Rituximab for the first-line treatment of stage III-IV follicular lymphoma (review of Technology Appraisal No. 110): a systematic review and economic evaluation

Aetiology

The causes of NHL in general, including FL, are unclear. There are a number of well-established risk factors, such as infectious agents [e.g. human immunodeficiency virus (HIV)],21 immunosuppression (e.g. post organ transplantation),22 genetic susceptibility (e.g. ataxia–telangiectasia)23 and environmental factors (e.g. exposure to agrochemicals). 24 Rare immunodeficiency conditions such as hypogammaglobulinaemia, Wiskott–Aldrich syndrome and ataxia–telangiectasia have been associated with as much as a 25% increased risk of developing lymphoma;25 however, the primary causes of NHLs remain elusive.

Pathology

First infusion

The recommended initial rate for infusion is 50 mg/hour; after the first 30 minutes, it can be escalated in 50 mg/hour increments every 30 minutes, to a maximum of 400 mg/hour. 85

Subsequent infusions

Subsequent doses of rituximab can be infused at an initial rate of 100 mg/hour, and increased by 100 mg/hour increments at 30-minute intervals, to a maximum of 400 mg/hour. 84 The prepared rituximab solution should be administered as an intravenous infusion through a dedicated line. It should not be administered as an intravenous push or bolus. 85

Infusion reactions

Infusion-related side effects (including cytokine release syndrome) are reported commonly with rituximab and predominantly occur during the first infusion and include symptoms such as fever and chills, nausea and vomiting, allergic reactions (such as rash, pruritus, angioedema, bronchospasm and dyspnoea), flushing and tumour pain. 86 Mild or moderate infusion-related reactions usually respond to a reduction in the rate of infusion, which can be increased on improvement of symptoms. Patients who develop severe reactions, especially severe dyspnoea, bronchospasm or hypoxia should have the infusion interrupted immediately. 85

Before each dose of rituximab, patients should be given an analgesic and an antihistamine to reduce these effects and consideration should be given to premedication with a corticosteroid. In all patients, the infusion should not be restarted until symptoms have resolved and laboratory values and chest radiographs appear normal. Patients who have experienced severe cytokine release syndrome should be closely monitored, as although they may show an improvement in symptoms, this may be followed by deterioration. Thus, such patients must be evaluated for evidence of tumour lysis syndrome and pulmonary infiltrations with chest radiography.

Fatalities following severe cytokine release syndrome (characterised by severe dyspnoea) and associated with features of tumour lysis syndrome have occurred 1–2 hours after infusion of rituximab. Patients with a high tumour burden and those with pulmonary insufficiency or infiltration are at increased risk and should be monitored very closely. 86

Anaphylactic and other hypersensitivity reactions have been reported following the intravenous administration of proteins to patients. Clinical manifestations of anaphylaxis may appear similar to clinical manifestations of the cytokine release syndrome. However, in contrast to cytokine release syndrome, true hypersensitivity reactions typically occurs within minutes after starting infusion. 85

Pregnancy and lactation

Rituximab should be avoided during pregnancy unless the potential benefit to the mother outweighs risk of B-lymphocyte depletion in the fetus. It is also contraindicated in women who are breastfeeding. Effective contraception is required during treatment and for 12 months after treatment. 86

Cardiovascular disease

Rituximab should be used with caution in patients who are receiving cardiotoxic chemotherapy or who have a history of cardiovascular disease because exacerbation of angina, arrhythmia and heart failure have been reported. Transient hypotension occurs frequently during infusion and antihypertensive drugs may need to be withheld for 12 hours before infusion. 86

Infections

Serious infections, including fatalities, can occur during therapy with rituximab. Physicians should exercise caution when considering the use of rituximab in patients with a history of recurring or chronic infections or with underlying conditions that may further predispose patients to serious infection. 86

Search strategy

The search aimed to systematically identify all literature relating to the clinical effectiveness of (1) the intervention: rituximab in combination with chemotherapy or (2) the comparators, i.e. chemotherapy alone for the treatment of FL. The searches were conducted in September and October 2010.

Sources searched

Eleven electronic databases were searched from inception: MEDLINE including MEDLINE In-Process & Other Non-Indexed Citations (Ovid); CINAHL; EMBASE; The Cochrane Library including the Cochrane Database of Systematic Reviews (CDSR), Cochrane Central Register of Controlled Trials (CENTRAL), Database of Abstracts of Reviews of Effects (DARE), NHS Economic Evaluation Database (NHS EED) and Health Technology Assessment (HTA) databases; Science Citation Index (SCI); and BIOSIS.

Ongoing research was searched using clinical trials databases and registers including NIHR Clinical Research Network Portfolio; National Research Register (NRR) archive 2000–7; Current Controlled Trials (CCT) and ClinicalTrials.gov.

Relevant conference proceedings were searched, including the American Society of Clinical Oncology (ASCO), European Society of Clinical Oncology (ESCO), American Society of Hematology (ASH), the British Society for Haematology (BSH) and the European Hematology Association (EHA).

In addition, the reference list of relevant articles and the MS62 was handsearched. The review team also contacted experts in the field and scrutinised the bibliographies of retrieved papers to identify relevant evidence.

Search terms

A combination of free text and thesaurus terms were used. ‘Intervention’ terms (e.g. rituximab, MabThera, Rituxan) or chemotherapy terms (CHOP, CVP, etc.) were combined with ‘population’ search terms (e.g. lymphoma, non-Hodgkin’s). Copies of the search strategies used in MEDLINE are included in Appendix 5 (these were adapted for use in other databases).

Search restrictions

Searches were not restricted by language or publication date. Where possible, a filter was applied in order to limit search results to systematic reviews/meta-analyses, economic/cost evaluations, quality-of-life studies or RCTs. Examples of the RCT filter, cost-effectiveness filter and quality-of-life filter are provided in Appendix 5.

Inclusion and exclusion criteria

Number of studies identified

The search retrieved 7858 unique citations relating to clinical effectiveness (4223 duplicates were removed). Of these, 7600 articles were excluded at title/abstract stage, 243 articles were examined at full-text level, and 15 articles were unobtainable from the interlibrary loans service (see Appendix 7). In addition, 1032 articles were examined from ongoing trials registers and conference proceedings.

Number and type of studies included

Four RCTs were included: M39021 trial by Marcus et al. ,95,96 GLSG-2000 by Hiddemann et al. ,91,92 OSHO-39 trial by Herold et al. 93 and the FL2000 trial by Salles et al. 94 Overall, 24 published reports were identified which related to the four included studies, and these are listed in Appendix 8. The principal source/sources for each study are listed in Table 6.

Number and type of studies excluded

In total, 212 citations were excluded from the full text selection (see Appendix 9). Studies that could potentially have provided head-to-head data for the interventions and comparators accounted for 108 excluded articles; 44 were excluded because they were not RCTs, i.e. case reports, literature reviews, commentaries and single-arm interventions; 29 studies were excluded because the interventions used were not relevant; 13 studies were excluded because the patient group was clinically heterogeneous and data for patients with FL were not reported separately; nine studies were excluded because patients did not have FL (e.g. Hodgkin’s disease or NHL unspecified) or had aggressive disease; six studies did not provide first-line treatment; five non-English-language studies were excluded; two were study protocols; and one did not provide relevant outcome data.

One hundred and four citations that were potential candidates to inform a network meta-analysis were excluded. Fifty-four were excluded because the participants did not have FL (e.g. NHL not specified) or the disease was not indolent; 21 were excluded because the population was heterogeneous and data relating to FL were not reported separately; 15 were excluded because the interventions were not relevant; eight were excluded because they were not RCTs; four were excluded because they were non-English-language reports; one was excluded as outcome data were not relevant; and one study was not included as it did not report on first-line treatment.

Studies identified for a potential network meta-analysis

Three additional studies (corresponding to seven references – see Appendix 10 for list) met the criteria for providing evidence within a network meta-analysis, i.e. the population included FL (with analysis for FL presented separately), the therapy being investigated was either a relevant intervention or comparator (as stated in the decision problem – see Chapter 2) and appropriate outcomes were reported (as stated in the decision problem – see Chapter 2) (Figure 3).

FIGURE 3.

Network of evidence. F, fludarabine; R-B, rituximab and bendamustine; R-CHVP, rituximab, cyclophosphamide, doxorubicin, etoposide, prednisolone and interferon-alpha.

Incorporating these three studies into a network of evidence would facilitate the comparison of interventions when a direct head-to-head trial was not available (as depicted in Figure 3). However, the network meta-analysis was not undertaken, as it was not deemed appropriate given that treatment efficacy is not the only factor in terms of choice of chemotherapy selection (see Chapter 1 for discussion of other factors). Additionally, head-to-head data were available to inform a comparison between a chemotherapy regimen and that regimen with the addition of rituximab. It is noted that NICE has a strong preference for evidence from head-to-head RCTs that directly compare the technology with the appropriate comparator in the relevant patient groups as stated in the NICE methods guide (p. 15). 97

Ongoing trials

Seven ongoing studies were identified (Table 7). 70,77–80,98,99 Four studies are investigating one R-chemotherapy against another R-chemotherapy; one study is closed [a randomised phase III study of the STiL (Study Group Indolent Lymphomas)] with study follow-up complete and initial results reported as a conference abstract;78 one study is ongoing but not recruiting (ML17638)98 and two studies are ongoing and still recruiting [Purine-Alkylator Combination In Follicular lymphoma Immuno-Chemotherapy for Older patients (PACIFICO80) and Polish Lymphoma Research Group 479 (PLRG4)]. The study population in the PACIFICO trial81 is patients with FL aged > 60 years or aged < 60 years but with an anthracycline-based therapy contraindicated. Two ongoing studies are investigating the use of rituximab in maintenance following first-line induction therapy; one study is closed with follow-up completed [Primary Rituximab and Maintenance (PRIMA) study71], whereas the other study (ML17638)98 is ongoing but not recruiting. One study99 is investigating one chemotherapy compared with another chemotherapy regimen [British Lymphoma Investigation Group (BNLI) MCD vs FMD].

Population

Baseline demographic data are provided in Table 9. The target population were advanced-stage patients with FL who were symptomatic and requiring treatment (detailed eligibility criteria for each study are presented in the data extraction tables in Appendix 11). The M3902195,96 and GLSG-2000 trials91,92 recruited patients with stage III–IV FL, whereas the FL2000 trial94 recruited patients with stage II-IV FL. The OSHO-39 trial93 included CD20-positive patients with indolent NHL, which included lymphoplasmacytic lymphoma or mantle cell lymphoma (MCL); however, the primary analysis population was defined as the population of patients with FL. The OSHO-3993 and GLSG-200092 trials limited to grade 1 or 2 FL (WHO classification); the M39021 trial95,96 included grade 1–3 FL; and the FL2000 trial94 included grades 1, 2 and 3a FL.

The median age of patients randomised across the trials ranged from 52 to 61 years. Two trials presented the percentage of participants aged over 60 years: 26% in the M39021 trial95,96 and 52% in FL2000 trial. 94 The majority of patients had stage IV FL (69–77% in the three studies that reported these data). Most participants had an ECOG performance status of 0–1, ranging from 91% to 97%. Bone marrow involvement was present in 62–74% of patients, and 22–44% presented with one or more B symptoms (defined as fever, weight loss or night sweating). Elevated LDH levels (a marker of aggressive disease) were recorded in 26–37% of patients.

Within the individual studies, the treatment groups were well balanced with respect to demographic and disease characteristics, with the exception of gender in OSHO-39 trial92 (more males in the R-MCP group; no p-value reported) and the GLSG-2000 trial (higher proportion of males in the CHOP arm; p = 0.027). The populations were reasonably similar when compared across the four studies, although there were some differences, including younger median age (52–53 years) in the M39021 trial,95,96 and larger proportion of patients aged > 60 years and inclusion of stage II participants in the FL2000 trial. 93 The study populations included were generally reflective of the general FL population, with the exception of age – the median age of participants in the trials being younger than seen in clinical practice (70% are aged > 60 years when diagnosed). 10 The younger median age of trial participants meant that ECOG performance status was better than that seen in clinical practice. In addition, the M3902195,96 and OSHO-3993 trials excluded patients with an ECOG performance status of > 2.

Interventions and comparators

The interventions in each of the four studies were a R-chemotherapy combination; each trial used a different chemotherapy agent. The comparator within each trial was the chemotherapy regimen minus rituximab. These are described in Table 10. Two studies provided subsequent treatment following response to first-line treatment. The OSHO-39 trial93 planned to provide all responders with interferon-alpha maintenance [3 × million international units (MIU)/week] until disease progression. The GLSG-2000 trial92,93 randomised responding patients who were aged < 60 years to a high-dose chemotherapy regimen followed by autologous stem cell transplant (ASCT) or interferon-alpha maintenance treatment (3 × 5 MIU/week until disease progression of intolerable AEs). Patients aged ≥ 60 years received interferon-alpha maintenance.

Outcomes

The clinical efficacy outcomes reported in the four studies91–96 are shown in Table 11; primary outcomes are highlighted in grey. All four studies91–96 included the appropriate outcome measure of OS; defined as the time from randomisation to the date of death by any cause. The OSHO-39 trial93 was the only trial to report PFS, defined as randomisation to disease progression or death from NHL. All four studies91–96 appropriately reported response rates (according to the International Workshop criteria described by Cheson et al. 89). Two studies91–93 did not use the category of ‘unconfirmed complete responder’ (CRu), instead counting such patients within the PR category. The FL2000 trial94 and M39021 trial95,96 used the category of CRus and presented the numbers separately from CRs and PRs. No studies reported the duration of disease remission, although the studies did report a number of time-to-event outcomes which approximated disease remission, for example all four studies reported response duration as an outcome.

