A Monte Carlo Microsimulation (MCM) model was developed to compare the cost-effectiveness of 5 frequently used oral atypical antipsychotics in the usual care of schizophrenia in the United States. Results are based upon a simulation of 1,000,000 patients. The target patient population was community-dwelling adult patients with schizophrenia who had a history of schizophrenia. The model compares oral olanzapine with generic oral risperidone (primary comparator), quetiapine, ziprasidone, and aripiprazole in the treatment of patients with schizophrenia for a 1-year study period. Health care costs are evaluated from the perspective of a public or private third party health care payer in the United States. The model simulates the dynamic nature of usual care where patients switch, continue, discontinue, and restart their antipsychotics in quarterly cycles. The choice of quarterly cycles is based on previous cost-effectiveness research 58 and expert consensus that the duration of an "adequate antipsychotic treatment trial" 255859 is 3–8 weeks if there is no response and 5–12 weeks if there is a partial response before switching to another pharmacologic strategy. The MCM model captures clinical outcomes and estimates third-party payers' costs. The MCM model allows for a number of input parameters including: adherence levels, relapse with and without hospitalization, health state utilities, treatment discontinuation by reason, treatment-emergent adverse events, health care resource utilization, and health care costs, including medication costs. Key clinical outcomes predicted include psychiatric inpatient hospitalization rates and quality-adjusted life years (QALYs). Costs are expressed in U.S. dollars based on 2007 values. The MCM model assumes an intent-to-treat approach that attributes all estimated direct medical costs to the initial therapy.

Although schizophrenia is a chronic illness that requires long-term treatment, we chose a 1-year timeframe for the MCM model because 1 year is the time period the typical third-party payer is responsible for covering medical costs of a covered life. In addition, the dynamic nature of the treatment for schizophrenia with its high rate of medication switching and discontinuation makes it difficult to directly relate the initial treatment selection to the final cost-effectiveness outcomes in a multiyear study period. Furthermore, projections of total medical costs from a third-party payer perspective may not be very useful beyond a 1-year time horizon due to shifts in drug pricing, reimbursement rates, turnover of plan membership, and changes in benefit design.

Figure 1 presents a conceptual overview of the usual treatment for patients living in the community where patients are initiated on specific antipsychotic medications and manifest various adherence levels (fully adherent, partially adherent, or nonadherent). Depending on their adherence level, the patients may (a) remain stable, (b) suffer relapse(s) requiring hospitalization, or (c) relapse(s) not severe enough to warrant psychiatric hospitalization. The patients could potentially experience treatment-emergent adverse events: extrapyramidal symptoms (EPS), clinically significant weight gain (≥ 7%), diabetes, or hyperlipidemia. Depending on benefits and/or adverse events on the initiated medication, the patients and/or their treating physicians decide whether to continue or discontinue the medication. Medication discontinuations involve either a switch to another antipsychotic or discontinuing antipsychotic treatment for awhile. The model takes into account switching patterns, incorporating the primary reason for medication discontinuation (poor efficacy, intolerability, patient decision, or other reasons). As patients with schizophrenia are at a high risk of suicide, the model also incorporates the risk of attempted and completed suicide 60. The patient's health state at the end of the first quarter constitutes the base for the patient's health state in the next quarter until the end of the fourth quarter (1 year). In addition, certain adverse events (i.e., diabetes and hyperlipidemia) were assumed to remain "with" the patient for the remaining periods, since these adverse events may not disappear within the 1-year timeframe and, therefore, contribute to treatment costs for the remainder of the study period.

The MCM model is designed to capture clinically relevant variables for patients with schizophrenia in the usual care setting. However, important clinical variables do not always impact total treatment costs or cost-effectiveness results due to low incidence, low cost, or both. As a result, we used sequential bifurcation 61 to screen all model inputs to determine those variables impacting total treatment costs that warrant focus in sensitivity analyses. Sequential bifurcation is a process that iteratively samples inputs within relevant input ranges and assesses the impact of each input against a predetermined threshold value. For each of the iterations, factors that impact results at or above the threshold value are used in the next iteration. This process continues until there remains no new factor that impacts model outputs by the specified threshold value. Overall, the analyses tested 16 groups with 11 distinct variables examining the impact of variation in over 120 different input assumptions.

