Abstract

Background/Aims. We aimed to examine the cost-effectiveness of mycophenolate mofetil (MMF) and azathioprine (AZA) as maintenance therapy for patients with Class III and Class IV lupus nephritis (LN), from a United States (US) perspective. Methods. Using a Markov model, we conducted a cost-utility analysis from a societal perspective over a lifetime horizon. The modeled population comprised patients with proliferative LN who received maintenance therapy with MMF (2 gm/day) versus AZA (150 mg/day) for 3 years. Risk estimates of clinical events were based on a Cochrane meta-analysis while costs and utilities were retrieved from other published sources. Outcome measures included costs, quality-adjusted life-years (QALY), incremental cost-effectiveness ratios (ICER), and net monetary benefit. Results. The base-case model showed that, compared with AZA strategy, the ICER for MMF was $2,630,592/QALY at 3 years. Over the patients’ lifetime, however, the ICER of MMF compared to AZA was $6,454/QALY. Overall, the ICER results from various sensitivity and subgroup analyses did not alter the conclusions of the model simulation. Conclusions. In the short term, an AZA-based regimen confers greater value than MMF for the maintenance therapy of proliferative LN. From a lifelong perspective, however, MMF is cost-effective compared to AZA.

1. Introduction

Lupus nephritis (LN) is a serious and costly cause of kidney disease worldwide [1]. An analysis of United States (US) medical expenditures found that the annual costs per patient among those with LN exceeded $46,000 (USD) versus matched controls and $42,000 versus systemic lupus erythematosus (SLE) patients without nephritis [2]. These findings suggest that LN is a key driver of economic burden in the SLE population.

The Kidney Disease Improving Global Outcomes (KDIGO) practice guidelines for initial or induction therapy for LN are well accepted [3]; however, they do not indicate a preference for maintenance therapy with azathioprine- (AZA-) or mycophenolate mofetil- (MMF-) based regimens. The Task Force Panel of the American College of Rheumatology recommended that either AZA or MMF be used for maintenance [4]. These recommendations were based, in large part, on two randomized controlled trials of long-term maintenance therapies for LN. In the MAINTAIN Nephritis Trial, a predominantly Caucasian cohort was randomized to MMF 2 gm/day or AZA 2 mg/kg/day as maintenance therapy after induction with a fixed, low dose intravenous (IV) cyclophosphamide (CYC) regimen [5]. After a mean follow-up of 4 years, this European-based study found that MMF was not superior to AZA in preventing renal flares, without significant differences in adverse events except for higher rate of cytopenias in the AZA group. In the larger Aspreva Lupus Management Study (ALMS) trial, a multinational population was randomized to MMF 2 gm/day or AZA 2 mg/kg/day after response to initial induction therapy [6]. After 3 years, MMF was superior to AZA as maintenance therapy, based on the primary composite end point of death, end stage renal disease (ESRD), doubling of the serum creatinine, renal flare, or requirement for rescue therapy.

To our knowledge, a cost-effectiveness analysis of maintenance therapy for proliferative LN from a US perspective has not been reported. We conducted a cost-utility analysis from a societal perspective to evaluate the cost-effectiveness of the 3-year maintenance regimens (MMF versus AZA) for proliferative LN over a lifetime horizon.

2. Methods

2.1. Study Design

We constructed a Markov state transition model to estimate the quality-adjusted life-years (QALY) and costs associated with maintenance therapy with MMF versus AZA. Markov models analyze uncertain events over time and are suited to decisions where the timing of events is important and when events are recursive in nature [7]. While decision trees model uncertain events at chance nodes, Markov models analyze these events as transitions between health states. Markov models are suited to modelling chronic conditions, where costs and outcomes (QALY) are spread over a long period of time.

Our model encompasses an initial 3-year treatment phase after which simulated patients are no longer on immunosuppressive agents and followed long term (Tables 1(a) and 2(a); Supplemental Data Sources) (see Supplementary Material available online at http://dx.doi.org/10.1155/2015/917567). The time horizon in Markov models is divided into discrete time periods, called cycles. Each cycle length in our model is 6 months for the first 3 years and 12 months thereafter, reflecting the natural history of renal flares and remissions in LN. Analyses were performed using TreeAge Pro 2012 (TreeAge Software, Williamstown, MA) and Microsoft Excel 2010 (Microsoft Corp., Redmond, WA). Institutional review board approval was not required for this study. We adhered to the CHEERS (Consolidated Health Economic Evaluation Reporting Standards) reporting guidelines in our study [8].

2.2. Target Population

The modeled population is a hypothetical cohort of 1,000 patients with proliferative LN receiving maintenance therapy, having responded to their induction regimen. The starting age ranges from 20 to 40 years old, and various racial/ethnic groups are represented in the model, reflecting the demographic characteristics of study participants in the clinical trials.

