Clinical effectiveness and cost-effectiveness results from the randomised, Phase IIB trial in previously untreated patients with chronic lymphocytic leukaemia to compare fludarabine, cyclophosphamide and rituximab with fludarabine, cyclophosphamide, mitoxantrone and low-dose rituximab: the Attenuated dose Rituximab with ChemoTherapy In Chronic lymphocytic leukaemia (ARCTIC) trial

Standard therapy for chronic lymphocytic leukaemia

Fludarabine combined with cyclophosphamide is one of the more frequently used combinations of drugs for treating CLL in second and subsequent line use. The MD Anderson Cancer Center reported the use of fludarabine and cyclophosphamide combined with rituximab (Mabthera^®, Roche Products Ltd) (FCR) in both previously untreated and refractory CLL. 2,3 The response rates for FCR are very impressive and compare extremely positively with historical controls treated with fludarabine, either alone or in combination with cyclophosphamide. In previously untreated patients, complete remission was demonstrated in 217/300 (72%) patients, nodular partial remission in 31 (10%), partial remission in 37 (12%), no response in 13 (4%) and early death in 2 (< 1%) patients. 2 The same group also reported their experience with FCR in 284 patients with previously treated CLL. 3 The estimated median progression-free survival (PFS) was 21 months, with a median overall survival (OS) of 47 months. The median number of prior treatments was two: 67 patients were alkylating agent refractory, 52 were fludarabine refractory and 98 patients had prior rituximab. Using National Cancer Institute (NCI) criteria, 30% of patients achieved a complete remission rate, 14% achieved nodular partial remission and 30% had partial response (PR), giving an overall response rate (ORR) of 74%.

The German CLL Study Group (GCLLSG) completed the German CLL8 trial, which compared FCR with fludarabine and cyclophosphamide (FC) in patients with CLL who had previously been untreated and required therapy according to conventional criteria. 4 It was reported that 811 patients were entered into the GCLLSG CLL8 trial and randomly assigned to receive either FC or FCR. The ORR was significantly higher in the FCR arm (95%; 370/390 patients) than the FC arm (88%; 328/371) (p = 0.001). The complete response (CR) rate of the FCR arm was 52% compared with 27.0% in the FC arm (p < 0.0001). PFS was 65% at 3 years in the FCR arm and 45% in the FC arm (p < 0.0001). Updated data showed that at a median follow-up of 5.9 years, the PFS was 38% in the FCR group compared with 27.4% in the FC group (p < 0.0001). A total of 69.4% of the patients were alive in the FCR group versus 62.3% in the FC group. The median OS was 86 months in the FC group but the median OS was not reached in the FCR arm (p < 0.001). 5 In 2009, the European Medicines Evaluation Agency granted a product licence for rituximab combined with FC in previously untreated CLL.

Rituximab dose

The dose of rituximab has not been established systematically in CLL. It has been extrapolated from the earlier trials with use of rituximab in B-cell malignancies. However, rituximab monotherapy at a dose of 375 mg/m² induced an ORR of 13% in previously treated CLL/small lymphocytic lymphoma (SLL). 6,7 This poor response was thought to be attributable to low CD20 expression on CLL cells and binding of rituximab to CD20 positive cellular debris. Subsequent studies investigating thrice weekly doses of rituximab (375 mg/m²) and higher weekly doses of rituximab (500–2250 mg/m²) in previously untreated patients induced modest ORRs of 43% and 40%, respectively. 8–10 The combination of intravenous FC along with rituximab at variable dose (375 mg/m² in cycle 1 and 500 mg/m² in cycles 2–6) was used in the Phase III CLL8 trial, showing an excellent ORR, PFS and OS. 5 However, the rationale of using higher doses of rituximab has not been formally assessed.

Rituximab binds specifically to the transmembrane antigen CD20, a non-glycosylated phosphoprotein located on pre-B and mature B lymphocytes. The antigen is expressed on > 95% of all B-cell non-Hodgkin’s lymphomas. CD20 is found on both normal and malignant B-cells, but not on haematopoietic stem cells, pro-B-cells, normal plasma cells or other normal tissue. The phenomenon of CD20 shaving on CLL cells with rituximab has been established in CLL. Most of the CLL cells were cleared after 30 mg of rituximab followed by recrudescence of CLL cells which have lost > 90% of CD20 expression. These data suggested that low-dose rituximab thrice weekly at much lower doses of 20–60 mg/m² may promote enhanced clearance of CLL cells by preserving CD20 expression. 11 Subcutaneous rituximab at a dose of 20 mg three times a week resulted in the reduction of CD20 expression on CLL cells, but sufficient expression was maintained during the course of 6–12 weeks in another study. 12 A combination of low-dose rituximab (20 mg/m² three times a week), alemtuzumab (Lemtrada^®, Genzyme Therapeutics) and pentostatin (Nipent^®, Hospira UK Ltd) in high-risk CLL showed that this low dose of rituximab is able to opsonise and clear the majority of circulating cells, but the loss of CD20 is less pronounced. There was also evidence of complement activation owing to C3d deposition on CLL cells and natural killer cell activation owing to down-modulation of CD16, up-regulation of CD54 and a decrease in the number of natural killer cells. 13 Hence, there is considerable evidence that rituximab at doses as low as 20 mg/m² can be effective and can reduce the phenomenon of CD20 shaving, as seen with the higher dosing of rituximab used in CLL.

Rituximab has also been used in lower doses in a variety of autoimmune conditions, such as refractory systemic lupus erythematosus and rheumatoid arthritis, where it is standard to use two intravenous doses of 1000 mg 2 weeks apart. 14–17 Rituximab at a dose of 100 mg once a week for 4 weeks has been used in autoimmune haemolytic anaemia and immune thrombocytopenic purpura with relative similar efficacy to the standard dose of 375 mg/m², although there are no randomised controlled trials to compare the two doses. 18,19 Furthermore, two infusions of 250 mg/m² of rituximab in mixed cryoglobulinaemia are as effective as four infusions of standard-dose rituximab. 20

In summary, the dose of rituximab in the treatment of CLL has not been systematically established and there is good evidence to suggest that low-dose rituximab would be effective in combination with chemotherapy.

Addition of mitoxantrone

Mitoxantrone is a synthetic anthracenedione that is structurally similar to doxorubicin and daunorubicin. It was synthesised with the aim of reducing side effects, especially cardiotoxicity. It is indicated, either in combination therapy or as a single agent, in the treatment of acute non-lymphocytic leukaemia, metastatic breast cancer, hepatoma, lymphoma and paediatric sarcoma.

The addition of mitoxantrone to the fludarabine-based therapy has been found to result in high response rates in a variety of indolent lymphoproliferative disorders, including follicular lymphoma21 and mantle cell lymphoma. 22 The combination of fludarabine, cyclophosphamide and mitoxantrone (FCM) has been reported in 60 patients who have relapsed or resistant CLL. 23 The ORR in this series was of 78% with 30 patients (50%) achieving a complete remission. It was of considerable importance that 10 of the patients in CR had an eradication of detectable minimal residual disease (MRD) by a sensitive four-colour flow cytometric test, and that these patients had a significantly prolonged survival compared with the other patients in this series. In addition, FCM plus rituximab (FCM-R) appears to be a very promising combination in Phase II trials for CLL. The Barcelona group have reported the use of FCM-R in a non-randomised Phase II trial reporting a complete remission rate of 82% and an ORR of 93% in previously untreated CLL. 24 In this study, 46% of the CR patients had undetectable MRD. The National Cancer Research Institute (NCRI) CLL subgroup has recently completed a randomised Phase II study including FCM and FCM-R in previously treated patients with CLL. This study recruited 52 patients, with 26 in each arm, and reported a 65% CR rate for FCM-R compared with a 58% CR rate for FCM, with five and three patients, respectively, achieving eradication of MRD following FCM-R and FCM. 25

Inclusion criteria

• At least 18 years of age.

• B-CLL with a characteristic immunophenotype, including SLL.

• Binet’s stage A progressive or B, or stage C.

• Requiring therapy by the IWCLL criteria in that they must have at least one of the following:

  • evidence of progressive marrow failure as manifested by the development of, or worsening of, anaemia and/or thrombocytopenia

  • massive (i.e. 6 cm below the left costal margin) or progressive or symptomatic splenomegaly

  • massive nodes (i.e. 10 cm in longest diameter) or progressive or symptomatic lymphadenopathy

  • progressive lymphocytosis with an increase of more than 50% over a 2-month period or lymphocyte doubling time of < 6 months as long as the lymphocyte count is over 30 × 109/l

  • a minimum of any one of the following disease-related symptoms must be present:

    • unintentional weight loss more than or equal to 10% within the previous 6 months

    • significant fatigue (i.e. Eastern Cooperative Oncology Group performance status 2 or worse; cannot work or unable to perform usual activities)

    • fevers of greater than 38.0 °C for 2 or more weeks without other evidence of infection

    • night sweats for more than 1 month without evidence of infection.

• No prior therapy for CLL.

• World Health Organization (WHO) performance status of 0, 1 or 2.

• Able to provide written informed consent.

Exclusion criteria

• Prior therapy for CLL.

• Active infection.

• Past history of anaphylaxis following exposure to rat- or mouse-derived complementarity determining region-grafted humanised monoclonal antibodies.

• Pregnancy, lactation or women of child-bearing potential unwilling to use medically approved contraception while receiving treatment and for 12 months after treatment has finished.

• Men whose partners are capable of having children but who are not willing to use appropriate medically approved contraception while receiving treatment and for 12 months after treatment has finished, unless they are surgically sterile.

• Central nervous system involvement with CLL.

• Mantle cell lymphoma.

• Symptomatic cardiac failure not controlled by therapy or unstable angina not adequately controlled by current therapy (in patients with a significant cardiac history the left ventricular function should be assessed and patients with severe impairment should be excluded).

• Other severe, concurrent diseases or mental disorders.

• Known to be human immunodeficiency virus (HIV)-positive.

• Patient has active or prior hepatitis B or C.

• Active secondary malignancy excluding basal cell carcinoma.

• Persisting severe pancytopenia (neutrophils < 0.5 × 10⁹/l or platelets < 50 × 10⁹/l) or transfusion-dependent anaemia unless attributable to direct marrow infiltration by CLL.

• Active haemolysis (patients with haemolysis controlled with prednisolone at a dose of 10 mg or less per day can be entered into the trial).

• Patients with a creatinine clearance of < 30 ml/minute (either measured by or derived from the Cockcroft–Gault formula).

Rituximab-related infusion reactions

The infusion was temporarily stopped until the reaction was resolved and then restarted at half the speed of infusion.

Impaired renal function

Fludarabine was not to be given to participants with a creatinine clearance of < 30 ml/minute. Participants with a creatinine clearance of < 30 ml/minute could have a delay of treatment for up to 4 weeks but were withdrawn from the trial treatment if their creatinine clearance did not improve. Participants with a creatinine clearance of between 30 and 60 ml/minute were permitted to have a 50% dose of fludarabine at the discretion of the treating clinician.

Neutropenia

If neutrophils were < 1.0 × 10⁹/l owing to trial chemotherapy rather than bone marrow involvement, treatment was delayed for up to 2 weeks, with a 25% dose reduction of FC in subsequent treatment cycles. Participants who had a neutrophil count of < 1.0 × 10⁹/l at day 28 of any cycle of therapy received granulocyte colony-stimulating factor {GCSF [lenograstim (Granuocyte, Chungai Pharma UK Ltd)]} at a recommended dose of 263 µg/day from days 7–13 for the next and all subsequent cycles of chemotherapy. Further dose reductions were permitted if neutropenia recurred after the 25% dose reduction. If the neutrophil count recovered to > 1.0 × 10⁹/l the doses of chemotherapy were re-escalated with continuing GCSF support.

Other haematological toxicities

If platelets were < 75 × 10⁹/l as a result of trial chemotherapy rather than bone marrow involvement, treatment was delayed for up to 2 weeks, with a 25% dose reduction of FC in subsequent treatment cycles. If on subsequent cycles of therapy platelets had recovered to over 100 × 10⁹/l, the chemotherapy doses were re-escalated. If further haematological toxicity occurred after the 25% dose reduction further dose reductions were permitted.

Primary outcome measure

• Proportion of participants achieving a CR at 3 months post therapy, as assessed by IWCLL criteria. 26

Secondary outcome measures

• Proportion of participants with eradicated MRD at 3 months post therapy: MRD is defined as negative or undetectable owing to the presence of < 0.01% CLL cells in the blood or bone marrow by IWCLL criteria. 26

• Overall response rate at 3 months post therapy: defined as complete or partial remission by IWCLL criteria. 26

• Safety and toxicity: reported based on AEs (as graded by CTCAE V3.0), SAEs, SUSARs and treatment-related mortalities within 3 months of discontinuing protocol treatment. Determined by routine clinical assessments at each centre.

• Economic evaluation: quality-adjusted life-years (QALYs) were used to measure health benefit. Health-related quality of life was estimated using responses to health economics participant questionnaires, which included the European Quality of Life-5 Dimensions (EQ-5D™) and Short Form questionnaire-12 items (SF-12) [converted to Short Form questionnaire-6 Dimensions (SF-6D)]. 28

• Progression-free survival at 2 years: time from randomisation to first documented evidence of disease progression or death. Participants without evidence of disease progression at the time of analysis are censored at the last date on which they were known to be alive and progression free. The initial analysis was planned once all participants had been followed for 2 years post randomisation.

• Overall survival: time from randomisation to date of death. Participants still alive at the time of analysis are censored at the date on which they were last followed up. The initial analysis was planned once all participants had been followed for 2 years post randomisation.

