Cost-effectiveness analysis of rituximab with methotrexate, cytarabine and thiotepa for the treatment of patients with primary central nervous system lymphoma

Jaclyn M. Beca, Kaiwan Raza, Elena Mow, James Keech & C. Tom Kouroukis

To cite this article: Jaclyn M. Beca, Kaiwan Raza, Elena Mow, James Keech & C. Tom Kouroukis (2020): Cost-effectiveness analysis of rituximab with methotrexate, cytarabine and thiotepa for the treatment of patients with primary central nervous system lymphoma, Leukemia & Lymphoma, DOI: 10.1080/10428194.2020.1711902
To link to this article: https://doi.org/10.1080/10428194.2020.1711902

Published online: 13 Jan 2020. Image Submit your article to this journal

l Terms & Conditions of access and use can be found at https://www.tandfonline.com/action/journalInformation?journalCode=ilal20

Jaclyn M. Becaa,b, Kaiwan Razaa, Elena Mowc, James Keechc and C. Tom Kouroukisc,d,e
aPharmacoeconomics Research Unit, Cancer Care Ontario, Toronto, Canada; bCanadian Centre for Applied Research in Cancer Control, Toronto, Canada; cProvincial Drug Reimbursement Programs, Cancer Care Ontario, Toronto, Canada; dJuravinski Cancer Centre and Hospital, Hamilton, Canada; eDepartment of Oncology, McMaster University, Hamilton, Canada

ABSTRACT
The International Extranodal Lymphoma Study Group-32 (IELSG32) randomized patients with pri- mary central nervous system lymphoma (PCNSL) for induction treatment with methotrexate- cytarabine, methotrexate-cytarabine-rituximab, or methotrexate-cytarabine-thiotepa-rituximab (MATRix) and reported significantly improved complete remission with the MATRix regimen. This study assessed cost-effectiveness among these three induction strategies for PCNSL. A Markov model was developed based on the IELSG32 trial over a 20 year time horizon from the Canadian health care system perspective. Costs for induction, consolidation, inpatient treatment adminis- tration, follow-up, adverse events, relapsed disease, and palliative care were included. Methotrexate-cytarabine-rituximab was subject to extended dominance by the other two strategies. The MATRix regimen compared to methotrexate-cytarabine produced 3.05 quality- adjusted life year (QALY) gains at added costs of $75,513, resulting in an incremental cost-effect- iveness ratio of $24,758/QALY gained. The MATRix regimen was the optimal strategy in the majority of simulations (98% probability at willingness-to-pay of $50,000/QALY gained) and results appeared robust across sensitivity analyses.
ARTICLE HISTORY
Received 27 September 2019
Revised 20 December 2019
Accepted 26 December 2019

KEYWORDS
Primary CNS lymphoma; cost-effectiveness analysis; Markov; chemotherapy

Introduction
Primary central nervous system lymphoma (PCNSL) is a rare form of extranodal non-Hodgkin lymphoma (NHL) comprising 2.2% of all central nervous system (CNS) tumors [1]. It encompasses lymphoma exclu- sively involving the brain, spinal cord, eyes, meninges, and cranial nerves, with 90–95% classified histologi- cally as diffuse large B-cell lymphoma (DLBCL).

The International Extranodal Lymphoma Study Group-32 (IELSG32) was a phase 2 study with a double randomization and two phases (induction and consoli- dation) – the first phase randomized patients
<70 years old with newly diagnosed PCNSL in a 1:1:1
ratio for induction treatment with methotrexate-cytar- abine (Group A), methotrexate-cytarabine-rituximab (group B) or the MATRix regimen (group C) [2]. The primary outcome of the first randomization of IELSG32 was centrally assessed complete remission rate after induction treatment, which was significantly improved in favor of patients who received the MATRix regimen. Complete remission was achieved in 23% of patients
in group A, 30% of patients in group B and 49% of patients in group C, and the hazard ratio (HR) for com- plete remission for group C vs A was 0.46 (95% CI:0.28–0.74, p ¼ .0007); group C vs B was 0.61 (0.40–0.94, p ¼ .020), and group B vs A was 0.74 (0.43–1.29, p ¼ .29).
Secondary outcomes included progression-free sur- vival (PFS) and overall survival (OS), which were also improved for patients receiving the MATRix regimen compared to standard care. At a median follow-up of 30 months, the 2-year PFS was 36% (95% CI: 31–41%) for group A, 46% (40–52%) for group B, and 61%
(55–67%) for group C, and the HR for PFS for Group B vs A was 0.52 (0.32–0.86, p ¼ .051), Group C vs A was 0.38 (0.24–0.61, p ¼ .00089), and Group C vs B was 0.72 (0.46–1.13, p ¼ .12). The 2-year OS was 42% (36–48%) in
group A, 56% (50–62%) in group B, and 69% (64–74%) in group C, with a HR for Group B vs A of 0.63 (0.42–1.02, p ¼ .095), Group C vs A of 0.41 (0.25–0.68,
p ¼ .0015), and Group C vs B of 0.78 (0.48–1.26,
p ¼ .12). Based on these data, results with the MATRix

