PubMed Health. A service of the National Library of Medicine, National Institutes of Health.

Comparative Clinical and Cost-Effectiveness of Drug Therapies for Relapsing-Remitting Multiple Sclerosis [Internet]. Ottawa (ON): Canadian Agency for Drugs and Technologies in Health; 2013 Oct. (CADTH Therapeutic Review, No. 1.2B.)


3.1. Systematic Review

3.1.1. Literature search strategy

The literature search was performed by an information specialist using a peer-reviewed search strategy (APPENDIX 3).

Published literature was identified by searching the following bibliographic databases: MEDLINE with In-Process records and daily updates via Ovid; Embase via Ovid; and PubMed. The search strategy consisted of both controlled vocabulary, such as the National Library of Medicine’s MeSH (Medical Subject Headings), and keywords. The main search concepts were relapsing-remitting multiple sclerosis and interferon beta-1a/1b, natalizumab, glatiramer acetate, fingolimod, teriflunomide, dimethyl fumarate, and alemtuzumab.

Methodological filters were applied to limit retrieval to health technology assessments, systematic reviews, meta-analyses, randomized controlled trials (RCTs), and safety studies. Where possible, retrieval was limited to the human population. Retrieval was not limited by publication year but was limited to English language results. Conference abstracts were excluded from the search results.

The initial search was completed on November 9th, 2012. Regular alerts were established to update the search until October 2013. Regular search updates were performed on databases that do not provide alert services.

Grey literature (literature that is not commercially published) was identified by searching relevant sections of the Grey Matters checklist (, which includes the websites of regulatory agencies, health technology assessment agencies, clinical trial registries, and professional associations. Google and other Internet search engines were used to search for additional web-based materials. These searches were supplemented by reviewing the bibliographies of key papers and through contacts with appropriate experts.

3.1.2. Selection criteria and methods

Trials were included in the systematic review based on the pre-specified selection criteria (Table 2). Active and placebo-controlled trials were selected for inclusion if they were published in English, involved patients with RRMS, had treatment arms consisting of currently available or emerging disease-modifying agents, and reported any of the specified outcomes related to clinical efficacy and safety. Trials that included mixed populations of MS were also included if the proportion of RRMS patients was more than 50% of the total population. For interventions currently approved by Health Canada for the treatment of RRMS, only approved formulations and doses were included in the systematic review. Interventions not yet approved by Health Canada for the treatment of RRMS, but expected to enter the Canadian market shortly, were not restricted to specific doses or formulations.

Table 2. Inclusion and Exclusion Criteria for Primary Studies.

Table 2

Inclusion and Exclusion Criteria for Primary Studies.

Two reviewers independently screened titles and abstracts relevant to the clinical research questions regarding available and emerging agents for the treatment of patients with RRMS. Full texts of potentially relevant articles were retrieved and independently assessed for possible inclusion based on the pre-determined selection criteria. The two reviewers then compared their chosen included and excluded studies; disagreements were discussed until consensus was reached. The study selection process was presented in a Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) flowchart (APPENDIX 6).

3.1.3. Data extraction strategy and critical appraisal of included studies

One reviewer performed data extraction for each article, using a pre-drafted data extraction form covering the following items:

All extracted data were checked for accuracy by a second reviewer. Any disagreements were resolved through discussion until consensus was reached. A quality assessment of RCTs was performed independently by two reviewers using a standardized table based on major items from the SIGN-50 instrument for internal validity. Additional critical appraisal was performed based on input from clinical experts.

Clinical outcomes included relapse (annualized relapse rate [ARR] and proportion of patients remaining relapse-free) and disability (proportion of patients with sustained disability progression, mean change of EDSS, and mean change of Multiple Sclerosis Functional Composite [MSFC]). Disability is measured by EDSS change. The definitions of relapse and sustained disability progression from individual studies are presented in APPENDIX 9. MSFC comprises the average of the scores on the timed 25-foot walk, the nine-hole peg test, and the paced auditory serial-addition test with a three-second interstimulus interval, with higher scores (Z-score) representing improvement.57

MRI outcomes included a proportion of patients with GdE lesions, mean number of GdE lesions, proportion of patients with new or enlarging T2-hyperintense lesions, and mean number of new or enlarging T2-hyperintense lesions.

Safety outcomes included serious adverse events, discontinuation of treatment because of serious adverse events, total withdrawal, and common adverse events.