• Time to MRD relapse in participants who are MRD negative 3 months post treatment: time from the 3-month post-treatment visit, for those participants who became MRD negative, to when the participant became MRD positive. Participants who were alive and MRD negative at the time of analysis, or participants dying from causes unrelated to CLL, were censored. If participants were lost to follow-up, their MRD relapse-free survival time was censored at the time at which they were last known to be alive and MRD negative.

Analysis populations

All analyses were conducted on the intention-to-treat (ITT) population, in which participants were included according to the treatment they were randomised to.

For the outcome measures assessing response, participants without an available response assessment, who had not withdrawn for toxicity or died, were excluded from the denominators. This was felt to be appropriate as it was strongly assumed that response end point data would be missing completely at random, given that participants were unlikely to refuse to have assessments performed owing to their level of response or treatment allocation. Assessments were more likely to be unavailable as a result of samples being un-assessable or missed in error. Reasons for missing response data were monitored and have been summarised (see Table 24).

A per-protocol (PP) analysis was planned for the primary end point assessment in addition to the ITT, where only participants who received at least one cycle of treatment in line with the protocol, who were not major eligibility violators and for whom primary end point data were available are included. For the primary analysis, equal weighting is given to both the ITT analysis and the PP analyses, as the ITT analysis is likely to be the least conservative approach when assessing non-inferiority.

Safety end points were assessed based on the safety population, which included participants according to the treatment they actually received and who had been exposed to at least one dose of the study treatment.

Missing data handling

In the evaluation of response for the primary end point, the analysis was based on the centrally reviewed data at 3 months post treatment. Participants without an assessment of response were treated as non-responders in the ITT and PP analyses if they either:

1. died from CLL or protocol treatment prior to the 3-month post-treatment assessment or

2. discontinued treatment early owing to non-response or toxicity.

Trephine data are required to confirm a CR in participants who are known to be at least a PR. For participants with at least a PR but with missing trephine data, the following rules, defined in advance within the Statistical Analysis Plan, were applied:

1. For participants assessed as at least a PR with no evidence that were not CR/CRi:

   1. MRD-negative participants were reported as ‘CR/CRi’

   2. MRD-positive participants were reported as ‘PR’

   3. Participants with missing MRD status were excluded from the analysis.

2. Participants assessed as PR with evidence that they were not a CR or CRi by the IWCLL criteria, or participants with stable/progressive disease (PD), were treated as non-responders in the analysis.

Evaluation of MRD was based on an assessment of the bone marrow, performed centrally at the Haematological Malignancy Diagnostic Service (HMDS), St James’s University Hospital, Leeds, at 3 months post treatment. For participants with a missing MRD assessment, the next available observation based on the peripheral blood was carried backwards and imputed in place of the missing 3-month observation. This was considered to be a conservative approach as participants are not expected to improve over time without treatment.

Frequency of analyses

Interim reports presenting recruitment, demographic, safety and toxicity data along with treatment and protocol compliance were presented by treatment arm to the DMEC in strict confidence at approximately yearly intervals. In addition to the full annual reports, safety data were presented to the DMEC on a 6-monthly basis. The DMEC reported its recommendations regarding the continuation of the trial to the TSC.

A single formal interim analysis was planned on the short-term efficacy data when half the number of participants (103) had reached their primary end point; this was reported to the DMEC in September 2012. A separate formal analysis plan was written for the interim analysis and signed off before the final data download. The results and outcome of the meeting are reported below (see Interim analysis).

Final analyses were carried out on all but the survival end points when the response data became available for all participants, approximately 9 months after the close of recruitment. The survival end points were analysed 2 years after the close of recruitment and will be updated as appropriate.

Interim analysis

The interim analysis was carried out at a stringent alpha level in order to retain an overall 5% level (two-sided) for the final analysis. The O’Brien and Fleming32 alpha spending function was used to adjust for multiple testing, requiring an alpha level of 0.005 (two-sided) for the interim analysis, and an alpha level of 0.048 (two-sided) for the final analysis.

Primary end point analysis

An overall one-sided 2.4% significance level was used for the final primary response analysis.

The proportion of participants who achieved at least a CR are summarised by treatment group, and the lower limit of the 95.2% CI (one-sided type I error rate of 2.4%) for the difference in the proportions of participants achieving a CR between the treatment groups reported. This was obtained using one-sided binary logistic regression to adjust for the minimisation factors: Binet stage, sex and age, but excluding centre. The treatment estimate is presented along with the odds ratio (OR) and 95.2% Wald CI around the OR estimate. In order to determine whether FCM-miniR was non-inferior to FCR, defined as FCM-miniR being no more than 10% worse in terms of response rates, the lower limit of the CI for the treatment effect is compared with the non-inferiority margin of 10%, expressed as an OR using the following formula:

where π₁ = the proportion of responders in the FCR arm and π₂ = proportion of responders in the FCR arm minus 10%.

For the primary end point analysis, the ITT and PP populations are of joint primacy. In the event that the analyses do not concur (i.e. one demonstrates non-inferiority and the other does not), non-inferiority cannot be concluded.

Sensitivity analyses assess the robustness of results of the primary analysis, and the assumptions regarding missing data:

1. treating all participants will miss primary end point data as non-responders

2. treating all participants will miss primary end point data as responders

3. excluding all participants will miss trephine data from the analysis.

Secondary end point analyses

A two-sided 5% significance level was used for all secondary superiority efficacy end point comparisons.

The proportion of participants with undetectable MRD following treatment, and the proportion of participants who achieved an OR of at least a PR by IWCLL criteria following treatment, are each summarised by treatment group. The differences in the proportions and exact 95% CIs are reported. Binary logistic regression is used to provide treatment estimates with corresponding standard errors (SEs) and p-values, along with ORs and 95% CIs around the OR estimates, after adjusting for the minimisation factors, excluding centre.

Cox regression analysis is used to analyse time to MRD relapse, progression and death, both overall and between treatment arms, accounting for the minimisation factors, excluding centre. Treatment and covariate estimates, SEs, hazard ratios (HRs) and corresponding 95% CIs and p-values are presented for all variables incorporated in the models. Median PFS, OS and time to MRD relapse and corresponding 95% CIs are also presented per treatment arm and overall. The proportional hazards assumption is assessed by plotting the hazards over time (i.e. the log-cumulative hazard plot) for each treatment arm, after adjusting for the minimisation factors. In addition, MRD relapse, PFS and OS curves are calculated using the Kaplan–Meier method.

Safety analyses summarise the number of AEs, SAEs and SUSARs occurring after randomisation. Safety data are presented by treatment group using the safety population. Summaries of the total numbers of SAEs/SUSARs reported and numbers of participants experiencing each event are presented, along with details of the suspected relationship with trial medication or other causality, duration of recovered SAEs/SUSARs, seriousness criteria, event outcome and Medical Dictionary for Regulatory Activities (MedDRA) body system coding. The number and causes of deaths occurring from randomisation until 3 months post treatment are summarised, and the proportion of participants with each cause of death is calculated. No statistical testing is performed between the two groups.

Subgroup analyses

Exploratory subgroup summaries are presented to assess the heterogeneity of the treatment effect among the following subgroups of interest for the primary end point and, where relevant, PFS and OS:

• sex (male, female)

• age group (≤ 65 years, > 65 years)

• Binet stage (A progressive or B, C)

• creatinine clearance levels (30–60 ml/minute, > 60 ml/minute)

• β₂-microglobulin (β₂M) concentration (< 4 mg/l, ≥ 4 mg/l)

• number of cycles of treatment received (three or fewer, more than three)

• GCSF received (yes, no)

• 17p deletion (yes, no)

• 11q deletion (yes, no)

• heavy-chain variable-region (VH) mutation risk (poor risk, standard risk).

In addition, PFS and OS are analysed by IWCLL response and MRD response to treatment, both alone and by treatment arm.

The analyses carried out use the same populations as for the main analyses on the primary and secondary end points. Analyses on the PP population were considered unnecessary owing to the similarity with the ITT population. Subgroup analyses may, by chance, generate false-negative or false-positive results and must be interpreted with caution and treated as hypothesis generating.

Measurement of outcomes

The economic analysis used QALYs to measure health benefit. Health-related quality of life was estimated using responses to health economics participant questionnaires, which included the EQ-5D. 28 QALYs represent a quality-weighted survival value in which one QALY is equivalent to 1 year of full health. All participants were asked to complete these questionnaires at the following time points:

• baseline

• after three cycles of therapy

• at the end of therapy

• 3 months after the end of therapy

• every 3 months after the end of therapy until 24 months post randomisation (i.e. at 6, 9, 12, 18 and 24 months post randomisation).

Standard UK tariff values were applied to these responses at each time point to obtain participant utility values for the within-trial period. QALYs were calculated using the ‘area under the curve’ method and formed the primary outcome measure of the cost-effectiveness study. As NICE currently recommends the use of EQ-5D derived utilities in its reference case, EQ-5D utilities were used in all base-case analyses; sensitivity analysis using SF-12 (converted to SF6D) utilities was also conducted. 27

Measurement of costs

Participant-reported data on resource usage were collected in the trial using the health economics participant questionnaires at similar time points as the health outcomes, except at baseline. For the 3-month follow-up data, the recall period was 3 months (i.e. participants were asked to provide information on their use of health-care resources over the previous 3-month period). For all other cases, the recall period was either since entering the study or since the last questionnaire was completed. The questionnaires included the number and length of hospital inpatient stays, the number of outpatient visits, and the number of primary/community care visits. Participants were also asked to report on the use of PSS related to their treatment (such as the number of visits of carers and social workers).

Costs were estimated by combining participant-reported resource usage with unit cost data obtained from national databases such as the Personal Social Services Research Unit (PSSRU) Costs of Health and Social Care and British National Formulary (BNF). 33,34

The analyses took the perspective of the NHS including the costs of health and social care. All costs are reported in 2013 GBP (£) and future costs were discounted at an annual rate of 3.5%, as per the NICE Methods Guide. 27

Missing data

Missing data for participant-reported health-related quality of life were dealt with by using the multiple imputation method. 35,36 This method assumes that data are missing at random; missing data values are replaced with plausible substitutes based on the distribution of observed data, with uncertainty around the observed data values incorporated using iterative multivariable regression techniques. A set of baseline variables and cost data from that time point were used to impute missing health outcomes data. This approach is recommended for economic analyses alongside clinical trials as it reflects the uncertainty inherent in missing data. 37

Within-trial analysis

Decision economic model analysis

Decision analytical modelling was used to compare FCR and FCM-miniR therapies over a lifetime horizon. A discrete-time state-transition (modified-Markov) model was developed to estimate the long-term cost-effectiveness of treating participants with CLL with FCM-miniR compared with the standard treatment of FCR. In line with the within-trial economic evaluation, the model analysis adopts a UK NHS and PSS perspective and future costs and QALYs were discounted at an annual rate of 3.5% in line with NICE guidance. 27 The model was built in Microsoft Excel version 2010 (Microsoft Corporation, Redmond, WA, USA).

Model structure and parameters

The model structure is presented in Figure 1. The model included three possible health states: PFS, PD and death.

FIGURE 1.

Three-state Markov model.

Markov models describe patient progression over time through a pathway of health states, with movement between the health states being triggered by events such as disease progression or death. Resource use and costs are associated with each health state and patients accumulate costs and health benefits in each state over 3-monthly cycles. As 2-year individual patient cost and utility data were available from the trial data, the model runs from the end of the trial in order to estimate long-term costs and health benefits. The lifetime cost and QALY results are calculated as the sum of the trial 2-year results and the model lifetime estimates (truncated at age 100 years).

The model inputs were derived using information from a range of sources. Where possible, data from the trial were used directly, and published literature was used to inform remaining parameters. Where published literature was required, focused non-systematic reviews were conducted to identify potential sources of information. Appropriate distributions were applied using observed or published variance data; where no such data were available, the standard deviation was assumed to be equal to the mean. As for the within-trial analysis, the model base-case analysis excludes participants who crossed over from the FCM-miniR arm to the FCR arm; a sensitivity analysis was conducted to assess the impact of this. The proportion of participants beginning in each health state in the model was derived directly from the proportion of participants in the trial who remained progression-free and mortality-free at the last follow-up (24 months). Derivation of the post-24-month rate of transition between the progression-free state and progressed disease state required significant extrapolation beyond the relatively short follow-up period of the trial, which contained low numbers of progression events upon which to base the extrapolation. Owing to the short follow-up of the trial, the usual practice of fitting survival curves to the trial data in order to extrapolate results over time is likely to produce highly uncertain, and most likely implausible, results. An alternative approach was therefore adopted that involved fitting a parametric survival model using the Remission Duration Model (RDM) model. The RDM model is based on a plausible biological rationale for the mechanism of disease progression. 39 The model was fitted, using maximum likelihood estimation, to the control arm of the CLL8 trial conducted in Germany which represents a comparable patient population treated with a comparable FCR regimen. 4,40 The fitted model parameters were then calibrated to the PFS curve from the FCR arm of ARCTIC (Figure 2).

FIGURE 2.

Progression-free survival in the FCR arm.

As chemotherapy is expected to exert its effect on PFS only during the treatment period, which was captured within the trial follow-up period, the progression-free event rate beyond 2 years (i.e. in the model) was assumed to be the same between arms. In a sensitivity analysis, the transition rate for FCM-miniR was derived by applying the HR observed in the trial period to the FCR survival curve over various durations to allow for the possibility of a carry-over effect.

Adverse events (AEs) relating to each of the treatment strategies were assumed to occur only within the trial follow-up period (2 years post randomisation to treatment). The cost of AEs has therefore been captured in the within-trial analysis period (in which AEs were costed separately for each arm based on the trial individual patient data) and has not been included within the model follow-up period.

The probability of dying from the PF state was assumed to be equivalent to the general population age- and sex-specific mortality. The probability of dying from the PD state was derived from the literature and assumed to be identical between arms.