CONTACT Jaclyn M. Beca Image [email protected] ImagePharmacoeconomics Research Unit, Cancer Care Ontario, 620 University Ave, 11th floor, Toronto M5G2L7, Canada
© 2020 Informa UK Limited, trading as Taylor & Francis Groupregimen were consistently superior to the other treat- ment strategies.
Patients with adequate response to the induction treatment (i.e. at least stable disease) and fit for consoli- dation therapy with autologous stem cell transplant (ASCT) were subject to a second randomization (1:1) for consolidation with whole brain radiotherapy (WBRT) or ASCT. There were no significant differences in 2-year PFS by consolidation therapy, (80% [70–90%] in the WBRT group and 69% [59–79%] in the ASCT group, HR 1.50
[0.83–2.71, p ¼ .17]), suggesting survival benefits bygroup were associated with each respective induction therapy [3]. Overall, the trial provided evidence support- ing the addition of rituximab to standard induction treat- ment with methotrexate-cytarabine, and the preferred use with rituximab and thiotepa in the MATRix regimen as the new standard chemoimmunotherapy for patients aged up to 70 years with newly diagnosed PCNSL.
Clinical practice guidelines in Ontario and other jurisdictions have incorporated these trial findings to recommend methotrexate, cytarabine, thiotepa, and rituximab (MATRix regimen) for first-line treatment of PCNSL [4,5]. The purpose of this analysis was to assess the costs, effects and cost-effectiveness of MATRix and methotrexate-cytarabine-rituximab regimens compared to current standard of care with methotrexate-cytara- bine for induction treatment in patients with PCNSL.

Methods
Model framework
The economic analysis included cost-effectiveness and cost-utility analysis for patients with PCNSL receiving induction treatment with methotrexate-cytarabine (Group A), methotrexate-cytarabine-rituximab (Group B) or methotrexate-cytarabine-thiotepa-rituximab in the MATRix regimen (Group C). The starting age of the cohort was 57 years, based on the median age in the IELSG32 trial [2]. Outcomes included total costs, total life-years (LYs) gained, total quality-adjusted life-years (QALYs) gained, and incremental cost-effectiveness ratios (ICERs) per QALY gained and per LY gained. The model was developed from the perspective of the publicly-funded Canadian health care system over a 20 year time horizon using a one-week model cycle length. Future costs and outcomes were discounted at a rate of 1.5% per year [6].

Model structure
A four health state Markov model was developed with health states for (1) progression-free survival

ImageProgression-
free survival
Palliative
treatment
If relapse occurs >2 years, and if eligible
Salvage
Treatment
Death

Figure 1. Model structure for Markov model. Four health state Markov model was utilized with health states for (1) progres- sion-free survival (PFS), (2) salvage treatment, (3) palliative care and (4) death. Patients whose disease progresses during the first 2 years of receiving induction treatment are consid- ered to receive palliative therapy. Patients who remained pro- gression-free for over 2 years were potentially eligible for salvage treatment. Patients could die at any time within the model.(PFS), (2) salvage treatment, (3) palliative care and (4) death (Figure 1). All patients begin induction treat- ment on one of the three treatment strategies. Patients who remained progression-free after induc- tion were eligible to receive consolidation either with WBRT or ASCT. Patients in the IELSG32 trial had to be fit for ASCT to enter the second randomization; per the trial report, 50–60% of the starting cohort were still progression-free and eligible across the induction treatment arms. Thus, we assumed half of patients remaining progression-free may be suitable for ASCT, and the remainder would receive WBRT consolidation.
In consultation with Ontario hematologic cancer clin- ical experts, patients whose disease progresses or relap- ses within the first 2 years of receiving induction treatment are considered to have poor prognosis and unlikely to be fit for further active salvage treatments. These patients would be offered palliative treatment only. We also assumed 20% of patients who relapse after a 2-year progression-free interval to be ineligible for salvage treatment due to comorbidities, dementia, or other age-related reasons and also offered palliation only. The remainder of patients who remained progres- sion-free for over 2 years before relapsing were consid- ered likely to be eligible for salvage treatments, either WBRT or high-dose methotrexate. Patients who had received ASCT during initial consolidation could still receive salvage WBRT, while patients who received initial

eCOST-EFFECTIVENESS OF MATRix FOR PCNSL 3

WBRT during consolidation were likely previously unfit for ASCT, nor could be exposed to further radiation; thus, received high-dose methotrexate as their salvage treatment. Patients were assumed to stay in each of these respective health states until death, an absorbing state to which patients could transition from any state.