Other time-to-event outcomes reported by one or more of the studies were EFS, time to treatment failure (TTF), time to next antilymphoma treatment (TTNT), DFS and TTP. However, these outcomes were inconsistently defined by the four studies91–96 and thus not directly comparable across the four studies. For example, the M3902195,96 and GLSG-2000 trials91,92 measured TTF and both studies considered a treatment failure as disease progression. However, the M39021 trial95,96 additionally considered death by any cause, relapse after response, new antilymphoma treatment or stable disease after cycle 4 as treatment failures, whereas the GLSG-2000 trial91,92 also considered resistance to initial therapy and death not specified as treatment failures. In addition, when the definitions for each time-to-event outcome were cross-referenced against each other, no outcomes were directly comparable (e.g. we examined whether or not PFS as measured in the OSHO-39 trial93 may have matched the definition used for EFS as measured by FL2000 trial;94 however, this was not the case). Appendix 12 provides the definitions for each outcome described in the four studies. 91–96

All four studies reported data on AEs. The M39021, OSHO-39 and FL2000 trials93–96 graded AEs in accordance with the National Cancer Institute Common Toxicity Criteria (NCI-CTC) grading system,100 and the GLSG-2000 trial91,92 used the WHO toxicity criteria101 to record AEs. The GLSG-2000 and OSHO-39 trials93 reported data for grade 1, 2, 3 and 4 AEs separately,92,93 and the M39021 and FL2000 trials94–96 reported AEs for grades 3 and 4 combined. None of the studies reported HRQoL as an outcome.

Quality assessment

All four included studies91–96 were randomised and allocation was concealed using centralised allocation to treatment. Numbers randomised were stated in all four studies. None of the studies were blinded; all were open label and none of the studies reported attempting to conceal treatment allocation from the outcome assessors. Power calculations were undertaken by all four included studies. 91–96 At least 80% of patients were followed up in all four studies. All four studies reported baseline characteristics and were mostly balanced between treatment groups; with the exception of gender in OSHO-39 trial93 (greater number of males in the R-MCP group; no p-value reported) and the GLSG-2000 trial (higher proportion of males in the CHOP arm; p = 0.027). The M3902195,96 and FL-2000 trials94 reported no significant differences in baseline data. All studies specified eligibility criteria.

Co-interventions were used in three studies. 91–94 Interferon maintenance therapy was given to patients in the OSHO-39 trial93 achieving a partial or complete remission; this was initiated within 4–8 weeks after treatment completion. In the GLSG-2000 trial,91,92 patients < 60 years who had achieved either CR or PR were offered a second randomisation of Dexa-BEAM regime (salvage chemotherapy) followed by stem cell harvest and radiochemotherapy, or long-term interferon maintenance, whereas patients aged > 60 years were given interferon maintenance. Patients in both arms in the FL2000 trial94 were given interferon-alpha as part of initial treatment (6 months) and then as a consolidation treatment for a further 12 months. In addition, 11% of patients in the FL2000 trial94 had stage II FL. Reasons for withdrawals were unclear in the four studies. 91–96 Most withdrawals were stated as being a result of disease progression; however, withdrawals relating to AEs were not explicitly stated. All four studies91–96 reported using ITT analyses. See Figure 4 for overview of the quality assessment.

FIGURE 4.

Quality assessment of the included trials (+, yes; – , no; ?, unclear).

Response to treatment

Response to treatment is reported in Table 12. ORR was significantly improved for patients receiving R-chemotherapy than those who received chemotherapy alone in three studies91–93,95,96 (the FL2000 trial94 did not report a p-value). The ORR in the four studies ranged from 81% to 97% for the R-chemotherapy arm and from 57% to 91% for the chemotherapy-only arm. The difference in ORR between the treatment and comparator arms in each of the four studies ranged between 5% and 24%; the greatest difference was between the R-CVP and CVP arm. R-CHOP, R-CHVPi and R-MCP were the regimens that provided the highest ORR of 96%, 94% and 92%, respectively. CHOP alone provided a high ORR of 91%.

Difference in the CR rates between treatment and comparator arms in the four studies91–96 ranged from 2% to 25%, and was reported as significant in two studies. 93,95,96 The regimens providing the highest CR rates were R-CHVPi and R-MCP (51% and 50%, respectively). The number of CRs in the GLSG-2000 trial91,92 for both R-CHOP and CHOP (19% and 17%, respectively) were notably lower than those reported in the other studies. The greatest difference in CR between treatment and comparator arms was reported in the OSHO-39 trial93 between R-MCP and MCP (25%).

The difference in PR rate ranged from 2% to 11%. None of the four studies reported a p-value for the difference between treatment and comparator arms.

The GLSG-200091,92 and FL2000 trials94 reported low numbers of patients within the stable disease category. However, the M39021 trial95,96 reported greater numbers of patients with stable disease (7% in R-CVP and 21% in CVP). Meta-analysis of response rates in the four trials has been explored (see Meta-analysis, for further discussion).

Chi-squared test for response rates

The AG performed a chi-squared test on the response rate data to compare the numbers within each category of response between the two trial arms for each of the four trials. The results showed that there was a statistically significant difference in the numbers in the response categories for the R-chemotherapy arm compared with the chemotherapy-alone arm for R-CVP compared with CVP, R-MCP compared with MCP and R-CHVPi compared with CHVPi 6-month response rate (p < 0.001 for all comparisons). The difference between the categories of response was not statistically significant for R-CHOP compared with CHOP (p = 0.15) and for the R-CHVPi compared with CHVPi 18-month response rate (p = 0.12).

A second analysis was performed for each trial, which combined relevant categories of response (e.g. progressive disease or death) where necessary so that the number of observations within each category was greater than five per cell. Where grouping was performed, death was categorised with progressive disease. In one analysis, stable disease was categorised with disease progression and death, as clinical advice to the AG indicates that patients with stable disease are treated as patients with disease progression and not responders. In terms of statistical significance at the 5% level, the effects of grouping altered only on comparison of R-CHVPi and CHVPi at 18 months, which became statistically significant. Analyses are presented in full in Appendix 13.

Overall survival

The OS rate in the four studies91–96 ranged from 83% to 90% in the R-chemotherapy arms and from 77% to 84% in the chemotherapy-alone arms (Table 13). The difference in OS rate was significantly improved in three trials when R-chemotherapy was compared with chemotherapy alone; the exception being the FL2000 trial94 (p = 0.1552). The median OS was reported as not reached in three studies and was not reported in the FL2000 trial. 94 The OS data from the GLSG-200091,92 and OSHO-3993 trials were confounded owing to the effects of subsequent therapy provided to all responders to first-line treatment. The FL2000 trial94 also provided additional treatment (interferon-alpha) to both treatment arms during the 6-month remission induction phase. In addition, the FL2000 trial94 provided a further 12-month treatment phase in which the chemotherapy-alone arm received bimonthly CHVP and both treatment arms received interferon-alpha.

Overall survival: hazard ratios

The hazard ratios (HRs) for OS were not available in the manuscripts for each of the individual trials. The AG used Kaplan–Meier plot data provided in the health economic model in the MS,61 which provided a series of survival probability estimates at monthly time points for two of the four trials: M39021 and OSHO-39. 93,95,96 Visual inspection of these probability estimates alongside the Kaplan–Meier data provided in the publications for each trial indicated that these data were reasonable. Kaplan–Meier data for OS for the FL2000 trial94 and the most-up-to-date data for the GLSG-2000 trial91 were digitised by the AG using TechDig^© software to estimate survival probability estimates at time points along the Kaplan–Meier curve.

The corresponding HRs were calculated by taking the ratio of the cumulative hazard from the R-chemotherapy and chemotherapy arms from the OS Kaplan–Meier curves. The cumulative hazard was calculated by summing the negative log of the survival probabilities {H(t) = –Slog[S(t)]} for each treatment arm, restricted to the clinical follow-up reported in the publications. 102 There are limitations with this method of calculating HRs, namely that it relies on the data from the trial publications rather than patient-level data and that estimating survival probabilities from digitised curves are subject to inaccuracies. As such, these estimates provide an indication of the HR for OS rather than definitive values. Given resource constraints and data limitations, it was not possible to calculate the standard errors (SEs) and confidence intervals (CIs) to give an indication of the uncertainty in the data.

For all four trials,91–96 there was an increased likelihood of survival if receiving R-chemotherapy. For R-CVP compared with CVP, there was 36% increased survival benefit, for R-CHOP compared with CHOP there was a 42% increased survival benefit, for R-MCP compared with MCP a 60% increased survival benefit and for RCHVPi compared with CHVPi there was a 31% survival benefit. However, it is noted that the treatment effect on OS is confounded in the latter three trials owing to additional trial treatments administered after response to first-line treatment.

Progression-free survival

The median PFS was significantly prolonged in OSHO-39 trial93 for the R-chemotherapy arm (R-MCP) (28.8 months MCP vs median not reached R-MCP; p < 0.0001). PFS was not reported in the other three trials.

Other time-to-event data

Several other efficacy outcomes, namely time-to-event data, were reported in the four studies. 91–96 As stated above (see Summary of trials), these outcomes were inconsistently defined between the four studies and thus not directly comparable (see Appendix 12). In addition, the time-to-event data were confounded in GLSG-200091,92 and OSHO-3993 trials owing to the effects of subsequent treatment provided to responders to first-line treatment in these trials. However, we present a summary of the findings in Table 14.

The median response duration was significantly prolonged for the R-chemotherapy arm compared with the chemotherapy-alone arms (p < 0.001) in the M39021 and OSHO-39 trials. 93,95,96 Two studies reported the duration of response, which differed significantly between treatment and comparator arms; at 5 years in the GLSG-2000 trial91,92 (p < 0.0001) and the 4-year estimates presented in the FL2000 trial94 (p = 0.012). Significantly prolonged (p < 0.0001) median TTF was reported for the R-chemotherapy arm compared with chemotherapy-alone arm in the M39021 and GLSG-2000 trials. 91,92,95,96 Similarly, median EFS was significantly improved in the R-chemotherapy arms in two studies compared with the chemotherapy-alone arms [median EFS MCP 26 months, not reached in R-MCP (p < 0.0001); median EFS 35 months in CHVPi, not reached in R-CHVPi (p = 0.0004)]. 93,94 The M39021, GLSG-2000 and OSHO-39 trials reported a statistically significant difference in TTNT. 91–93,95,96 The M39021 trial95,96 reported significantly improved DFS and TTP for R-CVP compared with CVP.

Follicular Lymphoma International Prognostic Index score

All four studies91–96 presented analysis of treatment outcomes according to FLIPI score subgroups. The M39021 trial95 found after undertaking univariate analyses that median TTP was significantly improved in the R-CVP group at 53-month follow-up for all FLIPI groups (low, intermediate and high risk) (Table 15). Similarly, the GLSG-2000 study91 found significantly prolonged 5-year TTF associated with the addition of rituximab in all FLIPI subgroups [84% vs 46% for low risk (p = 0.0021); 73% vs 37% for intermediate risk (p < 0.0001) and 49% vs 23% for high risk (p < 0.0001)].

Marcus et al. 95 conducted a multivariate analysis (which included the FLIPI score as a composite along with other prognostic factors that are not incorporated in the FLIPI), which found that only the FLIPI low-risk and intermediate groups combined (0–2) compared with high-risk groups (3–5) was a significant prognostic parameter for TTP in addition to trial treatment.

The MS62 presented data on the OSHO-39 trial,93 which demonstrated that treatment with R-MCP significantly increased the 4-year PFS rate, as well as prolonging the median TTP or death in patients with intermediate (p = 0.0016), as well as high-risk (p = 0.0011) FLIPI subgroups. Among patients with high-risk disease, a significant improvement in OS was also seen among those treated with R-MCP compared with MCP (p = 0.0096). 104 No such significant improvement between treatment arms was noted for median OS for the FLIPI intermediate subgroup (p = 0.8607). These data are presented in the MS61 and are reproduced in Table 16; FLIPI 0–1 data were not presented.

In the FL2000 trial,94 when patients with either a low (n = 65) or an intermediate (n = 122) FLIPI score were grouped together, no significant difference in EFS or OS was seen between the treatment arms. However, significant improvements in 5 years’ EFS (p < 0.001) and OS (p = 0.025) were seen between the treatment arms in the high-risk FLIPI subgroup. Cox regression analysis, which included the FLIPI score (low and intermediate vs high) and the treatment arm, confirmed the impact of both parameters on EFS [FLIPI, HR = 2.08 (95% CI 1.6 to 2.8); R-CHVPi treatment, HR = 0.59 (95% CI 0.44 to 0.78)] and OS [FLIPI, HR = 4.11 (95% CI 2.34 to 7.23); R-CHVPi treatment, HR = 0.67 (95% CI 0.41 to 1.11)].

International Prognostic Index

Marcus et al. 95 conducted a univariate analysis of the M39021 trial data, which found significantly prolonged median TTP for all IPI risk groups (Table 17). Similarly analysis of the GLSG-2000 trial data91 found significantly prolonged TTF at 18-month follow-up by IPI risk group (Table 18).

Age

Eighteen-month follow-up data in the GLSG-2000 trial92 found that TTF was prolonged in the R-CHOP arm for patients of any age (Table 19). The relative risk (RR) of treatment failure in the R-CHOP arm compared with the CHOP arm was 0.417 (95% CI 0.233 to 0.747) for patients aged < 60 years and was 0.354 (95% CI 0.175 to 0.715) for patients aged ≥ 60 years.

Quality of response

Salles et al. 94 analysed the response duration for the subgroup of patients who were in CR/CRu at 18 months of treatment in the FL2000 trial. The response duration was significantly different between the two treatment arms, with 4-year estimates of 44% (95% CI 32% to 54%) compared with 64% (95% CI 55% to 72%) in the CHVPi and R-CHVPi arms, respectively (p = 0.012). Therefore, as well as rituximab and chemotherapy increasing the number of CR/CRus, patients are also more likely to have a longer response duration.

Other prognostic factors

Marcus et al. 95 conducted several univariate analyses for a number of prognostic factors (Table 20) in the M39021 trial. The R-CVP treatment arm was associated with a significant prolonged TTP when compared with CVP alone for all subgroups investigated including baseline histology, presence or absence of B symptoms, and presence or absence of bulky disease. A significant improvement in TTP was seen in patients with baseline-only haemoglobin of at least 12 g/dl; however, no difference in TTP was observed between the R-CVP and CVP arms in patients with baseline haemoglobin of < 12g/dl (p = 0.3941).

Marcus et al. 95 also undertook two multivariate analyses: one that included the IPI as a composite along with other prognostic factors not incorporated in the IPI and one that included the individual factors which make up the FLIPI and IPI, together with other prognostic factors. These analyses found that only haemoglobin level (< 12 g/dl) and number of nodal areas involved (> 1) were statistically significant predictors of TTP in addition to trial treatment.