Death was assumed to be associated with zero utility and zero cost. For the PF health state, the associated 3-month cost and utility values were derived directly from the trial second-year data, using available data on participants prior to disease progression. Cost and utility values were calculated using second-year data only in order to avoid using data from participants’ primary treatment period, which would bias the results. For the PD state, a combination of a lack of progression events and lack of completed questionnaires at the time of progression in the trial meant that cost and utility estimates for this health state had to be derived from the literature. For the cost of the PD, the 3-month health state cost was taken from a previously conducted cost–utility analysis which looked at the cost-effectiveness of bendamustine (Levact, Napp Pharmaceuticals) versus chlorambucil (Leukeven, Aspen) for the first line treatment of CLL in England and Wales. Woods et al. derived resource use for CLL health states (including PD) from an advisory board conducted in January 2010, which consisted of five haematologists who worked in the UK NHS and were experienced in treating CLL. The reported 3-month progressed disease state cost was £1924, based on 2009 NHS reference costs; this cost was inflated to a 2013 price for use in the current model. 41 For the PD state utility, a decrement was applied to the PF state, using data from Beusterien et al. 42 This was a cross-sectional study in which 89 members of the general population in the UK (England and Scotland) were asked to value health states describing CLL response status using standard gamble methodology. 42

A full list of model parameters and distributions applied in the model is given in Table 3.

TABLE 3 - Model input parameters

Standard deviation.

Model parameter		Mean	Distribution	SE	Source
Global parameters	Annual discount rate	0.035	Fixed	–	NICE27
	Start age, years	62.38	Fixed	–	ARCTIC trial data
	Proportion male, %	0.66	Fixed	–
Starting distribution	PF (FCR)	0.894	Dirichlet	0.03	ARCTIC trial distribution at 24-month follow-up
	PD (FCR)	0.064	Dirichlet	0.04
	Dead (FCR)	0.042	Dirichlet	0.02
	PF (FCM-miniR)	0.791	Dirichlet	0.05
	PD (FCM-miniR)	0.094	Dirichlet	0.06
	Dead (FCM-miniR)	0.115	Dirichlet	0.04
Health state costs (3 months), £	PF	268	Gamma	43	ARCTIC trial data
	PD	2146	Gamma	2146	Woods et al., 201241
	Dead	0.00	Fixed	–	–
Health state utilities (3 months)	PF (EQ-5D)	0.82	Beta	0.01	ARCTIC trial data
	PF (SF-6D)	0.70	Beta	0.03
	PD utility decrement applied to PF state value	–0.16	Log-normal	0.02a	Beusterien et al., 201042
	Dead	0.00	Fixed	–	–
Transition probabilities/effects (3 months)	Risk of progression (PF to PD) in FCR arm	Varies (drawn from survival curve figure)	Fixed	–	Survival analysis of ARCTIC within-trial data (see Figure 2)
	Risk of progression (PF to PD) in FCM-miniR arm	Varies (drawn from survival curve figure)	Fixed	–
	Mortality in PF state	Varies (age- and sex-dependent)	Fixed	–	Office for National Statistics, 2013 age- and sex-standardised rates43
	Mortality in PD state	0.14	Beta	0.03	Wierda et al., 201044
	Log of HR for risk of progression in FCM-miniR arm vs. FCR arm (used in SA only)	0.33	Normal	0.30	ARCTIC trial data

Sensitivity analyses

Probabilistic sensitivity analyses were conducted to assess the impact of joint parameter uncertainty on the results. Probabilistic analysis accounts for joint parameter uncertainty in non-linear models by assigning probability distributions to each of the input parameters and randomly drawing from these probabilities over 10,000 Monte Carlo model simulations to produce different cost and QALY estimates in each simulation of the model. As for the bootstrap within-trial analysis, the results are presented on the cost-effectiveness plane as a scatterplot, and using CEACs to show the probability that the two arms are cost-effective across different WTP per additional QALY thresholds.

Deterministic sensitivity analysis was used to assess the impact of individual parameter uncertainty on the results. Parameters were independently varied between upper and lower bands of plausible values, based on increasing and decreasing each parameter by 25% of its initial value.

Three additional sensitivity analyses were conducted to assess the impact of key model assumptions:

1. SF-6D utilities. The base-case analysis was conducted using utilities derived from participant responses to the EQ-5D questionnaire (which NICE recommends). A sensitivity analysis was conducted using utilities derived from participant responses to the SF-12 questionnaire. For a discussion on the relative merits of SF-12 versus EQ-5D please see Chapter 5, Economic evaluation discussion.

2. Intention to treat analysis. In the base-case analysis participants who crossed over from FCM-miniR to FCR as a result of the trial interim analysis were excluded from the cost-effectiveness analysis. A sensitivity analysis was conducted in which these participants were instead retained in the analysis of the FCM-miniR arm.

3. Treatment effect. As chemotherapy is expected to exert its effect on PFS only during the treatment period, which was captured within the trial follow-up period, the progression-free event rate beyond 2 years was assumed to be the same between arms in the base case. In a sensitivity analysis, the transition rate for FCM-miniR was derived by applying the HR observed in the trial to the FCR survival curve over various durations to allow for the possibility of a carry-over effect; in addition to the base-case analysis, in which a differential rate of progression occurs in the initial 2-year trial period only, analyses were conducted extending the observed HR over 3, 5, 10 years and a lifetime horizon in the FCM-miniR arm.

Value of information analysis

Any model is subject to uncertainty around model parameters and assumptions, which may result in a wrong decision being made based on the model results. If an incorrect decision is made, there will be consequences in terms of health benefit and resources lost. The maximum amount the NHS should be willing to invest to reduce uncertainty in the decision can be informed by the expected value of perfect information (EVPI). 45 The EVPI evaluates the expected opportunity cost of current uncertainty by assessing the probability that a decision based on current information is wrong, multiplied by the costs of making a wrong decision. It is calculated by applying non-parametric methods to the output from the Monte Carlo simulation of the model. 46 It is the difference between the expected value (E) of the decision made with perfect information across j interventions and θ parameters, E_θ max_j NB(j,θ), and the expected value of the decision made on the basis of existing evidence, max_j E_θNB(j,θ):

Additional research should be considered only if the EVPI exceeds the expected cost of research. The EVPI therefore provides an upper limit on the amount that the NHS should be willing to spend on further research in order to resolve current uncertainties.

The value of future research can be further explored using the expected value of perfect parameter information (EVPPI). The EVPPI is a calculation of the maximum value attributable to specific components (parameters) of the evidence base. The EVPPI is calculated as the difference between the expected value of the decision made with perfect information for a particular parameter or set of parameters, ϕ, across the remaining uncertain parameters, ψ, and the expected value of the decision made on the basis of existing evidence:

The EVPPI represents the maximum cost the NHS should be willing to spend on further research to resolve uncertainty for the given parameter or set of parameters evaluated. It is useful for isolating which specific parameters of the decision model are fuelling the overall EVPI, and thereby indicates in what direction future research should be focused.

The formulae above give the per-person EVPI and EVPPI values. To generate population value of information statistics, the per-person estimates must be multiplied by the population expected to be affected by the new treatment. This value is derived from the annual incidence of disease multiplied by the number of years the treatment decision is expected to be relevant. Based on cancer research statistics, the annual incidence of CLL across England and Wales is expected to be 2943. 47 In the absence of quantitative data it was assumed that the period over which the current decision problem will be relevant (i.e. the effective lifetime over which the new treatment will be used) was 10 years. An annual discount factor of 3.5% was applied to the population estimate.

Interim analysis recommendations

The primary aim of this Phase IIB trial was to determine whether FCM-miniR should continue to be investigated in a Phase III trial. Given the strength and robustness of the interim data, the DMEC felt that continuing this trial was futile, as it was highly unlikely that the data would warrant continued investigation of the experimental treatment. As such, they recommended that recruitment into the trial should stop with immediate effect, and these recommendations were ratified by the TSC. It was agreed that the local site investigators should be informed of the findings of the interim analysis as soon as possible. All participants receiving FCM-miniR were recommended to transfer to treatment with FCR for the remainder of their treatment cycles, although this was not mandated. The decision was to be made by the participant following detailed discussions with their clinician. It was also agreed that participants should continue to be followed up as per the protocol.

At the time of trial closure, 200 participants had been recruited. Twenty-one out of the 23 participants still receiving FCM-miniR at the time of trial closure chose to transfer over to receive treatment with FCR on discussion with their treating clinician. Where appropriate, the results will categorise participants into three treatment categories: FCR, FCM-miniR and FCM-miniR/FCR for the patients randomised to FCM-miniR who were recommended to transfer. It is not appropriate to include the participants who transferred to FCR in either of the other categories according to the ITT analysis, because the decision to stop FCM-miniR was on recommendation from the DMEC, rather than a decision from the treating clinician or participant, which violates the ITT assumption.

Treatment received

Of the first 103 participants randomised, 72 (69.9%) received all six cycles of treatment, with a higher proportion of participants coming from the FCR arm (n = 38, 74.5%) than the FCM-miniR arm (n = 34, 65.4%) (Table 16). Thirty-one participants (30.1%) had discontinued treatment early (FCR: 25.5%, n = 13; FCM-miniR: 34.6%, n = 18).

Table 17 summarises the number and proportion of participants experiencing at least one modification to their protocol-defined dose of treatment. A higher proportion of participants in the FCM-miniR arm experienced at least one dose omission, dose reduction and dose delay than in the FCR arm.

Efficacy

An independent central assessment of response was carried out in order to assess formally the primary end point data on response to treatment as defined by IWCLL criteria. 26 The interim analysis of the primary end point data was based on the ITT population, which included participants for whom written informed consent had been received and for which primary end point data were available. A PP analysis was not performed as, out of the first 103 participants, all received at least one cycle of their randomised treatment. The FCR participant who had been previously infected with hepatitis B was not excluded as this was not classed as a major protocol violation.

Of the 103 participants assessed in the interim analysis, 18 (17.5%) were excluded from the formal analysis of the primary end point, with a slightly higher proportion coming from the FCR arm (FCR: n = 10, 19.6%; FCM-miniR: n = 8, 15.4%) (Table 18). Fifteen participants had a missing trephine sample (7 FCR; 8 FCM-miniR), which is required by the IWCLL criteria to confirm a CR, with no further evidence that they were definitely not a CR/CRi. Of these, two trephine samples were taken but were inadequate for analysis. Thirteen samples were not taken owing to a sample being missed in error (n = 4); a clinician’s decision (n = 2); a participant’s decision (n = 2); a participant being unwell (n = 1); it being too painful/difficult to sample a participant (n = 4).

In the FCR arm, two participants withdrew their consent for further trial treatment and follow-up data collection prior to the assessment of the primary end point, and one participant discontinued treatment after their second cycle and was subsequently lost to follow-up.

Table 19 presents the number of participants achieving a CR and the difference in response rates between the treatment arms. Overall, 71.8% (n = 61) of participants achieved a CR, with a higher proportion in the FCR arm (n = 34, 82.9%) than in the FCM-miniR arm (n = 27. 61.4%). The difference in the CR rates was 21.6% in favour of the FCR arm, with a 99.5% CI (–48.0% to 4.8%).

A binary multivariate logistic regression model was used to assess formally the effect of treatment on the proportion of participants achieving a CR at 3 months post treatment, adjusting for the minimisation factors, excluding centre (Table 20). The O’Brien and Fleming32 alpha spending function was used to adjust for multiple testing, requiring an alpha level of < 0.005 (two-sided) to indicate significance, in order to preserve the alpha for the final analysis.

The OR for achieving a CR in the FCM-miniR arm compared with the FCR arm was 0.32 (99.5% CI: 0.07 to 1.48). With a p-value of 0.037, the difference between the treatment arms in terms of CR rates was not significant at the reduced 0.5% significance level, although it was noted that the experimental treatment was somewhat worse in terms of response.

Of the 103 participants included in the interim analysis, 51.5% (n = 53) had undetectable MRD at 3 months post treatment, with a higher proportion of participants in the FCR arm (n = 29, 56.9%) than in the FCM-miniR arm (n = 24, 46.2%) (Table 21).

Safety and toxicity

A total of 103 SAEs were reported from 60 participants out of the first 103 randomised into the trial. Of the 103 reported SAEs, 44 events (42.7%) were from 26 participants receiving FCR and 59 events (57.3%) were from 34 participants receiving FCM-miniR. The mean number of SAEs reported per participant (1.7) was the same in both treatment arms (Table 22).

Of the 103 SAEs reported, 85 (82.5%) were suspected to be related to trial treatment (51 reported from participants receiving FCM-miniR and 34 reported from participants receiving FCR). One SUSAR was reported from a participant in the FCR arm. The participant received all six cycles of treatment and was diagnosed with a ‘squamous cell carcinoma’ approximately 4 months after their last cycle of treatment. The event was felt by the Principal Investigator at the site to be related to trial treatment (F, C and R) and unexpected. The majority of SAEs required hospitalisation (n = 89, 86.4%); three participants died as a result of their SAE (one in the FCR arm and two in the FCM-miniR arm) (Table 23).

A total of 1469 AEs were reported from 101 participants. A total of 669 events were from 49 participants receiving FCR and 770 events from 52 participants receiving FCM-miniR.

Conclusions from the Data Monitoring and Ethics Committee

The difference in the CR rates between the treatment arms (82.9% FCR; 61.4% FCM-miniR), although not statistically significant, was deemed by the DMEC to be clinically relevant. In light of this, and the indication of an increased toxicity rate and increased number of dose omissions and reductions in the FCM-miniR arm, the DMEC recommended that the trial should close the recruitment with immediate effect, and all participants receiving FCM-miniR were recommended to transfer to treatment with FCR for the remainder of their treatment cycles. At the time of trial closure, 200 participants had been recruited into the trial.

