Comparison of Medicaid spending in schizoaffective patients treated with once monthly paliperidone palmitate or oral atypical antipsychotics.

Abstract Background Compared to oral atypical antipsychotics (OAAs), long-acting injectable antipsychotics require less frequent administration, and thus may improve adherence and reduce risk of relapse in schizoaffective disorder (SAD) patients. Objective To evaluate the impact of once monthly paliperidone palmitate (PP) versus OAAs on healthcare resource utilization, Medicaid spending, and hospital readmission among SAD patients. Methods Using FL, IA, KS, MS, MO, and NJ Medicaid data (January 2009–December 2013), adults with ≥2 SAD diagnoses initiated on PP or OAA (index date) were identified. Baseline characteristics and outcomes were assessed during the 12month pre- and post-index periods, respectively. Propensity score matching (PSM) and inverse probability of treatment weighting (IPTW) were used to reduce confounding and compare the estimated treatment effect for PP versus OAA. Results A total of 10,778 OAA-treated patients and 876 PP-treated patients were selected. Compared to OAAs, PP was associated with significantly lower medical costs (PSM: mean monthly cost difference [MMCD] = -$383, p < 0.001; IPTW: MMCD = -$403, p = 0.016), which offset the higher pharmacy costs associated with PP (PSM: MMCD = $270, p < 0.001; IPTW: MMCD = $350, p < 0.001) and resulted in similar total healthcare cost (PSM: MMCD = -$113, p = 0.414; IPTW: MMCD = -$53, p = 0.697) for PP versus OAA. Reduced risk of hospitalization (PSM: incidence rate ratio [IRR] = 0.85, p = 0.128; IPTW: IRR = 0.96, p = 0.004) and fewer hospitalization days (PSM: IRR = 0.74, p = 0.008; IPTW: IRR = 0.85, p < 0.001) were observed in PP versus OAA patients. Among hospitalized patients, PP was associated with a lower risk of 30 day hospital readmission compared to OAA (IPTW: odds ratio = 0.89, p = 0.041). Limitations The Medicaid data may not be representative of the nation or other states, and includes pre-rebate pharmacy costs (potentially over-estimated). Also changes in treatment over time were possible. Conclusions Total healthcare costs associated with the use of once monthly PP versus OAAs appeared comparable; higher pharmacy costs for PP users were offset by lower medical costs related to fewer and shorter inpatients visits.


Introduction
Schizoaffective disorder (SAD) is a condition that combines schizophrenia symptoms with mood disorder symptoms that can be of depressive type or bipolar type. The lifetime prevalence for SAD globally is estimated to range from 0.5 to 0.8 per 100 lives, and patients with this disorder have high risk of hospitalization, suicidality, and substance abuse 1 .
Studies examining specific treatments for SAD are few leaving it a relatively under-explored area of clinical research 2-4 . In practice antipsychotics (APs) are the cornerstone of treatment for SAD. Mood stabilizers and antidepressants are also frequently added to manage affective symptoms 5,6 . The primary goal of pharmacotherapy treatment with APs in patients with SAD is to relieve symptoms, prevent relapse, and reduce the severity of subsequent acute psychotic episodes over time 7 . While patients with SAD may achieve and maintain symptom stability with available oral therapies, long-term adherence to daily oral treatment regimens is often difficult to achieve due to challenges with patients' adherence. This may contribute to suboptimal outcomes and psychotic and mood relapses [8][9][10][11] . While APs are the cornerstone of treatment for SAD, oral paliperidone is the only oral AP approved by the FDA for the treatment of SAD 12 .
Previous studies have shown that long-acting injectable therapy (LAT) antipsychotic treatment given to patients with schizophrenia may improve adherence, reduce symptoms, and reduce the risk of relapse and re-hospitalization compared to treatment with oral APs [13][14][15] . When compared to patients treated with oral APs, patients treated with LATs have a lower likelihood of being nonadherent, of discontinuing treatment, and of having a hospital readmission 16 . Paliperidone palmitate (PP) is a once monthly LAT that was FDA approved for the treatment of schizophrenia in 2009, and more recently (2014) for the treatment of schizoaffective disorder 17 . Observational studies have been commonly used by medical researchers to assess the effect of treatments on outcomes in the real-world setting. However, in such studies participants' baseline demographics and clinical characteristics may not be balanced between treatment groups. Without appropriately controlling for baseline confounding and/or selection bias, the estimate of treatment effect will most likely be biased 18 . Therefore, the objective of this study was to use propensity score matching (PSM) and inverse probability of treatment weighting (IPTW) to adjust for baseline confounding and selection bias in assessing the impact of treatment initiation on PP versus OAAs on healthcare resource utilization and costs among Medicaid beneficiaries with schizoaffective disorder.
The study first described the characteristics of all patients who had a SAD diagnosis (Objective 1), and then evaluated the impact of PP versus OAA treatment on healthcare resource utilization and costs among patients who had a SAD diagnosis and received PP or OAA treatments using two different statistical approaches (Objective 2). New Jersey (1997Q1-2013Q1) were used. Medicaid databases contain information on medical claims (e.g., type of service; service unit; date; International Classification of Diseases, 9th revision  diagnoses; Current Procedural Terminology codes; physician specialty; and type of provider), prescription drug claims (e.g., days of supply; units; date of service; and National Drug Codes), and eligibility (e.g., age; gender; enrollment start and end dates; and date/year of death, if applicable). Cost information on prescription drug claims is pre-rebate, which may lead to an overestimation of total pharmacy costs.