Buske and Hoster105 conducted a multivariate analysis on the GLSG-2000 trial92 data at 20-month follow-up including the individual FLIPI risk factors. This found that a serum LDH level higher than the upper normal limit (RR 2.6, 95% CI 1.5 to 4.5) and a haemoglobin level of < 12 g/dl (RR 2.5, 95% CI 1.4 to 4.3) were independently associated with a shorter TTF in addition to trial treatment. However, age (≥ 60 years vs < 60 years; RR 0.9, 95% CI 0.5 to 1.5) and the number of nodal areas (> 4 vs ≤ 4; RR 1.5, 95% CI 0.8 to 2.6) did not significantly influence the TTF.

Treatment completion and withdrawals

The M39021, OSHO-39 and FL2000 trials93–96 reported data on the number of treatment cycles that were completed. No data were presented on the planned cycle completion, doses of study drugs administered and withdrawal numbers or reasons in the GLSG-2000 trial. 91,92

Overall, a greater proportion of patients in the R-chemotherapy arms received the planned number of cycles when compared with the chemotherapy-alone arm (Table 21). No differences in dose of chemotherapy received were noted between the R-chemotherapy and chemotherapy-alone arms, with the exception of cyclophosphamide in the M39021 trial. 95,96 Reasons for withdrawal from treatment appeared to be mostly owing to disease progression or treatment failure (e.g. failing to achieve a response to treatment after a defined number of cycles). However, there was a lack of transparency in the studies regarding withdrawals for other reasons such as AEs/reactions. This is considered in more detail by trial.

M39021 trial

A Consolidated Standards of Reporting Trials diagram108 was reported for the M39021 trial,95,96 which showed the flow of patients through the trial. This showed that 137/162 (85%) of patients in the R-CVP arm and 108/159 (68%) patients in the CVP arm completed eight cycles. 95,96 The MS62 provided further details on cycle completion, with 6/162 (4%) of patients in the R-CVP arm withdrawn before cycle 4 compared with 13/159 (8%) in the CVP arm. Thus, 19/162 (12%) patients in the R-CVP arm were withdrawn after cycle 4 compared with 38/159 (24%) in the CVP arm. The majority of patients appear to have been withdrawn owing to an insufficient treatment response (defined as disease progression or stable disease after cycle 4). However, a number of patients were withdrawn before cycle 4 for which the reasons are not made explicit. The authors note that two patients were withdrawn as a result of grade 3 or 4 rituximab infusion-related reactions and one patient withdrew consent and thus withdrew from the trial; however, this does not account for all patients.

Marcus et al. 96 report the proportion of patients in the M39031 trial who received the planned doses of chemotherapy. The proportion of patients that received > 90% of the planned dose of prednisolone and vincristine at each administered cycle was comparable between the R-CVP and CVP arms. However, the proportion of patients who received > 90% of cyclophosphamide was higher in the CVP group (> 94%) than the R-CVP group (> 85%). The authors state that this was ‘mainly due to dose modifications in the R-CVP group for NCI–CTC grades 3 and 4 neutropenia’. Clinical advice suggests this is now less of a problem as granulocyte-stimulating factor is routinely used to treat neutropenia. Ninety-six per cent of patients received > 90% of the planned dose of rituximab at each administered cycle. 96

OSHO-39 trial

In the OSHO-39 trial,93 88% of patients in the R-MCP arm and 67% in the MCP arm completed all eight cycles of treatment. Treatment failure owing to disease progression after two cycles occurred in three patients in the R-MCP arm and in 10 patients in the MCP arm. Failure to achieve at least a PR after six cycles occurred in seven patients in the R-MCP arm and 22 patients in the MCP arm. Numbers of patient withdrawals (n = 16) prior to the study drug administration and the associated reasons were reported; however, this includes patients with MCL as well as FL. The authors state that all other withdrawals were because of non-response/treatment failure during therapy (which was defined as disease progression after two cycles of therapy or failure to reach a PR or CR after six cycles of therapy). The authors do not state if there were any withdrawals because of AEs or reactions.

The mean dose of study drugs administered in the OHSO-39 trial92 were rituximab, 660−680 mg/cycle; mitoxantrone, 24−28 mg/cycle; chlorambucil, 68−81 mg/cycle and prednisolone, 226−231 mg/cycle. The authors stated that the dose intensity of the chemotherapy did not differ between treatment arms. 93 Interferon-alpha maintenance treatment (3 × 4.5 MIU/week until disease progression) was initiated in 97% and 92% of responding patients in the R-MCP and MCP arms, respectively.

FL2000 trial

In the FL2000 study,94 the MS62 noted that 95% of patients in the R-CHVPi arm and 94% of patients in the CHVPi arm received the initial six cycles of treatment. Among patients who did not progress during therapy, 161 (98%) and 153 (98%) of the patients received the planned chemotherapy courses during the first 6 months in the R-CHVPi and CHVPi arms, respectively. In the CHVPi arm, 116 (87%) of 134 patients without death or progression received the six planned cycles of chemotherapy consolidation; the R-CHVPi arm did not receive this chemotherapy consolidation. Two hundred and thirty-seven (66%) patients followed the interferon treatment according to the protocol, with dose adaptation (45 patients) or short (< 4 weeks) interruptions (55 patients), without significant differences in adaptation between the two study arms. In addition, interferon treatment was stopped in 50 patients resulting from disease progression (R-CHVPi arm, 19 cases, and CHVPi arm, 31 cases, respectively) and was interrupted either for > 1 month (16 cases) or definitively (72 cases) resulting from toxicity. These major interruptions were observed in 41 patients in the RCHVPi arm and 47 patients in the CHVPi arm. One patient withdrew consent after registration, and one patient had a major inclusion violation (registered at relapse) and thus were withdrawn from the treatment in the FL2000 trial. 94 No further details are provided on withdrawals in the FL2000 trial94 during treatment; although not all patients received the planned six cycles of initial treatment.

Adverse events of any grade

Adverse events of any grade were reported as more frequent in the R-MCP arm than in the MCP arm in the OSHO-39 trial93 (99% vs 86% of patients, respectively). However, the M39021 trial95,96 reported that the proportion of patients that reported at least one AE was comparable between the CVP (95%) and R-CVP (97%) groups. Marcus et al. 96 report that AEs associated with the gastrointestinal and nervous systems as well as general disorders and administration site reactions were the most commonly occurring types of events in both treatment groups in the M39021 trial. 95,96 Fatigue, neutropenia and back pain were the most common severe AEs and occurred at a slightly higher frequency in patients receiving R-CVP. These data were not available within the manuscripts95,96 reporting on the M39201 trial but appear to be confirmed by data presented in the MS,62 which reports on all grades of AEs in the M39201 trial.

Grade 1 and 2

The OSHO-3993 and GLSG-200092 trials reported grade 1 and 2 AEs. The authors in each trial reported that there were no significant differences between the treatment arms. The most common grade 1/2 AE in the OSHO-39 trial92 study was infection, which affected 42% of patients receiving R-MCP, and 35% receiving MCP. In the GLSG-2000 trial,92 the most commonly reported grade 1/2 AE was low haemoglobin level, with 50% of R-CHOP and 49% of CHOP patients affected. Neurotoxicity was another frequent grade 1 or 2 AE reported in the GLSG-2000 trial92 (R-CHOP 34%, CHOP 42%). Reduced platelet count was also a common AE, especially in the OSHO-39 trial93 (R-MCP 30%, MCP 33%), whereas the GLSG-2000 trial92 reported lower incidences for patients receiving the CHOP-based treatments (R-CHOP 17%, CHOP 16%). Nausea and vomiting was another frequent grade 1 or 2 AE in both trials (R-CHOP 45%, CHOP 44% in the GLSG-2000 trial and R-MCP 24%, MCP 15). 93 For a detailed list of grade 1 and 2 AEs see Table 22.

Grade 3 and 4 adverse events

All fours studies reported grade 3 and 4 AEs; the GSLG-200092 and OSHO-3993 trials reported grade 3 and 4 AEs separately, whereas the M3902195,96 and FL200094 trials combined the numbers of grade 3 or 4 AEs. The most common AEs observed in the four trials were related to the blood and bone marrow, including leucocytopenia, neutropenia and granulocytopenia. For two trials, the most common grade 3 and 4 AEs were reduced leucocyte (white blood cell) levels; this was observed in 69% of R-CHOP and 61% CHOP patients in the GLSG-2000 trial92 and 72% R-MCP and 58% MCP patients in the OSHO-39 trial. 93 The statistical significance of the difference in grade 3/4 leucopenia between the treatment arms in the OSHO-39 trial93 was not reported by the authors, whereas the difference between the R-CHOP and CHOP treatment arms in the GLSG-2000 trial92 was reported as not significant.

The most common AE in the M39021 trial95,96 was neutropenia (24% in R-CVP and 14% in CVP arms); however, the authors do not state if this was a statistically significant difference between treatment arms. In the FL2000 trial,94 the most common grade 3/4 AE was neutrophil toxicity (59% R-CHVPi and 62% in CHVP arms). However, the FL2000 trial94 only noted a significant difference in grade 3 or 4 AEs for neutrophil toxicity during the 12-month consolidation period, which was more frequent in the chemotherapy-alone arm than the rituximab-containing arm (p < 0.001) (results presented in the data extraction form for the FL2000 trial94 in Appendix 11).

There were a number of patients who had a low granulocyte count of grade 3 or 4 severity in the GSLG-2000 trial92 and the difference between the treatment arms was statistically significant (R-CHOP 63%, CHOP 53%; p = < 0.01). In addition, grade 3 or 4 alopecia was a frequently observed AE in both arms of the GLSG-2000 trial92 (R-CHOP 67%, CHOP 61%).

Blood or bone marrow AEs may be associated with infection. However, the difference in frequency of blood or bone marrow AEs between treatment arms is of minor clinical significance as they did not translate into a difference in infection rates between the treatment arms for all three studies. Infections of grade 3 or 4 were observed in 8% of the MCP group and 7% of the R-MCP group; 5% R-CHOP arm and 7% CHOP arm and 2% of the R-CHVPi arm and 0% CHVPI arm. 92–94 The MS61 reports all grades of infections for three trials, and follows a similar pattern (33% R-CVP and 32% CVP, 38% R-CHOP and 36% CHOP, 49% R-MCP and 43% MCP).

More detail on grade 3/4 AEs combined for the four trials and grade 3 or 4 AEs reported separately (only for the GSLG-200092 and OSHO-3993 trials) are reported in Tables 23 and 24, respectively.

Infusion-related reactions

Infusion-related reactions were observed in 7% of courses during the first infusion in the GSLG-2000 trial92 and early cessation of rituximab therapy was required in two patients. Fourteen (9%) patients in the M39201 trial95,96 had a grade 3 or 4 rituximab infusion-related reaction, and two of these patients were withdrawn from study treatment. More patients in the R-CVP group than in the CVP group experienced an AE within 24 hours of an infusion (71% vs 51%, respectively). One grade 3 infusion-related reaction was reported in the OSHO-39 trial93 in the MS62 and related to the full study population of FL and MCL.

Death and life-threatening adverse events

Overall, there were very few AEs reported as life-threatening or leading to death within the trials. The M39201 trial95,96 reported that five patients experienced a total of six life-threatening events following R-CVP; however, no treatment-related deaths occurred. The remaining three studies did not report whether or not AEs were either life-threatening or led to death.

The number of deaths reported for the chemotherapy-alone arms were consistently higher compared with the R-chemotherapy arms in all four trials. A total of 49 deaths were reported in the M39201 trial96 from 30-month follow-up96 (21 in the R-CVP arm and 28 in the CVP arm; patients may have received second-line therapy at this stage). Twenty-three deaths (17 CHOP and six R-CHOP) and 40 deaths (25 MCP and 15 R-MCP) occurred in study GLSG-200091 and study OSHO-39,93 respectively. In the FL2000 trial,94 a total of 45 patients had died at the time of the analysis at 42 months (16 R-CHVPi and 29 CHVPi). The majority of deaths were attributed to lymphoma progression. The GLSG-2000 study92 reported the additional reasons for death in detail (Table 25); however, the other three trials did not report this information.

Subgroup analyses

The MS61 reported data on the safety from the GLSG-2000 trial92 for the elderly population (≥ 60 years of age, n = 221). As for the whole trial population, the most common AEs were blood and bone marrow disorders, gastrointestinal disorders, skin toxicities, neurological disorders, cardiac disorders, infections and fever. Most of the AEs were mild to moderate in intensity, except for alopecia, leucopenia and neutropenia, which were mainly of grade 3/4 in intensity. The most common grade 3/4 AEs in the elderly population were blood and bone marrow disorders and alopecia. The remaining three trials did not provide AE data for subgroup populations.

Postmarketing data (taken from the manufacturer’s submission)

Over 1 million patients (length of exposure not known), predominantly NHL patients, have received rituximab since its first marketing authorisation. Worldwide safety data submitted to the Periodic Safety Update Reports (PSURs) (with a cut-off date of April 2007) have recorded 13,008 AEs. Of these reported AEs, 10,184 were classified as serious. For 7174 events, the report came from spontaneous sources (postmarketing experience). Other sources include clinical trials in oncology and rheumatoid arthritis (company- and investigator-sponsored trials). The MS62 presents a summary of AEs in the global safety database for rituximab (as of 30 April 2007) and this is presented in Table 26. The most frequently reported events were infection and infestation (15%), blood and lymphatic system disorders (14%), general disorders and administration site conditions (11%) and respiratory, thoracic and mediastinal disorders (10%).

The updated summary of product characteristics from the EMA84 also discusses cases of progressive multifocal leucoencephalopathy (PML) being associated with the use of rituximab. All patients treated with MabThera for rheumatoid arthritis must be given a patient alert card with each infusion, which contains important safety information for patients including signs and symptoms to watch out for. However, cases of PML reported during postmarketing use of rituximab in NHL are very rare (numbers/percentages are not reported).

Identified studies

The search retrieved 280 citations relating to cost-effectiveness (Figure 5). Two hundred and fifty-four articles were excluded at title stage, and 21 articles were excluded at abstract level. Four studies (corresponding to five references) were examined at full-text level111–115 and three studies (corresponding to four references) were identified as meeting the inclusion criteria of the systematic review of economic evaluations. 111–113,115 This included the Evidence Review Group (ERG) report submitted to NICE for TA110112 in which the addition of rituximab to CVP in first-line induction treatment was evaluated. Gomez et al. 114 was excluded from the review as this reference was unobtainable.

FIGURE 5.

Flow diagram of economic evaluation selection/exclusion.

An economic model was described in two studies: an HTA monograph113 and an ERG report. 112 Both studies are based on a critique undertaken by the ERG of the model submitted by the manufacturer (Roche) for TA110,83 a single technology appraisal (STA).