Central assessment of response by International Workshop on chronic lymphocytic leukaemia

Table 24 summarises the proportion of participants with a centrally reviewed IWCLL assessment of response. Of the 200 participants, a response assessment was available for 166 participants (83.0%), with the same proportion in each treatment arm. Twenty-four (12.0%) participants assessed as being at least a PR had a missing trephine sample, which is required by the IWCLL criteria to confirm a CR. Four participants in the FCR arm withdrew from follow-up and one FCM-miniR participant died prior to the assessment of response. In addition, in the FCM-miniR arm, one participant’s response to treatment was unable to be assessed and four participants (one in the FCR and three in the FCM-miniR arm) missed their 3-month post-treatment visit.

Numbers analysed and reasons for non-inclusion in the intention-to-treat population

Of the 200 participants randomised, 167 (83.5%) participants were included in the ITT analysis, with a higher proportion in the FCR arm (n = 92, 92.0% FCR; n = 75, 75.0% FCM-miniR) (Table 25). Participants who were randomised to FCM-miniR but transferred over to receive treatment with FCR after the interim analysis were excluded from the ITT population (n = 21). In addition to the 146 participants in the FCR and FCM-miniR arms with an assessment of response, a further 14 participants who were at least a PR but had missing trephine samples were included in the ITT analysis owing to further clinical evidence that confirmed that they were either a CR/CRi or PR, as per the pre-specified decision rules, specified in Chapter 2, Missing data handling. Two participants in the FCR arm withdrew consent for the trial owing to non-response and toxicity, and a further FCM-miniR participant died as a result of their CLL, all prior to the assessment of their primary end point (3-month post-treatment visit). Four participants (one in the FCR and three from the FCM-miniR arms) discontinued treatment early owing to non-response or toxicity and subsequently missed their 3-month post-treatment visit. All seven participants (three FCR, four FCM-miniR) were included in the ITT analysis as non-responders, as per the missing data decision rules.

A total of 33 participants were not able to be included in the ITT analysis, including the 21 treatment arm transfers:

• There were nine participants with a missing trephine sample (six in the FCR and three in the FCM-miniR arm), with no further clinical evidence they were definitely not a CR/CRi. Of these, three were missed in error; two were at the discretion of the treating clinician; two were the participants’ decisions; two were a result of it being too painful/difficult to sample participants.

• Two participants withdrew their consent for further trial treatment and follow-up data collection prior to the assessment of the primary end point, both randomised to FCR. Of these, one participant had a 17p deletion and on the advice of the treating clinician was treated with a ‘more appropriate regime’ off trial, prior to receiving any trial treatment. The second participant withdrew consent after receiving one cycle of treatment, for non-treatment-related reasons: ‘Due to other tests/treatment, the participant no longer wishes to remain on trial and feels they do not have time to continue with trial follow-up’.

• One participant, who received two cycles of treatment with FCM-miniR, was unable to be assessed for response owing to insufficient clinical evaluations performed at the 3-month post-treatment visit (i.e. missing haematology tests, liver and spleen examinations and blood/trephine samples), with no reason provided.

These 33 participants were excluded from the ITT population in line with the guidance within the Statistical Analysis Plan, which was signed off prior to any analysis being conducted.

In the PP population, an additional participant with an assessment of response in the FCR arm was excluded as a result of breaching the eligibility criteria and not receiving any of their randomised treatment. The participant in the FCR arm who had been previously infected with hepatitis B was not excluded as this was not classed as a major protocol violation.

Table 26 presents the baseline characteristics for the ITT population compared with the 33 participants who were not able to be included in the analysis. In the group of participants who were not included in the ITT analysis, participants tended to be older, with a higher proportion of females and Binet stage A progressive or B disease, although these were relatively small numbers for comparison. The excluded population contained a slightly higher proportion of poorer risk participants in terms of 17p and 11q deletions.

Formal analysis of the primary end point

Overall, 66.5% (n = 111) of participants in the ITT population achieved a CR, a higher proportion in the FCR arm (n = 70, 76.1%) than the FCM-miniR arm (n = 41, 54.7%) (Table 27). The difference between the arms was 21.4%, in favour of the FCR arm (95% CI –35.8% to –7.0%), indicating that FCM-miniR appears to be performing worse in terms of response.

A binary multivariate logistic regression model was used to assess formally the primary end point: the effect of treatment on the proportion of participants achieving a CR at 3 months post treatment, after adjusting for the minimisation factors, excluding centre (Table 28). The OR for achieving a CR in the FCM-miniR arm compared with the FCR arm was 0.37 (95% CI: 0.19 to 0.73), indicating that participants in the FCM-miniR are less likely to achieve a CR. The trial was powered on demonstrating that FCM-miniR was no more than 10% worse in terms of CR rates than FCR. With a CR rate of 76.1% in the FCR arm, a 10% reduction gives an OR of 0.61 (see Primary end point analysis for further details of the derivation). As the lower limit of the 95% CI for the treatment effect is < 0.61, and even the mean OR is below this level, there is very strong evidence that FCM-miniR is not non-inferior to FCR in terms of CR rates. As the upper limit of the 95% CI is also below 1, there is evidence that FCM-miniR is significantly inferior to FCR.

The 95% CIs around the ORs for the minimisation factors all contain 1, indicating that there is no evidence that any of the minimisation factors are significantly associated with response, although there is a suggested trend towards participants who are aged over 65 years and with Binet stage C performing worse.

The analysis of the primary end point using the PP population concurred with the outcome of the ITT analysis and demonstrated inferiority of FCM-miniR to FCR with an OR of 0.38 (95% CI 0.19 to 0.75) (Table 29).

The analysis of the primary end point on both the ITT and PP populations strongly demonstrated that FCM-miniR is not non-inferior to FCR in terms of CR rates and, in fact, there is evidence to suggest that FCM-miniR is significantly inferior to FCR.

Sensitivity analyses

Sensitivity analyses were performed in order to assess the reliability of the analysis of primacy and the assumptions regarding the missing primary end point data. The responses of participants that were not included in the ITT population were imputed, with the exception of those participants who received FCM-miniR/FCR after the FCM-miniR arm was closed (see Table 25).

The first sensitivity analysis assumed that participants who were not included in the ITT population were all non-responders, and thus these participants were treated as not having achieved a CR in the analysis of the primary end point (eight in the FCR arm and four in the FCM-miniR). A total of 179 participants were included in this analysis, with 62.0% (111/179) achieving a CR compared with 66.5% (111/167) in the analysis of primacy. A total of 70.0% (70/100) of FCR participants achieved a CR compared with 51.9% (41/79) in the FCM-miniR arm. The OR for achieving a CR in the FCM-miniR arm compared with the FCR arm was 0.46 (95% CI 0.24 to 0.86) (Table 30), indicating that FCM-miniR participants are less likely to achieve a CR. With a response rate of 70.0% in the control arm, a 10% reduction (i.e. the non-inferiority margin) gives an OR of 0.64. As the lower limit of the 95% CI for the treatment effect and the mean OR are both < 0.64, there is very strong evidence that FCM-miniR is not non-inferior to FCR in terms of CR rates, even after treating participants with missing data as non-responders.

The second sensitivity analysis assumed that the participants who were not included in the ITT population were all responders, and thus these participants were treated as having achieved a CR in the analysis of the primary end point (8 FCR, 4 FCM-miniR). A total of 179 participants were included in this analysis, with 68.7% (123/179) achieving a CR compared with 66.5% (111/167) in the analysis of primacy. A total of 78.0% (78/100) of FCR participants achieved a CR compared with 57.0% (45/79) in the FCM-miniR arm. The OR for achieving a CR in the FCM-miniR arm compared with the FCR arm was 0.37 (95% CI: 0.19 to 0.72) (Table 31) indicating that participants in the FCM-miniR are less likely to achieve a CR. With a response rate of 78.0% in the control arm, a 10% reduction (i.e. the non-inferiority margin) gives an OR of 0.60. As the lower limit of the 95% CI for the treatment effect and the mean OR are < 0.60, there is very strong evidence that FCM-miniR is not non-inferior to FCR in terms of CR rates, even after treating participants with missing data as responders.

The third sensitivity analysis excluded all participants with missing trephine data from the ITT population and thus the analysis of the primary end point (6 FCR, 8 FCM-miniR). A total of 153 participants were included in this analysis, with 69.9% (107/153) achieving a CR compared with 66.5% (111/167) in the analysis of primacy. A total of 79.1% (68/86) of FCR participants achieved a CR compared with 58.2% (39/67) of participants in the FCM-miniR arm. The OR for achieving a CR in the FCM-miniR arm compared with the FCR arm was 0.37 (95% CI: 0.18, 0.77) (Table 32), indicating that participants in the FCM-miniR are less likely to achieve a CR. With a response rate of 79.1% in the control arm, a 10% reduction (i.e. the non-inferiority margin) gives an OR of 0.59. As the lower limit of the 95% CI for the treatment effect and the mean OR are < 0.59, there is very strong evidence that FCM-miniR is not non-inferior to FCR in terms of CR rates, even after excluding participants with missing trephine data.

The results of the sensitivity analyses are consistent with the analysis of primacy and confirm that there is strong evidence that FCM-miniR is not non-inferior to FCR in terms of CR rates at 3 months post treatment. The results do not change considerably even after imputing the missing data, demonstrating the robustness of the conclusions. After excluding participants with missing trephine data, the results remained consistent with the analysis of primacy, supporting the approach to impute the response for participants with missing trephine data but who were known to be at least a PR (as described in Missing data handling).

Subgroup analyses

Pre-specified subgroup analyses were carried out to assess the heterogeneity of the treatment effect among subgroups of interest for the primary end point, using the ITT population. As the PP population was very similar to that of the ITT, subgroup analyses on this population were considered unnecessary. As the trial was not powered to look for differences in subgroups of participants, the results are to be treated as exploratory and should be used for hypothesis-generating purposes only.

Tables 33–36 present the subgroups of interest by whether or not a CR was achieved, the difference in response rates and 95% CIs.

Of the 163 participants included in the ITT analysis of the primary end point, a similar proportion of males and females achieved a CR (65.0% vs. 70.0%; Table 33). The difference in response rates between the two age groups was 11.7% (95% CI –3.3% to 26.8%), suggesting a trend towards younger participants (aged ≤ 65 years) being more likely to achieve a CR; as the CI contains zero this difference is not statistically significant. A higher proportion of participants with Binet stage A progressive or B achieved a CR than those with stage C (68.5% vs. 62.5%). The 95% CI contains zero, indicating that this difference is not statistically significant but suggesting a trend in favour of Binet stage A progressive and B participants.

The difference in response rates between the two creatinine clearance groups was –7.6% (95% CI –28.3% to 13.1%), suggesting a trend towards participants with a creatinine clearance level between 30 and 60 ml/minute being less likely to achieve a CR than those with a creatinine clearance level of > 60 ml/minute. As the CI contains zero, this difference was not statistically significant. A higher proportion of participants achieved a CR with a β₂ M concentration < 4 mg/l compared with a concentration ≥ 4 mg/l (71.7% vs. 61.9%). The 95% CI contains zero, indicating that this difference is not statistically significant but suggesting a trend in favour of the lower β₂M concentration group (Table 34).

A much higher proportion of participants who received more than three cycles of treatment achieved a CR than those who received three cycles or fewer (73.2% vs. 28.0%) (Table 35). The difference in response rates was –45.2% (95% CI –64.3% to –26.2%), indicating a significant trend in favour of participants who received more than three cycles of treatment. Of the participants who received treatment with GCSF at some stage during the first three treatment cycles, 67.9% achieved a CR, compared with 72.6% of those who did not. This difference (–4.8%) was non-significant, with 95% CIs that contain zero (–19.9% to 10.4%). Of the participants who received treatment with GCSF during any treatment cycle, 64.9% achieved a CR compared with 71.1% of those who did not receive treatment with GCSF. This difference (–6.1%) was non-significant, with 95% CIs that contain zero (–20.6% to 8.3%). This suggests that the use of GCSF was successful in allowing the delivery of more cycles of therapy and that this enabled participants to achieve better responses which were similar to those seen in participants not requiring GCSF.

All participants who were 17p deleted and who had an available assessment of response (n = 6) failed to achieve a CR. The 95% CI (–76.8% to –61.9%) is highly statistically significant. A higher proportion of participants who did not have an 11q deletion achieved a CR compared with those who did (67.7% vs. 58.3). The difference in response rates (–9.3%) was non-significant with 95% CIs which contain zero (Table 36). Of the 10 participants with an 11q deletion receiving FCR who had an available response assessment, a greater number achieved a CR compared with those participants receiving FCM-miniR [FCR = 8/10 (80.0%), FCM-miniR = 6/14 (42.9%)]. This might be expected as patients with 11q-deleted CLL classically have a relatively high expression of CD20 and are very sensitive to rituximab when combined with FC. Of the participants who were in the ‘poor’ VH mutation risk group, 62.1% achieved a CR compared with 69.2% of those who were in the ‘standard’ risk group. This difference (–7.2%) was non-significant with 95% CIs that contain zero (–23.3% to 9.0%).

Overall response rate

At 3 months post treatment, of the 197 participants randomised and with a response assessment, a total of 184 (93.4%) achieved at least a PR, including 10 participants with missing trephine data for whom it was known that they were at least a PR (Table 37) but for whom it could not be confirmed if they were a CR for the primary analysis. A slightly higher proportion of participants in the FCR arm achieved at least a PR compared with the FCM-miniR arm (n = 94, 95.9% FCR; n = 69, 88.5% FCM-miniR) with all participants receiving FCM-miniR/FCR achieving at least a PR. Thirteen participants (6.6%) did not achieve an overall response, and there were three participants with missing data (Table 37).