3.1.4. Data analysis methods

Direct pairwise meta-analyses were performed for all outcomes to assess consistency with network meta-analysis (NMA) results when NMA was undertaken, and to obtain summary estimates for outcomes that were not analyzed by NMA.

Review Manager 4.2 was used for all statistical analyses of direct comparisons of dichotomous and continuous outcomes in the clinical review. Where the quantitative pooling of results was appropriate, the random-effects model was used to compute treatment efficacy between interventions across studies, based on the assumption that treatment effects follow a distribution across studies.

Dichotomous data were summarized using relative risk (or risk ratio), which compares the proportion of patients having the event between two treatment groups. In our study, the dichotomous outcomes that were measured included:

Continuous data with means and standard deviations were summarized using mean differences. Where standard deviations were not reported, they were obtained from standard errors, confidence intervals, t values, or P values.58 Where no variance was reported, a value of standard deviation was imputed using the coefficient of variation, which was calculated based on studies with similar population, study design, and intervention.59 The continuous outcomes that were measured in this study included:

Relapses were considered as count data and were summarized using a Poisson approach to obtain the relative ARR or rate ratio from the total number of relapses and patient-years. The analyses were performed using the Comprehensive Meta-Analysis software.

The heterogeneity between studies was assessed using I2 statistics, which quantifies the percentage of variation across studies that is because of heterogeneity rather than chance.60 Heterogeneity is considered to be low when I2 is less than or equal to 25%, moderate when I2 is between 25% and 75%, and high when I2 is greater than or equal to 75%. Attempts were made to explain substantial statistical heterogeneity (I2 ≥ 50%) by subgroup analyses or elimination of outliers. Where statistical heterogeneity remained present in the subgroup analyses, clinical outcomes were presented separately for each study and were reviewed qualitatively. The I2 statistics, however, do not provide evidence about clinical heterogeneity in study design, treatments, and baseline demographics and characteristics of patient population.

The planned subgroup analyses included age (≤ 40 years or > 40 years), baseline EDSS score (0 to 3, or > 3), GdE lesions at screening (0 ≥ 1), gender (female or male), and number of relapses in the previous year before screening (1, 2, or ≥ 3).

3.2. Indirect Comparisons

Bayesian NMAs were conducted for two outcomes: relapse and disability. The selection of the outcome-specific measures for the NMA (ARR and the proportion of patients with sustained disability progression) was based on input from clinical experts. NMAs were not conducted for other efficacy outcomes (MRI findings and health-related quality of life) because data were sparsely reported, and, in the case of MRI, eight out of 14 studies reporting MRI outcomes were subsets of randomized populations with unclear selection criteria for MRI scans (Table A10.2). NMAs were not conducted for adverse events data (serious adverse events, and withdrawal because of adverse events) because the occurrence of events was low.

WinBUGS software (MRC Biostatistics Unit, Cambridge, UK) was used for all NMAs. Posterior densities for all unknown parameters were estimated using Markov Chain Monte Carlo methods. Prior distributions for overall effects of interest and study-specific effect estimates were assigned vague normal prior distributions centred at zero, with adequately large variances to allow the collected data to drive the calculation of pooled estimates. Model diagnostics including trace plots, autocorrelation plots, and the Brooks-Gelman-Rubin statistic were assessed to ensure model convergence. Assessment of model fit for NMA comprised the assessment of deviance information criterion and comparison of residual deviance to the number of unconstrained data points. Measures of effect were estimated according to the WinBUGS routine developed by the Evidence Synthesis Group, consisting of experts from the universities of Bristol and Leicester (the code is available from the website). Median estimates were reported, along with corresponding 95% credible intervals ([CrI]; Bayesian confidence interval). For comparative purposes, both fixed-effects and random-effects NMAs were conducted.

Regarding the interpretation of NMA estimates, if a 95% CrI for a risk ratio comparing two interventions did not include the value 1, this was interpreted as an indication that there is a less than 5% probability that there was no difference in effect between treatments.

3.2.1. For ARR

The Poisson distribution is a discrete distribution and is appropriate for modelling counts of observations or events that occur in a given interval of time (or space). In this review, ARR was modelled as a Poisson outcome based on the total number of relapses observed within a treatment group and the total number of person-years of follow-up for that treatment group as the input data.