Overall, three different economic models were identified.

Summary of published cost-effectiveness studies

The three identified economic models111–113,115 were similar and used a Markov approach. There were differences in the comparators used between the studies. Dundar et al. 112,113 and Hornberger et al. 115 evaluated the cost-effectiveness of the addition of rituximab to CVP only. Ray et al. 111 reported the cost-effectiveness of the addition of rituximab to a CVP, CHOP, MCP and CHVPi regimen.

Ray et al. 111 and Dundar et al. 112,113 adopted the perspective of the UK NHS and Personal Social Services (PSS) with costs and benefits discounted at an annual rate of 3.5%. Hornberger et al. 115 conducted an economic evaluation in the USA, with costs and benefits discounted at 3.0%.

The impact of main model parameters was examined in univariate sensitivity analyses (SAs)in all economic evaluations identified by the AG. 111–113,115 Probabilistic sensitivity analyses (PSAs) were performed in the two UK models only. 111–113

The two UK economic evaluations produced broadly similar incremental cost-effectiveness ratios (ICERs) for the comparison of R-CVP with CVP. Dundar et al. 112,113 reported a cost per quality-adjusted life-year (QALY) gained of £8290 for the addition of rituximab to a CVP regimen in the MS62 model. Ray et al. 111 reported an ICER of £8613 per QALY gained for the same comparison and reported an ICER of £10,676, £7455 and £8498 per QALY gained for the addition of rituximab to a CHOP, MCP and CHVPi regimen, respectively. The two UK economic evaluations111–113 showed that the addition of rituximab to chemotherapy compared with chemotherapy alone has a cost per QALY gained under £20,000. In the US, Hornberger et al. 115 reported a cost per QALY gained of US$28,565 for the comparison between R-CVP and CVP.

A tabulated summary of key features and data sources for studies included in the review is presented in Table 27.

A full description of each of the three cost-effectiveness studies along with a quality assessment checklist is presented below.

Critical appraisal of economic evaluation

The included cost-effectiveness studies;111–113,115 were assessed against a critical appraisal checklist adapted from the Drummond and Jefferson109 and Eddy110 checklists (Table 28).

Description and results of the published economic evaluations

Overview

The MS62 used a state-transition model with individuals moving between four possible HSs: PFS/first-line induction treatment (PFS1); PFS/second-line treatment (PFS2); progressive disease; and death (Figure 6). The model compared the cost-effectiveness of the addition of rituximab to CVP, CHOP, MCP and CHVPi for patients with advanced FL in the UK. The starting age in the model was 60 years and patients were followed up for 25 years. The study adopted the perspective of the UK NHS, with costs and QALYs discounted at 3.5%. A tabulated summary of key features and data sources of the economic model included in the MS62 is presented in Table 31.

FIGURE 6.

Model structure included in the MS62 (reproduction of figure 3 from pp. 104 of the MS62).

Summary of effectiveness data

The effectiveness in first-line induction treatment was derived from four randomised Phase III clinical trials in patients with FL comparing the addition of rituximab to CVP,95,96 CHOP,91,92 MCP93 and CHVPi. 94 Individual patient-level data from the M39021 trial from journal manuscripts95,96 and the MS62 were used to estimate the rate of progression among patients treated with CVP or R-CVP in first-line induction assuming a log-logistic distribution. Individual patient-level data for the trials that compared CHOP with R-CHOP, MCP with R-MCP and CHVPi with R-CHVPi91–94 were not available to the manufacturer and therefore only publicly available data were used. A similar methodology to Ray et al. 111 was used by fitting a Weibull or exponential distribution (to the digitised data from the papers) to patients treated in first line with chemotherapy alone. The exponential distribution was selected for CHOP and MCP, whereas the Weibull distribution was chosen for CHVPi based on the R2-value. A HR was then applied to the estimated curves for the first 53 months to estimate the reduction in the risk of progression for patients receiving rituximab in addition to chemotherapy. Deaths in PFS1 were derived from the number of deaths and follow-up duration from the M39021 trial. 95,96

The effectiveness in PFS2/second-line treatment was based on data from the EORTC 20981 trial74,75 conducted among patients treated with CHOP or R-CHOP with or without maintenance rituximab in second line. Digitised data from the paper74,75 were used in the absence of individual patient-level data. The manufacturer used exponential distributions to estimate the risk of progression. The manufacturer stated that:

In order to avoid overcomplicating the model, the transition probabilities of progressing from PFS2 were not varied over time. Varying the probabilities over time would require tracking patients’ progression within the model and would result in an exponential increase of the size and complexity of the model with limited impact to the cost effectiveness of rituximab in first-line.

The most up-to-date data from the EORTC 20981 trial74 were used to estimate the progression rate from PFS2 to the progressive HS, and from the progressive HS to death [post-progression survival (PPS)]. The PPS have been calculated as a function of PFS and OS, assuming that the rate of progression in PPS equalled the sum of the rate of progression in OS and PFS. The manufacturer also attempted to apply a rule so that patients treated with rituximab in first-line induction and who relapse within 6–12 months would not receive rituximab in second-line induction.

Utilities were extracted from a study commissioned by the manufacturer (Oxford Outcomes Study). 118,119 The following utility values were used in the economic model; PFS1 = 0.88 (disease free); PFS2 = 0.79 (remission/full response); progressive disease = 0.62. AEs were not included in the MS. 62

Summary of resource utilisation and cost data

Drug costs were taken from the British National Formulary (BNF),86 using the planned dose from the trials. Administration costs were taken from NHS reference costs. The manufacturer also assumed that rituximab treatment was administered as a hospital day case.

The cost associated with monitoring/surveillance after induction treatment was derived from a study commissioned by the manufacturer. Supportive care costs for patients in the progressive HS (£500.53 per month) were derived from the cost used in the MS62 for an ongoing NICE appraisal124 from the post-protocol treatment from the EORTC 20891 trial74,75 and the cost of palliative care in the UK. 125

Summary of cost-effectiveness

Two analyses were presented for R-CVP compared with CVP. The first analysis fitted separate curves to each arm using individual patient-level data, whereas the second analysis assumed a HR (for R-CVP) for the first 53 months and fitted a parametric curve to CVP using the same approach as for CHOP, MCP and CHVPi.

In the base-case analysis, the addition of rituximab to CVP, CHOP, MCP and CHVPi led to a gain of 0.867/0.443, 1.096, 1.289 and 0.675 discounted QALYs compared with chemotherapy alone, respectively. The incremental discounted cost of the addition of rituximab to chemotherapy was estimated to be £1325/£2486, £6312, £6268 and £6247, respectively. Thus, the addition of rituximab to CVP, CHOP, MCP and CHVPi compared with chemotherapy alone resulted in an ICER of £1529/£5611, £5758, £4861 and £9251 per QALY gained, respectively (Table 32).

One-way SAs showed that the results were robust to parameter changes with none of the SAs increasing the ICER above £20,000 per QALY gained. PSAs were conducted. The PSA results indicated that the addition of rituximab to chemotherapy compared with chemotherapy alone was highly cost-effective assuming a willingness-to-pay (WTP) threshold of £20,000 per QALY gained. No incremental analysis was presented to compare across chemotherapy regimens.

Review of previous analyses and quality-of-life data

No economic review or quality-of-life reviews were included in the MS. 62

Sources of effectiveness for CHOP, MCP and CHVPi with or without rituximab

The trials used91–94 were likely to overestimate the effect of rituximab given that responders to first-line induction treatment received subsequent treatments with interferon maintenance or SCT (see Chapter 3, Summary of trials). This issue was not discussed in the MS. 62

Method used to estimate the rate of progression in the absence of patient-level data

Owing to the lack of individual patient-level data, the manufacturer assumed that the TTP was represented by either a Weibull or an exponential distribution, with these distributions estimated using ordinary least squares regression method. This approach is commonly used in health economic models when only data from manuscripts are available. However, it appears that there a number of errors and inconsistencies in the process used by the manufacturer to estimate the exponential distribution. By definition, the exponential distribution is composed of only one parameter (λ), as the rate is constant and does not vary with time. However, the manufacturer fitted a linear regression model (y = α × t + λ) to the transformed data (log scale) that contained two parameters: λ (constant) and α (variable time dependent). In some parts of the economic model, the rate of progression was calculated using λ only or the sum of λ and α. The inconsistency in the approach used limits the interpretation of the estimated coefficients used throughout the economic model. Furthermore, this approach is not correct, as this sometimes includes or excludes a time-dependent variable. The linear model has to be of the following form in order to estimate the parameter of the exponential distribution: y = α × t.

Inconsistencies and errors were identified between the risk of progression presented in the report and the risk of progression used in the economic model, notably in second line. In most cases, the fitted exponential distribution (using an ordinary least squares methodology) was not found to provide a reasonable fit to the data. Therefore, it appears that the manufacturer adjusted the parameters of the exponential distribution ‘manually’ by adding extra parameters in order to provide a reasonable visual fit to the data. This was not discussed by the manufacturer in the report and was identified by the AG only after review of the economic model. In some instances, the unadjusted coefficients were used (instead of the coefficient artificially adjusted to fit the data) in the economic model. For example, considering the PFS for patients treated with R-CHOP as induction in second line and maintenance with rituximab (Figure 7). The curve presented by the manufacturer in the report (grey line) was estimated after the addition of extra parameters (manual adjustment). However, in the economic model the dashed grey curve (before adjustment) was used (estimated by the AG), which provided a poorer fit.

FIGURE 7.

Kaplan–Meier plot and exponential distributions presented in the MS62 and used in the economic model for patients that respond to R-CHOP second-line induction treatment and receive maintenance rituximab. KM, Kaplan–Meier.

Approach used to estimate the hazard ratio

The HRs were calculated by taking the cumulative hazard (estimated by the sum of the negative log of the survival) from the PFS Kaplan–Meier curve estimated from the appropriate trials. The AG acknowledges that the approach was necessary in the absence of individual patient-level data. However, the AG note that such an approach might introduce bias, as the calculated cumulative hazard is dependent on the number of point estimates considered. A better approach would be to estimate the HR from the baseline parametric survival curve.

Duration of benefits

The manufacturer assumed that the treatment effect (HR) lasts 53 months based on the median follow-up duration in the M39021 trial. 95,96 However, the follow-up was different in other trials used. 91–94

Rule for patients previously treated with rituximab

The manufacturer wished to apply a rule whereby patients that relapsed within 6–12 months after first-line induction treatment with rituximab would not be eligible for rituximab in second-line treatment if previously exposed to rituximab. However, the decision of the manufacturer to simplify the economic model structure meant that several assumptions had to be made as the model was not able to track patients over time.

Treatment pathway

The manufacturer assumed that patients can receive only CHOP or R-CHOP in second-line treatment (followed, or not, by maintenance rituximab). Discussion with clinical experts indicated that CHOP-containing regimens are aggressive and therefore mainly used in younger patients. Older patients are likely to receive less aggressive chemotherapy regimens, such as FC with or without rituximab. Furthermore, clinical experts indicated that anthracycline-containing regimens (CHOP, MCP, CHVP) should only be used once in a lifetime and therefore patients previously treated with anthracycline regimens are likely to receive SCT in second line if they are fit enough or less aggressive chemotherapies (FC) if they are not considered to be sufficiently fit.

Source of effectiveness for patients treated in second line

The manufacturer used data from the EORTC 20981 trial74,75 to estimate the risk of progression for patients treated with CHOP or R-CHOP in second line with or without maintenance. However, patients in this study were rituximab naive, i.e. not previously treated with rituximab. The applicability of outcomes from this study to patients previously treated with rituximab is unclear. Furthermore, because data from second randomisation (i.e. after response induction) were used, the time spent in second-line induction (where the risk is zero for responders) was missing from calculations of PFS and OS. Furthermore, outcomes for non-responders were missed.

Estimation of post-progression survival

The manufacturer estimated PPS as a function of PFS and OS from the EORTC 20981 trial. 74,75 The AG, however, has some concerns about the approach used by the manufacturer. The manufacturer calculated PPS as the additive risk of OS and PFS (using the coefficients of the exponential distribution) so that PPS = OS + PFS. It is unclear why the addition of the coefficient of PFS and OS would be equal to the coefficient of PPS. Furthermore, the manufacturer used direct coefficients of the exponential distribution to estimate PPS before their ‘manual adjustment’ (curves are artificially modified to fit the data). This means that the curves for OS and PFS used to calculate PPS no longer fitted the data (Figures 8 and 9). Finally, the manufacturer used the combined data for patients randomised to observation or maintenance, therefore implying that the PPS would be the same following CHOP or R-CHOP induction.

FIGURE 8.

Kaplan–Meier plot and exponential distribution used to calculate PPS reported in the MS62 and actually used in the economic model for patients receiving observation after response to second-line induction treatment. Exp, exponential distribution; KM, Kaplan–Meier.

FIGURE 9.

Kaplan–Meier plot and exponential distribution used to calculate PPS reported in the MS62 and actually used in the economic model for patients receiving maintenance rituximab after response to second-line induction treatment. Exp, exponential distribution; KM, Kaplan–Meier. Note: The curve for OS used in the model is superimposed on the curve for OS presented in the report as this was derived correctly by the MS. 62

Model structure

Although Markov models are commonly used for oncology treatments, the Markov approach requires assumptions and can be inflexible. The manufacturer used exponential distributions to ‘avoid over-complicating’ the model. However, in most cases, the exponential distribution did not fit the data well.

Adverse events

The MS62 did not include the impact of AEs either in terms of costs or impact on quality of life, stating that there is no clinically significant difference between the rates and/or severity of AEs observed in the rituximab arms of each of the four first-line clinical trials91–96 when compared with the respective comparator arms. However, the clinical effectiveness review indicated that a greater number of blood and bone marrow AEs occurred in the R-chemotherapy arm than in the chemotherapy-alone arm, for example neutropenia, leucocytopenia. Despite these AEs not resulting in a difference in infection rates and thus being clinically significant, they would still incur costs to treat and their exclusion might bias the cost-effectiveness in favour of rituximab.

Treatment/management costs

Several errors/inconsistencies were identified by the AG after review of the economic model. First, the planned number of cycles in the EORTC 20981 trial74,75 (used to represent second-line treatment) is six cycles of CHOP or R-CHOP. Assuming a cost per cycle of £1462 for R-CHOP (estimated by the manufacturer), the maximum cost that a patient can incur is £8772 (£1462 × 6). In the economic model, the cost for patients treated with R-CHOP (accounting for the fact that some patients receive less than the planned dose owing to progression) was estimated to be £11,305 by the MS. 62 This is because of an error in the translation between month and cycle. The same error was found for the calculation of the cost of administration in second line.