Table 38 presents the proportion of participants achieving an OR (at least a PR) at 3 months post treatment and the difference in the response rates between the arms, after excluding those FCM-miniR participants (n = 21) who received treatment with FCR post-interim analysis. The ORR is high at 92.6% (n = 163), with a slightly higher proportion of participants in the FCR arm achieving at least a PR compared with participants in the FCM-miniR arm (95.9% FCR; 88.5% FCM-miniR). Approximately 7.5% fewer participants achieved at least a PR in the FCM-miniR arm (95% CI –15.6% to 0.6%). As the CI contains zero, the difference between the arms is not statistically significant, although it is bordering on significance.

When the responses to treatment are broken down into categories, as per IWCLL, 128/200 (64.0%) participants achieved a CR, with 33.5% achieving complete remission and 30.5% achieving complete remission with incomplete marrow recovery at 3 months’ treatment. Forty-six participants (23.0%) were assessed as achieving partial remission. Three participants were recorded as having stable disease (one FCR, two FCM-miniR) and three as having PD (three FCM-miniR). A total of 15 participants had missing response data owing to being lost to follow-up prior to the 3-month post-treatment visit (n = 4), insufficient clinical evaluations to be able to assess response (n = 1) and missing trephine data (n = 10). The participants with missing trephine data were known to have achieved at least a PR and so are included in the ORR analysis (Table 39). The participant who died as a result of their CLL, two of the participants who withdrew owing to toxicity and non-response and the four participants who discontinued treatment early owing to toxicity/non-response and were lost to follow-up are included in the CR and OR analyses as non-responders.

Minimal residual disease

Minimal residual disease was assessed in the bone marrow at 3 months post treatment. At this time, 85/200 participants (42.5%) were MRD negative and 81/200 (40.5%) were MRD positive (Table 40). A total of 29 participants (14.5%) had missing MRD data with proportions reasonably balanced across the treatment arms (n = 15, 15.0% FCR; n = 13, 16.5% FCM-miniR). Reasons for missing MRD data included: procedure attempted but unable to get sample (n = 1); participant refused to have sample taken (n = 3); sample not taken as participant unwell (n = 2); sample not taken owing to investigator’s discretion (n = 2); sample not taken owing to an administrative error or clinical omission (n = 10); missed visit (n = 3); reasons unknown (n = 8).

For participants with missing MRD data at 3 months post treatment, the missing value was imputed using the next available observation, if available, as described in Chapter 2, Missing data handling. This was the case for 14 participants (10 FCR; three FCM-miniR; one FCM-miniR/FCR). Twelve MRD-negative results were imputed (eight FCR; three FCM-miniR; one FCM-miniR/FCR) and two MRD-positive results (two FCR). Of the four participants in the FCR arm who withdrew from follow-up data collection prior to the assessment of response, two did so owing to toxicity and non-response. These participants and the participant in the FCM-miniR arm who died early as a result of CLL (prior to 3 months post treatment) have been classed as ‘MRD positive’.

Of the 183 participants with an assessment of MRD after imputation, over half (53.0%) were MRD negative at 3 months post treatment (Table 41). A total of 57.0% (n = 53) of FCR participants were MRD negative compared with 46.4% (n = 32) of FCM-miniR participants and 57.1% (n = 12) of those participants randomised to FCM-miniR who swapped over to receive treatment with FCR.

Table 42 presents the proportion of participants with undetectable MRD (MRD negative) at 3 months post treatment and the difference in rates between the arms, excluding those FCM-miniR participants (n = 21) who received treatment with FCR following the interim analysis. A higher proportion of participants in the FCR arm were MRD negative than participants in the FCM-miniR arm (n = 53, 57.0% FCR; n = 32, 46.4% FCM-miniR). Approximately 11% fewer participants were MRD negative in the FCM-miniR arm (95% CI –26.1% to 4.9%). As the CI contains zero, there is no evidence of a significant difference between the treatment arms, although the overall trend is in favour of the FCR arm.

A binary multivariate logistic regression model was used to assess formally the effect of treatment on the proportion of participants achieving MRD negativity at 3 months post treatment, after adjusting for the minimisation factors, excluding centre (Table 43).

The OR for achieving MRD negativity in the FCM-miniR arm compared with the FCR arm was 0.63 (95% CI 0.34 to 1.20), indicating that there is a non-significant trend towards FCM-miniR participants being less likely to achieve MRD negativity at 3 months post treatment than participants in the FCR arm.

Progression-free survival

The final participant’s 24-month clinic visit (after which they are followed up annually) was carried out on 13 October 2014; the data lock for the analysis of the long-term follow-up data was performed on 15 October 2014. At the time of analysis, the median follow-up time of participants who were event-free and still in follow-up was 37.3 months (range: 24.7–58.1 months). PFS was defined as the time from randomisation to the first documented evidence of disease progression or death. Overall, 38 participants (19.0%) had progressed, 17 in the FCR arm (17.0%), 19 in the FCM-miniR arm (24.1%) and two participants who received both FCM-miniR and FCR. A total of 49 (24.5%) participants had either progressed or died: 24 in the FCM-miniR arm (30.4%) compared with 22 in the FCR arm (22.0%) (Table 44).

Figure 5 presents the Kaplan–Meier curves for PFS in months by randomisation allocation. Participants still alive and progression-free at the time of analysis were censored at the last date on which they were known to be alive and progression-free, indicated by a cross. Participants who were randomised to FCM-miniR but transferred over to receive treatment with FCR were excluded from the analysis (n = 21).

FIGURE 5.

Kaplan–Meier plot of time to progression by randomisation allocation.

The PFS curves indicate an overall trend towards an improvement in PFS for participants in the FCR arm. More early progressions are seen with FCM-miniR than with FCR within the first 12 months, but after that the curves remain roughly proportional. At 24 months post randomisation, the PFS probability in the FCR arm is 89.4% compared with 79.1% in the FCM-miniR arm. There is a high number of censored observations around and beyond the 24-month point, the time to which the majority of participants have been followed up, so care must be taken at this stage not to over-interpret the results until longer-term follow-up data are acquired.

A log-rank test was used to compare the differences between the PFS curves. There was no evidence of a significant difference between the treatment arms at the 5% significance level with respect to time to progression (p = 0.279). The Wilcoxon rank-sum test, which gives greater weight to earlier differences, was also non-significant (p = 0.1081) (Table 45).

In a formal Cox regression analysis, after adjusting for the minimisation factors, excluding centre, there was no evidence of a significant difference in PFS between the treatment arms at the 5% significance level with a HR of 1.39 (95% CI 0.77 to 2.49; p = 0.2771) (Table 46), although it should be noted that the CIs are wide owing to the relatively short follow-up period. Although non-significant, a HR of 1.39 indicates that there is a trend in favour of the FCR arm, and that participants in the FCM-miniR arm are around 40% more likely to progress at each time point. The trial was not powered to detect a difference in PFS and CIs around the HR are wide, indicating a lack of precision of the HR estimate owing to the small number of events, and indicating that longer-term follow-up data are required.

The 95% CIs around the HRs for the minimisation factors all contain 1, indicating that there is no significant evidence that any of the minimisation factors are significantly associated with PFS, although it should be noted that the CIs are wide owing to the relatively short follow-up period.

Overall survival

At the time of analysis (October 2014), the median follow-up time for survivors was 37.7 months (range 24.7–58.1 months). OS was defined as the time from randomisation to death from any cause. In total, 24 participants (12.0%) had died: 10 in the FCR arm (10.0%), 13 in the FCM-miniR arm (16.5%) and one participant who received FCM-miniR followed by one cycle of FCR (Table 47). The most common primary cause of death was ‘overwhelming tumour load’ in 33.3% of cases (n = 8) and there was only one treatment-related death in the trial, occurring in the FCR arm.

Figure 6 presents the Kaplan–Meier curves for OS in months by randomisation allocation. Participants still alive at the time of analysis were censored at the last date on which they were known to be alive, indicated by a cross. Participants who were randomised to FCM-miniR but transferred over to receive treatment with FCR were excluded from the analysis (n = 21).

FIGURE 6.

Kaplan–Meier plot of OS by randomisation allocation.

The survival curves indicate an overall trend towards an improvement in OS for participants in the FCR arm. As with PFS, there are a greater number of earlier events in participants on FCM-miniR, with the curves becoming closer to parallel after 18 months. At 24 months post randomisation, the survival probability is 95.8% in the FCR arm compared with 88.5% in the FCM-miniR arm. There are a high number of censored observations around and beyond the 24-month point, the time to which the majority of participants have been followed up, so care must be taken at this stage not to overinterpret the results until longer-term follow-up data are acquired.

The log-rank test was used to compare the difference between the survival curves. There was no evidence of a significant difference between the treatment arms at the 5% significance level with respect to OS (p = 0.2779). The Wilcoxon rank-sum test, which gives greater weight to earlier differences, was also non-significant (p = 0.1013) (Table 48). Note, however, that there have been few events in either arm owing to the relatively short follow-up period.

In the formal Cox regression analysis, after adjusting for the minimisation factors, excluding centre, there was no evidence of a significant difference in OS between the treatment arms at the 5% significance level with a HR of 1.57 (95% CI 0.68 to 3.58; p = 0.2876) (Table 49). Although non-significant, a HR of 1.57 indicates that there is a trend in favour of the FCR arm, and that participants in the FCM-miniR arm are around 57% more likely to not survive at any time point. The trial was not powered to detect a difference in OS and the CIs around the HR are wide, indicating a lack of precision of the HR estimate owing to the small number of events, so longer-term follow-up data would be beneficial.

The 95% CIs around the HRs for the minimisation factors all contain 1, indicating that there is no evidence that any of the minimisation factors are significantly associated with OS, although there is only a limited period of follow-up at the time of reporting. There is a suggested trend in favour of participants aged 65 years or younger to have an improved survival than those aged over 65.

Time to minimal residual disease relapse

Participants who were MRD negative at 3 months post treatment were followed up 6-monthly until MRD relapse (i.e. until they became MRD positive) or until 24 months post randomisation. Longer follow-up was not possible owing to funding constraints. In the population of participants who were assessable for MRD at 3 months post treatment, 85/162 (52.5%) participants had achieved MRD negativity (see Table 42). At the time of analysis, it was reported that only seven participants had relapsed at the MRD level (four FCR, three FCM-miniR); however, there was a high proportion of missing MRD data at each of the follow-up visits. Owing to the small number of events, high proportion of missing data and lack of longer-term follow-up data, the time to MRD relapse analysis was unable to be performed.

Subgroup analyses

Subgroup analyses were carried out to assess PFS and OS based on IWCLL and MRD response groups, the minimisation factors, number of cycles of treatment received, GCSF usage and genetic risk groups. As the trial was not powered for survival end points or to look for differences in subgroups of participants, the results are to be treated as exploratory and used for hypothesis generating purposes only.

Progression-free survival by minimal residual disease response status

Figure 7 presents the Kaplan–Meier curves for time to progression by MRD status at 3 months post treatment. Of the 85 participants who were MRD negative at this time point, eight (9.4%) have reported an event (i.e. progression or death) compared with 30/77 (39.0%) who were MRD positive.

FIGURE 7.

Kaplan–Meier plot of time to progression by MRD status at 3 months post treatment.

The curves show clear divergence from 6 months post randomisation, the end of treatment visit, with an overall trend towards an improvement in PFS for participants who were MRD negative at 3 months post treatment. This difference was significant at the 5% level in favour of the MRD-negative participants (log-rank: χ² = 23.20; p < 0.0001). At 24 months post randomisation, the PFS probability is 96.4% for MRD-negative participants compared with 79.2% for the MRD-positive group.

Progression-free survival by minimal residual disease response status and randomisation allocation

Figure 8 presents the Kaplan–Meier curves for time to progression by randomisation allocation and MRD status at 3 months post treatment.

FIGURE 8.

Kaplan–Meier plot of time to progression by randomisation allocation and MRD status at 3 months post treatment.

For the participants who became MRD negative, the PFS curves are similar for the two treatment arms. For the participants who were MRD positive, participants in the FCR arm show a trend towards a slightly improved time to progression compared with participants in the FCM-miniR arm until around 24–30 months post randomisation, at which point there are a number of censored observations, indicating that longer term follow-up would be beneficial to interpretation. At 24 months post randomisation, the PFS probabilities for the MRD-negative participants were similar between the treatment groups (96.2% FCR, 96.9% FCM-miniR). For MRD-positive participants, the PFS probability was 83.7% in the FCR arm compared with 74.9% in the FCR-miniR arm.

Progression-free survival by complete response status

Figure 9 presents the Kaplan–Meier curves for time to progression by CR status at 3 months post treatment. Of the participants who had achieved a complete remission or CRi at this time point, 16/111 (14.4%) had an event (i.e. progression or death) compared with 27/56 (48.2%) who had not achieved at least a CR.

FIGURE 9.

Kaplan–Meier plot of time to progression by CR status at 3 months post treatment.

The curves show clear divergence from 6 months post randomisation, the end of treatment visit, with an overall trend towards an improvement in PFS for participants who achieved a CR at 3 months post treatment. This difference was significant at the 5% level in favour of participants who achieved at least a CR (log-rank: χ² = 29.41; p < 0.0001). At 24 months post randomisation, the PFS probability is 93.4% for participants who achieved a CR compared with 65.4% for those who did not.

Progression-free survival by complete response status and randomisation allocation

Figure 10 presents the Kaplan–Meier curves for time to progression by randomisation allocation and CR status at 3 months post treatment.

FIGURE 10.

Kaplan–Meier plot of time to progression by randomisation allocation and CR status at 3 months post treatment.