Where studies did not report the total number of relapses or exposure time (person-years) directly in the publication, imputations were performed to derive the respective values. Missing total number of relapses were derived using exposure time (in person-years) and the reported mean ARR values. For missing exposure time (in person-years), the values were imputed using treatment duration and number of patients completing the study (100% was assumed in cases where the percentage of completers was not reported).

3.2.2. For sustained disability progression

Patient sustained disability progression was analyzed as a binomial outcome, with the total number of patients with the event within a treatment group and the total number of patients randomized for that treatment group as the input data.

3.2.3. Exploring heterogeneity

NMA requires that studies be sufficiently similar in order for their results to be pooled. A wide range of patient and trial characteristics were recorded to allow for a qualitative assessment of the heterogeneity of included trials. However, the methodological limitations with this approach are recognized; assessment of heterogeneity is naturally limited to reported characteristics. For example, older trials did not report or indicate whether the patient population consisted solely of treatment-naïve patients or was inclusive of patients with history of a prior treatment. Assumptions based on the reported information were made to that aspect; consequently, the ability to explore the impact of heterogeneity between studies regarding patient population, in terms of treatment experience, in the NMA was limited.

Heterogeneity was further explored through selected meta-regressions and subgroup analyses based on patient covariates (baseline EDSS score, time since symptom onset, number of relapses in previous year, prior-treatment history) and trial characteristics (publication date and treatment duration). Meta-regressions were performed when the variable was continuous in order to incorporate the maximum amount of information available from trials. Subgroup analyses were performed when the variable could be dichotomized (e.g., patient population was treatment-naïve or mixed). Cut-offs defining the subgroups (e.g., trial publication date or treatment duration) were selected based on currently accepted conventions and clinical expert input.

3.3. Pharmacoeconomic Analysis

3.3.1. Type of economic evaluation

The analysis was in the form of a cost-utility analysis. The primary outcome was the number of QALYs, with treatments compared in incremental cost per QALY (ICUR).

3.3.2. Target population

The target population was Canadians with RRMS. For the base case analysis, a typical patient profile from the RCTs identified in the systematic review was adopted: an average age of 36 years, 68% of patients being female, time since onset of five years, and an initial discrete distribution of EDSS score with a mean score of 2.3.

3.3.3. Treatments

The currently available treatments that are approved and available in Canada were included in the primary analysis (Table 3).

Table 3. Available Treatments Included in Primary Analysis.

Table 3

Available Treatments Included in Primary Analysis.

Emerging treatments in RRMS (for which regulatory approval has not been granted) were included in an exploratory analysis (Table 4). As the costs of these treatments are unknown, it was assumed that the prices would follow the same patterns as in the US for base case. Given the uncertainty of the price for unmarketed agents, this assumption was tested in sensitivity analyses.

Table 4. Emerging Treatments Included in Exploratory Analysis.

Table 4

Emerging Treatments Included in Exploratory Analysis.

Due to a lack of clinical data exploring the sequential use of treatments following the failure of first-line treatment or switching, it was assumed that patients cannot switch between treatments in the model. Therefore, the only transition between interventions that is possible is from active treatment to no treatment (treatment discontinuation).

3.3.4. Perspective

This analysis was conducted from the perspective of a provincial Ministry of Health in Canada.

3.3.5. Time horizon

The analysis adopts a time horizon of 25 years as a base case, with a cycle length of three months. Alternative horizons of 10 years, 20 years, and 40 years (lifetime) were considered in sensitivity analyses. Although RRMS disease onset can occur in early life and has limited effect on life expectancy, a time horizon of 25 years has been implemented to account for the uncertainty regarding natural history of the disease, as well as the uncertainty regarding the long-term efficacy of the treatments.

3.3.6. Model structure

A Markov cohort approach was taken for the analysis, with the model developed in MS Excel. The model was based on a series of health states that reflect the progression of patients with RRMS. Time elapses explicitly in Markov models and transition probabilities are assigned for movement between these states over the three-month cycles. By attaching estimates of resource use and health outcome consequences to the health states, and running the model over a 25-year time horizon (100 cycles), it was possible to estimate the long-term costs and outcomes associated with the various treatments.

Health states were defined according to the Kurtzke EDSS, as well as based on severity of relapse. EDSS levels were grouped into five health states for modelling disease progression (Table 5). These five EDSS levels are generally regarded as the key markers for disability of patients with RRMS.61 This approach was implemented in other published economic models, such as Prosser,62 and clinical experts were in agreement with the approach (Figure 1).