The MS62 also used a complicated formula to estimate the cost associated with maintenance based on the area under the curve from the most up-to-date EORTC 20981 trial data. 74 The cost was then applied to the first cycle in the economic model. The AG had some difficulty in following the logic; however, we believe that costs were discounted twice.

Inconsistencies were also identified in the approach used to estimate the management costs in the ‘progressive’ HS. The manufacturer calculated a cost per month including the cost associated with the post-protocol treatment from the EORTC 20981 trial74,75 and the cost of palliative care. 125 This had the effect of inflating the cost for patients who spend a longer time in the ‘progressive’ HS and bias the cost-effectiveness in favour of rituximab.

The manufacturer also assumed no drug wastage. This might not be true if chemotherapies are not given in a large centre and vial sharing is not possible.

Utilities

The economic model included in the MS62 used utility values from the Oxford Outcomes Study. 118,119 The manufacturer assumed that the utility in PFS1 was similar to the utility of patients considered to be disease free (0.88, 95% CI 0.81 to 0.95). The utility for patients in remission/full response to therapy (0.79, 95% CI 0.72 to 0.86) was used to represent the utility for patients in PFS2. Finally, the utility for progressive disease was assumed to be 0.62 (95% CI 0.48 to 0.76). As suggested by the ERG in the ongoing appraisal for first-line maintenance,126,127 it seems inappropriate to assume that patients in PFS1 and PFS2 have different utility values given that these patients are in remission. This choice by the manufacturer to use the utility for patients considered to be ‘disease free’ to represent the utility in patients in PFS1 also appears to be inappropriate as these patients are in a ‘remission’ state and not ‘disease free’.

Other assumptions

The MS62 assumed that there were no resistance effects among patients previously treated with rituximab implying that the efficacy would be equal regardless of previous treatment. The MS62 referred to two studies to support the assumption of the absence of a resistance effect to rituximab. 128,129 However, the AG does not believe that the data from these two studies provide conclusive evidence that the resistance of rituximab is not a consideration. Further studies identified by the AG in other types of lymphoma130–132 suggest that there might be a resistance effect to rituximab.

Introduction

The review of published economic evaluations111–113,115 were used. The main limitations identified were the description of the treatment pathway, the sources of effectiveness and assumptions that were made.

Previous guidance by NICE (TA110) was issued for the use of rituximab in combination with CVP for the first-line induction treatment of FL. 83 Since this guidance was produced, the licence of rituximab was extended for use in combination with any chemotherapy-containing regimen. 85 In 2008, NICE issued guidance recommending the use of rituximab in combination with chemotherapy in second-line induction treatment and for rituximab monotherapy as maintenance treatment in patients responding to second-line induction chemotherapy with or without rituximab. 72 At the time of writing, NICE is currently considering the use of rituximab monotherapy for first-line maintenance treatment of patients responding to first-line induction treatment with rituximab in addition to chemotherapy. 127 The final guidance is expected to be issued after delivery of this assessment report. A summary of previous guidance issued by NICE is presented in Chapter 1 (see Current service provision and Table 4).

This section describes the development of a de novo economic model addressing the main limitations identified in existing economic evaluations. 62,111–113,115 The key objective of the economic assessment is to address the cost-effectiveness of the addition of rituximab to chemotherapy in previously untreated, stage III/IV, patients with FL in England and Wales in line with changes in the licensing of rituximab85 and previous guidance issued by NICE. 73,82

Population appraised

The population under assessment is previously untreated, symptomatic, stage III–IV patients with FL in England and Wales.

Interventions/comparators

A probabilistic decision analytic model was developed to estimate the costs and QALYs of the addition of rituximab to three chemotherapy regimens: CVP, CHOP and MCP. The choice of chemotherapies was primarily based on available data91–96 and the robustness of the evidence in order to address the NICE scope defined for this appraisal. 133

No comparison was provided for the addition of rituximab to a CHVP regimen with interferon. This was because of issues in the design of the FL2000 trial,94 which compares the addition of rituximab with a CHVP regimen with interferon. There were differences in the interventions in the FL2000 trial. 94 The control/comparator group received 12 courses of a CHVP regimen administered every 28 days for six courses and then every 56 days for an additional six courses combined with 18 months of interferon, whereas the active treatment group received only six courses of a CHVP regimen administered every 28 days in addition to rituximab, with interferon delivered for 18 months. Clinical opinion suggests that CHVP regimens are very rarely used in the UK, and that interferon might not be used because of toxicity.

Description of the de novo economic model

The main source of effectiveness data were obtained from the three main trials conducted in first-line induction treatment which compared CVP against R-CVP,95,96 CHOP against R-CHOP91,92 and MCP against R-MCP. 93

The economic model was programmed using R software^® (The R Foundation for Statistical Computing, Vienna, Austria), version 2.11.1, and uses a 25-year time horizon in the base case to capture costs and benefits as in the MS. 62 Shorter horizons (5 years, 10 years) and a lifetime horizon are presented in SAs. In accordance with the NICE guide for the methods of technology appraisal,97 the economic model adopts the perspective of the UK NHS and PSS with costs and benefits discounted at an annual rate of 3.5%.

Treatment pathway and clinical practice in the UK

The modelled treatment pathway incorporates guidance issued by NICE73,83 for the treatment of patients with FL in England and Wales and tries to replicate the treatment pathway observed in clinical practice. Owing to the possibility that first-line maintenance rituximab could be recommended by NICE, an alternative scenario including this option has been included.

Clinical opinion was sought and two clinicians completed a short questionnaire via a telephone interview. A summary of the answers is presented in Table 33. Overall, clinical opinion suggested that:

• In clinical practice, patients relapsing within 6–12 months after rituximab in combination with chemotherapy are not likely to be retreated with rituximab as recommended by the ESMO guideline. 19 An exception was for patients previously treated with R-CVP.

• Anthracycline-containing regimens (CHOP, MCP) can be given only once in a lifetime. Thus, in second-line treatment, patients previously treated with an anthracycline-containing regimen will be considered for alternative treatments with salvage therapy [high-dose chemotherapy (HDT)] with or without rituximab in addition to ASCT, if aged < 65 years and are fit enough. Older or unfit patients are likely to receive less aggressive chemotherapies with or without rituximab, such as FC (note: this reflects the view of the clinical experts consulted by the AG). Rituximab may not to be given in second line as part of the salvage treatment for those patients previously treated with rituximab that relapse within 6–12 months after first-line induction treatment.

• Patients who are not in complete or partial remission at the end of first-line induction treatment (i.e. stable disease) with chemotherapy with or without rituximab are likely to be offered second-line treatment despite the absence of progression.

Clinicians were only asked to define the treatment pathway in patients treated with CVP- or CHOP-containing regimens with or without rituximab. The pathway for MCP and R-MCP were assumed to be identical to CHOP and R-CHOP on the rationale that both were anthracycline regimens. The AG stresses that the treatment pathway defined in Table 33 is a simplification of treatment options given in second line and acknowledges that the treatment decisions taken includes other parameters, such as the presence of comorbidities and patient’s preferences.

The treatment pathways used in the economic model are presented in Figures 10–13, based on our discussions with clinical experts (see Table 33) and previous guidance issued by NICE. 72,82 Within the model, an age cut-off of 65 years was selected to classify eligibility for treatment in second line; however, the AG acknowledges that in clinical practice, patient age would not be the sole criteria as older patients who were fit enough may be eligible for SCT. Non-responders that did not progress at the end of treatment induction, were assumed to receive second-line treatment at the end of first-line induction treatment. Furthermore, we considered early relapse as relapse within 12 months after the start of treatment.

FIGURE 10.

Treatment pathways modelled in the economic model for CVP.

FIGURE 11.

Treatment pathways modelled in the economic model for R-CVP.

FIGURE 12.

Treatment pathways modelled in the economic model for CHOP/MCP.

FIGURE 13.

Treatment pathways modelled in the economic model for R-CHOP/R-MCP.

For the scenario analysis, clinical opinion was sought to determine the treatment pathway after first-line maintenance treatment in patients treated in first-line induction with rituximab. This is largely unknown, as first-line maintenance is not currently part of clinical practice. After discussion with clinical experts, the treatment pathway presented in Table 34 was used in the economic model for responders to first-line induction with rituximab in addition to chemotherapy.

Note that the choice of second-line treatment for patients treated with chemotherapies only (i.e. without rituximab) in first-line induction was not amended (see Figures 10–13), as first-line maintenance is only considered an option by NICE in the ongoing appraisal for patients treated with rituximab in addition to chemotherapy in first-line induction therapy.

In addition to the base case, a range of SAs were conducted exploring the impact of the treatment pathway. As described later (see Effectiveness in patients treated with CHOP with or without rituximab in second line), there is a gap between evidence available and the treatment in clinical practice. No robust evidence were available for the effectiveness of FC-containing regimens with or without rituximab in patients aged ≥ 65 years at the time of relapse after first-line induction treatment. There were also no trials identified providing a direct comparison of ASCT in addition to salvage therapy with HDT with or without rituximab in patients with relapsed FL. Finally, the identified studies in patients with relapsed FL74,75,134 were conducted in cohorts of patients with FL that were not previously treated with rituximab (see Resistance to rituximab in patients previously exposed to rituximab treatment).

The following assumptions were explored in SAs:

• Patients previously treated with R-CVP not being retreated with rituximab if relapsing < 12 months after the start of treatment (in the base case, those patients receive rituximab despite early relapse).

• Patients previously treated with an anthracycline-containing regimen and aged < 65 years old receiving CHOP or R-CHOP in second line (in the base case, those patients receive salvage therapy with or without rituximab ± ASCT).

• Patients aged > 65 years old receiving a CHOP-containing regimen (CHOP or R-CHOP) (in the base case, those patients receiving FC or R-FC).

• Patients receiving second-line treatment after progression only (in the base case, patients with stable disease at the end of treatment induction are considered to be non-responders and undergo further line of treatment).

An additional scenario is also presented assuming that patients responding to first-line induction treatment with rituximab in combination with chemotherapy receive first-line maintenance rituximab for up to 2 years. This scenario is presented to explore the potential impact of the addition of first-line maintenance into the treatment pathway if NICE issue positive guidance. No final guidance was issued by NICE at the time of writing of this report. 127

Definition of progression

In the economic model, the need for further treatments is driven by the presence of progression, i.e. that patient receive second-line treatment only after relapse/progression. However, trials use different definitions for the TTP (see Chapter 3, Summary of trials and Appendix 12). A comparison of TTF (that includes next antilymphoma treatment and stable disease at cycle 4 as event), TTP and TTNT curves from the M39021 trial95,96 suggests that some patients might have received further/second-line treatments before progression.

In the economic model, we used TTP from the M39021 trial,95,96 as patient-level data were available (data provided by Roche, 15 February 2011, personal communication) for this outcome. PFS or EFS have been used in second line according to the data available. The AG acknowledges the potential differences between the outcomes, and refers to progression outcomes as PFS for simplicity and consistency.

Structure of the economic model

The structure of the economic model developed by the AG is similar to the model included in the MS62 in terms of HSs, with patients moving between four possible HSs: PFS1 (first-line induction treatment/progression-free), PFS2 (second line/progression free), progressive disease (including subsequent lines of chemotherapy), and death.

Health states were selected to represent the natural history in patients with patients with FL to incorporate previous NICE guidance. 72 The AG acknowledges that patients are likely to receive more than two lines of therapy in clinical practice; however, there is no robust evidence available that would allow the effectiveness after second-line treatment to be modelled with accuracy.

The economic model developed by the AG for this appraisal differs from the economic model included in the MS62 in the following manner:

• use of a continuous time method over a traditional Markov process

• treatment pathways reflecting more accurately clinical practice in England and Wales (see Figures 10–13)

• responders and non-responders are modelled as two separate subgroups

• use of a different source of evidence to model the effectiveness in patients treated with CHOP or MCP with or without rituximab (see Progression-free survival in patients treated with CHOP in first-line induction with or without rituximab).

The economic model treats responders and non-responders as two separate subgroups and therefore does not use the PFS curve calculated for the whole trial population. This choice has been made after reviewing the evidence available in first-line induction for patients treated with CHOP or MCP with or without rituximab (see Effectiveness in patients treated with MCP with or without rituximab in first line). This choice of model structure allows the implementation of maintenance after first- or second-line induction treatment. 73,127

A simplified schematic of the model structure is provided in Figure 14. A cohort of 100,000 individual patients were simulated, each with individual demographic characteristics (age, gender and BSA). The age at death owing to non-FL causes was sampled from a Gompertz distribution estimated from life tables in the UK,135 conditional on the patient being alive at the start of the simulation. In PFS1, patients received CVP, CHOP or MCP with or without rituximab. Patients remaining in PFS1 at the end of the induction treatment were assumed to be monitored but to not receive any further treatments. For each of the therapies examined, the response rates from the applicable trials91–93,95,96 were used to classify patients into responders and non-responders.

FIGURE 14.

Economic model structure.

The TTP is then sampled according to the PFS curves for responders and non-responders as appropriate, with non-responders having a faster disease progression (see Figure 27). If the estimated time of progression is later than the estimated time to non-FL death, patients are assumed to die before progression. For patients progressing before the age at death from all causes, the event (relapse or death) is determined based on the proportion of progression attributable to death (see Death in progression-free survival after first line). Patients do not continue in the simulation if progression is attributable to death. Patients dying incur no further costs and accrue no further QALYs. Patients relapsing move to second-line treatment.

Patients treated in second line are classified as either responders or non-responders. Responders to CHOP, R-CHOP, FC or R-FC receive maintenance rituximab for up to 2 years at the end of the induction phase as per NICE guidance. 73 Patients responding to HDT or R-HDT receive ASCT. Patients remaining in PFS2 at the end of treatment induction, maintenance or ASCT are assumed not to receive further treatment but would be monitored. The TTP is sampled and patients who progress before the age at death from all causes receive further lines of therapies (third/subsequent line). The time to death from the receipt of second-line treatment is also calculated to identify the cause of death (FL or all causes). Patients dying incur no further costs and accrue no further QALYs. Patients relapsing move to progressive disease. Those patients are assumed to incur additional costs associated with palliative and terminal care as appropriate.

