Introduction
Chronic lymphocytic leukaemia (CLL) is characterised by accumulation of malignant B lymphocytes in the lymph nodes, bone marrow and blood [
1,
2]. CLL is the most common adult leukaemia in the developed world, with an annual incidence of 4.2 per 100,000, increasing to >30 per 100,000 among individuals over 80 years old [
3]. In 2014, 15,720 diagnoses and 4600 deaths were reported in the USA and 18,480 cases were estimated to have been diagnosed in the EU5 in 2013 [
4,
5]. As the average age of the global population increases, the incidence of CLL is expected to increase. In the USA, CLL diagnoses are estimated to increase by more than 50% by 2033 [
6].
Although chemoimmunotherapy is effective as a first-line therapy in CLL patients without
TP53 dysfunction and long-term remissions after fludarabine/cyclophosphamide/rituximab (FCR) in IGHV-mutated patients may indicate a potential cure of some patients [
7], CLL is normally considered incurable. Most CLL patients will eventually relapse from first-line treatment or become refractory to it [
3,
4]. Until recently, available salvage regimens had limited efficacy in patients with a poor prognosis [
8]. New molecular targets are being investigated in order to identify therapies to improve treatment outcomes in refractory CLL patients. Bruton’s tyrosine kinase (BTK) is a component of the B cell receptor (BCR) signalling pathway, which is critical in the maturation of B cells, and as such, BTK has emerged as a therapeutic target for B cell malignancies such as CLL [
9].
Ibrutinib is a first-in-class inhibitor of BTK approved for the treatment of adult patients with previously untreated CLL. Ibrutinib as a single agent or in combination with bendamustine and rituximab (BR) is also approved for the treatment of adult patients with CLL who have received at least one prior therapy.
Ibrutinib monotherapy has been evaluated in a phase 3 study (RESONATE) in previously treated CLL patients against ofatumumab monotherapy [
10]. The trial was a multicentre, open-label, phase 3 study, of 391 relapsed or refractory CLL patients receiving either ibrutinib orally at a dose of 420 mg daily until disease progression or standard dose of intravenous ofatumumab for up to 24 weeks. The RESONATE study demonstrated significant improvement with ibrutinib versus ofatumumab in progression-free survival (PFS) and overall survival (OS) in previously treated CLL patients. Long-term follow-up data for ibrutinib from a single-arm phase 2 study in treatment-naïve or previously treated CLL patients demonstrated a PFS rate of 69% and an OS rate of 79% at 2.5 years [
11]. An additional phase 2 trial explored ibrutinib in a cohort of patients with del(17p)/
TP53 mutation with an ORR of 83% [
12]. These data have been largely confirmed in two real-world setting studies performed in Sweden [
13] and the UK/Ireland [
14] but with significantly shorter PFS and OS among patients with del(17p) or
TP53 mutation in the Swedish study.
Health technology assessment bodies assessing new therapies require comparisons with a wide range of treatments. With the absence of direct head-to-head comparisons of single-agent ibrutinib with other widely used treatments in the previously treated CLL patient population, comparative evidence against previous standard of care in clinical practice can provide useful additional preliminary insights. However, naïve unadjusted comparisons of outcomes from different sources are prone to confounding bias due to lack of treatment non-randomisation and variation in prognostic factors between the treatment populations as well as being dependent on the generalizability of the control group.
The main objective of this study was to estimate the relative efficacy of ibrutinib versus previous standard-of-care treatments used in routine healthcare as used in the RESONATE trial in previously treated CLL patients. This estimate is based on a comparison of patient-level data from two different sources: the phase 3 RESONATE study and a retrospective, observational cohort of strictly consecutive patients from a well-defined geographical region [
13]. By using this patient sample, this study aims to minimise these issues as it utilised a well-defined cohort of consecutive patients with almost complete follow-up from the Stockholm region of Sweden with absence of external referrals and controlled for baseline prognostic factors. Within the limitations that follow with such a study design, it can provide preliminary information on outcome with new versus previous therapies for previously treated CLL patients.
Discussion
When evaluating the efficacy of a new class of therapy, previous standard of care and thus the appropriate comparator may differ between countries. There may also exist a wide range of treatment options for a particular disease; then, it may not be financially or logistically practical to compare the new therapy with all available treatment options in a randomised clinical trial setting. In situations where the effectiveness of a new therapy has not yet been directly and proactively assessed in a prospective phase 3 trial, adjusted multivariate analysis of retrospective data may provide a temporary solution. This preliminary comparative information may help assist in healthcare decisions and provide hypothesis-generating results for the next generation of phase 3 clinical trials. Our approach exemplifies a comparison of the efficacy of ibrutinib against previous standard of care in relapsed and refractory patients with CLL by pooling data from a randomised international clinical trial with data from a retrospective observational consecutive cohort of Swedish patients from the Stockholm region with almost complete follow-up and without influence on results from external referrals.