Efficacy inputs
PFS curves for each treatment strategy were digitized using Engauge Digitizer software (v4.1) and estimated patient-level data were recreated using published methods [7]. Parametric distributions were fitted to the recreated data in R statistical software assuming exponential, Weibull, log-logistic, log-normal, general- ized gamma and gompertz distributions [8,9]. The generalized gamma distribution was deemed best- fitting based on a combination of statistical criteria (AIC, BIC), visual inspection and clinical plausibility and used to estimate initial conditional transition probabil- ities leaving the PFS state for each treatment strategy. To avoid inflating the treatment benefit, we assumed the duration of treatment effect was truncated at 5 years after induction treatment (the observed duration of the study), using equal transition probabilities across treatments based on Group A for all cycles beyond 5 years. The resulting parameters are presented in Table 1 and curves are presented in Figure 2.
Since transitions from PFS represent either progres- sion events or deaths and patient-level data for each transition were not available, the proportion of first events (progressed disease or death) for each treatment strategy were estimated based on follow-up event data (after median follow-up of 30 months) from the clinical trial. The proportion of events that represented pro- gression (as opposed to death) was calculated to be 83.0%, 79.5%, 80.6% for Groups A, B and C, respect- ively, which was used to distribute PFS events between the subsequent health states [2]. Transition probabilities from salvage treatment were estimated assuming median survival of 16 months, based on a retrospective study evaluating the efficacy of salvage WBRT [10]. The transition probability from palliative therapy was based on clinical input assuming a median survival of two months. Mortality rates based on Statistics Canada life tables were also applied to reflect the probability of death due to age-related mortality [11].

Cost inputs
All costs presented in the model are in Canadian dol- lars and adjusted to reflect 2018 values.

Drug costs
—
Dosing information and treatment schedule was obtained from the IELSG32 trial [2]. Methotrexate was administered at a dose of 3.5 g/m2 on day 1, cytara- bine at 2 g/m2 twice daily on days 2 and 3, rituximab at 375 mg/m2 once daily on days 5 and 0, and thio- tepa 30 mg/m2 on day 4, every 21 days for up to four cycles. Drug prices were obtained from the pro- vincial cancer agency, Cancer Care Ontario (CCO). Dose-intensity of 100% was assumed. Patients were assumed to have an average body surface area of
1.8 m2. Negligible wastage for rituximab was assumed given high volume of use for other lymphomas within the institutions.

Hospitalization costs
Methotrexate requires inpatient administration to min- imize toxicity and promote drug clearance. As induc- tion is administered in a mix of inpatient and outpatient settings, hospitalization and outpatient clinic costs were included over the first four treatment cycles based on the dosing schedules and clinical input. Clinical experts indicated that patients may be admitted for the day 0 rituximab administration and that the first cycle of rituximab-based regimens might require additional hospitalization starting from the first
rituximab dose (day —5). Otherwise, experts suggested
there was unlikely to be any difference between regi- mens, with hospitalization from days 1 to 4. Patients were anticipated to spend 16, 25, and 25 inpatient and 0, 3, and 3 outpatient days for Groups A, B, and C, respectively.
The inpatient hospitalization costs were estimated from Ontario Case Costing Initiative (OCCI) data [12]. Diagnoses incorporated for hospitalization costs included antineoplastic pharmacotherapy procedures among DLBCL, other NHL not specified, and malignant CNS neoplasm diagnoses. Since rituximab is adminis- tered on an outpatient basis during cycles 2–4, admin- istration costs incorporating routine laboratory tests, supplies, pharmacists, nursing, and administrative (clerical and management) staff were estimated from CCO outpatient chemotherapy clinic costs. Additional administration costs, including the staging work-up and pretreatment tests were assumed to be similar between the three groups and therefore not included.

Adverse event (AE) management cost
The OCCI database was used to obtain the number of cases and total inpatient and ambulatory care costs

4 Image J. M. BECA ET AL.

Table 1. Model inputs.
Parameter Input Source Efficacy inputs
Survival curve parameters Group A (Generalized gamma distribution)
Mu 1.064 Estimated from parametric regression based on

Sigma 1.777
—
Q 1.36
Survival curve parameters Group B (Generalized gamma distribution)
survival curves from Ferreri et al. [2,9]

Mu 2.003 Estimated from parametric regression based on

Sigma 2.23
—
Q 1.054
Survival curve parameters Group C (Generalized gamma distribution)
survival curves from Ferreri et al. [2,9]

Mu 2.3421 Estimated from parametric regression based on

Sigma 3.5212
—
Q 1.3148
PFS progression event proportion (vs. deaths)
survival curves from Ferreri et al. [2,9]

Group A 83.02% Ferreri et al. [2]
Group B 79.49%
Group C 80.65%

Proportion of patients receiving ASCT (vs WBRT) during consolidation treatment
Duration of remission required to be eligible for subsequent treatment
Proportion of patients who progress after 2 years who are ineligible for salvage treatment
Duration patients remain stable during salvage treatment (average duration in health state)
Duration patients remain stable during palliation (average duration in health state)
Costs