Data sources
All data collected were de-identified in compliance with the patient confidentiality requirements of the Health Insurance Portability and Accountability Act (HIPAA).

Study design and patient selection
A retrospective longitudinal cohort design was used, as depicted in Figure 1. To assess the two study objectives, two different study populations were formed. To fulfill Objective 1, patients with a SAD diagnosis (claim with ICD-9-CM code: 295.7x) were identified from the database for each state. The date of the first SAD diagnosis was termed as the index date, and adult patients (age !18 years old) who had at least 12 months of continuous insurance coverage prior to the index  date formed the study population for Objective 1 (this population is referred to as Population 1).
To evaluate the impact of PP versus OAAs on healthcare resource utilization and costs (Objective 2), patients who received PP or OAA since 1 January 2010 (five months after the PP approval date, defined a priori to ensure that PP was largely distributed) were identified. To be eligible for the study, patients had to have at least two claims for PP or the same OAA within 90 days (the date of the first claim was defined as the index date), and could have no evidence of prior use of the PP/OAA received at index date during the 12 months before the first claim. In addition, eligible patients had to be at least 18 years of age at the index date, have continuous Medicaid enrollment for at least 12 months prior to and after the index date, and have at least two SAD diagnoses between January 2009 or the start of insurance coverage (if later) and the end of insurance coverage. The period of 12 months preceding the index date was then referred to as the baseline period. The end of the observation period for each patient was fixed at 12 months after the index date. This population was referred to as Population 2.
Note that although Population 1 represents the overall SAD population and Population 2 represents SAD patients who were treated with PP or OAAs since January 2010, Population 2 is not a complete subset of Population 1 due to the different study design elements that were used (such as index date and inclusion criteria).

Treatment, outcomes, and covariates
The treatments of interest in this study included PP and the nine atypical OAAs (i.e., aripiprazole, asenapine maleate, iloperidone, lurasidone, olanzapine, paliperidone, quetiapine fumarate, risperidone, and ziprasidone). They were identified using the generic product identifier (GPI) code and/or the Healthcare Common Procedure Coding System (HCPCS) code. Patients were classified into PP or OAA cohorts based on the treatment received at the index date. Notably, four of the nine atypical OAAs (i.e., olanzapine, quetiapine fumarate, risperidone, and ziprasidone) had generic formulations commercially available at the time this study was conducted.
The outcomes of interest included all-cause and mentalhealth-related (claims with ICD-9-CM between 290.xx and 319.xx) healthcare resource utilization and costs, which were evaluated during the 12 month period after the index date. Healthcare resource utilization and costs were further classified into inpatient visits, outpatient visits, one-day mental-health institute visits, long-term care admissions, mentalhealth institute admissions, emergency room visits, home care, and other medical ancillary services. Total pharmacy costs and costs for AP use were also reported. Costs were calculated based on the amounts paid by state Medicaid programs without supplemental rebates for drugs, and were adjusted to 2013 US dollars according to the medical care component of the Consumer Price Index. The rate of hospital readmission at 30, 60, and 90 days among the subset of patients with at least one hospitalization during the follow-up period was also evaluated.
Demographic and clinical characteristics assessed during the 12 months prior to the first SAD diagnosis for Objective 1 or during the baseline period for Objective 2 included age, gender, race, state, region, number of unique mental health diagnoses, number of unique APs, adherence status (proportion of days covered [PDC] < 0.8 and !0.8), use of concomitant medication, types of schizophrenia disorder, Charlson Comorbidity Index (CCI), and individual comorbidities. Healthcare resource utilization and costs during those periods were also assessed.