For the participants who achieved a CR, the curves are similar for the two treatment arms, with a slight trend towards the FCM-miniR participants doing better. For the participants who did not achieve a CR, participants in the FCR arm have an improved time to progression compared with FCM-miniR participants. At 24 months post randomisation, the PFS probabilities for the participants who achieved a CR were similar between the treatment groups (92.6% FCR, 94.8% FCM-miniR). For participants who did not achieve a CR, the PFS survival probability was 74.3% in the FCR arm compared with 60.3% in the FCR-miniR arm.

Overall survival by minimal residual disease response status

Figure 11 presents the Kaplan–Meier curves for OS by MRD status at 3 months post treatment. Of the participants who were MRD negative at this time point, 4/85 (4.7%) had died compared with 14/77 (18.2%) who were MRD positive.

FIGURE 11.

Kaplan–Meier plot of OS by MRD status at 3 months post treatment.

The curves show clear divergence from the end of treatment visit, with an overall trend towards an improvement in OS for participants who were MRD negative at 3 months post treatment. This difference was significant at the 5% level in favour of the MRD-negative participants (log-rank: χ² = 9.25; p = 0.003). At 24 months post randomisation, the OS probability for MRD-negative participants is 97.6% compared with 91.9% for the MRD-positive group.

Overall survival by minimal residual disease response status and randomisation allocation

Figure 12 presents the Kaplan–Meier curves for OS by randomisation allocation and MRD status at 3 months post treatment.

FIGURE 12.

Kaplan–Meier plot of OS by randomisation allocation and MRD status at 3 months post treatment.

For the participants who became MRD negative, the OS curves are similar for the two treatment arms. A similar survival pattern was observed for MRD-positive participants receiving FCR until around 30 months post randomisation, at which point survival worsened. OS was poorest for MRD-positive participants who received FCM-miniR. At 24 months post randomisation, the OS probabilities for the MRD-negative participants were similar between the treatment groups (98.1% FCR, 96.9% FCM-miniR), and for MRD-positive participants receiving FCR (97.3%). The FCM-miniR, MRD-positive group showed the poorest OS probability at 24 months (86.5%).

Overall survival by complete response status

Figure 13 presents the Kaplan–Meier curves for OS by CR status at 3 months post treatment. Of the participants who had achieved a CR at this time point, 8/111 (7.2%) had died compared with 15/56 (26.8%) who had not achieved a CR.

FIGURE 13.

Kaplan–Meier plot of OS by CR status at 3 months post treatment.

The curves show clear divergence from 6 months post randomisation, the end of treatment visit, with an overall trend towards an improvement in OS for participants who achieved a CR at 3 months post treatment. This difference was significant at the 5% level in favour of participants who achieved a CR (log-rank: χ² = 16.92; p < 0.001). At 24 months post randomisation, the OS probability for participants who achieved a CR is 96.4% compared with 82.9% for those who did not.

Overall survival by complete response status and randomisation allocation

Figure 14 presents the Kaplan–Meier curves for OS by randomisation allocation and CR status at 3 months post treatment.

FIGURE 14.

Kaplan–Meier plot of OS by randomisation allocation and response status at 3 months post treatment.

For the participants who achieved a CR, the curves are similar for the two treatment arms. A similar survival pattern was observed for participants who did not achieve a CR who were receiving FCR until around 30 months post randomisation, at which point survival worsened. OS was poorest for participants who did not achieve a CR and who received FCM-miniR. At 24 months post randomisation, the OS probabilities for the participants who achieved a CR were similar between the treatment groups (95.6% FCR, 97.6% FCM-miniR), and for participants who did not achieve a CR receiving FCR (95.0%). The FCM-miniR participants who did not achieve a CR showed the poorest OS probability at 24 months (75.8%).

Progression-free survival and overall survival by minimisation factors at baseline: sex

Figure 15 presents the Kaplan–Meier curves for time to progression by participant sex. At the time of analysis, of the 123 male participants, 32 (26.0%) had reported an event (i.e. progression or death) compared with 14/56 (25.0%) female participants.

FIGURE 15.

Kaplan–Meier plot of time to progression by sex.

At 24 months post randomisation, the PFS probability for male participants was 84.4% compared with 85.5% for females. The difference between the PFS curves for sex was non-significant (log-rank: χ² = 0.045; p = 0.833).

Figure 16 presents the Kaplan–Meier curves for OS by participant sex. At the time of analysis, of the 123 male participants, 17 (13.8%) had died compared with 6/56 (10.7%) female participants.

FIGURE 16.

Kaplan–Meier plot of OS by sex.

At 24 months post randomisation, the OS probability for male participants was 92.4% compared with 92.8% for females. The difference between the OS curves for sex was non-significant (log-rank: χ² = 0.411; p = 0.522).

Progression-free survival and overall survival by minimisation factors at baseline: age group

Figure 17 presents the Kaplan–Meier curves for time to progression by age group (≤ 65 years, > 65 years). At the time of analysis, of the 113 participants aged ≤ 65 years, 31 (27.4%) had reported an event (i.e. progression or death) compared with 15/66 (22.7%) participants aged over 65 years.

FIGURE 17.

Kaplan–Meier plot of time to progression by age group.

At 24 months post randomisation, the PFS probability for participants aged ≤ 65 years was 87.0% compared with 80.9% for those older than 65 years. Although there appears to be a trend in favour of younger participants performing better in terms of PFS, the overall difference between the PFS curves for age group was non-significant (log-rank: χ² = 0.015; p = 0.902).

Figure 18 presents the Kaplan–Meier curves for OS by age group (≤ 65 years, > 65 years). At the time of analysis, of the 113 participants aged ≤ 65 years, 13 (11.5%) had died compared with 10/66 (15.2%) participants aged over 65 years.

FIGURE 18.

Kaplan–Meier plot of OS by age group.

At 24 months post randomisation, the OS probability for participants aged ≤ 65 years was 94.5% compared with 89.0% for those over 65 years of age. The difference between the OS curves for age group was non-significant (log-rank: χ² = 1.106; p = 0.293), although there is a suggested trend for younger participants performing better in terms of OS.

Progression-free survival and overall survival by minimisation factors at baseline: Binet stage

Figure 19 presents the Kaplan–Meier curves for time to progression by Binet stage (A progressive or B, C). At the time of analysis, of the 118 participants who were Binet stage A progressive or B, 32 (27.0%) had reported an event (i.e. progression or death) compared with 14/61 (23.0%) participants who were Binet stage C.

FIGURE 19.

Kaplan–Meier plot of time to progression by Binet stage.

At 24 months post randomisation, the PFS probability for participants with Binet stage A progressive or B was 84.8% compared with 84.7% for Binet stage C participants. The difference between the PFS curves for Binet stage was non-significant (log-rank: χ² = 0.217; p = 0.641).

Figure 20 presents the Kaplan–Meier curves for OS by Binet stage (A progressive or B, C). At the time of analysis, of the 118 participants who were Binet stage A progressive or B, 16 (13.6%) had died compared with 7/61 (11.5%) participants who were Binet stage C.

FIGURE 20.

Kaplan–Meier plot of OS by Binet stage.

At 24 months post randomisation, the OS probability for participants with Binet stage A progressive or B was 91.2% compared with 95.0% for Binet stage C participants. The difference between the OS curves for Binet stage was non-significant (log-rank: χ² = 0.030; p = 0.863).

Progression-free survival and overall survival by treatment received

Figure 21 presents the Kaplan–Meier curves for time to progression by number of treatment cycles received (three or fewer, more than three). At the time of analysis, of the 31 participants who received three cycles or fewer, 16 (51.6%) had reported an event (i.e. progression or death) compared with 30/148 (20.3%) participants who received more than three cycles of treatment.

FIGURE 21.

Kaplan–Meier plot of PFS by number of treatment cycles received.

The curves show clear divergence from the end of treatment with an overall trend towards an improvement in PFS for participants who received more than three cycles of treatment. At 24 months post randomisation, the PFS probability for participants receiving more than three cycles of treatment was 92.3% compared with 43.9% for those receiving three or less. The overall difference between the PFS curves was significant at the 5% level (log-rank: χ² = 51.629; p < 0.0001).

Figure 22 presents the Kaplan–Meier curves for OS by number of treatment cycles received (three or fewer, more than three). At the time of analysis, of the 31 participants who received three cycles or less, 8 (25.8%) had died compared with 15/148 (10.1%) participants who received more than three cycles of treatment.

FIGURE 22.

Kaplan–Meier plot of OS by number of treatment cycles received.

The curves show clear divergence from the end of treatment with an overall trend towards a worse OS for participants who received three or fewer cycles of treatment. At 24 months post randomisation, the OS probability for participants receiving more than three cycles of treatment was 95.9% compared with 74.1% for those receiving three or fewer. The overall difference between the OS curves was significant at the 5% level (log-rank: χ² = 14.167; p = 0.0002).

Progression-free survival and overall survival by granulocyte-colony stimulating factor usage

Figure 23 presents the Kaplan–Meier curves for time to progression by GSCF usage during treatment. At the time of analysis, of the 82 participants who had received GCSF at some stage during their treatment, 21 (25.6%) had reported an event (i.e. progression or death) compared with 23/87 (26.4%) participants who had not received any GSCF.

FIGURE 23.

Kaplan–Meier plot of PFS by GCSF usage.

At 24 months post randomisation, the PFS probability for participants who had received GCSF at some stage during their treatment was 87.6% compared with 83.3% for participants who had not received any GCSF. The difference between the PFS curves for GCSF usage was non-significant (log-rank: χ² = 0.249; p = 0.618).

Figure 24 presents the Kaplan–Meier curves for OS by GCSF usage during treatment. At the time of analysis, of the 82 participants who had received GCSF at some stage during their treatment, 11 (13.4%) had died compared with 12/87 (13.8%) participants who had not received any GSCF.

FIGURE 24.

Kaplan–Meier plot of OS by GCSF usage.

At 24 months post randomisation, the OS probability for participants who had received GCSF at some stage during their treatment was 95.1% compared with 89.5% for participants who had not. The difference between the OS curves for GCSF usage was non-significant (log-rank: χ² = 0.067; p = 0.796).

Progression-free survival and overall survival by genetic risk factors: 17p and 11q

Given that there were only five participants with 17p deletion this analysis was not felt to be appropriate.

Figure 25 presents the Kaplan–Meier curves for time to progression by whether or not participants were 11q deleted. At the time of analysis, of the 26 participants who had an 11q deletion, six (23.1%) had reported an event (i.e. progression or death) compared with 37/142 (26.1%) participants who reported no event.

FIGURE 25.

Kaplan–Meier plot of PFS by 11q deletion.

At 24 months post randomisation, the PFS probability for participants who were 11q deleted was 79.8% compared with 84.5% for participants who were not. The difference between the PFS curves was non-significant (log-rank: χ² = 0.032; p = 0.858).

Figure 26 presents the Kaplan–Meier curves for PFS by whether or not participants were 11q deleted and by treatment group. The numbers are small, with only 10 and 16 participants in the FCR and FCM-miniR groups, respectively; however, it can be seen that only one of the FCR patients with 11q deletion progressed after over 3 years, compared with 5/16 (31.3%) of 11q-deleted patients on FCM-miniR. Excluding these 11q-deleted patients from the PFS curves brings the FCR and FCM-miniR curves closer together than those in Figure 5, although it can be seen that there is still a trend towards a PFS advantage with FCR over FCM-miniR.

FIGURE 26.

Kaplan–Meier plot of PFS by 11q deletion and treatment group.

Figure 27 presents the Kaplan–Meier curves for OS by whether or not participants were 11q deleted and by treatment group. The numbers are small, with only 10 and 16 participants in the FCR and FCM-miniR groups, respectively. No FCR patients with 11q deletion died, compared with 2/16 (12.5%) of 11q-deleted patients on FCM-miniR. After excluding these 11q-deleted patients from the OS curves, there is still a trend towards a survival advantage in the FCR group.

FIGURE 27.

Kaplan–Meier plot of OS by 11q deletion and treatment group.

Figure 28 presents the Kaplan–Meier curves for OS by whether or not participants were 11q deleted. At the time of analysis, of the 26 participants who had an 11q deletion, 2 (7.7%) had died compared with 21/142 (14.8%) participants who had not.

FIGURE 28.

Kaplan–Meier plot of OS by 11q deletion.

At 24 months post randomisation, the OS probability was the same for participants who were 11q deleted and those who were not (92.0%). The difference between the OS curves was non-significant (log-rank: χ² = 0.648; p = 0.421).

Progression-free survival and overall survival by genetic risk factors: VH mutation risk

Figure 29 presents the Kaplan–Meier curves for time to progression by VH mutation risk, where standard risk indicates that the participant had a VH mutation not involving the VH3–21 gene and poor risk indicates that the participant did not have a VH mutation, or that the VH3–21 gene was involved. At the time of analysis, of the 91 participants with a poor VH mutation risk, 27 (29.7%) had reported an event (i.e. progression or death) whereas 11/58 (18.0%) participants with a standard VH mutation risk had reported an event.

FIGURE 29.

Kaplan–Meier plot of PFS by VH mutation risk.

At 24 months post randomisation, the PFS probability for the poor risk group was 82.6% compared with 87.4% for the standard-risk group. Although there appears to be a slight trend in favour of standard-risk participants performing better in terms of PFS, the difference between the PFS curves was non-significant (log-rank: χ² = 1.672; p = 0.1960).

Figure 30 presents the Kaplan–Meier curves for OS by VH mutation risk. At the time of analysis, of the 91 participants with a poor VH mutation risk, 16 (17.6%) had died whereas 5/58 (8.6%) participants with a standard VH mutation risk had died.

FIGURE 30.

Kaplan–Meier plot of OS by VH mutation risk.

At 24 months post randomisation, the OS probability was 91.0% for the poor-risk group compared with 94.6% for the standard-risk group. The difference between the OS curves was non-significant (log-rank: χ² = 2.172; p = 0.141), although there is a suggested trend for standard-risk participants performing better in terms of OS.