Drug costs (full treatment – four cycles)
50.00% Ferreri et al. [2]

2 years Clinical expert input

20.00% Clinical expert input

16 months Clinical expert input

2 months Clinical expert input

Rituximab $25,408.08 Price: CCO

Methotrexate $831.60
Cytarabine $3957.12
Thiotepa $7415.28
Adverse event management cost
Dosing: Ferreri et al. [2]

Group A $4073.32 Ontario Case Costing Initiative [12]
Group B $4413.52
Group C $5200.49
Hospital/outpatient administration costs
Days in hospital
Group A 16 Dosing protocol, clinical input
Group B 25
Group C 25
Inpatient hospitalization cost (per day) $1297.36 Ontario Case Costing Initiative [12]

Outpatient chemotherapy clinic costs (per day)
Consolidation Treatment
Autologous stem-cell transplant (ASCT), (conditioning, procedure, follow-up)
Whole brain radiation therapy (WBRT) (25 fractions)
Post-treatment follow-up care costs
$329.58 CCO

$112,959.70 CCO

$4872.51 Earle et al. [13]

Magnetic resonance imaging $1031.62 Kim et al. [14]
Physician visits $77.20 Physician Schedule of Benefits [15]
Salvage treatment
Radiation therapy (25 fractions) (see above) Earle et al. [13]

High-dose methotrexate (including four inpatient hospitalization days)
$2691.83/week CCO

Palliative care cost $814.82/week Chan et al. [17] Utility values
Induction treatment utility
Group A 0.8450 Calculated from adverse event disutility
Group B 0.8309
Group C 0.8000
Consolidation treatment utility
—
ASCT decrement 0.213 Forsythe et al. [23]
WBRT 0.70 Lester-Coll et al. [24]
Post-consolidation (short-term)
Duration of post-ASCT recovery (weeks) 13 Clinical expert input
Post-ASCT Treatment 0.60 Clinical expert input
WBRT cognitive decline decrement
Stable post treatment
Stable post-consolidation (long-term) —0.25
0.88 Lester-Coll et al. [24]

Calculation
(continued)

Table 1. Continued.
Parameter Input Source
Progressed disease
Salvage treatment
0.40
Clinical expert input
Palliative care 0.30 Clinical expert input
Adverse event disutility estimates
Neutropenia Thrombocytopenia —0.0897 Nafees et al. [25] Tolley et al. [26]

COST-EFFECTIVENESS OF MATRix FOR PCNSL Image 5
—0.1080
Anemia —0.0900 Beusterien et al. [27]
Cardiotoxicity —0.1540 Sullivan et al. [28]
Coagulopathy/deep vein thrombosis —0.1100 Bamber et al. [29]
Gastrointestinal —0.1100 Tielemans et al. [30]
Group A ¼ Methotrexate-cytarabine.
Group B ¼ Methotrexate-cytarabine-rituximab.
Group C ¼ Methotrexate-cytarabine-thiotepa-rituximab (MATRix).

Figure 2. Kaplan-Meier and parametric progression-free survival (PFS) curve for each treatment strategy extrapolated over time horizon (20 years). Extrapolated PFS curves using generalized gamma distribution shown to validate well against observed Kaplan Meier data for the duration of the observed data. Group A ¼ Methotrexate-cytarabine; Group B ¼ Methotrexate-cytarabine-rituxi-
mab; Group C ¼ Methotrexate-cytarabine-thiotepa-rituximab (MATRix)for diagnoses corresponding to grade 3/4 AEs occurring at a higher frequency in the MATRix regimen group in the IELSG32 clinical trial. The weighted costs for each diagnosis were combined with the proportion of each AE experienced by patients in each treatment arm and applied to the first twelve weeks of the model.

Consolidation treatment
ASCT costs were estimated from CCO data and included all costs associated with transplant, including clinic follow-up visits, psychosocial care, and readmis- sions up to 1 year after the treatment. Additionally,
the cost of expensive drugs most commonly used in conditioning regimens (carmustine (BiCNU) and thio- tepa) were also included. Costs for WBRT were esti- mated using costs per fraction determined from a 1999 Canadian analysis of costs associated with radi- ation therapy [13]. The average number of fractions used for costing was obtained from the IELSG32 trial, in which patients received five fractions per week with a fraction size 180 centigray (cGy), using between 36 Gy up to an additional 9 Gy in patients with partial response. Within the base-case analysis, 25 fractions were used at an average cost of $139.25 per fraction.

Post-treatment follow-up costs
Based on input from clinicians, follow-up care includes annual visits to the specialist and biannual MRI for 5 years post-induction. MRI costs were obtained from a Canadian cost-effectiveness analysis of diagnostic imaging [14]. Patients are also assumed to visit their physician every 3 months [15]. Additionally, baseline per capita age-specific health care costs were obtained from Canadian Institute for Health Information (CIHI) for patients in long-term remission [16].