The results of the sequential bifurcation tests demonstrated that not all variables that are clinically relevant impact economic outcomes. The suicide rate for patients with schizophrenia is an example of a clinically relevant input, but the sequential bifurcation confirms that it does not impact economic outcomes because of its relatively low incidence rate. In addition, the sequential bifurcation test found that the majority of the costs associated with failed suicide attempts are captured in the treatment cost of an inpatient relapse. Further, cost incurred after a completed suicide are mainly societal and as such, generate no additional costs in our model, and the simulation ends for that patient. Therefore, input assumptions for the suicide rate are modifiable in the MCM model, but this variable is not included in the sensitivity analyses.

The sequential bifurcation tests indicate that the key economic outcomes of the MCM model include the number/cost of unit health care resources, relapse rates, initial adherence rates, and conditional probabilities of relapse given a history of relapse. As a result, we conducted single variable sensitivity analyses to examine the impact of discrete changes in the value of these variables on the model's results. Specifically, we performed the following 5 analyses:

1. Sensitivity on adherence rates;

2. Sensitivity on adverse event rates;

3. Sensitivity on relapse rates expressed as inpatient hospitalization risk ratios;

4. Sensitivity for olanzapine versus risperidone, changing CATIE relapse risk ratio to achieve desired ICER result.

5. Variation in the cost per day of therapy for generic risperidone.

It should be noted that 1-way sensitivity analysis was not conducted on key input variables that did not vary between the 5 antipsychotic medications, such as the cost of most health care resources.

We conducted 2 multivariable PSAs to examine the uncertainty in the model and the stability of the results. The first PSA allowed the input values for adherence rates, relapse rates, treatment discontinuation rates, and the generic cost of risperidone to be randomly drawn from independent distributions of possible input values. With the exception of the generic cost of risperidone, the range of possible input values was created by setting the minima and maxima of the range to be 50% and +50% of the base case value. The second PSA extended the first analysis by adding distributions around the number and cost of resources consumed for stable patients (no relapse), patients experiencing inpatient relapse, and patients experiencing outpatient relapses. In both PSAs, the results were based on 1,000 cohorts of 1,000 patients each.

The sequential bifurcation analysis identified a number of key clinical and economic inputs. The remainder of this section reviews the development of the baseline assumption for these key inputs, which were based, when possible, on evidence reported in peer-reviewed articles. Information reported in these articles is used to derive baseline assumptions for each of the 5 antipsychotic medications.

Adherence to antipsychotic therapy in the MCM model is based on the annual medication possession ratio (MPR), the number of days with the medication prescribed by the total number of days in a given period 1628303132. The MCM model allowed for patients to be categorized into 1 of 3 adherence levels: fully adherent (MPR >/= 80%), partially adherent (60% </= MPR < 80%), or nonadherent (MPR < 60%) 22. The baseline assumptions of the proportion of patients who fall into the full, partial, or nonadherent categories are based on the information contained in the only published latent class analysis reporting adherence rates of an antipsychotic medication for patients in the United States 62. In order to derive differential adherence distributions (for fully, partially, or nonadherent patients) for the 5 antipsychotic medications, we made the following assumptions: 1) the results for haloperidol, a typical antipsychotic reported in Ahn 62, represent the lower bound of adherences for the MCM model because the findings are based on Medicaid patients; 2) we then used the annual MPR ratios reported in Ascher-Svanum 31 by medication (olanzapine = 75%; risperidone = 69%; quetiapine = 61%, and haloperidol = 49%) to produce an adjustment factor for each adherence level for these medications; 3) proportion of patients at each adherence level for ziprasidone and aripiprazole were assumed to be equal to quetiapine as in a previous cost-effectiveness study 18. Table 1, Part A, reports the MCM model's baseline adherence rates by adherence category for each study medication.