2.3. Model Structure

The general structure of the model is shown as a state transition diagram in Figures 1(a) and 1(b) and Markov cycle trees (Supplemental Model Structure, Model Assumptions, and Supplemental Figures and ). We did not include the costs and QALY with induction therapy for proliferative LN given that this was a study of the differences between two maintenance treatment strategies and inclusion would not alter the conclusion of the analysis. We attempt to simulate patient-oriented outcomes and treatment strategies that are typically utilized in “real-world” clinical practice. For both strategies, after model entry each patient would progress through five potential health states, in 6-month cycles:(1)Remission state on MMF or AZA as maintenance therapy;(2)relapse of LN requiring MMF rescue therapy (escalation of MMF dose if maintained on MMF);(3)relapse of LN despite MMF rescue therapy, requiring monotherapy with IV CYC;(4)ESRD due to LN;(5)death.

Upon completing the 3-year maintenance therapy, patients in each arm are assumed to be off the immunosuppressive medications and would progress through four potential health states in the lifetime model, in 12-month cycles:(1)Remission;(2)relapse of LN;(3)ESRD due to LN;(4)death.

2.4. Interventions

We evaluated MMF (2 gm/day) and AZA (150 mg/day) as maintenance therapy for LN. The model accounted for sequential rescue therapy during 3 years of maintenance therapy. There is a paucity of clinical trial data on the treatment of LN flares. Therefore, the treatment approach in our model reflects the current recommendations of national and international experts [9, 10].

2.5. Costs

Costs of healthcare products and services were undertaken from a societal perspective. All costs were adjusted for inflation to 2013 US dollars by using the Consumer Price Index for Medical Care [11]. Drug costs are based on average wholesale prices (AWP) [12]; other cost items were obtained from previous literature and public sources. Tables 1(b) and 2(b) show the components of direct and indirect costs incurred during the 3-year maintenance therapy with either MMF or AZA and thereafter in the lifetime model (Supplemental Costs). As noted above, patients are assumed to be off immunosuppressive therapy after 3 years; therefore, costs of MMF, AZA, and CYC are not included in the lifetime model.

2.6. Utilities (QALY)

QALY is the product of the utility score and the number of years spent in a particular health state. A utility score reflects preference of a surveyed sample of individuals for a particular health state; a preference score of 1.0 represents perfect health, whereas a 0 score represents death. Tables 1(c) and 2(c) show the various utility weights of the health states in the model, obtained from previous literature (Supplemental Utilities).

2.7. Outcome Measures

The first outcome measure is the incremental cost-effectiveness ratios (ICER) which is the difference in costs between two strategies divided by the difference in effectiveness [7]: where is the cost of strategy 1, is the cost of strategy 2, is the QALY of strategy 1, and is the QALY of strategy 2.

The second outcome measure is the net monetary benefit (NMB) which represents the difference between the monetary value of an incremental QALY and the cost of achieving the benefit. The strategy with the highest NMB is the most cost-effective given a WTP parameter [7]where is effectiveness (QALY), is WTP, and is cost.

WTP is the amount society that is willing to pay for an additional QALY. We used a WTP of $50,000–$100,000 per QALY gained, often cited as the cost-effectiveness threshold in the literature [30].

2.8. Data Analysis

Our model is based on Reference Case analysis, a standard set of methodological practices for cost-effectiveness analysis [31]. We conducted a two-dimensional simulation via a combination of probabilistic sensitivity analysis (PSA) and microsimulation [32] (Supplemental Data Analysis). We conducted sensitivity analyses to assess uncertainty in our model (Supplemental Sensitivity Analysis). We also conducted value of information analyses, using NMB calculations from the 3-year base-case model, to estimate the expected benefit of future research [32] (Supplemental Expected Value of Perfect Information). Total costs and QALY were calculated after six 1/2-year cycles in the 3-year model and after forty 1-year cycles in the base-case lifetime model.

3. Results

3.1. Model Validation

In assessing external validity, we compared predicted outputs from the 3-year model with observed data, which were generally comparable and within standard deviations (Supplemental Model Validation, Assessment of External Validity, and Supplemental Table ). We also compared simulated 10-year and 15-year survival rates from the lifetime model with actual event data [33, 34]. Overall, the predicted outcomes from the lifetime model approximated observed data from these studies (Supplemental Assessment of External Validity).

3.2. -Year Model
3.2.1. Base-Case Analysis (Cochrane Data)

(i) Cost-Effectiveness. Compared with an AZA-based regimen, MMF had an incremental cost of $17,611 and gain of 0.0067 QALY, with an ICER of $2,630,592 per QALY (Table 3(a)).