Treatment for relapsed disease and palliative care
Salvage WBRT therapy was assumed similar to WBRT used for consolidation. High-dose methotrexate was administered at a dose of 3.0 g/m2 every 14 days until death with hospital admission for four days. Palliative care costs were obtained from a previously conducted pharmacoeconomic analysis and applied to relapsed and palliative care states [17].

Health state utilities
Since stratified health-related quality of life data dur- ing and post-induction for each treatment strategy were not presented and preference-based utilities were not collected in the clinical trial, utilities were obtained from the literature and validated through clinician input. Utility values were estimated for patients undergoing induction, undergoing consolida- tion treatment, post-consolidation short-term and long-term, salvage treatment, and palliation. Utility decrements were obtained from the literature for treatment-related toxicities associated with each treat- ment strategy and used to estimate utilities during induction and consolidation. Based on clinical input, we assumed that patients who undergo treatment with ASCT experience a dramatic decrement in utility for approximately 3 months, but improve with long- term remission, while WBRT was associated with more minor short-term impact but can result in long-term cognitive decline, including worsening memory, confu- sion, headaches, and personality changes. Patients who relapsed were assumed to be in poor health based on clinical expert opinion, with patients relaps- ing after more than 2 years having slightly better utility compared to patients relapsing quickly (<2 years).

Base-case and sensitivity analysis
Probabilistic analysis was conducted to incorporate uncertainty in the model parameters together at once.
Beta distributions were used for model parameters with values between 0 and 1 (probabilities, propor- tions and utilities), normal distributions for population values (e.g. body surface area, weight), and gamma distribution for costs. For the correlated uncertainty in the extrapolation parameters, we used normal distri- butions and the Cholesky decomposition. The results of 5000 simulations were used to calculate net benefit
(NB ¼ benefit ω willingness-to-pay – cost) and the
probability that each strategy was optimal (i.e. associ- ated with the largest NB) over a range of willingness- to-pay values. Furthermore, we conducted one-way sensitivity analyses on all input parameters and several scenario analyses to explore assumptions, including the long-term PFS benefits between treatments, pro- portions of patients eligible for consolidation and sal- vage treatments, and survival post-relapse.

Results
Base-case results
The ICER comparing Group B vs Group A was larger than the ICER from the subsequent comparison of Group C vs Group B, indicating Group B was subject to extended dominance by the other two strategies (Table 2). Removing Group B from the analysis due to extended dominance, the ICER for Group C compared to Group A was $24,758/QALY gained, as a result of added costs of $75,513. The cost-effectiveness accept- ability curve demonstrates the MATRix regimen to be the optimal strategy beyond willingness-to-pay of
~$25,000/QALY gained (Figure 3). At a willingness-to-
pay threshold of $50,000/QALY gained, the MATRix regimen was associated with 98% probability of being the optimal treatment, and only 2% probability that the methotrexate-cytarabine-rituximab regimen would be optimal.

One-way deterministic sensitivity analysis and scenario analyses
As Group B was subject to extended dominance in all cases, results are presented for Group C compared to Group A. In one-way sensitivity analysis, model was most sensitive to a shorter time horizon, which would reduce LY and QALY gains (Figure 4). Assuming a 5 year time horizon that only accounts for the benefits observed in the trial and ignores all future benefits produced an ICER of $54,000/QALY gained. The model was also sensitive to starting age of the cohort, as younger ages produce longer opportunity for benefit from curative interventions. Beyond these two

Base-case results from Markov Model from probabilistic analysis.
COST-EFFECTIVENESS OF MATRix FOR PCNSL 7

Group A
Group B
Group C Group B vs Group A Group C vs Group B Group C vs Group A
COSTS $117,105 $172,773 $192,618 $55,667 $19846 $75,513
PFS $88,773 $141,260 $160,561 $52,487 $19,301 $71,789
Salvage $22,334 $26,723 $28,197 $4389 $1474 $5863
Palliation
LIFE YEARS (LY) $5999
3.99 $4790
5.51 $3860
7.50 —$1209 —$930 —$2138
3.51
PFS 3.60 5.11 7.10 1.51 1.99 3.50
Salvage 0.24 0.29 0.31 0.05 0.02 0.06
Palliation 0.14
QUALITY-ADJUSTED LIFE YEARS (QALYS) 3.24 0.11
4.56 0.09
6.30 —0.03 —0.02 —0.05
3.05

1.52 1.99

1.32 1.74
PFS 3.11 4.41 6.15 1.30 1.74 3.04
Salvage 0.10 0.12 0.12 0.02 0.01 0.03
— — —
Palliation 0.04 0.03 0.03 0.01 0.01 0.02
ICER (Cost per LY gained) $36,523 $9986 $21,504
¼
ICER (Cost per QALY gained) $42,332 $11,438 $24,758
Group A Methotrexate-cytarabine.
¼
Group B Methotrexate-cytarabine-rituximab.
¼
Group C Methotrexate-cytarabine-thiotepa-rituximab (MATRix).
ICER: incremental cost-effectiveness ratio; PFS: progression-free survival.