The MCM model requires a series of assumptions concerning patients' adherence levels and relapse rate for each of the study medications. One study – sponsored by the National Institute of Mental Health (NIMH) – the Clinical Antipsychotic Trials of Intervention Effectiveness (CATIE) 23 provided the data on relapse rates. This large, randomized, double-blind study provides relapse rates for 4 of the 5 antipsychotics included in our model for an 18-month study period. This independent study is the only one to provide relapse data for each of the studied 4 atypical antipsychotics in the treatment of chronically ill schizophrenia patients in the United States. Results from the primary phase of CATIE, phase 1 23, found significant differences among the antipsychotics for relapses requiring hospitalization, with olanzapine therapy having the lowest risk of relapse (number of hospitalizations/total person-year of exposure). The reported hospitalization risk ratios for the 4 medications of interest were 0.29× for olanzapine, 0.45× for risperidone, 0.66× for quetiapine, and 0.57× for ziprasidone. Table 2, Part A, presents the MCM model's baseline assumptions for the risk of an initial relapse resulting in an inpatient hospitalization by adherence category for each medication. We used the following 3-step process to estimate these relapse rates. First, a baseline relapse rate by adherence level was adopted from a study by Gilmer and colleagues 16 among Medicaid patients. Second, the relapse rates for olanzapine, quetiapine, risperidone, and ziprasidone were derived using the hospitalization risk ratios reported from CATIE phase 1 23. Consistent with a prior model comparing the cost-effectiveness of antipsychotics in the treatment of schizophrenia 18, we also assumed that the rates of relapse for aripiprazole are equivalent to ziprasidone. This was done because no comparative data are available for aripiprazole versus the other 4 studied atypicals on relapse rates as the CATIE study did not include aripiprazole. Finally, we assumed a constant proportion of inpatient-to-outpatient rates of relapse by adherence level; 1.0 for fully adherent; 1.13 for partially adherent; and 1.11 for nonadherent for all antipsychotic medications studied 18.

In addition, the MCM model requires a set of conditional probabilities to allow for: 1) multiple outpatient relapses within a single quarter, 2) multiple inpatient relapses within a single quarter, and 3) higher rates of inpatient relapse given a history of inpatient relapse. First, we assumed if a patient had an inpatient relapse, there was a 20% probability of the occurrence of another inpatient relapse during the same quarter 18. If the first event was an outpatient relapse, then there was a 75% chance of another outpatient relapse during that quarter 18. Second, the probabilities of having an inpatient relapse given 1 inpatient relapse in a previous quarter across adherence categories was adjusted to reflect the impact of adherence on relapse found in prior research 6566 which reported that in the 3 months following a relapse, 19% of fully adherent (> 80% MPR) and 43% of nonadherent patients (< 80% MPR) experienced relapses. We set the probability of a second relapse at 19% for patients fully adherent and distributed the probability of a second relapse (43%) between the partially adherent and nonadherent groups weighted by the mean baseline proportion of individuals in each group. These steps result in the baseline assumptions reported in Table 2, Part B. It should be noted that using these baseline rates in the MCM model results in a weighted average number of relapses that is nearly identical to the crude rate of relapse for individuals with a history of 1 relapse reported in the literature (0.47 vs. 0.46) 36.