Table 5. Description of Health States.

Table 5

Description of Health States.

Figure 1. Model Diagram.

Figure 1

Model Diagram. EDSS = Expanded Disability Status Scale.

During one cycle, patients can remain in the current health state; progress to the next, more severe state; improve to a less severe state; transition to a secondary-progressive health state; withdraw treatment; or die (Figure 2).

Figure 2. Model Diagram (continued).

Figure 2

Model Diagram (continued). EDSS = Expanded Disability Status Scale; RRMS = relapsing-remitting multiple sclerosis; SPMS = secondary-progressive multiple sclerosis. Note: Patients progressing to SPMS will transition across SPMS EDSS scores based on SPMS (more...)

The progression to more severe states was based on natural history data for MS from a London, Ontario cohort study.47 The fluctuating nature of RRMS — i.e., a possibility of patients moving in both directions along the EDSS scale — has recently been frequently recognized as a clinical phenomenon in the early stage of the disease. Therefore, improvements in lower health states (health state 1 and 2) in the EDSS score were also modelled, based on the study by Tremlett et al. that was conducted using the British Columbia MS database.63 However, since the evidence on improvement on the EDSS scale is mixed, a scenario analysis was conducted, where no improvements on the EDSS scale were modelled (i.e., once patients progressed, they could not transition back to a less severe EDSS state, which is consistent with the older modelling studies64).

Based on the conclusion of the study by Wong et al.,65 there are no significant differences in adherence among the disease-modifying agents for RRMS. Therefore, a constant annual rate of discontinuation was assumed across all treatments for the first two years of 15%, based on the withdrawal rates in the clinical trials included in the systematic review, as well as in line with some of the observations by clinical experts in Canada. After two years, the discontinuation rate was assumed to be zero, assuming that all patients who discontinue treatment would have done so by the end of the second year. Sensitivity analysis was conducted, varying the discontinuation rates and the number of years that discontinuation rates were applied to, to address the variability of discontinuation that might be present in different settings.

The stopping rules for RRMS therapies vary across Canadian public drug plans, ranging from an EDSS score of 5.5 to 7.0, as per clinical experts’ opinion. For the base case scenario, a conservative assumption was made that once patients progress to an EDSS of 7.0 or secondary-progressive multiple sclerosis, they would withdraw treatment. Given the differences in stopping rules for therapies across the Canadian provincial plans, a sensitivity analysis was conducted, varying the EDSS score from 5.0 to 7.0, as well as exploring the scenario if the no-stopping rule has been implemented.

As the model assesses the cost-effectiveness of the treatments in RRMS, progression to secondary-progressive multiple sclerosis (SPMS) in the base case scenario led to treatment discontinuation. However, this assumption was also tested in the scenario analysis.

Relapses were assumed to occur only in patients in the health states 1 and 2 (EDSS 0.0 to 5.5). Although relapses may occur for states with EDSS greater than or equal to 6.0, as per clinical expert opinion, the severity of disability may prevent detection of relapses (i.e., acute increase in disability due to relapse and increase in sustained disability arising from disease progression may not be easily differentiated). Relapses were assumed to last for 45 days for mild or moderate relapses, and 90 days for severe relapses, based on clinical opinion and published literature.

A half-cycle correction was implemented to adjust both costs and QALY gains, so that they are calculated halfway through each cycle, as opposed to the end of each cycle.

3.3.7. Data inputs

To the extent that data inputs of the given model are estimated, they will be subject to uncertainty regarding their true value, known as parameter uncertainty.66 This can be achieved by implementing an informal Bayesian approach to cost-effectiveness analysis by specifying relevant parameters as probability distributions rather than point estimates. This technique allows for the estimation of the likelihood of various output values based on a wide number of sets of input parameters generated by sampling from their probability density functions, and was implemented in the probabilistic sensitivity analysis.

a. Natural history

Disability progression

Ideally, the model would use transitional probabilities derived from one of the large Canadian cohort studies;46,47 however, none of these data were directly available or easily accessible; therefore, the transitional probabilities were based on estimates reported in the published literature.