Incremental cost-effectiveness ratio ($/QALY gained)
Figure 4. Tornado diagram with one-way deterministic sensitivity and scenario analyses, showing the impact on the ICER for the comparison between Group C vs Group A. All other parameters not shown changed the ICER by less than $1000/QALY gainedparameters, no scenario or parameter range changed the ICER by more than $10,000/QALY gained, and all ICERs remained below $30,000/QALY gained.

The model was somewhat sensitive to parameters regarding salvage treatment, such as the duration of remission after which patients may become eligible for salvage treatment. Reducing this value from 2 years in the base-case to zero years allowed the majority of patients to be offered salvage treatment – in other words, 20% of patients were assumed to be unfit and the remainder assumed to receive salvage treatment after disease progression, regardless of its timing. This decreased the ICER to $17,170/QALY gained due to considerably increased salvage treat- ment costs, particularly among Group A given the higher rate of early progression in this group. On the other hand, increasing the value to 5 years had very minor impact, since differences in progression risk between treatment groups by that time were small. Removing salvage therapy altogether also reduced the ICER, as did assuming treatment benefit from induc- tion was not truncated beyond the observed 5 years of data. Choice or cost of consolidation therapy also had minor impact. Finally, given uncertainty in survival durations from salvage and palliation health states, we
tested the low end of the range for salvage therapy using median survival duration of 10.9 months from a study evaluating the efficacy of salvage WBRT in immunocompetent patients [18]. Similarly, we tested upper bound for survival during palliation with the low end of the salvage therapy range (10.9 months), producing more palliative care costs for less effective strategies. Both scenarios had very minor impact.

Discussion
The use of cost-effectiveness analysis to inform deci- sion-making about newer and more expensive treat- ment options is necessary to achieve good value for money with limited public health care system resour- ces. To our knowledge, no cost-effectiveness analysis has been conducted evaluating the MATRix regimen for PCNSL. The purpose of our analysis was to esti- mate the cost-effectiveness of the MATRix regimen and methotrexate-cytarabine-rituximab compared to the current standard of care (methotrexate-cytarabine) for the induction treatment of PCNSL. We found that while treatment with novel strategies increased drug costs and required additional health care resources (such as additional inpatient and outpatienthospitalization visits), they also provided substantial life-year gains, such that the benefit estimated from the clinical trial appears to justify the incrementally higher costs.

The primary effect driver was the observed differences in PFS between regimens. Given uncertainty, we attempted to remain conservative in our modeling approach by assuming an equal risk of progression between strategies 5 years after induction treatment, thus assuming no additional net carryover benefit attributable to novel induction treatment regi- mens beyond the results observed in the trial. Clinical experts have noted that most relapses occur within the first few years after diagnosis and that patients who respond well to initial treatment have the poten- tial for long remission. Therefore, it is plausible that achieving response during induction may lead to improved long-term clinical outcomes. We found higher LYs gained using the MATRix regimen across all scenarios tested, including when considering only the observed period of 5 years from the trial. We also con- servatively assumed costs for additional inpatient hos- pitalization days with the novel regimens, and included future unrelated health system costs for patients who are in long-term remission. While the inclusion of unrelated health care costs are debated in the literature [6], we chose to include these costs to avoid any bias from assuming accrual of future benefit without associated health care costs from prolonged survival, noting the ICERs were more favorable without these costs included.

Only one other cost-effectiveness study has exam- ined induction therapy for PCNSL, comparing induc- tion chemotherapy with or without radiation, which acknowledged the tradeoff between superior disease control and neurotoxicity, particularly in elderly patients [17]. The combination of chemotherapy and WBRT consolidation showed significant benefit for younger patients that also produced cost-savings by avoiding relapse, however. The evolution of the treat- ment paradigm for PSCNL has included age stratifica- tion and novel chemo-immunotherapy combinations to improve disease control without compromising safety and tolerability [2,5]. The IELSG32 trial both sup- ported the addition of active therapies against aggres- sive lymphomas, particularly thiotepa, which can cross the blood–brain barrier and increase the cytotoxicity of antimetabolites without higher rates of severe com- plications, as well as adding support to the use of ASCT consolidation as a valid alternative to WBRT to avoid long-term neurological complications [2,3].

It is possible that characteristics of patients in the trial differed from real-world populations of patientswith PCNSL. Though literature is limited for this condi- tion, reports of consecutively diagnosed PCNSL patients in several centers have involved somewhat older patients with median ages in the 60 s (vs 57 in the IELSG32 trial, which restricted eligibility to<70 years) and tended to have higher proportions with poorer performance status (ECOG > 1); however,other prognostic characteristics (lactate dehydrogen- ase level, CSF protein concentration, and involvement of deep regions) [19] and resulting composite risk scores tended to be similar or somewhat lower in real world settings [20–22]. Rate of ASCT eligibility was also similar in the most recent study to that of the trial [21]. Thus, though trials often restrict to younger patients, there may not otherwise be dramatic differ- ences in overall prognostic risk nor transplant-eligibil- ity. Sensitivity analysis exploring the impact of up to average age of 70 suggested higher ICERs that may still be considered economically attractive.