Serious adverse events

A total of 183 SAEs [80 (43.7%) FCR; 81 (44.3%) FCM-miniR; 22 (12.0%) FCM-miniR/FCR] have been reported from 104 (52.5%) participants. In the FCR arm, 49/100 (49.0%) participants reported at least one SAE compared with 46/79 (58.2%) receiving FCM-miniR and 9/22 (47.4%) receiving FCM-miniR followed by FCR (Table 50). Of the participants experiencing at least one SAE, the mean number of SAEs reported was 1.8, with a similar number for the FCR (mean = 1.6) and FCM-miniR (mean = 1.8) arms, as for rates for all participants.

Of the 183 SAEs reported, 145 (79.2%) were suspected to be related to protocol treatment (SARs), with a slightly higher proportion in the FCM-miniR arm (n = 67, 82.7%) than in the FCR arm (n = 62, 77.5%) (Table 51). SARs were reported from 89 participants (44.9%), with a higher proportion of SARs reported from participants receiving FCM-miniR (n = 39, 49.4%), compared with FCR (n = 41, 41.0%), and from nine participants (47.4%) receiving FCM-miniR followed by FCR. There was one SUSAR reported in the trial, from a participant in the FCR arm. This SUSAR was reported to the regulatory authorities (Medicines and Healthcare products Regulatory Agency), sponsor and the main REC within the required timelines for expedited reporting. The participant received all six cycles of treatment and was diagnosed with a ‘squamous cell carcinoma’ approximately 4 months after their last cycle of treatment. The event was felt by the Principal Investigator at the site to be related to trial treatment (FCR) and unexpected. Further details on this event, and all SAEs, are provided in Appendix 1.

Of the 145 SAEs suspected to be related to protocol treatment, in the FCR arm the majority (51.6%) were suspected to be related to all three IMPs, F, C and R, with 33.9% suspected to be related to just F and C, and 11.3% suspected to be related to rituximab only. In the FCM-miniR arm, SAEs were most commonly (44.8%) suspected to be related to all four IMPs, F, C, M and reduced-dose rituximab (miniR) with 31.3% suspected to be related to F, C and M.

The majority of SAEs required (prolonged) hospitalisation (89.1%), with 92.5% coming from the FCR arm and 84.0% from the FCM-miniR arm (Table 52).

Six SAEs in the FCM-miniR arm were deemed to be life-threatening or resulted in death, compared with three in the FCR arm. In the FCR arm, one participant died as a result of an ‘infection’ which was suspected to be related to trial treatment within 5 months of their last course of treatment. This participant received two cycles of treatment. In the FCM-miniR arm, one participant died as a result of a ‘Bilateral pneumonia’ which was suspected to be related to trial treatment within 9 months of their last course of treatment. The participant received all six cycles of treatment. A further FCM-miniR participant died as a result of ‘Neutropenic sepsis and infected shoulder’, which was not suspected to be related to trial treatment, within 3 months of their last course of treatment. This participant received four cycles of treatment. Further information on the SAEs that resulted in death are provided in Appendix 1.

All but three SAEs had recovered at the time of reporting.

The majority of SARs required (or prolonged) hospitalisation (90.3%) (93.5% in the FCR arm compared with 86.6% in the FCM-miniR arm). All but three SARs had recovered at the time of reporting (Table 53).

Of the 198 participants in the safety population, 96 (48.5%) required hospitalisation during the trial as a result of an SAE, with a higher proportion in the FCM-miniR arm (n = 41, 51.9%) than the FCR arm (n = 46, 46.0%) (Table 54).

Of the 180 SAEs that were not ongoing, the median duration of an event was 5 days (range: < 1–303 days) with a similar duration in each of the treatment groups (Table 55). The median duration of SARs that were suspected to be related to protocol treatment was the same, although there was variability in the mean durations between the treatment arms.

Of the 145 SAEs suspected to be related to protocol treatment (SARs), the majority (n = 90, 62.1%) were classed as ‘infections and infestations’ according to the MedDRA System Organ Class system (Table 56). A higher proportion of events reported in the FCR arm were classed as ‘general disorders and administration site conditions’ than those reported in the FCM-miniR arm [10 (16.1%) FCR; 6 (9.0%) FCM-miniR].

Adverse events

A total of 2163 AEs have been reported from 192 (97.0%) participants. A total of 1117 events have been reported from 96 participants receiving FCR, 863 events from 77 participants receiving FCM-miniR and 183 from 19 participants who received FCM-miniR followed by FCR. The mean number of AEs reported was similar for the FCR and FCM-miniR treatment groups for the population of participants experiencing at least one event (Table 57). Line listings of all AEs by treatment received are presented in Appendix 2, Adverse event listings.

New AEs most commonly occurred during the first treatment cycle, with the number of new AEs occurring gradually declining with the more treatment cycles received, although this trend will be influenced by the declining number of participants receiving each treatment cycle (Table 58). The proportion of new AEs occurring at each treatment cycle appears to be reasonably well balanced between FCR and FCM-miniR.

Table 59 presents the maximum CTCAE grade experienced for each AE by treatment received. The majority of AEs were reported as a maximum CTC grade 1 (50.5%). A higher percentage of CTC grade 3/4 AEs were experienced in the FCM-miniR arm (n = 193, 22.4%) than the FCR (n = 168, 15.0%) and FCM-miniR/FCR arms (n = 27, 14.6%).

Treatment-related mortalities within 3 months of ending protocol treatment

There were no treatment-related mortalities within 3 months of the protocol treatment ending.

One trial participant, receiving FCM-miniR, died within 3 months of discontinuing the protocol treatment but the cause was not suspected to be related to trial treatment. This participant discontinued treatment after their fourth cycle owing to ‘Disease progression, requiring further treatment’. After discontinuing treatment, and prior to death, they reported a SAE of ‘Neutropenic sepsis and infected shoulder’, which was not suspected to be related to the trial treatment. The participant gradually deteriorated and died within 3 months of ending protocol treatment. The primary cause of death was given as ‘Infection due to CLL’ and the participant’s disease status at the time of death was recorded as ‘stable disease’.

Secondary cancers

The following tables summarise the secondary cancers that were reported at follow-up.

Table 60 shows that the incidence of secondary cancers was similar across the trial arms, with 11.1% of patients reporting a secondary cancer. The most common types of secondary cancer were skin- and haematological-related cancers.

Table 61 further summarises the type of secondary cancer reported. Note that the results are not mutually exclusive, as one patient was diagnosed with basal cell carcinoma twice, and another with both squamous cell carcinoma and melanoma.

Table 62 shows the mean, median and ranges of when the secondary cancers were diagnosed from randomisation and from end of treatment.

Unit cost of resource use

Individual-level resource use was combined with unit costs to calculate the total health-care use cost for each participant in the trial. In order to convert resource usage figures into costs, unit cost figures were assigned from national sources such as the PSSRU Unit Costs of Health and Social Care 2013. 33 Table 63 presents the summary of unit costs.

Unit cost of medications

Participants were randomised to receive six cycles of either FCR or FCM-miniR. Each cycle was 28 days. Participants were evaluated after three cycles of chemotherapy and if no response or disease progression was observed they were stopped from receiving further therapy. These participants would still attend follow-up assessments until 24 months after randomisation. Unit costs for medications were obtained from the BNF 67th edition of September 2013. 34 Details for medication dosages given within the six treatment cycles were collected. The total medication costs using the dosage and the frequency provided by the participant were calculated. If the dose of the drug was not recorded, it was assumed that the participant received the same dosage as other participants who were given the same treatment. If the quantity was not recorded, the average quantity for that drug as reported in the data was applied. Table 64 shows all the unit costs for the drugs that were reported in the trial.

Treatment costs

In addition to the cost of medications used during chemotherapy, an additional cost was identified which was the cost of administering the drug. The costs are shown in Table 65.

Cost-effectiveness results: base case

Resource use and QALY data were available for 162 participants, with 92 participants being treated with FCR and 70 participants with FCM-miniR.

Health-care resource use

The total costs associated with resource use during the trial are shown in Tables 66 and 67. The mean total NHS and PSS resource use costs during therapy are £15,492 for FCR and £9049 for FCM-miniR. Costs were significantly higher for FCR (+£6444; p = 0.00). In the 18 months following the therapy period, the difference in resource use costs between the two arms diminished, with a mean total cost of £1756 for FCR and £1570 for FCM-miniR (p = 0.71). There were no significant differences in NHS and PSS resource use costs between the treatment arms following the end of therapy according to the t-tests.

Health outcomes

Table 68 presents the EQ-5D scores at baseline, 3 months after therapy ended, and 12, 18 and 24 months post randomisation with missing values imputed. Both treatment groups showed increasing EQ-5D scores from baseline up to 3 months after the end of therapy, with a slight dip at 12 months and 18 months post randomisation in both arms before increasing again.

On average, the difference between arms was marginal. Independent sample t-tests indicated that the changes in EQ-5D score over time were not statistically significant. The average total QALYs gained over the 24 months was marginally higher in the FCR arm (1.61) than in the FCM-miniR arm (1.55) (p = 0.40).

Cost-effectiveness results within the NHS and Personal Social Services perspectives

Table 69 shows the total costs and EQ-5D-generated QALYs for each of the treatment arms. Differences in QALYs between groups were minimal and suggested marginal health decrements in the FCM-miniR arm compared with the FCR arm. The FCR group had the highest EQ-5D-generated QALYs over the trial period. The mean total cost was significantly higher for the FCR group. The high standard deviation (SD) for the deterministic cost estimates indicates the presence of a few outlying individuals who incurred significant health service costs.

Table 70 below provides the probabilistic cost-effectiveness results, showing the incremental costs and benefits as well as the ICER. The results suggest that FCM-miniR is associated with an incremental cost saving of £6619 and an incremental QALY loss of 0.059. Note that, owing to the fact that FCM-miniR is associated with both negative cost and QALY increments, the ICER result cannot be interpreted in the usual way as the cost per additional QALY associated with FCM-miniR; the ICER instead represents the cost saved per QALY lost. That is, if the NHS adopts FCM-miniR, it could expect to save £112,193 per QALY lost. As this cost saving outweighs the value of a lost QALY based on a WTP of £20,000 per QALY, the overall net benefit associated with FCM-miniR is positive (£5349) and FCM-miniR is therefore expected to be cost-effective over a 24-month time horizon.

The uncertainty around the cost-effectiveness estimate is represented graphically on the cost-effectiveness plane (Figure 31) using bootstrapping with 10,000 iterations. This method samples at random with replacement from each of the 10 imputed data sets, producing 10,000 incremental cost and incremental QALY estimates.

FIGURE 31.

Cost-effectiveness plane showing the incremental costs and QALYs for FCM-miniR compared with FCR from the bootstrap analysis of the within-trial results over 24 months (EQ-5D, NHS and PSS perspectives).

Figure 31 shows that all the points were below the x-axis, indicating that FCM-miniR is cost-saving compared with FCR, and most points were to the left of the y-axis, indicating that FCM-miniR produces fewer QALYs than FCR. The majority of points lie below the £20,000/QALY threshold line, indicating that FCM-miniR is cost-effective.

The CEAC showing the probability that FCR is cost-effective is presented below in Figure 32 with a range of cost-effectiveness WTP thresholds values. The probability that FCM-miniR is cost-effective is high and remains above 60% up to a threshold value of £100,000 per QALY.

FIGURE 32.

Cost-effectiveness acceptability curves at 12 months (NHS and PSS perspectives, using EQ-5D generated QALYs).

Sensitivity analyses within trial in the NHS and Personal Social Services perspective

In order to test the robustness of the within-trial analysis a number of different cost-effectiveness analyses were completed.

Sensitivity analysis of health utility measurement

A similar cost-effectiveness analysis was completed using SF-6D-generated QALYs. When the equivalent SF-6D figures were analysed, a different pattern was observed. There was a decrease in mean SF-6D values between baseline and 3 months post randomisation, with fluctuations in either direction at the follow-up time points. The average total QALYs gained were slightly higher in the FCM-miniR arm (1.211) than the FCR arm (1.195) (p = 0.81), in contrast to the EQ-5D analysis in which FCR produced a higher overall QALY value. A statistically significant difference in the SF-6D utility values was observed at the 24 months post randomisation time point only (Table 71).

TABLE 71 - Short Form questionnaire-6 Dimensions index scores at baseline and follow-ups, and QALYs of CLL participants by treatment arm (imputed data)

Max., maximum; min., minimum.

Statistically significant at the 95% confidence level.

Parameter		FCR (n = 92)	FCM-miniR (n = 89)	Difference: p-value of t-test
Baseline	Mean (SD)	0.651 (0.290)	0.670 (0.269)	0.713
	Median (min.–max.)	0.723 (–0.01–1.00)	0.723 (–0.128–1.00)
3 months after end of therapy	Mean (SD)	0.606 (0.261)	0.593 (0.242)	0.645
	Median (min.–max.)	0.6592 (–0.024–1.00)	0.634 (–0.074–1.00)
12 months post randomisation	Mean (SD)	0.715 (0.238)	0.706 (0.237)	0.814
	Median (min.–max.)	0.770 (–0.127–1.00)	0.770 (–0.060–1.00)
18 months post randomisation	Mean (SD)	0.670 (0.239)	0.722 (0.219)	0.157
	Median (min.–max.)	0.711 (–0.075–0.965)	0.777 (–0.071–0.965)
24 months post randomisation	Mean (SD)	0.612 (0.289)	0.763 (0.170)	0.000a
	Median (min.–max.)	0.704 (–0.059–1.00)	0.808 (–0.014–0.965)
Total QALYs	Mean (SD)	1.195 (0.414)	1.211 (0.412)	0.812
	Median (min.–max.)	1.242 (0.038–1.850)	1.305 (0.126–1.820)

Table 72 below shows the total costs and SF-6D-generated QALYs for each of the treatment arms. FCM-miniR is associated with an incremental cost saving of £6492 and an incremental QALY gain of 0.016 compared with FCR; FCM-miniR therefore dominates FCR, being more effective and less costly, with a positive INB of £6805.