Statistical analysis
Descriptive analyses were conducted to illustrate patient characteristics during the 12 months prior to the first SAD diagnosis among SAD patients (Objective 1) and to compare patients' baseline characteristics between PP and OAA cohorts (Objective 2); p-values were calculated using two-sided chi-square tests for categorical variables and Wilcoxon-Mann-Whitney tests or Student's t-test for continuous variables. A test with a p-value 0.05 was considered statistically significant.
Baseline confounding and selection bias were controlled for by using an IPTW or a PSM approach. First, the propensity score (PS), defined as the probability of initiating PP treatment, was estimated using a multivariate logistic regression model conditional on the baseline covariates (i.e., age, gender, race, state, region, number of unique mental health diagnoses, number of unique APs, adherence status, use of concomitant medication, types of schizophrenia disorder, Charlson Comorbidity Index [CCI], and individual comorbidities). IPTWs were then calculated as the inverse of patients' estimated probabilities of having their observed initiation treatment. Specifically, IPTWs were calculated as 1/PS for the PP group and 1/(1 -PS) for the OAA group. Finally, the normalized IPTWs were calculated by dividing each IPTW by the overall mean IPTW. In the PSM approach, an algorithm was used to match each OAA patient to a single PP patient (i.e., 1:1 matching). All OAA patients with a PS within the same percentile as the PS of a PP patient were eligible for matching. If no OAA patient had a PS in the same percentile, that PP patient was excluded from the PSM analysis.
Since the PSM analysis retained only OAA patients that resembled PP patients, results should be interpreted as the average treatment effect among those treated (ATT) and retained in the analysis (i.e., the effect of being treated with PP when compared to OAA patients that resembled PP patients). In contrast, since the IPTW analysis retained all patients and reweighted them so they are more similar to the overall study population, results should be interpreted as the average treatment effect (ATE) among the overall population.
Poisson regression models were used to calculate incidence rate ratios (IRRs) and to estimate the impact of initial treatment (PP vs. OAA) on healthcare resource utilization. Logistic regression models were used to calculate the odds ratios (ORs) of having a hospital readmission and estimate the impact of initial treatment (PP vs. OAA). Linear regression models were used to calculate mean monthly cost differences (MMCDs) and estimate the impact of initial treatment (PP vs. OAA) on healthcare costs. A nonparametric bootstrap method was used to obtain the 95% confidence intervals (CIs) and p-values for both IPTW and PSM approaches. No adjustment was made for multiplicity. All study analyses were performed with SAS software, Version 9.3 (SAS Institute, Cary, NC, USA).

Results
Demographic and clinical characteristics among study population Figure 2 presents the study flow chart. Among 136,757 patients with !2 SAD diagnoses, 71,198 adult patients with !12 months of continuous insurance enrollment prior to the first SAD diagnosis were identified as the SAD cohort. Of the 20,870 patients initiated on PP or an OAA since January 2010, 876 eligible patients were classified into the PP cohort and 10,778 eligible patients were classified into the OAA cohort.
Descriptive statistics of patient characteristics of the SAD cohort, the PP cohort, and the OAA cohort are presented in Table 1. Mean age of patients in the SAD cohort was 44.6 years (standard deviation [SD] 14.8), and 52.9% were female. During their baseline period, they had on average 5.2 (SD 6.3) mental health diagnoses and 69.1% of patients in the SAD cohort had mental health medication use (APs: 63.9%, antidepressants: 50.5%, mood stabilizers: 33.3%, and anxiolytics: 43.1%; see Appendix 1 for the complete list of antidepressants, mood stabilizers, and anxiolytics). Moreover, among patients in the SAD cohort, 88.9% had an outpatient visit and 70.5% had an inpatient visit during the observation period,  ). 1 ICD-9 ¼ 295.7x. 2 OAA agents include aripiprazole, asenapine maleate, iloperidone, lurasidone, olanzapine, paliperidone, quetiapine fumarate, risperidone, and ziprasidone. 3 Initiation of an antipsychotic agent is defined as having !2 claims for the same agent within 90 days and no claim of the same agent during the 12 months before the first claim.

762
Y. XIAO ET AL. Note that since the index date for 'All SAD cohort' is the date of the first SAD diagnosis, there was no patient who had a SAD diagnosis at baseline (i.e., the proportion of SAD diagnosis is 0%). On the other hand, for the PP and OAA cohorts, we required !2 SAD diagnoses during the study period (anywhere), and thus not all patients had to have a SAD diagnosis at baseline (i.e., the proportion of SAD diagnosis could be <100%).
resulting in an average of 0.