Patient characteristics

A patient’s baseline characteristics were derived from registry data in England3,135 and Wales (data provided by the Welsh Cancer Intelligence & Surveillance Unit, 200816) and the demographics from the trials conducted in patients with FL.

Baseline age

The baseline age is derived from registry data in England3 and Wales (data provided by the Welsh Cancer Intelligence & Surveillance Unit, 200816) using a two-stage process. For consistency, a 5-year age band was assumed for patients aged ≥ 85 years (Figure 15). We then estimated the age within each age band, assuming a uniform distribution (i.e. equal probability). First, the age band of the patients was sampled. Then the precise age was estimated assuming a uniform distribution within the age band.

FIGURE 15.

Age distribution of patients diagnosed with FL in England3 and Wales. (Data provided by the Welsh Cancer Intelligence & Surveillance Unit, 2008. 16)

Response rate after first-line induction treatment

In the economic model, patients are separated into responders and non-responders according to the response rates after first- or second-line induction treatment. The response rates in first-line induction treatment were extracted from the proportion of responders observed in the three main first-line remission induction trials (Table 36). 91–93,95,96

Owing to absence of relevant data for PFS by response category, no distinction was made between partial and CRs.

Progression-free survival in patients treated with CVP in first-line induction with or without rituximab

Progression-free survival in responders to CVP and R-CVP

Individual patient-level data from the M39021 trial95,96 have been provided by the manufacturer after a request from the AG (Roche, personal communication). The manufacturer provided the Kaplan–Meier plots from first randomisation (i.e. from start of treatment) and consequently, the Kaplan–Meier curve is flat for responders for the first 6 months corresponding to the initial period of induction treatment. Because of this, it was not appropriate to fit a distribution the entire Kaplan–Meier curve. Consequently, in the economic model, we assumed no progression for responders during treatment induction (196 days for eight cycles of 21 days + 28 days), with a distribution fitted from the end of this period.

To preserve the correlation between treatments in the PSA, the AG fitted a parametric distribution to all responders using treatment as a covariate. This was shown to provide an adequate fit to the data (Figures 16–18). The parametric distribution was selected through an iterative process after evaluating goodness-of-fit criteria, the visual plot of the curve to the observed data, the plausibility of the extrapolation at the end of clinical evidence and the plot of the hazard.

FIGURE 16.

Plot of the observed Kaplan–Meier data and predicted distributions for patients treated with CVP. 95,96 Analysis of individual patient-level data provided by the manufacturer (Roche, personal communication). KM, Kaplan–Meier.

FIGURE 17.

Plot of the observed Kaplan–Meier data and predicted distributions for patients treated with rituximab in addition to CVP. 95,96 Analysis of individual patient-level data provided by the manufacturer (Roche, personal communication). KM, Kaplan–Meier.

FIGURE 18.

Plot of hazard for responders treated with CVP with or without rituximab. 95,96 Analysis of individual patient-level data provided by the manufacturer (Roche, personal communication).

Different parametric models incorporate different hazard functions. Exponential models are only suitable if the observed hazard is approximately constant and positive. Weibull and Gompertz models incorporate monotonic hazards, whereas the logged models (log-logistic, log-normal) can incorporate non-monotonic hazards, but typically have long tails owing to a reducing hazard as time increases beyond a certain point. 136

The Akaike information criterion (AIC) and Bayesian information criterion (BIC) were calculated, and suggest that the log-normal model provides the best fit to the data (Table 37). Broadly similar AIC and BIC values were observed for the log-logistic and Gompertz distribution. However, goodness-of-fit criteria provide an indication of the goodness of fit to only the observed period and do not categorically indicate that one distribution should be preferred to the remaining distributions. The observed Kaplan–Meier data were plotted against the five fitted parametric distributions (exponential, Weibull, Gompertz, log-logistic and log-normal). The Gompertz, log-logistic and log-normal distributions provided a plausible fit to the observed data (see Figures 16 and 17). Similarly, visual inspection of the plot of the hazard (see Figure 18) suggests that the log-normal, log-logistic and Gompertz distributions were suitable, as the plot was broadly linear.

TABLE 37 - Goodness-of-fit criteria for the risk of progression among responders to CVP first-line induction with or without distribution95,96

df, degrees of freedom.

Analysis of individual patient-level data provided by the manufacturer (Roche, personal communication).

Model	Observed	ll(null)	ll(model)	df	AIC	BIC
Exponential	221	–333.225	–316.846	2	637.692	644.488
Weibull	221	–330.528	–315.636	3	637.271	647.466
Gompertz	221	–322.177	–309.133	3	624.266	634.460
Log-logistic	221	–323.495	–307.567	3	621.134	631.328
Log-normal	221	–320.175	–304.582	3	615.165	625.359

As single parametric distributions provided reasonable and plausible fit to the data, the AG did not considered other methodologies, such as the use of piecewise exponentials.

From the values of the AIC/BIC, the visual inspections of the fit to the observed period and hazards, the AG believes that the Gompertz, log-logistic and log-normal distributions provided a reasonable and plausible fit to the data. However, the AG believed that the log-normal distribution provided a more plausible long-term extrapolation compared with the Gompertz distribution (see Figure 44). The risk of progression using the Gompertz distribution flatten out after about 60 months, implying that about 40% of responders would never progress. FL is considered as an incurable disease and therefore the use of the Gompertz distribution may be implausible. In the base case, the log-normal distribution was selected by the AG as this was believed to be the most plausible parametric extrapolation. The Weibull and Gompertz distributions have been used in SA as these provided a different extrapolation. The AG did not test the log-logistic as the curve was very similar to the log-normal distribution. The log-normal regression model and variance–covariance matrix are presented in Table 38.

TABLE 38 - Log-normal regression model for responders to CVP-containing regimen with or without rituximab95,96

Analysis of individual patient-level data provided by the manufacturer (Roche, personal communication).

No. of subjects = 221			No. of observations = 221
No. of failures = 136
Time at risk = 5913.331
			LR χ2(1) = 31.18
Log-likelihood = –304.582			Prob > χ2 = 0
_t	Coef.	SE	z	p > z	95% CI
trt	1.16341	0.204687	5.68	0	0.76223 to 1.56459
_cons	2.591318	0.152575	16.98	0	2.292276 to 2.89036
/ln_sig	0.335348	0.065793	5.1	0	0.206395 to 0.464301
sigma	1.398427	0.092007			1.229239 to 1.590901
Variance–covariance matrix
	_t:	_t:	ln_sig:
	trt	_cons	_cons
_t:trt	0.041897
_t:_cons	–0.02258	0.023279
ln_sig:_cons	0.001846	0.001042	0.004329

Progression-free survival in non-responders to CVP and R-CVP

A similar process to that detailed for responders to CVP and R-CVP has been used to estimate the risk of progression among non-responders to CVP and R-CVP; however, Kaplan–Meier data from the start of treatment induction was used95,96 (data provided by Roche, personal communication). The goodness-of-fit criteria (Table 39), visual plot of the Kaplan–Meier to the observed period (Figures 19 and 20) and the plot of the hazard (Figure 21) indicate that the Gompertz, log-logistic and log-normal distributions again provide a plausible fit to the data. In the base-case analysis, the log-normal distribution was selected (Table 40), with other distributions tested in SA.

TABLE 39 - Goodness-of-fit criteria for the risk of progression among non-responders to CVP first-line induction with or without distribution95,96

df, degrees of freedom.

Analysis of individual patient-level data provided by the manufacturer (Roche, personal communication).

Model	Observations	ll(null)	ll(model)	df	AIC	BIC
Exponential	93	–158.965	–158.490	2	320.979	326.044
Weibull	93	–156.174	–155.819	3	317.637	325.235
Gompertz	93	–149.751	–149.509	3	305.019	312.616
Log-logistic	93	–147.759	–147.431	3	300.863	308.461
Log-normal	93	–147.316	–147.070	3	300.139	307.737

FIGURE 19.

Plot of the observed Kaplan–Meier data and predicted distributions for non-responders treated with CVP. 95,96 Analysis of individual patient-level data provided by the manufacturer (Roche, personal communication). KM, Kaplan–Meier.

FIGURE 20.

Plot of the observed Kaplan–Meier data and predicted distributions for non-responders treated with rituximab in addition to CVP. 95,96 Analysis of individual patient-level data provided by the manufacturer (Roche, personal communication). KM, Kaplan–Meier.

FIGURE 21.

Plot of hazard for non-responders treated with CVP with or without rituximab. 95,96 Analysis of individual patient-level data provided by the manufacturer (Roche, personal communication).

TABLE 40 - Log-normal regression model for non-responders to CVP-containing regimen with or without rituximab95,96

Analysis of individual patient-level data provided by the manufacturer (Roche, personal communication).

No. of subjects = 93			No. of observations = 93
No. of failures = 80
Time at risk = 1425.807
			LR χ2(1) = 0.49
Log-likelihood = –147.07			Prob > χ2 = 0.4823
_t	Coef.	SE	z	p > z	95% CI
trt	–0.20573	0.292553	–0.7	0.482	–0.77912 to 0.367666
_cons	2.273996	0.16551	13.74	0	1.949602 to 2.59839
/ln_sig	0.254605	0.081224	3.13	0.002	0.095409 to 0.413802
sigma	1.289952	0.104776			1.100108 to 1.512558
Variance–covariance matrix
	_t:	_t:	ln_sig:
	trt	_cons	_cons
_t:trt	0.085587
_t:_cons	–0.02731	0.027394
ln_sig:_cons	–0.00035	0.000945	0.006597

Progression-free survival in patients treated with CHOP in first-line induction with or without rituximab

Progression-free survival in responders to CHOP and R-CHOP

In the GLSG 2000 trial,91,92 patients of < 60 years of age achieving CR or PR following first-line induction treatment were randomised to either SCT or maintenance with interferon. Patients aged ≥ 60 years received maintenance with interferon. Consequently, the reported effectiveness in responders is confounded by the effect of maintenance interferon or SCT.

The AG believes that data from the GLSG 2000 trial91,92 would lead to an overestimate of the absolute gain of the addition of rituximab to CHOP because of the additional treatments provided to responders. Alternative sources of effectiveness have therefore been considered to model the risk of progression among responders to CHOP first-line induction with or without rituximab.

The PRIMA study71 provides data on the progression rate of patients responding to first-line induction with chemotherapy in combination with rituximab only (R-CVP, R-CHOP, R-FCM). Patients were randomised to maintenance rituximab or observation up to 2 years from the end of first-line treatment induction (R-CHOP, R-CVP or R-FCM). The majority of patients (90%) had stage IV FL and most of patients received R-CHOP as first-line induction treatment (74%).

Individual patient-level data from the PRIMA study71 were made available to the AG by the manufacturer (Roche, personal communication). The Kaplan–Meier curves for the responders randomised to observation for R-CHOP and R-CVP from the end of treatment induction have been compared (Figure 22). Although apparently visually different, the difference between the two curves was not statistically significant (p = 0.0970). However, the AG acknowledges that the absence of statistical differences might be attributable to small sample sizes (R-CVP, n = 113; R-CHOP, n = 386) and that this does not necessarily means that the two curves are similar. 71

FIGURE 22.

Comparison of the Kaplan–Meier data for responders to R-CHOP and R-CVP in the PRIMA study. 71 Analysis of individual patient-level data provided by the manufacturer (Roche, personal communication).

No robust sources of effectiveness were identified for the risk of progression for patients treated with CHOP first-line induction without rituximab. Most of the studies identified have been conducted in populations with other lymphomas, used a different study designs (retrospective) or were confounded by subsequent therapies for patients in remission. 137,138 Clinical opinion was sought about the mechanism of action of rituximab. This suggested that the addition of rituximab might provide the same relative benefit compared with chemotherapy alone, irrespective of the choice of chemotherapy.

Although patient-level data from the PRIMA study71 (data provided by Roche, personal communication) could have been used, the AG was not comfortable to use direct data from the trial owing to the high degree of censoring, which was noted by the ERG in the ongoing appraisal on rituximab for first-line maintenance treatment. 126 Furthermore, if a parametric function is fitted to patient-level data from the PRIMA study,71 the curve between R-CHOP and R-CVP curves would cross, as the curve for R-CVP becomes relatively flat after about 50 months. It is unclear if this is only an artefact of the limitation in the data used. 71,95,96

Given the limited evidence available on the progression for patients treated with CHOP and R-CHOP in first-line induction, the absence of a statistically significant difference for the risk of progression among responders to first-line induction with R-CVP and R-CHOP (p = 0.0970) and the suggestion by clinicians of a similar mechanism of action of rituximab for the different type of chemotherapies assessed, the AG used patient-level PFS data from the M39021 trial95,96 (data provided by Roche, personal communication) as a proxy of the PFS for patients responding to CHOP and R-CHOP, respectively.

The assumptions made were supported by additional analyses comparing the risk of progression among responders to R-CVP from the PRIMA study71 (data provided by Roche, personal communication) and responders to R-CVP from the M39021 trial95,96 (data provided by Roche, personal communication). Overall, the PFS from end of treatment induction was found to be broadly similar between the two trials (Figure 23).

FIGURE 23.

Plot of the risk of progression among patients responding to first-line induction treatment with R-CVP from the PRIMA71 study and M3902195,96 trial. Analysis of individual patient-level data provided by the manufacturer (Roche, personal communication).

Comparison of the modelled R-CHOP by the Assessment Group against data from an alternative randomised controlled trial77

Rummel et al. 78 report data from a Phase III trial comparing R-CHOP with R-bendamustine in patients treated for FL, Waldenström’s, marginal zone lymphoma, small lymphocytic lymphoma and MCL. Fifty-four per cent of patients had FL and patients treated with R-CHOP received a maximum of six cycles. The median age was 63 years and 77% of patients had stage IV disease. Thirty-three per cent and 48% of patients randomised had a FLIPI score of 2 or 3/> 3, respectively. The median observation time was 36 months. The response rate for all patients randomised to R-CHOP was 91.3% (all lymphoma types) and the median overall PFS (from randomisation) was 46.7 months in patients with FL who are treated with R-CHOP in first-line induction (which included all patients with FL). Although patients’ characteristics for patients with FL are not presented separately, patients’ characteristics for the whole trial population randomised to R-CHOP78 are broadly similar to the characteristics of the population included in other first-line induction trials for FL. 91–93,95,96

The PFS for patients with FL from Rummel et al. 78 was compared with our estimated combined PFS (responders and non-responders) for patients treated with R-CHOP assuming a response rate of 91.3% and that patients receive up to six cycles of treatment in the induction phase. Overall, the PFS predicted by the AG for R-CHOP is broadly similar to the PFS reported in Rummel et al. 78 (Figure 24).