Moreover,clinicians are not likely to offer the MATRix regimen to elderly (>70 year old) patients. The trial was fairly inclusive of the patients who would be considered fortreatment using the MATRix regimen in clin- ical practice.
There are several limitations to this study. Publications describing the health-related quality of life (HRQoL) of patients with PCNSL are limited and no studies have included preference-based utilities. As patients with PCNSL often present with neurocognitive impairment, they may also experience particular chal- lenges with respect to their ability to recall or accur- ately report their HRQoL. Since the IELSG32 study did not measure utility or report HRQoL results for each induction treatment, we had to make assumptions about utility values associated with the different health states. In some cases, we had to use utility val- ues extrapolated from other types of cancers, or val- ues based on clinical expert opinions, which does not necessarily capture the HRQoL for a given patient.

This is an area that requires active research to find reliable preference-elicitation methods in the presence of functional cognitive impairments from diseases with central nervous system involvement. Additional limita- tions included the open-label study design of the clin- ical trial, the small sample size and the lack of formal planned comparisons in the phase II study. However, the IELSG32 study is the largest randomized trial to- date comparing different induction regimens and con- solidation treatments in patients with PCNSL and these data represent the best available evidence at this time. Finally, the lack of robust data for down- stream salvage and palliative therapy was also alimitation but results were robust across scenarios testing a variety of inputs and assumptions. Importantly, variability in the management of patients with poor outcomes after initial therapy have limited impact on the findings, given limited anticipated benefit from salvage treatments. Future studies are needed to identify additional effective therapies and to further the knowledge of optimal treatment path- ways in this patient population.
Despite the uncertainties introduced as a result of limited data, in almost all simulations the MATRix regi- men appears to be the optimal strategy. Results from the sensitivity and scenario analyses suggested these findings to be robust. Overall, the MATRix regimen appears to the optimal induction treatment option from both a clinical and economic standpoint for patients with PCNSL.

Acknowledgements
We would like to acknowledge the helpful contributions of members of Cancer Care Ontario’s Hematology Drug Advisory Committee in validating clinical and costing assumptions used within this analysis.

Disclosure statement
No potential conflict of interest was reported by the authors.

Funding
This study was conducted with the support of Cancer Care Ontario, which is funded by the Ontario Ministry of Health and Long-Term Care. The opinions, results, view, and conclu- sions reported in this publication are those of the authors and do not necessarily reflect those of Cancer Care Ontario. No endorsement by Cancer Care Ontario is intended or should be inferred.