In the published literature, the most common outcome in the natural history of disease studies was time to reaching EDSS 6, which was not granular enough to be used for modelling the disability progression. The only available data reported in a format that could readily be used were from the London, Ontario cohort study reported by the Centre for Bayesian Statistics in Health Economics of the University of Sheffield, School of Health and Related Research (ScHARR) in its final report to the National Institute for Health and Care Excellence.67 The ScHARR report contained hazard rates for disability progression within RRMS, transitioning from RRMS to SPMS, as well as disability progression within SPMS (Table 6, Table 7, and Table 8).

Table 6. Hazard Rates on Progression Rates Within RRMS Health States.

Table 6

Hazard Rates on Progression Rates Within RRMS Health States.

Table 7. Hazard Rates on Progression Rates Within SPMS Health States.

Table 7

Hazard Rates on Progression Rates Within SPMS Health States.

Table 8. Hazard Rates on Progression Rates from RRMS to SPMS States.

Table 8

Hazard Rates on Progression Rates from RRMS to SPMS States.

Hazard rates were calculated as:

λi=number of people leaving state ij=1nduration in state i
var(λi)=number of people leaving state i(j=1nduration in state i)2
where n is the number of individuals, j is each individual leaving state i, and i = EDSS states 0 to 10.

These hazard rates were further transformed into transitional probabilities using standard methodology:

where t is the cycle length. These transitional probabilities were used to inform the model. By using quarterly transitional probabilities, it is possible for patients to transition to a maximum of four EDSS states during one year.

The authors of the ScHARR model noted that, although there is some evidence that disability improvements can occur up to 12 months after the progression is observed, the improvements are not reported in the long-term natural history data used in this model.64 Tremlett et al.63 recently conducted a large study based on 2,961 patients in the British Columbia MS database and concluded that disability improvements in MS over one or two years are not unusual. That is, the authors of the study reported greater than or equal to 2-point improvements on the EDSS score in 2.2% EDSS intervals per year, greater than or equal to 1-point improvements in 8.3%, and greater than or equal to 0.5-point improvements in 14.9%. To capture the fluctuating nature of RRMS, which is frequently recognized as a clinical phenomenon, these were included in the model by assuming that a maximum of 2 EDSS-point improvements could be achieved. The rates of annual disability improvements were transformed into quarterly rates resulting in 0.5% of 2-point improvements and 1.49% of 1a -point improvement per quarter, and were applied only to the first two health states.

Relapse rate

Based on London, Ontario cohort data, ScHARR reported a mean relapse rate of 0.835 and 1.423 for EDSS 0 to 2 and 3+, respectively, over the first two years since onset.67

However, there is available evidence suggesting that the frequency of relapse is affected by a patient’s age and disease duration,68 and therefore it is time-dependent.5 A prospective study by Patzold and Pocklington reported relapse rates over 19 years, showing a decrease over time.69 This study reported the correlation of the mean annual relapse rate and the duration of disease through a logistic regression analysis (r = 0.9466, P < 0.01).

y = 1.613 − 0.512Log(x),
where x is the duration of disease.

Based on the regression analysis from the Patzold and Pocklington study, the estimate for relapse rate after two years since onset closely matches the estimate of relapse rate reported in the ScHARR report for patients in EDSS of 3+. Therefore, the regression analysis from the Patzold and Pocklington study was used as the basis for estimating the decrease of the relapse rate over time for patients in health state 2 (EDSS 3.0 to 5.5), adjusting such that the patients enter the model with an average time since disease onset of five years, as in the RCTs identified in the systematic review.

The base estimate from ScHARR of 0.835 for EDSS 0 to 2 in combination with the rate of decrease by Patzold and Pocklington was used to estimate the relapse rate for health state 1 (EDSS 0 to 2.5) for patients with five years since onset and onwards (Table 9).

Table 9. Annual Relapse Rates.

Table 9

Annual Relapse Rates.

Regarding the severity of relapses, it was assumed that 23% of relapses are severe.62 The average length of mild or moderate relapse was assumed to be 45 days, while the length of severe relapse was assumed to be 90 days, based on clinical expert opinion and published literature.

b. Treatment efficacy

The clinical efficacy of disease-modified treatments on both disability progression and relapse rates were included in the model. The comparative data were based on the NMA conducted as part of the CADTH systematic review (Table 10, Table 11). The relative rate of annual relapse and the relative risk of disability progression versus placebo were included, as the transitional probabilities describing the natural history of the disease are based on the London, Ontario dataset of untreated patients.