TABLE 72 - Cost-effectiveness results (NHS and PSS perspectives, deterministic)

Strategy	Total cost, £	Total QALY (based on SF-6D)	Incremental cost, £	Incremental QALY	ICER	INB, £
FCR	17,248	1.1949
FCM-miniR	10,756	1.2105	–£6492	0.016	FCM-miniR dominates	6805

Using the NICE WTP threshold of £20,000, FCM-miniR is expected to be 100% cost-effective over a 24-month time horizon. This is illustrated in Figure 33 by the fact that all of the simulated points in the probabilistic sensitivity analysis lie under the WTP threshold.

FIGURE 33.

Cost-effectiveness plane generated from bootstrapped mean cost and SF-6D-generated QALY differences over 24 months (NHS and PSS perspectives).

Summary of within-trial cost-effectiveness analysis

Using the EQ-5D to generate QALYs, the results showed that participants with CLL, treated in the FCM-miniR arm, did not show a higher QALY gain when compared with participants in the FCR arm (1.552 vs. 1.610). However, FCM-miniR presented negative incremental costs (£6443), indicating that treatment with FCM-miniR would lead to cost savings. This difference was driven by the higher cost of delivery of FCR, with cost differences stemming from treatment costs during the six cycles of chemotherapy.

Based on imputed data, the analysis found the incremental QALY gain to favour FCR, despite the difference being minimal at 0.059 (roughly equivalent to 21.5 days of full health). This result is robust to sensitivity analysis and stochastic bootstrapping.

However, using SF-6D to generate QALYs, the results showed that participants with CLL treated with FCM-miniR showed a slightly higher QALY gain than participants in the FCR arm (1.211 vs. 1.195). This gain, together with the incremental cost saving of £6805, suggests that FCM-miniR dominates FCR and is cost-effective at a £20,000 WTP threshold. Despite this difference in QALY findings between the two quality-of-life measures, FCM-miniR was found to be cost-effective in both analyses.

The within-trial analysis is conducted on data collected over a 24-month time period. The majority of CLL participants go on to live much longer than this; therefore, the within-trial analysis is unlikely to capture all of the relevant differences in long-term costs and health outcomes between the two participant groups. Hence, there is a need for decision model analysis to extrapolate these results to lifetime horizons.

Base-case model results

Results of the base-case analyses are presented in Table 74 (deterministic results) and Table 75 (probabilistic results). Taking into account uncertainty around the model parameters (i.e. using the probabilistic results), FCM-miniR is associated with an expected lifetime cost saving of £7723 and a lifetime QALY loss of 0.73 compared with FCR; an incremental loss of over two-thirds of a healthy life-year. As the incremental cost and QALYs are both negative, the ICER does not represent the incremental cost per additional QALY, but rather represents the cost saved per QALY lost associated with adoption of FCM-miniR. The ICER therefore indicates that one QALY will be lost for every £10,624 saved by adopting FCM-miniR. As the amount saved is less than the societal WTP for a QALY (£20,000), the overall expected INB of FCM-miniR is negative. This indicates that adoption of FCM-miniR would lead to a lifetime loss of benefit for the NHS and PSS, and FCM-miniR is therefore not expected to be cost-effective. The expected lost net benefit associated with FCM-miniR is –£6780, which is equivalent to a net health loss of 0.34 QALYs (assuming that QALYs are valued at £20,000 per QALY). There is significant uncertainty around this result, as can be seen by the large SD value for the INB (SD = 7907).

The results of the probabilistic analysis are presented in Figure 34. Each point on the graph represents the result of one probabilistic simulation of the model and indicates a potential incremental cost and QALY for FCM-miniR compared with FCR. The diagonal line represents the currently accepted WTP per QALY threshold of £20,000 per QALY. Points that lie below the threshold line are considered cost-effective; points above the threshold line are not considered to be cost-effective. In this analysis, the points are widely distributed, with many points lying in both cost-effective and non-cost-effective regions. This indicates that there is significant uncertainty around the cost-effectiveness of FCM-miniR compared with FCR over a lifetime horizon. The mean of the simulated values lies above the threshold line, indicating that, on average, FCM-miniR is not expected to be cost-effective.

FIGURE 34.

Scatterplot of model lifetime cost-effectiveness results for FCM-miniR vs. FCR.

The CEAC shows the proportion of model simulation points that lie under the cost-effectiveness threshold plane across different threshold values, which indicates the probability that each treatment is cost-effective at given WTP values. The CEAC for FCM-miniR versus FCR is presented in Figure 35. At low cost per additional QALY thresholds, FCM-miniR is associated with a high probability of being cost-effective. However, as the threshold value increases the probability of FCM-miniR being cost-effective rapidly diminishes, and FCR is associated with an increasing probability of cost-effectiveness. At a threshold value of £20,000 per QALY, FCM-miniR is associated with 19% probability of being cost-effective; increasing the threshold value to £30,000 per QALY results in a drop to a 12% probability of being cost-effective.

FIGURE 35.

Cost-effectiveness acceptability curve for model lifetime analysis.

Sensitivity analyses

One-way deterministic sensitivity analysis

Results of the one-way deterministic sensitivity analysis are presented in Figure 36. The vertical line indicates the base-case deterministic INB value for FCM-miniR versus FCR (–£5609). The horizontal bars indicate the extent to which this base-case value is altered when individual model parameter values are increased or decreased by 25% of their base-case value. From the diagram it appears that altering the discount rate and utility of the progression-free state results in the greatest change to the cost-effectiveness result. However, FCM-miniR is associated with a negative INB in all of the analyses considered, indicating that FCM-miniR is not expected to be cost-effective in any of the analyses. Changes in the input parameters therefore have no effect on the overall result of the analysis.

FIGURE 36.

Tornado plot of one-way deterministic sensitivity analysis results.

Value of information analysis

Results of the population EVPI analysis are shown in Figure 37. The EVPI value represents the absolute maximum that the NHS should be willing to spend on further research on the cost-effectiveness of FCM-miniR versus FCR. At a £20,000 per QALY threshold, and assuming a 10-year effective time horizon for the new treatment, the expected value of information is £21M. The maximum EVPI value (£43M) is reached at a £10,600/QALY threshold value (equivalent to the ICER value for FCM-miniR), as this is the point at which uncertainty around parameter estimates in the model has the greatest influence on the decision of whether or not to adopt the new treatment.

FIGURE 37.

Population EVPI.

Results of the EVPPI analysis are shown in Figure 38. Uncertainty around the starting distributions of the model (i.e. the proportion of participants who begin the model with progressed disease as opposed to being disease free) is found to be the most influential factor, with the associated EVPPI values for these parameters being substantial. These parameters are directly related to the efficacy of each of the treatments in terms of delaying disease progression, so it is unsurprising that these have a large impact on the model results.

FIGURE 38.

Population EVPPI for base-case model (£20,000/QALY threshold, 10-year effective lifetime and annual incidence of 2943).

The value of information results suggest that there is the potential for further research into the cost-effectiveness of FCM-miniR compared with FCR to be of value to the NHS. However, the EVPI and EVPPI values provide only a necessary (but not sufficient) condition for the decision to fund further research (i.e. a positive EVPI/EVPPI value indicates that there may be value in conducting further research). The decision of whether or not to invest in further research and what design that research should take requires further analysis of the cost of research and the value of particular trial designs.

Predicted overall survival

The predicted OS curves simulated by the base-case model are presented in Figure 39. The difference in OS between the two arms is greatest at the beginning of the model simulation; this is attributable to the fact that the model begins from the end of the trial 24-month period, and the proportion of participants surviving in each arm was taken directly from the trial follow-up data which showed the given discrepancy between the two arms. Subsequently in the model, however, it is assumed that the risk of mortality in both the progression-free and progressed disease states is the same in both arms. In the future, validation of the model could be achieved by comparing the predicted OS with real-world data; currently these data are unavailable and we therefore show the OS curves for illustrative purposes only.

FIGURE 39.

Overall survival curves simulated from the lifetime decision analytic model.

Summary of decision model lifetime cost-effectiveness analysis

The results of the decision analytic model indicate that FCM-miniR is not expected to be cost-effective over a lifetime horizon. Compared with FCR, FCM-miniR is associated with an average lifetime cost saving of £7723 per patient, and a lifetime QALY loss of 0.73. The resulting expected INB associated with FCM-miniR is –£6780 (equivalent to a net health loss of –0.34 QALYs), indicating that if FCM-miniR were to be adopted instead of FCR the NHS would be worse off as a whole.

Results of the deterministic sensitivity analyses found that the results were robust to changes in individual model parameters, with FCM-miniR remaining non-cost-effective in all of the analyses conducted. However, when joint parameter uncertainty is considered (using probabilistic sensitivity analysis), there is some uncertainty around the results: at a £20,000 per QALY threshold, there is a 19% chance that FCM-miniR is cost-effective. The expected value of information analysis indicates that there is potential benefit to conducting further research into the cost-effectiveness of FCM-miniR; however, additional considerations (such as the cost of research and specific research design) would need to be considered before a definitive recommendation for further research could be made.

Economic evaluation discussion

In both the within-trial and the decision model analyses, FCM-miniR was found to be associated with negative incremental costs and benefits compared with FCR. Care needs to be taken when interpreting the results of both analyses, given that ICERs have a different meaning when both cost and benefit increments are negative. Typically, a new intervention is found to be more expensive and more effective than the standard-care treatment (with results of probabilistic analyses lying in the north-east quadrant of the cost-effectiveness plane). In such cases, the ICER represents the additional cost required to be spent on the new treatment in order to gain an additional QALY compared with the standard-care treatment. However, in the case of new treatment that is less costly and less effective than the comparator treatment (with results of probabilistic analyses lying in the south-west quadrant of the cost-effectiveness plane), the ICER instead represents the incremental cost saved per QALY lost compared with the standard-care treatment. For example, in the economic analysis FCM-miniR is associated with an ICER of £10,651, indicating that for every £10,651 saved by adopting FCM-miniR, one QALY will be lost compared with FCR. Assuming a WTP of £20,000 per QALY, this means that FCM-miniR is not cost-effective, which may not be immediately clear from the ICER result. For this reason, it is preferable to look at the INB results rather than the ICERs, as interpretation of net benefit results does not depend on the direction of the incremental cost and QALYs. If the INB associated with a new treatment is positive, then that treatment is cost-effective, whereas if it is negative the treatment is not cost-effective. Thus, for the case of the decision model analysis, FCM-miniR is associated with an INB of –£6780, indicating that FCM-miniR is not cost-effective over a lifetime horizon.

Treatments that are both cost- and QALY-decreasing can still be cost-effective because the cost saved can be spent on treatments or interventions elsewhere in the NHS in order to gain additional QALYs, which may outweigh the QALY loss associated with the given intervention. That is, the overall INB of the new treatment compared with the standard-care treatment may be positive (indicating cost-effectiveness) if the cost saving associated with the new treatment is sufficient to outweigh the QALY loss. In the case of the within-trial analysis, FCM-miniR was associated with a mean cost saving of £6619 and a mean QALY loss of –0.056. In this case, the cost saved outweighs the QALYs lost, resulting in a positive INB. However, for the decision model analysis, the cost saving associated with FCM-miniR (£7723) was unable to outweigh the QALY loss (–0.73), resulting in a negative INB (i.e. –0.735 × 20,000 + 7723 = –£6780; equivalent to a net health loss of –0.735 + 7723/20,000 = –0.34 QALYs).

The discrepancy in the short- and long-term cost-effectiveness analyses results (i.e. the fact that FCM-miniR is found to be cost-effective in the within-trial analysis but not in the model lifetime analysis) is attributable to the fact that all cost-savings associated with FCM-miniR occur in the short term (i.e. when therapy costs are incurred). In the model it is assumed that costs in the progression-free and progressed disease states are the same for both treatments, so the benefit of FCM-miniR in terms of reducing treatment costs is contained in the initial trial 24-month period of the model. Over the lifetime analysis, the cost savings associated with FCM-miniR are, therefore, diluted and are no longer sufficient to outweigh the QALY losses associated with the new treatment.

For the within-trial analysis, the results of QALY calculations were not consistent across health utility measures. The base-case EQ-5D analysis found FCM-miniR to be inferior to FCR in terms of QALY production (resulting in a –0.059 QALY decrement), whereas the SF-6D analysis found FCM-miniR to be superior to FCR (resulting in a +0.016 QALY increment). The reason for this discrepancy is unclear. The SF-12 questionnaire used to derive the SF-6D QALYs asks more detailed questions regarding patients’ emotional states, and t-test results showed significant differences using SF-6D at 24 months post randomisation, whereas no significant differences between arms were shown using EQ-5D. This could suggest that the SF-12 form is more sensitive to changes in health states for this population. Nevertheless, despite the difference in expected QALY values between the EQ-5D and SF-6D analyses, both analyses indicate that FCM-miniR is associated with a positive INB value and therefore represents a cost-effective investment for the NHS. Overall, there is some uncertainty around the model results (see Generalisability) which will not have been captured in the deterministic or probabilistic sensitivity analyses. It is unclear in which direction this uncertainty is likely to impact on the results; therefore, some caution should be maintained when drawing conclusions from the model results.

Summary

In summary, we have shown that FCM-miniR does not have a non-inferior CR rate compared with FCR for the frontline therapy of CLL and that, over the lifetime of the patients, FCM-miniR was not found to be cost-effective compared with FCR. Therefore, there is no evidence against FCR remaining the gold-standard treatment for patients with CLL who require therapy and are considered fit for fludarabine-based combinations.