Demographic and clinical characteristics among the PS-matched dataset and the IPTW-weighted dataset
The PSM and IPTW methods were used in the study to control for baseline confounding and selection bias. Table 2 shows the distribution of patient characteristics for the PP and OAA cohorts after matching and after weighting using the PSM and IPTW methods, respectively. As expected, the distribution of all the baseline covariates between patients who were initiated on PP and those who were initiated on an OAA among the matched population was balanced (Table 2). However, by matching individual patients, that is each OAA patient to a PP patient, only 1692 patients were included in the analysis. In contrast, the IPTW-weighted dataset retained all patients (N ¼ 11,654), but controlled for confounders through assigning different weights to patients with different characteristics. The median IPTW was 3.33 for PP patients and 0.54 for OAA patients, with interquartile range of (1.68-7.16) and (0.53-0.56), respectively. Table 2 shows that although there are some differences in the distribution of a few covariates, most covariates are evenly distributed between PP patients and OAA patients among the weighted population.
Association between initial treatment (PP vs. OAA) and monthly healthcare costs Table 3 presents the association between initial treatment and monthly healthcare costs estimated using linear regression models with either PSM (the first three columns) or IPTW (the last three columns) to control for baseline confounders.  Association between initial treatment (PP vs. OAA) and healthcare resource utilization . Similar results on the association between healthcare resource utilization and initial treatment were also found for the mental-health-related medical services. Figure 3 presents the association between initial treatment and hospital readmission estimated using logistic regression models with IPTW to control for confounders. Among the subset of patients with at least one hospitalization during follow-up (N ¼ 416 in PP cohort and N ¼ 5914 in OAA cohort), PP was associated with a lower risk of hospital readmission compared to OAA: the odds ratio (95% CI) of hospital readmission for PP relative to OAA within 30, 60, and 90 days 764 Y. XIAO ET AL.