Table 10. Relative Risk of Disability Progression Across Treatments.

Table 10

Relative Risk of Disability Progression Across Treatments.

Table 11. Relative Rates of Annual Relapse Across Treatments.

Table 11

Relative Rates of Annual Relapse Across Treatments.

Disability progression

The transition matrix for patients on treatments was derived from the transition matrix for untreated patients by multiplying the transitional probability to a higher EDSS state by the relative risk of sustained disability progression for each treatment. The relative risk is 1 for the no-treatment strategy, and there is a lower relative risk for each of the treatments, which represent the effect of slowing disability progression.

The relative risks of sustained disability progression were applied to the transitional probabilities of patients moving to a higher health state, as well as to progressing to SPMS. Once patients progressed to SPMS, the transition between health states during the SPMS phase was unaffected by the relative risks; i.e., patients transitioned as per natural history of disease transitional probabilities in the SPMS state. The probability of staying in the same health state was then increased by the percentage of patients who did not progress because of the treatment effects, so that the sum of the transitional probabilities remained 1.

Patients who discontinue treatment will progress according to rates for natural disability progression, but will retain benefits received.


The treatment effects on the relapse rates are modelled by applying the relative rate of relapse on the average number of relapses experienced while on no treatment, as presented in Table 11. As with progression, the magnitudes of these effects differ between treatments.

c. Treatment safety

Due to the transient nature of most of the adverse events related to the RRMS treatments (such as injection site reactions), as well as some of them potentially being related to the disease process (fatigue, depression), the implications of including them in the model (the costs of treating and decrements in quality of life) were expected to be negligible. Although the difference in safety profiles might be determinant for patients and physicians on choice of treatment, the costs and effects were expected to be similar among therapies.

PML has been identified by physicians and decision-makers as an important concern, and consequently the risk of PML associated with natalizumab was included in the model. Based on a recently published article by Hunt and Giovannoni,70 there is a risk of developing PML (which is associated with a mortality rate of 18.5%) for 0.15% of patients on natalizumab.

However, given different concerns with some of the treatments, monitoring costs were included to capture some of the differences for resource use. The input for necessary monitoring associated with each of the treatments was obtained from two clinical MS experts (Table 12).

Table 12. Monitoring Associated With Treatments.

Table 12

Monitoring Associated With Treatments.

d. Mortality

RRMS is disabling, but not a life-threatening disease, and there is only a small impact on mortality, captured by EDSS = 10. Therefore, the model assumes that these treatments have no survival benefit.

All-cause mortality is calculated using the Statistics Canada life table for the data-years 2000 to 2002.71,72

The data include year-on-year mortality rate distinguished by sex. As the model does not take the sex of the cohort into account, a weighted average has been calculated based on the assumption that the percentage of female patients with RRMS is 68%, as per RCTs included in the CADTH systematic review. These probabilities represent the probability of dying from causes other than MS during any given cycle.

e. Costs

The costs included in the model are drug costs, monitoring costs, and costs associated with MS care (excluding drugs) by EDSS scores.

Drug costs were obtained from the Ontario Drug Benefit Formulary (2013). For the drugs for which Canadian prices were not available at the time the analyses were conducted, information was obtained from the US, where the ratio of prices for the new agents compared with existing treatments was calculated and used to determine the hypothetical price for the new drugs. For drugs that are not approved in Canada and for which no international price is available, a conservative assumption was made that the cost will be equal to the highest-cost treatment. Sensitivity analysis regarding drugs costs was performed.

The annual drug cost was calculated based on recommended doses (Table 13) and detailed in a cost table (Table 44).

Table 13. Base Case Drug Costs.

Table 13

Base Case Drug Costs.

A systematic review of literature was conducted to identify Canadian studies reporting the cost associated with EDSS health states, as well as cost per relapse. Two major studies were found, Grima et al.74 and Karampampa et al.75 In addition, a study by Patwardhan reporting a systematic review of the cost of MS by level of disability was identified.76

The study by Grima et al. was based on a patient survey of RRMS patients recruited at MS clinics at the Montreal Neurological Institute and the London Health Sciences Centre: 153 patients in remission and 42 patients in relapse.74 The study reported cost per EDSS scores, and it included both direct costs (outpatient resources, prescription medications) and indirect costs. This study included only ambulatory patients and, therefore, patients with an EDSS score higher than 6 were excluded.

