitalization used clozapine in the following year; therefore, none of the patients in the matched control group used clozapine.
The unadjusted logit model odds ratios (ORs) indicated that the probability of hospital readmission was significantly lower 60 days after the index hospitalization for patients with a sole diagnosis of schizophrenia who were treated with LAIs versus oral antipsychotics (
Table 2). A significantly lower probability of hospital readmission with LAIs versus oral antipsychotics was also demonstrated with the unadjusted model for all patients at 60 days. When analyses adjusted for age, race and ethnicity, comorbid conditions, use of clozapine as the first oral antipsychotic, and preindex antipsychotic use (oral, LAIs, or SAIs), the odds of hospital readmission among the schizophrenia-only subpopulation were significantly lower with LAIs than with oral antipsychotics at 60 days after the index hospitalization. Consistent with this finding, adjusted ORs for all patients also indicated a significantly lower probability of rehospitalization at 60 days among those who received LAIs versus oral antipsychotics.
Table 3 shows the absolute mean difference in probability of hospital readmission between the LAI and oral antipsychotic patient groups calculated from the PSM model. PSM reduced differences in baseline characteristics between patients who were first prescribed an LAI versus an oral antipsychotic [
online supplement]. In the subpopulation with schizophrenia only, the probability of hospital readmission at 60 days after the index hospitalization was marginally significantly lower (within a .1 significance level) among LAI-treated patients compared with oral antipsychotic–treated patients and significantly lower in the all-patient analysis (
Table 3). In the all-patient analysis, the absolute difference was greater for the 60-day hospital readmission (−5.0%), translating into a relative reduction of readmission rate by 16%.
Discussion
This study analyzed data from a large sample of medical claims of patients with schizophrenia and used PSM to minimize selection bias. Medicaid patients with schizophrenia who received an LAI formulation as the first antipsychotic medication after a serious mental illness–related hospitalization were less likely than those who first received oral antipsychotics to be readmitted within 60 days. The effect was more pronounced at 60 days than at 30 days after the initial hospitalization, and these results were consistent in a multivariate logistic regression model and a PSM model accounting for patient characteristics.
Although the significance levels for the absolute rehospitalization rates differed between the two models for the all-patients sample, the results were qualitatively similar. The PSM model is more demanding of the data, and restricting the data set to the schizophrenia-only population necessarily provided fewer matched patients for the analysis. Nevertheless, the direction of effect for schizophrenia-only patients in the PSM model was the same as that for all schizophrenia patients, although the estimates were less precise. Lack of a significant effect at 30 days may reflect the need for a longer follow-up period to detect the benefit of improved adherence with an LAI formulation. Alternatively, it may result from a longer time needed to reach the therapeutic or steady-state concentration with LAI versus oral formulations (
23,
24). The benefit of LAIs appeared to be slightly stronger for patients with a sole diagnosis of schizophrenia compared with all schizophrenia patients, a group that included patients with comorbid bipolar disorder or major depressive disorder. Additional studies are required to fully describe the effectiveness of LAIs in treating patients with multiple psychiatric conditions.
The results from this study are consistent with reports from two previous retrospective studies assessing the risk of hospital readmission after first or second schizophrenia-related hospitalizations. A case-linkage study analyzing data from a large national database in Finland that included 2,588 patients with a sole diagnosis of schizophrenia reported that risk of readmission during a mean follow-up of two years for patients receiving an LAI after hospitalization was approximately 40% of that for patients receiving oral formulations (hazard ratio [HR]=.36, 95% confidence interval [CI]=.17–.75, p=.007) (
25). This strong protective effect of LAIs compared with oral antipsychotics may be at least partially attributed to low medication adherence; less than half of the patients continued initial treatment for ≥30 days. Finnish guidelines for treatment of schizophrenia recommend using oral antipsychotics for treatment of patients with acute and chronic disease and using LAIs for patients with inadequate insight and treatment adherence; in the reported nationwide cohort, about 8% of patients used LAIs as the first medication in outpatient care, a slightly greater proportion than in this study (6.3%). The patients included in the analysis might have been less severely ill than those in our cohort. The patients were hospitalized for the first time with a schizophrenia diagnosis, and the rate of rehospitalization during a mean follow-up of two years was 57.8%.
In the second study, which analyzed data from 3,828 patients in a large hospital-based database in the United States, hospital readmission rates over a mean follow-up period of 30 months were significantly lower for patients who switched to LAIs from oral antipsychotics after a first schizophrenia-related rehospitalization compared with those who continued oral therapy when matched on propensity scores (HR=.81, 95% CI=.76–.87, p<.001) (
26). The reduction in the risk of recurrence of hospitalizations associated with LAIs (19%) is comparable with that of this study (14%−16%), although over a much longer follow-up period.
In contrast to these results, LAIs were reported to be not superior to oral antipsychotics in reducing one-year rehospitalization rates among 14,610 patients frequently hospitalized for schizophrenia treatment (two or more psychiatric hospitalizations within two consecutive years) based on an analysis of a population-based cohort from a national database (2003–2008) in Taiwan (
27). Flupenthixol, used with 55.7% of LAI-treated patients in that study, was associated with odds of rehospitalization that were significantly greater than the odds found with the reference LAI, haloperidol (OR=1.44, 95% CI=1.01–2.06); flupenthixol LAI was not used for treatment of patients included in this study. Also, the Taiwanese study might have excluded a large proportion of patients (27.8%) lost to follow-up who tended to have poor adherence (
27). Alternatively, the risk of hospitalization of patients with unstable mental illness may be independent of the formulation of antipsychotic treatment (
28). Although our study did not assess differences based on frequency of past hospitalization, future analyses may attempt to explain the discrepancies between our findings with a Medicaid database and those previously reported for unstable patients or those with frequent hospitalizations (
27,
28). However, the rate of rehospitalization at 30 days observed in our population closely resembles that of a database including all-payer, all-cause 30-day hospital readmissions among patients with schizophrenia (
6), suggesting that the patients in our database may be representative of the broader population.
The patients in this study who received an LAI antipsychotic first after hospital discharge demonstrated reduced rehospitalization rates during the following 60 days compared with those who received oral antipsychotics. Although the absolute differences may appear relatively small (approximately 5%), they can be of great value to patients and payers alike, considering the critical importance of preventing such events.
This study was designed in part to minimize selection bias and analyzed data from a large sample of medical claims; however, unobserved patient heterogeneity may still result in residual confounding. The sample size for LAIs was much smaller than the sample for oral antipsychotics, indicating that LAIs were prescribed far less often than oral antipsychotics as the first antipsychotic after an index hospitalization for serious mental illness. However, the Truven Health MarketScan Database includes a limited number of states and may not be generalizable to the full population of Medicaid patients or the broader non-Medicaid population. Performing analyses with more recent or more comprehensive databases may more closely reflect the current treatment landscape and may provide a broader understanding.
The study analysis also had several limitations. The measured outcome—rehospitalization for any cause (psychiatric or other)—can reflect medication tolerability and safety in addition to efficacy. Although hospitalization represents an appropriate proxy for relapse (
2), it may not capture other clinical indicators related to relapse or patient-centered outcomes. The analysis was based on first antipsychotic medication prescribed after the index hospitalization and did not examine adherence or medication refills in the following year. Further, the inpatient use of LAIs versus oral antipsychotics can have an impact on rehospitalization rates—information not collected in this study. Last, certain baseline characteristics that can affect hospitalization were not taken into account in this study (
29).