FIGURE 24.

Comparison of the PFS from Rummel et al. 78 and predicted using the AG method.

Effectiveness in patients treated with MCP with or without rituximab in first line

As with CHOP-containing regimens, data from the first-line trial for R-MCP and MCP93 are confounded by responders receiving subsequent maintenance therapy with interferon-alpha. No robust alternative sources were identified by the AG.

To provide an estimation of the cost-effectiveness of rituximab in addition to MCP, a scenario analysis is presented, assuming that the PFS for responders and non-responders treated with MCP with or without rituximab are identical to the PFS in patients treated with CVP/CHOP with or without rituximab.

It is commented that although the PFS for responders and non-responders are assumed equal for R-CVP, R-CHOP and R-MCP, and are assumed equal for CVP, CHOP and MCP, the differences in response rates (see Table 36), number of cycles and time between cycles (see Table 44) result in different prognoses between interventions (see Figure 25).

FIGURE 25.

Progression-free survival for all patients treated with CVP, CHOP and MCP with or without rituximab.

Summary of modelled progression-free survival in first-line induction

The modelled combined PFS (including both responders and non-responders) for patients treated with CVP, CHOP and MCP with or without rituximab is presented in Figure 25.

Effectiveness of rituximab first-line maintenance for patients that respond to first-line induction with chemotherapy in combination with rituximab (scenario analysis)

First-line maintenance was incorporated into the economic model by altering the risk of progression for patients responding to first-line induction with R-chemotherapy. The HR from the PRIMA study71 was used to alter the risk of progression (observation vs maintenance). Although there were differences in the HR for patients treated with R-CHOP (HR 0.51, 95% CI 0.39 to 0.65) and R-CVP (HR 0.68, 95% CI 0.45 to 1.02), we used data for the whole randomised population as differences might have been attributable to small sample sizes. Consequently, a HR of 0.55 (95% CI 0.44 to 0.68) was applied to the rate of progression for responders to R-chemotherapy for the first 42 months as clinical opinion suggests that the lasting effect ranges between 36 and 48 months. 127 SAs were conducted varying the lasting effect of first-line maintenance rituximab between 36 and 72 months.

Response rates in patients receiving second-line chemotherapy

The response rates for patients treated with CHOP and R-CHOP second-line induction treatment were extracted from the EORTC 20981 trial (Table 41). 74,75 The response rates were not available for FC-containing regimens used in older patients. As FC-containing regimens are less aggressive therapies, a lower effectiveness is expected. In the absence of evidence, we arbitrarily assumed that FC is 20% less effective than CHOP. SAs were conducted varying the response rates for patients treated with FC with or without rituximab.

Effectiveness in patients treated with CHOP with or without rituximab in second-line

Data from the EORTC 20981 trial74,75 were used to model the PFS in patients with FL who were treated with CHOP and R-CHOP in second-line induction, with or without rituximab maintenance. Patients were included in the EORTC 20981 trial74,75 if they had relapsed but had no more than two previous non-anthracycline-containing chemotherapy regimens. The study was conducted before the introduction of rituximab and therefore patients are rituximab naive, i.e. not previously exposed to rituximab. The initial results of the EORTC 20981trial75 were updated in a second publication74 that included 6 years of follow-up data. Patients were randomised to second-line induction treatment with either CHOP or R-CHOP; those patients who achieved a CR or PR had a second randomisation to either maintenance treatment with rituximab (once every 3 months) or observation for 2 years or until relapse.

Where possible, data from the latest follow-up duration74 were used in the economic model. The PFS and OS curves for responders to CHOP and R-CHOP second-line induction treatment were extracted from the latest follow-up of the EORTC 20981 trial. 74 However, the PFS and OS curves for non-responders to treatment induction were extracted from data presented by the manufacturer in a previous submission to NICE. 73

van Oers et al. 74 reported only OS data for all responders regardless of whether treatment induction was CHOP or R-CHOP randomised to either maintenance treatment with rituximab or observation. Data by treatment induction have been presented by the manufacturer in a previous NICE appraisal;73 however, this used a shorter follow-up duration (median = 39.4 months from first randomisation). These data indicated that the OS curves for patients randomised to observation or maintenance rituximab were broadly similar whether patients received CHOP or R-CHOP in second-line treatment induction (see figure 10 in MS62 for TA137). In the economic model, it was assumed that the OS for patients treated with CHOP or R-CHOP was the same, although patients receiving observation did less well than those who had maintenance with rituximab.

The PFS and OS for responders using the latest follow-up data from the EORTC 2098174 are presented from second randomisation, i.e. from the end of treatment induction. Consequently, the risk of PFS and OS are assumed to be zero during treatment induction in the economic model. A summary of data used in the economic model is presented in Table 42.

It was not possible to have access to individual patient-level data from the EORTC 20981 trial,74,75 and therefore only data available in the public domain were used.

The digitised Kaplan–Meier curves included in the MS62 were used to fit several parametric distributions to represent the risk of progression or the risk to death. In the absence of individual patient-level data, the distributions have been fitted using the Solver function within Microsoft Excel in order to find the parameter values that minimise the root-mean-square error (RMSE) between the observed and predicted Kaplan–Meier. The best distribution was selected using an iterative approach after analysing the visual plot of the curve, the hazard plot and the RMSE. Overall, the Weibull and exponential distributions provided the poorest fit to the data. The Gompertz and log-logistic distribution provided a reasonable fit to only part of the data. The log-normal distribution fitted all the data reasonably well.

The plot of the PFS Kaplan–Meier and predicted log-normal distribution for patients responding to second-line treatment induction with CHOP and R-CHOP are presented in Figures 26 and 27.

FIGURE 26.

Plot of the Kaplan–Meier data and log-normal distribution for patients responding to CHOP second-line induction (from the end of treatment induction) with or without maintenance rituximab. 63,74 Obs, observation; Rx, maintenance rituximab.

FIGURE 27.

Plot of the Kaplan–Meier data and log-normal distribution for patients responding to R-CHOP second-line induction (from the end of treatment induction) with or without maintenance rituximab. 63,74 Obs, observation; Rx, maintenance rituximab.

The plot of the OS Kaplan–Meier and log-normal distribution for patients responding to second-line treatment induction is presented in Figure 28.

FIGURE 28.

Plot of the OS Kaplan–Meier data and log-normal distribution for responders to second-line induction (from end of treatment induction) with or without maintenance rituximab. 63,74 Obs, observation; Rx, maintenance rituximab.

Finally, the plot of the OS and PFS Kaplan–Meier for non-responders to CHOP and R-CHOP in second-line treatment is presented in Figure 29.

FIGURE 29.

Plot of the OS and PFS Kaplan–Meier data and log-normal distribution for non-responders to second-line induction treatment (from start of induction treatment). 73

However, the distribution that provided the best fit to the data (the log-normal) hampered uncertainty analysis. In the PSA, we varied the mean PFS and OS by ± 5% by changing the mean parameter of the log-normal distribution but assuming the same standard deviation (SD). PFS and OS curves were sampled independently; however, the same random number was used to preserve the correlation between OS and PFS.

Effectiveness in patients treated with FC in combination or not with rituximab in second-line treatment

Clinical opinion sought by the AG suggested that FC or R-FC would be used for patients that cannot tolerate aggressive therapy (such as CHOP or HDT with or without ASCT), in particular older patients.

The published literature was searched for potential sources to estimate the effectiveness of FC-containing regimens with or without rituximab in patients with relapsed FL aged > 65 years; however, no data were identified. The AG was aware of a trial conducted in second-line treatment that compared fludarabine, cyclophosphamide, mitoxantrone (FCM) with R-FCM,139 with or without maintenance rituximab. The median age of patients randomised to FCM or R-FCM was approximately 60 years. This trial also had a different maintenance schedule compared with that of van Oers et al. ,74,75 which compared CHOP with R-CHOP in second-line induction. A previous NICE technology appraisal73 reported that the data were not mature in the R-FCM compared with FCM trial but, despite this limitation, the outcomes for R-FCM and R-CHOP are broadly similar.

The PFS and OS curves for responders and non-responders to CHOP and R-CHOP73–75 in second line (see Effectiveness in patients treated with CHOP with or without rituximab in second line) have been used a proxy for the risk of progression for patients treated with FC and R-FC. However, because we assumed a lower response rate for FC-containing regimen (20% lower) and the shorter induction period for FC/R-FC (four cycles instead of six), the overall modelled effectiveness for FC-containing regimens will be reduced compared with CHOP-containing regimens. SAs were conducted varying both the response rate and PFS curves (see Appendix 15).

Effectiveness in patients receiving salvage therapy with high-dose therapy with or without rituximab and autologous stem cell transplantation in relapsed patients with follicular lymphoma

Clinical advice sought by the AG indicated that patients previously treated with an anthracycline-containing regimen (CHOP, MCP) would not be retreated with an anthracycline regimen and would probably receive salvage therapy with HDT with or without rituximab before ASCT in cases for those that respond to chemotherapy.

Discussion with clinical experts suggested that the most commonly used HDT are up to four cycles of ESHAP (etoposide, methylprednisolone, cytarabine and cisplatin) or DHAP (dexamethasone, cytarabine and cisplatin) chemotherapy with or without rituximab. Stem cell harvest is then obtained for responders only, with patients for whom the harvest was successful eligible for BEAM (BCNU^®/carmustine, cytarabine, etoposide and melphalan) conditioning plus ASCT (Figure 30).

FIGURE 30.

Treatment pathway for patients treated with HDT with or without rituximab.

The literature was searched to identify studies that reported the impact of the addition of rituximab to salvage therapy before ASCT in patients with relapsed FL, although given the resource constraints it was not possible to perform a systematic search of the literature.

Sebban et al. 134 reported the impact of rituximab with or without HDT with transplant at the time of relapse in patients with FL. This retrospective study included patients that received CHVP alone or in addition to interferon in first-line induction. Relapsed patients receive salvage therapies, with the most used regimens being dexamethasone, high-dose cytarabine and cisplastin, ifosfamide, carboplatin and etoposide, mesna, mitoxantrone and etoposide, and fludarabine-based regimens. Rituximab was also offered to a proportion of patients with or without chemotherapy as part of the salvage treatment. Sebban et al. 134 reported that the 5-year EFS after first relapse (EFSR) was 52% in patients receiving rituximab as part of the salvage therapy (with or without chemotherapy) and 29% in patient receiving salvage therapy without rituximab. The 5-year survival after first relapse (SAR) rate was 81% and 44%, respectively.

Clinical opinion was sought regarding the validity of using evidence from this study134 to model the effectiveness of salvage therapy in addition to ASCT with or without rituximab. Overall, the clinical experts found the study appropriate, but cautioned that there were potential limitations in the study design. The addition of rituximab to salvage therapy is associated with considerable benefit, although it is unclear if the magnitude of the observed improvement is because of the retrospective nature of the study. 134 The study was also conducted in a pre-rituximab era, and therefore patients were not previously exposed to rituximab. Also, the proportion of patients that responded to HDT (for whom the harvest was successful) is unclear from the study, as well as the proportion of patients that received ASCT in both arms.

Despite these potential limitations, data from Sebban et al. 134 were used in the economic model to represent the effectiveness of salvage therapy with or without rituximab. Data for EFSR and SAR after salvage therapy with or without rituximab were taken from figure 3 in Sebban et al. 134 TechDig^© software was used to estimate the data points and allow parametric distributions to be fitted. We examined different distributions using the Solver function within Microsoft Excel and overall, the log-normal was found to provide the best fit to the data. The plot of the Kaplan–Meier and estimated log-normal data is presented in Figure 31 for EFS and Figure 32 for OS.

FIGURE 31.

Event-free survival for patients treated with ASCT and salvage chemotherapy with or without rituximab. 134

FIGURE 32.

Overall survival for patients treated with ASCT and salvage chemotherapy with or without rituximab. 134

The mean effectiveness was varied by ± 5% in the PSA, with the SD of the log-normal distribution assumed constant.

Resistance to rituximab in patients previously exposed to rituximab treatment

A key assumption of the economic model submitted by the manufacturer is the absence of resistance in patients previously treated with rituximab.

Evidence of resistance in patients with relapsed FL have been estimated in cohorts of patients that have not been previously exposed to rituximab, although clinical opinion expressed in a previous NICE appraisal of rituximab73 suggested that there might be little or no loss of efficacy for retreatment with rituximab, given its mechanism of action. In the MS,62 two studies are referenced to support the assumption of the absence of a resistance effect to rituximab. 128,129 However, the AG does not believe that the data from these two studies128,129 provide conclusive evidence that resistance to rituximab can be discounted.

Johnston et al. 128 report that second-line response rates were only marginally reduced in patients with FL when compared with first-line response rates (ORR 88% to 76%, CR 52% to 44% and PR 36% to 32% in first line and second line, respectively). However, a comparison between patients who had received chemotherapy alone in first and second line and patients who had received R-chemotherapy in first and second line demonstrated that PFS following the second-line treatment was no different between the two patients groups, indicating that the second rituximab treatment had little benefit. There were several problems, however, with the study undertaken by Johnston et al. 128 in terms of its ability to prove or disprove resistance to rituximab. First, the number of patients with FL (n = 50) was small and the patients were not representative of UK patients with FL (median age at start of second treatment was young: 59 years). In addition, the comparisons being made were not ideal in determining the existence of rituximab resistance: R-chemotherapy (first line) and R-chemotherapy (second line) were compared with chemotherapy alone (first line) and chemotherapy alone (second line). The correct comparison would be R-chemotherapy (first line) + R-chemotherapy (second line) compared with chemotherapy alone (first line) and R-chemotherapy (second line). A substantial number of patients were also receiving R-monotherapy, which is not recommended in the UK unless all other options have been exhausted.

Coiffier et al. 129 presented results from a small sample of patients (n = 59) who received one of the following combinations: R-monotherapy/R-monotherapy; R-chemotherapy/R-chemotherapy, R-monotherapy/R-chemotherapy; and R-chemotherapy/R-monotherapy. The findings showed that the second-line response rate and TTP did not appear to be affected by rituximab in patients who had received rituximab in first line. However, the number of patients who received R-monotherapy is unknown and the participants in the study were patients diagnosed with a B-cell lymphoma, thus the numbers of patients with FL within the study is unknown.