The study by Karampampa et al. was based on a web-based questionnaire including 241 MS patients in Canada.75 Of these, 235 patients had an EDSS score less than 7, and only six patients with an EDSS score of 7 or higher were included in the study. The costs included in the study were related to in-patient care, outpatient care, consultations, investigations, MS treatments, prescribed co-medication and OTC drugs, investments or modifications, professional care, informal care, and indirect costs.

Because this analysis was conducted from the perspective of a public payer, only direct costs were included. The study by Grima et al.74 was used as a primary source for the health state 1 and 2. Since the study by Grima et al. included patients with an EDSS score up to 6, the costs for health states higher than 6 were calculated based on exponential extrapolation. This assumption was based on the aforementioned study by Patwardhan et al. which, based on the systematic review, concluded that costs rose at an exponential rate with increasing MS disability levels.76 Further, the costs reported by Grima et al. did not include professional care needed for patients with more severe disability. To account for this, information was obtained from Karampampa et al.75 and added to the total costs. Costs per health state were derived by averaging across the costs by disability levels, inflated to 2012 costs using Bank of Canada Consumer Price Index information (Table 14).

Table 14. Cost Estimates by Health State.

Table 14

Cost Estimates by Health State.

The cost of mild or moderate relapse was based on the study by Grima et al.,74 as this study included only ambulatory patients who were interviewed during their visit. The cost per severe relapse was estimated based on the Patwardhan et al. study,76 which reported that the cost of severe disabilities in RRMS is 240% higher than the cost of mild or moderate disability.

f. Utilities

Several sources for quality-of-life data in RRMS were identified based on a systematic review of the literature.62,75,7779 In the base case, the utilities values by Prosser were used,62 because it considered the same health state definitions and was based on community-based preferences. The study collected both patients’ and the general public’s preferences by using the standard gamble method. Based on CADTH guidelines for economic evaluation ― which state that preferences measured directly using a representative sample of the general public, who are suitably informed about the health states being valued, are preferred80 ― the results for the community-based group were used in the base case (Table 15).

Table 15. Base Case Utility Estimates.

Table 15

Base Case Utility Estimates.

Alternative sources were also included, such as Kobelt et al.,77 ScHARR,78 Earnshaw et al.,79 and Karampampa et al.75 (Table 16), and the impact of using these sources was tested in sensitivity analysis. It should be noted that, with the exception of Earnshaw, these alternative sources did not consider the same definitions of health states as in this model; therefore, the utility values were averaged across EDSS scores to reflect the health states in the model. Consequently, these utility estimates are used only in exploratory analysis

Table 16. Alternative Utility Estimates.

Table 16

Alternative Utility Estimates.

3.3.8. Assumptions within the economic model

The following assumptions were made for the base case:

Fixed discontinuation rate of 15% across treatments for the first 2 years, followed by no discontinuation thereafter
Adverse events, except PML, do not affect the ICUR (and were not included)
PML has impact on mortality rates, and no cost impact
Patients discontinue treatment once they reach EDSS = 7.0
Patients discontinue treatment once they progress to SPMS
Treatments have no effect on the transition between SPMS states
Treatment benefits are accrued only during the treatment period
Neutralizing antibodies are not included because of lack of data and confirmation from clinical experts that results are still controversial
Treatments have no survival benefit
Background costs related to EDSS states rise exponentially with increasing MS disability levels
Patients can progress by a maximum of one EDSS score per cycle (3 months)
Relapses have no residual effect
Patients cannot switch among treatments
Treatments not marketed in Canada are assumed to be in line with international pricing. Where international pricing is not available, the price is assumed to be in line with the highest-priced drug

EDSS = Expanded Disability Status Scale; ICUR = incremental cost-utility ratio; MS = multiple sclerosis; PML = progressive multifocal leukoencephalopathy; SPMS = secondary-progressive multiple sclerosis.