References
[1] Dolecek TA, Propp JM, Stroup NE, et al. CBTRUS statis- tical report : primary brain and central nervous system tumors diagnosed. Neuro Oncol. 2012;14(suppl 5): v1–v49.
[2] Ferreri AJM, Cwynarski K, Pulczynski E, et al. Chemoimmunotherapy with methotrexate, cytarabine, thiotepa, and rituximab (MATRix regimen) in patients with primary CNS lymphoma: results of the first ran- domisation of the International Extranodal Lymphoma Study Group-32 (IELSG32) phase 2 trial. Lancet Haematol. 2016;3(5):e217–e227.
[3] Ferreri AJM, Cwynarski K, Pulczynski E, et al. Whole- brain radiotherapy or autologous stem-cell transplant- ation as consolidation strategies after high-dose methotrexate-based chemoimmunotherapy in
patients with primary CNS lymphoma: results of the second randomisation of the International Extranodal Lymphoma Study Group-32 phase 2 trial. Lancet Haematol. 2017;4(11):e510–e523.
[4] Fraser G, Varela NP, Dudebout J, et al. Management of primary central nervous system diffuse large B-cell lymphoma (PCNS DLBCL). Toronto (ON): Cancer Care Ontario; 2017. Available from: https://www.cancercar- eontario.ca/sites/ccocancercare/files/guidelines/sum- mary/pebcSCT8s.pdf
[5] Fox CP, Phillips EH, Smith J, et al. Guidelines for the diagnosis and management of primary central ner- vous system diffuse large B-cell lymphoma. Br J Haematol. 2019;184(3):348–363.
[6] Guidelines for the Economic Evaluation of Health Technologies: Canada. 4th ed. Ottawa: CADTH; 2017. Available from: https://www.cadth.ca/sites/default/ files/pdf/guidelines_for_the_economic_evaluation_of_ health_technologies_canada_4th_ed.pdf
[7] Guyot P, Ades AE, Ouwens M, et al. Enhanced second- ary analysis of survival data: reconstructing the data from published Kaplan-Meier survival curves. BMC Med Res Methodol. 2012;12(1):9.
[8] Latimer NR. Survival analysis for economic evaluations alongside clinical trials–extrapolation with patient- level data: inconsistencies, limitations, and a practical guide. Med Decis Making. 2013;33(6):743–754.
[9] Jackson C. flexsurv: a platform for parametric survival modeling in R. J Stat Softw. 2016;70(8):1–33.
[10] Hottinger AF, DeAngelis LM, Yahalom J, et al. Salvage whole brain radiotherapy for recurrent or refractory primary CNS lymphoma. Neurology. 2007;69(11): 1178–1182.
[11] Statistics Canada. Life expectancy at various ages, by population group and sex, Canada (Table 13-10-0134- 01). Available from: https://doi.org/10.25318/ 1310013401-eng
[12] Ontario Case Costing Cost Analysis tool. Health Data Branch, Ontario Ministry of Health. Available from: https://hsimi.ca/occp/occpreports/
[13] Earle C, Coyle D, Smith A, et al. The cost of radiother- apy at an Ontario regional cancer centre: a re-evalu- ation. Crit Rev Oncol Hematol. 1999; 32(2):87–93.
[14] Kim JSM, Dong JZ, Brener S, et al. Cost-effectiveness analysis of a reduction in diagnostic imaging in degenerative spinal disorders. Healthc policy. 2011; 7(2):e105–21.
[15] Ministry of Health and long-term care. Schedule of benefits: physician services. 2016;2015:71. Available from: http://www.health.gov.on.ca/en/pro/programs/ ohip/sob/schedule_master.html
[16] Canadian Institute for Health Information. National Health Expenditure Trends, 1975 to 2017. Data Tables
– Series E1. Ottawa (ON): CIHI; 2017. Available from: https://secure.cihi.ca/free_products/nhex2017-trends- report-en.pdf
[17] Prica A, Chan K, Cheung M. Combined modality ther- apy versus chemotherapy alone as an induction regi- men for primary central nervous system lymphoma: a cost-effectiveness analysis. Neuro-oncology. 2014; 16(10):1384–1391.
[18] Nguyen PL, Chakravarti A, Finkelstein DM, et al. Results of whole-brain radiation as salvage of metho- trexate failure for immunocompetent patients with primary CNS lymphoma. JCO. 2005;23(7):1507–1513.
[19] Ferreri AJM, Blay JY, Reni M, et al. Prognostic scoring
system for primary CNS lymphomas: the International Extranodal Lymphoma Study Group experience. JCO. 2003;21(2):266–272.
[20] Abrey LE, Ben-Porat L, Panageas KS, et al. Primary central nervous system lymphoma: the memorial sloan-kettering cancer center prognostic model. JCO. 2006;24(36):5711–5715.
[21] Sanders S, Chua N, Larouche JF, et al. Outcomes of
consecutively diagnosed primary central nervous sys- tem lymphoma patients using the Alberta lymphoma clinical practice guideline incorporating thiotepa- busulfan conditioning for transplantation-eligible patients. Biol Blood Marrow Transplant. 2019;25: 1505–1510.
[22] Zeremski V, Koehler M, Fischer T, et al. Characteristics
and outcome of patients with primary CNS lymphoma in a “real-life” setting compared to a clinical trial. Ann Hematol. 2016;95(5):793–799.
[23] Forsythe A, Brandt PS, Dolph M, et al. Systematic review of health state utility values for acute myeloid leukemia. CEOR. 2018;10:83–92.
[24] Lester-Coll NH, Dosoretz AP, Hayman JA, et al. Health state utilities for Thiotepa patients with brain metastases. Cureus. 2016;8:e667.
[25] Nafees B, Stafford M, Gavriel S, et al. Health state util- ities for non small cell lung cancer. Health Qual Life Outcomes. 2008;6:1–15.
[26] Tolley K, Goad C, Yi Y, et al. Utility elicitation study in the UK general public for late-stage chronic lympho- cytic leukaemia. Eur J Health Econ. 2013;14(5): 749–759.
[27] Beusterien KM, Davies J, Leach M, et al. Population preference values for treatment outcomes in chronic lymphocytic leukaemia: a cross-sectional utility study. Health Qual Life Outcomes. 2010;8:1–9.
[28] Sullivan PW, Ghushchyan V. Preference-based EQ-5D index scores for chronic conditions in the United States. Med Decis Making. 2006;26(4):410–420.
[29] Bamber L, Muston D, McLeod E. Cost-effectiveness analysis of treatment of venous thromboembolism with rivaroxaban compared with combined low molecular weight heparin/vitamin K antagonist. Thromb J. 2015;13(1):1–12.
[30] Tielemans MM, Jaspers Focks J, van Rossum LGM, et al. Gastrointestinal symptoms are still prevalent and negatively impact health-related quality of life: a large cross-sectional population based study in The Netherlands. PLoS One. 2013;8(7):e69876–e69877.