Discussion
The study first described and compared patient characteristics among the overall SAD patients and among those SAD patients who were initiated on either PP or an OAA since January 2010. Compared to patients in the PP cohort, patients in the OAA cohort were similar to the overall SAD population with regard to demographic characteristics (age, gender, and race). However, in general, patients initiated on PP or OAA represented more severe cases of the overall SAD population, which was reflected by more unique mental health diagnoses among patients in the PP or OAA cohorts, compared to the overall SAD cohort. The study then compared healthcare resource utilization and Medicaid spending among SAD patients who were initiated on either PP or an OAA. Our results show that PP patients were different from OAA patients in terms of several baseline characteristics, including age, gender, race, and comorbidity profile, as well as prior AP use and adherence. This indicated a need to control for baseline confounding to estimate the association between treatment and outcomes. Two statistical approaches were used in this study to control for baseline confounding: PSM and IPTW. The PSM analysis retained only OAA patients that resembled PP patients, and thus yielded estimates for treatment effect among treated patients, while the IPTW analysis reweighted all patients so that they were more similar to the overall study population and hence resulted in estimates for treatment effect in the overall population.
Both PSM and IPTW analysis results showed that PP patients had lower mean total medical costs but had higher mean pharmacy costs, compared to OAA patients. The main driver of lower medical costs was lower inpatient costs resulting from fewer hospital visits and shorter length of hospital stay. PP patients also had a lower mean number of ER visits, resulting in lower corresponding mean ER costs. It is possible that the lower inpatient and ER costs observed among the SAD patients treated with PP may be due in part to more frequent mental-health institute visits among PP patients, since more frequent scheduled mental-health institute visits may reduce the risk of relapse and thus avoid potential ER or inpatient visits. This study also demonstrated that, among the subset of patients with at least one hospitalization during follow-up, PP was associated with a lower risk of hospital readmission compared to OAA. This is an important result, as hospital providers may be penalized for their excess readmission rates.
Other studies on schizophrenia have found similar results when comparing hospitalization rates and risk of relapse among PP and OAA cohorts. Baser et al. found that schizophrenia patients treated with PP had lower inpatient costs as well as lower inpatient admission rates, compared to a matched cohort of OAA patients 19 . They also found higher pharmacy costs for patients treated with PP and similar total costs between the two cohorts. Lafeuille et al. showed that PP versus OAA treatment was associated with lower all-cause re-hospitalization or ER visit rates and lower institution costs in hospital settings 20 . A previous study that used four state Table 3. Association between monthly healthcare costs and treatment (PP vs. OAA), estimated using linear regression models a .   In the current study, while PP treatment was shown to be associated with more mental-health institute admissions and longer mental-health institute stays compared to OAA treatment, it was not associated with higher mental-health institute costs. These interesting contradictory results might suggest that some mental-health institute visits and admissions among PP-treated patients were scheduled events rather than the result of a relapse or increase in symptoms. Information on the reason of admission and whether admission was voluntary was not available in the Medicaid database, but might help explain these findings.
In this study, the PSM and IPTW approaches yielded similar results in general. However, since OAA patients were matched 1:1 to PP patients in the PSM approach, the sample size was dramatically reduced among the matched population. This in turn decreased the study power and resulted in more nonsignificant results as evident from fewer statistically significant resource utilization IRRs. In addition, the PS-matched dataset retained only OAA patients that resembled PP patients, thus yielding the average treatment effect among the treated (i.e., the effect of being treated with PP when compared to OAA patients that resembled PP patients). In contrast, the IPTWweighted dataset retained all patients, but assigned high weights to patients who were initiated on PP but had similar characteristics to those who would be expected to be initiated on an OAA. Similarly, patients initiated on an OAA that had similar characteristics to those who would be expected to be initiated on PP were also assigned higher weights. The resulting IPTW-weighted pseudo-population was therefore more similar to the overall study population and yielded the average treatment effect among patient in the overall population.
This study has certain limitations. The Medicaid data used in this study came from only six states and may not be representative of the nation or other states, or of non-Medicaid patients. The data were also subject to billing inaccuracies and missing data. In addition, Medicaid claims data contain pre-rebate Medicaid spending; consequently, the pharmacy costs may have been overestimated in this study. While it is uncertain whether the impact of rebates affected one cohort more than the other, it seems reasonable to think that there would be a smaller impact in the OAA cohort and a larger impact on the PP cohort since branded oral medications are less expensive than PP and much of the OAA use consists of generics where no rebates are available. This may mean that a greater decrease in pharmacy costs would be observed post-rebate in the PP cohort than in the OAA cohort, and this difference could lead to total healthcare cost savings for the PP relative to the OAA cohort. Furthermore, as cohorts were determined at the index date in this study ('intent-to-treat' approach), interpretation of the estimates may become difficult if a large proportion of participants cross over between the treatment arms. In this study, 30.6% of PP patients switched to an OAA during the 12 months of observation and 3.9% and 8.0% of OAA patients switched to PP or another LAT during that same period, respectively. The higher switch rates to an OAA among the PP cohort likely makes these results conservative as the benefits observed in the PP cohort may have been muted. Also, this study did not assess whether the diagnosis of SAD patients changed overtime. Consequently, it is possible that such change occurred, so further research would be needed to understand the impact of PP versus OAA on healthcare costs, healthcare resource utilization and hospital readmission in patients with a change in diagnosis. Finally, as with all retrospective administrative claims data, the study results may be subject to residual confounding due to unmeasured confounders. Nonetheless, even though health insurance claims data present such shortcomings they remain a valuable source of information because they contain a fairly valid large sample of patients' characteristics and outcomes in a real-world setting.

Conclusions
This study revealed that significantly lower medical costs driven by fewer and shorter inpatient visits were associated with once monthly PP treatment, compared to OAAs. The lower medical costs offset the higher pharmacy costs associated with PP and resulted in comparable total healthcare costs for PP-and OAA-treated patients. In addition, PP was also shown to be associated with a lower rate of hospital readmission within 30, 60, and 90 days relative to OAAs. Both PSM and IPTW analysis approaches yielded similar results. CMRO peer reviewers on this manuscript have received an honorarium from CMRO for their review work. CMRO peer reviewer 1 has disclosed that he has received speaker honoraria from Janssen-Cilag, Eli Lilly, Sanofi-aventis, Otsuka and Lundbeck. He has also received travel or hospitality payments from Janssen-Cilag, Eli Lilly, Lundbeck, Johnson & Johnson, Pfizer, Bristol-Myers-Squibb, AstraZeneca and Novartis. He has participated in clinical trials for Janssen-Cilag, Eli Lilly, Lundbeck, Johnson & Johnson, Pfizer, Bristol-Myers-Squibb, AstraZeneca, Novartis, Servier, Pierre Fabre, Roche, Organon, and Merck. He has also participated on advisory boards or being a consultant to Janssen-Cilag, Eli Lilly, Lundbeck, Johnson & Johnson, Roche and Teva. CMRO peer reviewer 2 has no relevant financial or other relationships to disclose.