The MCM model requires assumptions about the likelihood of patients experiencing 4 types of potential treatment-emergent adverse events: EPS, clinically significant weight gain (≥ 7% weight gain from baseline weight), diabetes, and hyperlipidemia for each medication. Table 3 reports all baseline assumptions concerning adverse events by medication. EPS rates for olanzapine and risperidone are based on results from an integrated analysis of 23 clinical trials that compared incidences of EPS, dystonic, parkinsonian, and akathisia events 67. EPS rates for quetiapine and ziprasidone are based on package insert information, while the rate for aripiprazole is based on a 1-year randomized, double-blind study comparing olanzapine and aripiprazole in the treatment of patients with schizophrenia 68. Baseline assumptions concerning potentially clinically significant weight gain for all treatments except aripiprazole are based on the CATIE phase 1 results 23. Baseline assumptions for event rates for emergent diabetes for olanzapine, risperidone, and quetiapine are based on Lambert et al. 69. Due to the lack of data for treatment-emergent diabetes for ziprasidone and aripiprazole, we make the assumption that their rates are the lowest rates reported in the Lambert et al. study 69 (equal to typical antipsychotics). The rates for treatment-emergent hyperlipidemia were based on baseline rates reported for all CATIE participants 23 adjusted to rates reported in 2 California Medicaid studies 7071. The differential in baseline rates for EPS and potentially clinically significant weight gain for aripiprazole were based upon results of a double-blind, randomized comparative study of aripiprazole versus olanzapine 68. Finally, the MCM model requires a baseline assumption concerning the proportion of patients developing coronary heart disease (CHD) overall and conditional on having diabetes or metabolic syndrome. The MCM model used a quarterly baseline rate of 0.25% for the probability of developing CHD, calculated to be consistent with the model's 1-year timeframe using the Framingham risk equation 237273 and assumed a relative risk of 2.67 of CHD given diabetes 74 and 4.47 relative risk of CHD given metabolic syndrome 74.

The MCM model requires a set of assumptions regarding the switching patterns that takes into account the reason for the switch and attempts to choose subsequent treatments that relate to that reason. For example, discontinuation due to EPS would result in a switch to treatments with a more favorable EPS profile. The same approach was used to estimate switching patterns for clinically significant weight gain, diabetes, hyperlipidemia, lack of medication efficacy (a relapse), or patient decision. As such, the options for treatments to "switch to" are dependent on the treatment a patient is "switched from" and are consistent with the comparative efficacy and tolerability of the antipsychotics studied and reported for the CATIE 232425 and other research 1975. Table 5 presents the medication-switch patterns (the medication one is switched from and the medication one is switched to) for each of the 5 reasons for the switching.

Disease-specific utility values for 8 schizophrenia disease states have been reported by Lenert and colleagues 76 using the Positive and Negative Syndrome Scale. Table 6 reports the baseline utility values assigned to each of the 9 possible combinations of adherence levels (full, partial, or nonadherence) and the relapse results (stable, outpatient relapse, or inpatient relapse) required by the MCM model. A panel of 12 independent schizophrenia experts was used to develop these values as follows. First, we surveyed (via email) the panel of experts to determine which of Lenert and colleagues' 8 possible health states best matched the utility of a schizophrenia patient in each of the MCM model's 9 possible adherence/relapse outcomes. Next, we rounded the averaged survey response to the nearest whole number and assigned this number the appropriate utility value reported by Lenert and colleagues 76. Table 6 also reports baseline assumptions concerning disutility among patients experiencing 1 of the model's 4 treatment-emergent adverse events: EPS, clinically significant weight gain, diabetes, and hyperlipidemia. The disutility multipliers reported for EPS and clinically significant weight gain were derived from those reported by Lenert and colleagues 76. We assumed that utilities among patients experiencing diabetes or hyperlipidemia were equal to that of patients experiencing EPS, as we are unaware of any peer-reviewed utility information for patients with schizophrenia experiencing diabetes or hyperlipidemia.