(ii) Sensitivity Analyses. In a one-way sensitivity analysis, the MMF-based regimen was the favored strategy if the 6-month cost of MMF 2 gm/day was <$954.13 at WTP $50,000/QALY (Supplemental Figure  3). This is equivalent to $1.33 per 500 mg MMF tablet and represents 20.0% of the actual AWP. As shown in Table 3(b), we conducted other sensitivity analyses by excluding indirect costs, varying utility weights or changing model assumptions; the ICER (MMF versus AZA) of these analyses far exceeded the standard WTP $50,000–$100,000/QALY thresholds.

(iii) Tornado Analysis. At a WTP $50,000/QALY, the model was most sensitive to (1) indirect costs during remission; (2) utility weight of the remission state; (3) drug price of AZA 150 mg/day (Figure 2). These three parameters accounted for 82.4% of the total model uncertainty.

(iv) Scenario Analysis. Despite simulated conditions biased against AZA, the MMF-based regimen remained cost ineffective compared to its alternative at 3 years, with an ICER $709,870 per QALY (Table 3(b)).

(v) Probabilistic Sensitivity Analyses. The incremental cost-effectiveness (ICE) scatterplot and the cost-effectiveness acceptability curve (CEAC) showed that an AZA-based regimen had a near 100% probability of being cost-effective over a 3-year time frame, at WTP thresholds of $50,000 and $100,000/QALY (Figure 3(a), Supplemental Figure 4).

(vi) Expected Value of Perfect Information (EVPI). The population EVPI represents the upper bound on the expected gain on investment on further data collection, which we calculated to be $2,058,206 at WTP $100,000/QALY in the US population, assuming a period of 10 years with 3% discount rate (Supplemental Expected Value of Perfect Information).

3.2.2. Subgroup Analysis

Based on ALMS data, MMF had an ICER of $700,001 per QALY compared with an AZA-based regimen (Table 3(a)). Furthermore, AZA was both cost-saving and more effective than MMF using data from the MAINTAIN trial (Table 3(a)).

3.3. Lifetime Model (40 Years)
3.3.1. Base-Case Analysis (Cochrane Data)

(i) Cost-Effectiveness. Compared with an AZA-based regimen, MMF had an incremental cost of $5,976 and gain of 0.9260 QALY, with an ICER of $6,454 per QALY (Table 4(a)).

(ii) Probabilistic Sensitivity Analyses. The CEAC showed that an MMF-based regimen had a near 100% probability of being cost-effective over a 40-year time frame, at WTP thresholds of $50,000 and $100,000/QALY (Supplemental Figure  5).

(iii) Sensitivity Analyses. As shown in Table 4(b), the ICER (MMF versus AZA) decreased over time such that MMF became cost-effective compared to AZA at 10 years postmaintenance therapy (Figures 3(b) and 3(c)). We also conducted sensitivity analyses by varying the probability of ESRD in the relapse state, demonstrating that the higher the risk of ESRD, the greater the cost-effectiveness of MMF versus AZA. Given the higher baseline risk of ESRD on AZA maintenance therapy, any incremental increase in this risk would disproportionately affect AZA (higher costs and lower QALY) as compared to MMF, resulting in a lower ICER (MMF versus AZA). We conducted other sensitivity analyses by excluding indirect costs, varying utility weights or discount rates, with the ICER (MMF versus AZA) well below the WTP $50,000/QALY threshold (Table 4(b)).

(iv) Scenario Analysis. The ICER of the base-case ($6,454/QALY) was based on the assumption that the treatment effect of MMF and AZA during the trial phase would persist over a lifetime. As shown in Table 4(b), MMF remained cost-effective over lifetime even if the treatment effect of both therapies diminished by 1% or 2% per year. However, assuming no treatment benefit after 3 years of maintenance therapy with either agent, MMF was not cost-effective compared to AZA ($428,894/QALY).

3.3.2. Subgroup Analysis

MMF had favorable ICER compared to AZA over lifetime using ALMS ($4,394/QALY) and MAINTAIN data ($54,891/QALY), below the WTP $50,000–$100,000/QALY (Table 4(a)).

4. Discussion

MMF and AZA are the most widely used therapeutic agents for long-term maintenance therapy of proliferative LN [35]. However, there is no consensus on the agent of choice, reflected by current clinical practice guideline recommendations [3, 4]. To evaluate the cost-effectiveness of MMF versus AZA-based regimens, we developed a Markov model to simulate patient-oriented outcomes, both from short-term and from lifetime horizon.