From a non-systematic review of the literature via web searching, the AG identified further studies conducted in other types of lymphoma suggesting that retreatment with rituximab might be associated with a loss of efficacy. 130–132

Borgerding et al. 130 reported a very low response rate in a cohort of 28 patients with DLBCL after prior exposure to rituximab. The authors reported an ORR of 32% (9 of 28 patients). Furthermore, Weide et al. 132 examined the use of bendamustine in combination with mitoxantrone and rituximab (BMR) in patients with stage III/IV relapsed or refractory indolent lymphomas and MCL with or without prior rituximab-containing chemoimmunotherapy treatment. Fifty-seven patients were recruited, 39% of whom had received prior R-chemotherapy. The median age was 66 years (range 40–83 years). Approximately 50% of patients had FL. The ORR was 89% (35% CR and 54% PR). ORR in R-chemotherapy pre-treated patients was lower at 76% (38% CR and 38% PR).

Similarly, Martin et al. 131 report a Phase III trial comparing the response rates to R-ICE (rituximab, ifosfamide, carboplatin and etoposide) and R-DHAP salvage therapy followed by HDT with ASCT (CORAL trial) for patients with relapsed or refractory DLBCL. Martin et al. 131 report that prior exposure to rituximab was associated with a significant loss of efficacy. Patients in the rituximab group had a significantly worse PFS (17% vs 57% at 3 years) and OS (38% vs 67% at 3 years) compared with patients who were not previously treated with rituximab. Prior exposure to rituximab was an independent adverse prognostic factor for both PFS (RR 2.0, 95% CI 1.2 to 3.3, p = 0.008) and OS (RR 2.2, 95% CI 1.3 to 3.9, p = 0.004). The AG acknowledges that the effectiveness between patients previously treated with rituximab and those naive to rituximab might be confounded by the level of disease aggressiveness; those relapsing early on rituximab might have a more aggressive disease.

Overall, the two studies128,129 reported by the manufacturer do not provide conclusive evidence to prove or disprove rituximab resistance. Further studies identified by the AG130–132 suggest that there might be a resistance effect to rituximab. The AG sought clinical advice on this issue, which indicated that resistance of rituximab is unknown; however, the clinicians believed that there is little or no loss of effectiveness considering its mechanism of action.

In the base case, no resistance is assumed. SAs are conducted, exploring the potential development of resistance after rituximab retreatment. SAs are conducted by increasing the rate of progression in patients receiving rituximab in second line when they had previously been treated with rituximab.

Incorporation of adverse events in the economic assessment

The economic model includes the impact of AEs in terms of management costs and impairment in quality of life. Only grade 3 and 4 AEs were included, as these are deemed of clinical and economic importance by the AG. Furthermore, only those that occurred in the first-line induction setting were included because of the lack of robust data in patients treated in second-line and subsequent lines of treatment.

After reviewing the relative frequency of AEs within patients treated with chemotherapy with and without rituximab and the likely management cost and impact on HRQoL, the AG included the following AEs in first-line induction: leucopenia, granulocytopenia, neutropenia, anaemia, alopecia, infection, cardiac arrhythmia and cardiac dysfunction.

Only grade 3/4 neutropenia and leucopenia have been included in first-line maintenance as these were the most commonly reported grade 3 or 4 AEs in the PRIMA study. 71

The management costs associated with the treatment of AEs were extracted from the costs used in a submission by the manufacturer for an ongoing NICE appraisal. 124 It is also assumed that grade 3 and 4 AEs would incur the same costs. We further assumed that each AE led to a reduction in HRQol by 15% for 45 days. It was not possible to independently estimate the management costs of AEs and the effect on HRQoL owing to resource constraints. The management costs were varied by ± 20% in SAs. The disutility was also varied in SAs.

The AG acknowledges the limitations of the inclusion of AE in the economic model, in that it is very simplistic. However, SAs presented later indicated that AE had a limited impact on the ICER (see Appendix 15). Table 43 provides a summary of AEs included in the economic model.

Drug acquisition and administration costs

The planned dose from the three main trials92,93,95,96 were used to calculate the drug acquisition cost in the absence of detailed information about dose reduction/increase for each separate arms in the trials.

The planned number of cycles were also used in the economic model. Patients treated with CHOP or R-CHOP were assumed to receive a maximum of eight cycles in first-line induction and six cycles in second-line induction. A SA was conducted assuming that patients received a maximum of six cycles of CHOP and R-CHOP in first-line induction. Patients treated with FC or R-FC were assumed to receive a maximum of four cycles in second-line induction. A SA was conducted assuming that patients treated with FC-containing regimens would receive a maximum of six cycles. The planned dose and maximum number of cycles used in the economic model are summarised in Table 44.

In the economic model, the number of cycles a patient receives is calculated from the PFS curve to account for patients that withdraw before the end of planned treatment owing to progression. Withdrawal from toxicity was not modelled; however, this was shown to be uncommon in the first-line trials. 91–93,95,96

The acquisition costs of the intervention are calculated from the protocol defined/planned dose, the BSA (see Table 35) and unit costs extracted from the BNF. 86 No vial sharing is assumed.

The costs associated with the administration of each cycle of treatment are derived from NHS Reference Costs 2009/10140 and assumptions included in the MS. 62 Chemotherapies are assumed to be administered on a day-case basis. The unit costs and Healthcare Resource Groups used are presented in Table 45. In addition to the administration costs from the NHS reference costs, patients who receive rituximab are assumed to incur additional pharmacy costs based on the costs included in the MS62 (£15.54). A SA is conducted assuming a cost of £32 as used by the manufacturer in an ongoing NICE appraisal for maintenance rituximab. 124 Pharmacy costs were included separately because the manufacturer stated that other treatment costs (i.e. chemotherapy drugs, including any pharmacy dispensing costs and associated drugs to manage the side effects of the chemotherapy) are excluded from NHS reference costs. Finally, the cost associated with transport is also included assuming that 30% of patients require NHS transportation. 62

A summary of drug acquisition and administration costs by chemotherapy cycle in first-line induction per patient is presented in Table 46, assuming a BSA of 1.80.

It is not clear from Sebban et al. 134 which salvage therapies or which rituximab regimens was used. It is also unclear what were the proportion of patients that responded to salvage therapy, the proportion that had a successful harvest and the proportion of patients that receive ASCT.

In the economic model, we assumed that patients receive two cycles of ESHAP with or without rituximab before ASCT with BEAM. The planned dose has been extracted from the clinical policies and protocol document from Surrey, West Sussex and Hampshire Cancer Network,141 presented in Table 47. We assumed that rituximab is administered at 375 mg/m2. The cost of salvage therapy with or without rituximab in patients with relapsed FL is estimated from the BNF. 85

In the base case, we assumed the response rates for HDT with or without rituximab to be 10% higher than the response rates for CHOP and R-CHOP in second-line treatment. 74,75 We further assumed that 80% of patients have a successful harvest after response to HDT. The AG stresses that these assumptions have been made with extremely limited supportive data. SAs were conducted varying both the response rate for HDT and proportion of patients with successful harvest.

For patients responding to HDT with or without rituximab and for whom the harvest was successful, the cost of ASCT + BEAM was assumed to be £30,400, based on a costing exercise commissioned by the London Specialised Commissioning Group. 142 The cost includes pre-transplant mobilisation, stem cell harvest and storage, pre-transplant assessment, patient work-up, transplant admission and cost up to 1 year after discharge.

Management at the end of treatment induction/maintenance: monitoring and surveillance cost

The management of the disease at the end of treatment induction and/or maintenance is adapted from the monitoring reported in the MS62 after discussion with our clinical experts. Compared with the monitoring reported in the MS,62 the monitoring defined by our clinical experts (Table 48) was less intensive, particularly with regard to scanning and imaging.

The AG comments that the monitoring used in the economic model is simplistic, but that SAs indicated that the results were not markedly influenced by this parameter (see Appendix 15).

After first- and second-line induction treatment the monitoring was separated into two phases:

• first 6 months after the end of treatment induction

• remaining months.

The monitoring after maintenance treatment with rituximab has also been separated into two phases:

• first 24 months after the end of maintenance

• remaining months.

Unit costs have been extracted from the NHS Reference Costs 2009/10 and costs used in the Sheffield Teaching Hospital Trust (2005–6, personal communication). Costs are summarised in Table 49.

Health service costs associated with management in third/subsequent lines

Patients that progress after second-line treatment with CHOP, R-CHOP, FC or R-FC (induction or maintenance) and who are still alive are assumed to undergo third/subsequent lines of therapy. A one-off cost was applied in the economic model according to the choice of treatment received in second-line (induction and maintenance).

The management costs were estimated from the post-protocol treatments observed in the EORTC 20981 trial. 74,75 The frequency of resources used for patients treated with CHOP only, R-CHOP only, CHOP in addition to maintenance rituximab, and R-CHOP in addition to maintenance rituximab74 were multiplied by the unit costs used by the manufacturer in a previous NICE appraisal (Table 50). 73 Unit costs were not inflated as main costs were drug and procedure costs.

Patients treated with HDT with or without rituximab are assumed to go directly on to palliative care and no costs were applied for the further lines of treatments. This assumption was made in the absence of data about the post-progression treatment after HDT with or without ASCT and the assumption that fewer treatments are available after relapse to ASCT or HDT. A SA was conducted assuming no costs for third-line treatment for all patients.

Health service costs associated with palliative and/or terminal care

The costs associated with palliative care were estimated from the cost of palliative care for different type of advanced cancers (breast, colon, lung, uterus, ovary, prostate, stomach/oesophagus) from the start of strong opioid treatment until death. 143 The average cost per month was calculated excluding the cost of hospitalisation, as it is likely that hospitalisation costs represent terminal care. The costs per month have been inflated to 2010 prices and are estimated to be £180.68 per month.

In addition to the cost of palliative care, the cost associated with terminal care, i.e. the management before death, was included. This cost was applied only to patients whose cause of death is attributable to FL. The cost of terminal care is sourced from the NICE clinical guidance on cancer palliative/supportive care125 and includes the cost of support provided by specialist hospital/community palliative care teams, including hospice type care, day care, hospital inpatient/outpatient support, bereavement services and continuous support for dying patients. The cost per cancer death is assumed to be £4077 (£3236 inflated to 2010 prices). 125

The AG acknowledges that it is possible that there might be double-counting, as two separate sources have been used. SAs were conducted assuming no cost for terminal care.

Death in progression-free survival after first line

We used PFS as a proxy for progression; however, PFS includes both relapse and death as an event. The MS62 reported that seven deaths occurred in the CVP arm and three deaths in the R-CVP arm. At the end of the trial follow-up period, it was estimated that the number of events (death and/or progression) were 136 and 98, respectively, based on the Kaplan–Meier curves and number of patients randomised. Consequently, we estimated that 5.15% (CVP) and 3.06% (R-CVP) of progression events were attributable to death. The rate of death in CVP was applied to CHOP and MCP. The rate of death in R-CVP was applied to R-CHOP and R-MCP. The rate is then varied using a beta distribution in the PSA.

Health-state utilities

This section of the report presents a systematic review of HS utilities in patients with FL and describes the assignment of utilities in the economic model.

Systematic review of health-state utilities in patients with follicular lymphoma

Methods

A systematic search was performed to identify studies addressing the impairment in quality of life in patients with FL. Full papers and abstracts were included in the review. Only studies conducted in patients with FL or studies conducted in a mix of similar patients when the majority of patients had FL have been included. As the AG was aware of data using the EQ-5D in patients with FL and, given resource constraints, only studies assessing the quality of life using the EQ-5D have been considered for the review, as this is the preferred valuation method of HRQol by NICE. 97 The AG acknowledges that this may be a limitation.

The following databases were searched for relevant published literature: MEDLINE including MEDLINE In-Process & Other Non-Indexed Citations (Ovid); CINAHL; EMBASE; The Cochrane Library including the CDSR, CENTRAL, DARE, NHS EED and HTA databases; SCI; and BIOSIS. Ongoing research have been searched using clinical trials databases and registers, including NIHR Clinical Research Network Portfolio; National Research Register (NRR) archive 2000–7; Current Controlled Trials and ClinicalTrials.gov. Finally, relevant conference proceedings were searched, including the ASCO, ESCO, ASH, BSH and the EHA. Full details of the main search strategy for this review are presented in Appendix 5. In addition, the MS62 was handsearched62 to identify relevant references.

Studies were selected for inclusion through a two-stage process. Titles and abstracts were examined for inclusion by one reviewer. Full manuscripts of selected citations have been retrieved and assessed by one reviewer.

Results

The search retrieved 712 citations relating to quality of life (Figure 33). Six hundred and sixty-nine articles were excluded at title stage, and 28 articles were excluded at abstract level. Fifteen studies have been examined at full-text level and two studies (corresponding to three references) were identified meeting the criteria for the systematic review of quality-of-life data. The study conducted by Wild et al. 118,119 is unpublished and was commissioned by the manufacturer. The full report was made available to the AG and is referred as the ‘Oxford Outcome Study’. The second study, by Friedlich et al. 144 was available in only the abstract form and was conducted in a mix of patients with follicular and other indolent lymphomas. A summary of included studies is below. Reasons for exclusion were the absence of EQ-5D data (use of other instruments or EQ-5D data not presented), Q-TWiST analysis or utilities estimated in a different population.

FIGURE 33.

Flow diagram of quality-of-life review selection/exclusion.

Analytic methods

Results are presented in terms of mean undiscounted LYs, discounted lifetime costs and discounted QALYs.

The following strategies were compared and the ICER was calculated for:

• CVP against R-CVP

• CHOP against R-CHOP

• MCP against R-MCP.

Incremental analyses to determine the most cost-effective combination of chemotherapy with or without rituximab were not conducted by the AG as this was not considered relevant. Discussions with our clinical experts suggested that the choice of chemotherapy was based on additional factors such as patient’s disease characteristics and/or the presence of comorbidities as well as the efficacy of the regimen.

A range of scenarios were presented varying the main model assumptions to identify parameters that had the greatest impact on the ICER.

Probabilistic sensitivity analysis were also carried out using Monte Carlo simulation. The uncertainty in each parameter was represented using a probability distribution. The distribution with the key model parameters are presented in Table 54. The decision uncertainty was shown as the probability that each intervention is the most cost-effective at a given cost-effectiveness threshold. The probability of being the most cost-effective intervention was provided for WTP thresholds of £20,000 and £30,000 per QALY gained.

Base-case analysis assuming no first-line maintenance in patients responding to R-chemotherapy first-line induction