The cost of atypical antipsychotic medication is related to daily dose levels, which in turn are linked to patients' illness severity. In order to use comparable medication doses for the treatment of patients with schizophrenia who manifest similar illness severity profiles, we used daily dose levels reported in published, randomized, controlled, schizophrenia studies 237778. Table 7, Part A, reports baseline model assumptions concerning dosing and cost for each medication. With the exception of generic risperidone, medication costs reflect 2007 net wholesale price (NWP) 79. We used NWP instead of average wholesale price (AWP) because most third-party payers negotiate price discounts. In addition, we conducted a separate PSA that allowed medication costs to range from 20% above AWP to 50% below AWP for each study medication. These results are not reported because they did not materially change key cost-effectiveness results. Since the cost of generic risperidone is fluctuating at present, we estimated its average cost during the first year post-patent expiry to be at a 58% discount from its 2007 NWP 19.

The model requires resource cost assumptions for 3 types of acute health care services (inpatient hospitalization per day, day hospital treatment per day, and emergency room visit) and 5 outpatient health care services (physician visits, mental health clinic visits, home care hours, group intervention hours, and nutritionist visits). These baseline cost assumptions are reported in Table 7, Part C. All unit costs assumptions are inflated to reflect the value of 2007 U.S. dollars using the medical services component of the consumer price index 83.

The MCM model also captures the direct health care cost associated with treating 3 types of treatment-emergent adverse events: diabetes, hyperlipidemia, and EPS. The MCM model assumes that the quarterly cost of all health care utilization associated with the treatment of emergent diabetes is $600 per quarter based on the findings of Leslie and Rosenheck 84. The baseline assumption for the quarterly costs of statins for hyperlipidemia therapy is $225 and is based on a 50% market share of 40 mg generic statins and a 50% market share of branded statins 80. The baseline cost of treating EPS with anticholinergics is assumed to be $12 per quarter based on the cost of benztropine (2 mg/day) 18. Finally, the MCM model assumes all patients, regardless of initiated antipsychotic, undergo metabolic monitoring per published expert consensus guidelines 81 and include lab costs for fasting glucose level or hemoglobin A1c at the time of initiation, 4 months after starting and at 12 months. However, patients with potentially clinically significant weight gain are assumed to incur the cost of undergoing metabolic monitoring every 4 months.

The MCM model provides estimates of 3 key clinical outcomes: the proportion of patients experiencing outpatient relapse, those experiencing inpatient relapse, and those without an inpatient or outpatient relapse (stable or no relapse). The model also yields estimates of the mean QALYs per patient by medication.

The main economic outcome of the MCM model is total annual direct health care costs. The model also reports mean total direct health care costs for 4 selected outcomes: cost of stable days, cost of inpatient relapse, cost of outpatient relapse, and cost of adverse events. Finally, the MCM model reports the total annual medication cost of each antipsychotic medication.

The cost-effectiveness measure in the model is the incremental cost-effectiveness ratio (ICER), which was calculated as the difference in costs between 2 comparators divided by their difference in QALYs. In this study, we assumed that any ICER in the $50,000–$100,000 range was cost-effective 85.

ICERs have traditionally been calculated relative to no treatment. However, "no treatment" is not an appropriate option for persons with schizophrenia. As a result, the model calculates the ICER for every nondominated treatment.

Figures 2 and Figure 3 display a complete set of base case clinical results for the 5 comparators in the model. For the base case, Figure 2 indicates that olanzapine would result in more patients never experiencing relapse, fewer inpatient relapses, and fewer outpatient relapses. In addition, Figure 3 indicates that olanzapine would result in a lower number of relapses per patient and the highest QALY of any medication studied.

Figure 4 displays base case estimates of the total mean direct medical cost, as well as predicted costs for each of the key cost components for all comparators in the model. For total mean direct medical cost, the base case of the model predicted that olanzapine results in the least costly option among the 5 comparators. Figure 4 also demonstrates the mean annual direct costs varied by selected cost component by medication. For example, olanzapine had the highest annual medication acquisition cost but the lowest annual mean cost of treating relapse of the 5 comparators.