We found poor cost-effectiveness of MMF versus AZA-based therapy at 3 years, with an ICER $2,630,592/QALY. The ICER of MMF versus AZA remained substantially elevated in sensitivity analyses, even in conditions biased against AZA. Over a lifetime, however, our base-case analysis demonstrated MMF to be cost-effective compared to AZA, with an ICER $6,454/QALY. Overall, the ICER results from various sensitivity analyses did not alter the conclusions of the lifetime model, except in an unlikely scenario where the treatment effect was nil after 3 years of maintenance therapy. In contrast to the initial 3-year time period, subgroup analysis of ALMS and MAINTAIN trials showed that the MMF-based strategy was cost-effective compared to AZA from a lifetime perspective, at WTP $50,000–$100,000/QALY.

To our knowledge, there are only two published cost-effectiveness analyses of LN treatment. Wilson et al. estimated the cost-utility of MMF versus IV CYC as induction therapy for 6 months from the perspective of the National Health Service in the United Kingdom (UK) [36]. Their analysis suggested that MMF was likely to result in better quality of life and be less expensive than IV CYC as induction therapy. More recently, Mohara et al. conducted a lifetime cost-utility analysis of four different immunosuppressive regimens for LN patients in Thailand [37]. The study demonstrated that, from a Thai perspective, induction with IV CYC followed by AZA was the most cost-effective regimen of all the alternatives. Our study reached different conclusions due to notable mutual differences in the model structure and assumptions, setting (US versus Thailand), target population, cost and utility parameters, and transition probabilities; and the Cochrane meta-analysis was not used as data source in Mohara’s study.

Our study has certain limitations. First, lifetime modeling required extrapolation of data beyond the period observed in clinical trials which could lead to inconsistent results. We therefore assessed the uncertainty of future treatment benefit by conducting sensitivity analyses based on established guidelines [38]. Second, we assume that patients in our model receive immunosuppressive maintenance therapy for 3 years based on published clinical trials [13]. Due to lack of data from long-term randomized studies of maintenance therapy in patients with proliferative lupus nephritis, we did not model scenarios whereby patients are kept on maintenance therapy for >3 years. Modeling such scenarios based on incomplete medical evidence would compromise face validity. Furthermore, we would not be able to test the model for external validity (comparing predicted results from the model with actual event data) [39]. Third, the total costs of each strategy are likely underestimated since cost data are based on the Tri-Nation Study which included lupus patients from the US, Canada, and UK [18, 19]. This study estimated that direct and indirect costs in the US are 20% and 29% higher, respectively, than Canada. However, this underestimation of total costs does not change the conclusions of our analysis which is based on incremental calculations. Fourth, our model included utility scores that were measured by VAS [17, 23] which does not involve a trade-off that a subject must choose between the health states, in contrast to the standard gamble and time trade-off techniques. However, VAS was demonstrated to be a valid and reliable measure of health related quality of life in a SLE cohort [40]. Lastly, we incorporated major infection in the model as the most severe side effect of immunosuppressive therapy but did not consider gastrointestinal disturbance, leukopenia, alopecia, or infertility.

Acknowledging these limitations, our study does suggest that, from a cost-effectiveness standpoint, an AZA-based regimen confers greater value than MMF for the maintenance therapy of proliferative LN in the short term. Value of information analysis suggests a population EVPI of $2,058,206 at WTP $100,000/QALY which represents the expected maximum gain on investment on further research. The implication is that spending more than this amount on additional data collection would represent a poor investment of limited research funds. In contrast to the short-term perspective, MMF is cost-effective compared to AZA at the standard WTP threshold in the US over the patients’ lifetime. Despite the relatively higher upfront costs of MMF during the 3-year maintenance phase, its salutary effects (lower risk of LN relapse and progression to ESRD compared to AZA) make MMF a cost-effective option over the long term. Our analysis is consistent with the general notion that the time frame of a model should be sufficiently long to capture future differences in costs and health outcomes between treatment strategies.

Given the substantial economic burden of LN in our healthcare system, the findings of this study should be an important factor in selecting the optimal maintenance regimen for patients with proliferative LN. Furthermore, these findings may provide useful information to support more individualized therapy.

Disclaimer

The views expressed in this paper are those of the authors and do not necessarily reflect the official policy or position of the Department of the Army, the Department of the Navy, the Department of Defense, or the United States government.

Conflict of Interests

The authors report no conflict of interests.

Acknowledgment

The authors would like to acknowledge Annie Nguyen, PharmD, for her assistance in obtaining data for drug costs.

Supplementary Materials

The following Supplementary Material includes additional text, tables and figures that are not presented in the main manuscript. The document provides more detailed and technical descriptions of the models and data analysis.

  1. Supplementary Material