3.3.9. Sensitivity analyses

a. Deterministic sensitivity analyses

Extensive univariate sensitivity analyses were conducted to test the effect of changes in underlying parameter values and assumptions within the models. The analyses conducted were:

  1. Parameter uncertainty ―
    • costs of treatments currently not marketed in Canada
    • natural history of disability progression
    • background MS costs
    • cost of relapse
    • utility values
    • disutility associated with relapse
    • rates of PML associated with natalizumab.
  2. Structural uncertainty ―
    • earlier discontinuation of treatment when patients progress to an EDSS score of 5.0 and 6.0 (base case assumes discontinuation upon progression to EDSS score of 7.0)
    • no discontinuation of treatment because of progression to SPMS (base case assumes discontinuation because of progression to SPMS)
    • time horizon of 10 years, 30 years, and 40 years (base case implements time horizon of 25 years)
    • no improvements in EDSS scores (base case assumes improvements in EDSS scores)
    • relapse rate being static (base case implements relapse rate as being time-dependent variable).

b. Probabilistic sensitivity analysis

Probabilistic sensitivity analyses were conducted using Monte Carlo simulations, such that probability distributions related to natural history parameters, relative risks, costs, and utilities were incorporated into the analysis. The analysis adopted standard methods for defining uncertainty regarding parameters.66 Transition probabilities were characterized by beta distributions, relapse rates were characterized by gamma, and relative risks were characterized by log-normal distributions. Utility values were characterized by beta distributions, while costs were characterized by gamma distributions. Drug costs were assumed fixed. Probability distributions were parameterized using empirical data; except for parameters where no measures of dispersion were available, in which case a coefficient of variation of 25% was assumed.

Estimates of incremental costs and QALYs were obtained by re-running the model employing values from the related probability distributions. In this study, 5,000 replications were conducted; i.e., a set of 5,000 outcome estimates was obtained. Cost-effectiveness acceptability curves were derived, which present the probability that each treatment is cost-effective given different values of willingness to pay for an additional QALY.

c. Value of information analysis

In addition to the deterministic and probabilistic sensitivity analyses, expected value of information analysis was conducted resulting in estimates of expected value of partial perfect information (EVPPI) for each uncertain input parameter. Expected value of perfect information is an information-based measure of the reduction in opportunity loss associated with obtaining perfect information (no uncertainty) on a parameter, and can be seen as a measure of decision sensitivity.81 The decision sensitivity is determined by the probability that a decision based on existing information will be wrong, and the cost consequences if the wrong decision is made, as perfect information could eliminate the possibility of making a wrong decision.82

Therefore, the application of EVPPI is twofold. First, EVPPI can provide estimates of the value of conducting further research in this area, given the underlying uncertainty, and can be interpreted as the expected benefit by completely resolving uncertainty around an individual input parameter. Second, EVPPI can also be used as an importance measure identifying the contribution of uncertain input model parameters to output uncertainty.

In this analysis, because of the large number of input parameters, a screening method was applied which identified the input parameters that are candidates to having high EVPPI. Dominance measure was applied as a screening method.83 Next, a novel algorithm for the calculation of a single EVPPI proposed by Sadatsafavi et al84 has been applied. The method only relies on the data generated through Monte Carlo simulations (MCS), and one set of simulations is enough to generate EVPPIs for each uncertain parameter of the model. The EVPPI is the approximation of the expected value of the difference between the net benefit of the optimal treatment and the maximum net benefits across all treatments.

3.3.10. Model validation

The model has extensively been validated. The face validity of the model has been confirmed by two independent clinical experts experienced in treating patients with RRMS, such that the model structure, model assumptions, and data inputs have been evaluated and confirmed that they reflect the available evidence and are consistent with the medical science. The internal validity of the model has been confirmed by an external technical reviewer consultant/health economist, with all mathematical calculations examined and confirmed to be performing correctly. As well, the model was confirmed to be free from computational errors. Cross- validation of the model has been performed by the primary modeller, with reports for other models in RRMS examined and compared. External validation tests of the model has been performed, with a specific emphasis on the natural history of disease, confirming that the outcome of the model have been consistent with the reported results of the natural history of disease studies.

Copyright © CADTH March 2013.

Except where otherwise noted, this work is distributed under the terms of a Creative Commons Attribution-NonCommercial- NoDerivatives 4.0 International licence (CC BY-NC-ND), a copy of which is available at

Cover of Comparative Clinical and Cost-Effectiveness of Drug Therapies for Relapsing-Remitting Multiple Sclerosis
Comparative Clinical and Cost-Effectiveness of Drug Therapies for Relapsing-Remitting Multiple Sclerosis [Internet].
CADTH Therapeutic Review, No. 1.2B.


PubMed Health Blog...

read all...

Recent Activity

Your browsing activity is empty.

Activity recording is turned off.

Turn recording back on

See more...