The incremental cost per QALY gained of 1 comparator versus another is a widely recognized metric of cost effectiveness. A treatment that produces more QALY at a higher cost may be cost effective if the resulting ICER is below a stated threshold. As seen in Figure 3 and Figure 4, olanzapine was predicted to result in the highest mean QALY per patient and the lowest total mean direct medical cost. As a result, from a cost-effectiveness perspective, olanzapine was the dominant therapy in terms of cost/QALY gained because it is predicted to produce more QALYs at a lower cost. Moreover, Figures 3 and 4 show that olanzapine was dominant in 1-to-1 comparisons with each comparator in the same manner. Finally, risperidone dominated quetiapine, ziprasidone, and aripiprazole producing more QALY at a lower cost. One-way and probabilistic sensitivity analyses were conducted to examine the stability of the base case results.

One-way sensitivity analysis was performed to assess the impact of important variables on model outcomes and patient subgroups. First, Table 8, Test 1, reports the results of the sensitivity analysis for adherence subgroups simulated in the model. These results were calculated by assuming that all patients had the same adherence level. First, we simulated the model with a cohort of fully adherent patients and recorded the mean total direct medical cost by medication and the cost per QALY for olanzapine. The same process was applied to a cohort of partially adherent patients and then a cohort of nonadherent patients. The results from this sensitivity test indicate that risperidone has the lowest total cost if all patients remain fully adherent ($7,932) or partially adherent ($9,365), but that olanzapine has the lowest total cost if all patients are nonadherent ($6,699). In addition, the results for cost per QALY indicate that olanzapine is the dominant choice for nonadherent patients and a cost-effective option for partially adherent patients (ICER of $44,718) and for fully adherent patients (ICER of $82,519).

Table 9, Test 2 displays results for 1-way analysis was performed to assess the impact of 2 treatment-emergent adverse event rates on model outcomes: diabetes and hyperlipidemia. Literature-based rates 84 for treatment-emergent diabetes were used as upper bounds for all comparators. The upper limits for hyperlipidemia were based upon a California Medicaid Study 71. The model results indicated that olanzapine remained the dominant strategy (highest mean QALY and lowest mean direct cost) when incorporating alternate adverse event rates that were higher than the base case rates. Sequential bifurcation indicated that EPS and treatment-emergent clinically significant weight gain did not directly impact model results and were not included in the sensitivity analysis.

Table 10, Test 3, reports the results of sensitivity analysis associated with changing the rates of relapse expressed as hospitalization risk ratios. The first column of this table shows the base case hospitalization risk ratio reported in the CATIE, phase 1 23. This sensitivity analysis was performed by increasing the hospitalization risk ratio of olanzapine to each comparator which has the effect of increasing the costs of olanzapine versus each comparator. The values in the second column show the hospitalization risk ratios for each comparator required to make the total mean direct cost of olanzapine therapy roughly the same as the comparator. The values in the third column are the hospitalization risk ratios at which olanzapine was more costly, but relatively cost-effective in terms of cost per QALY (approximately $50,000).

Table 11, Test 4, reports results of the first of 2 head-to-head analyses between olanzapine and risperidone. The results in Table 11, Test 4, answer the following questions: At what relative rates of inpatient relapse does risperidone become cost-neutral when compared to olanzapine, and at what relative rate of inpatient relapse does risperidone dominate olanzapine? The first row of Table 11, Test 4, reports the base case results for comparison. The second row indicates that the hospitalization risk ratio of risperidone compared to olanzapine would need to drop to 1.54 for the 2 medications to be cost-neutral. The final row indicates that the hospitalization risk ratio of risperidone compared to olanzapine would have to be 0.70 for risperidone to begin dominance of olanzapine.

Table 12, Test 5, presents results of the second head-to-head analysis between olanzapine and risperidone and reports the sensitivity analysis associated with changing the generic price of risperidone below baseline rate of $5 per day. The second row of this table assumes that risperidone is reduced to 13¢ per day (initiatives by major U.S. retailers). At this price, olanzapine remains cost-effective ($10,246 per QALY) relative to risperidone. Finally the last row indicates that olanzapine remains cost-effective even if risperidone is acquired at zero cost.