Since some baseline imbalances exist between the cohorts (Table
1), a multivariate Cox proportional hazards regression model was developed which included baseline factors as covariates to adjust for confounding bias related to these differences.
The nature of the Swedish healthcare system means that comprehensive records of both treatment and long-term follow-up of CLL are available for all patients. Additionally, external referrals to the Stockholm region are rare, meaning that all records in the Regional Cancer Registry for the Stockholm region are for patients from one, defined geographical area with minimal external influence, thus minimising selection bias and being a strong representation of the general population. The RESONATE trial took place in multiple countries and adhered to a strict study protocol; the patient population was clearly derived from many geographical regions, which is likely to have introduced heterogeneity in patient characteristics, though both Swedish and RESONATE cohorts were heavily pre-treated. Patients from RESONATE and the historical Stockholm cohort differed in terms of patient characteristics at baseline (Table
1); Swedish patients tended to be older and with higher ECOG scores, and a larger proportion of patients were refractory to prior treatment, making naïve comparisons prone to confounding bias. The adjusted analyses, using multivariate statistical modelling, adjusted for these observed differences between both cohorts. The adjusted HR for OS was less in favour of ibrutinib when compared to results from the “naïve” (unadjusted) comparison, reflecting the fact that the analyses adjust for the higher degree of severity of the Stockholm cohort.
While patients in RESONATE were treated between 2012 and 2015, the Stockholm cohort included patients treated between 2002 and 2013. To explore the potential impact of this difference on the treatment effect estimates, we conducted a sensitivity analysis including only patients in the Stockholm cohort treated 2012 or later. Results for both PFS (HR = 0.15 [0.09; 0.24]) and OS (HR = 0.31 [0.15; 0.63]) were consistent with the main analyses and do not suggest any bias related to the different timeframes for both data sources. Additionally, we recently showed that CLL patients who had received second-line treatment in two time periods (2003–2007 and 2008–2013) displayed a trend of improving PFS over time, but no difference in OS was shown [
13]. Taken together, these and other reports on previous generation of salvage therapies [
8,
19] suggest that OS for R/R CLL patients was not clearly affected until kinase inhibitors became available.
Although the results obtained from our modelling should be viewed with caution, the data suggest that ibrutinib may provide longer PFS and OS compared with historical standard of care during the time period studied in patients with previously treated CLL. The difference remained intact even when the latest time period in the Stockholm cohort was compared separately. HRs reached statistical significance for most comparisons, even though comparisons versus specific previous generation treatment regimens were based on a small number of patient for a number of treatment regimens.
The adjusted HRs that have been reported in our analysis should be interpreted as estimates for the average treatment effect across the entire patient population included in RESONATE and the historical Stockholm cohort. To what extent the relative treatment effect between ibrutinib versus previous standard of care varied across patients according to their baseline characteristics was explored by additionally including interaction terms for treatment with all baseline characteristics in the statistical models. Results suggest that the effect of ibrutinib on PFS was more pronounced in patients between 60 and 74 compared to younger and older patients. OS effect for ibrutinib versus previous standard of care was significantly higher in Binet stage A patients (compared to stages B and C) and especially in ECOG 1 patients (versus ECOG 0). As none of the patients on ibrutinib had ECOG above 1, it is unclear whether this trend exists in these patients. Importantly, all such subgroup analyses shall be regarded as preliminary and hypothesis-generating only.
In this analysis, it is observed that the relative treatment effects for both PFS and OS of ibrutinib versus the ofatumumab arm within RESONATE and versus the Stockholm cohort are similar. A preliminary interpretation of this finding would be that the outcome observed in the ofatumumab arm within RESONATE can be considered as representative for the outcome of previous standard of care as observed in real clinical practice [
10]. The results of the adjusted comparison in this report are in line with the comparison versus the ofatumumab arm within RESONATE study [
10]. Additionally, these results are supported by other recent and preliminary reported analyses, where RESONATE trial data were compared with outcome data for R/R CLL patients from other data sources in various ways. A similar statistical modelling approach using patient-level data was applied to compare PFS and OS between ibrutinib monotherapies from RESONATE with bendamustine-rituximab (BR) from the HELIOS trial (comparing ibrutinib plus BR versus BR) [
20]. The adjusted HR in that report for ibrutinib versus BR was 0.13 for PFS and 0.45 for OS, which are in line with values reported in our analysis (HR = 0.25 [95% CI 0.14, 0.42] and HR = 0.30 [95% CI 0.16, 0.60], respectively). Doubek et al. compared PFS and OS data drawn from RESONATE with a cohort of R/R CLL patients from academic centres in Czech Republic and reported HRs for PFS (HR = 0.10 [95% CI 0.06, 0.16]) and OS (HR = 0.15 [95% CI 0.08, 0.28]) [
21]. Finally, our results are in line with HRs for ibrutinib versus physicians’ choice for PFS (HR = 0.07 CI 0.04; 0.13) and OS (HR = 0.27 CI 0.12; 0.58), based on the Bucher method of adjusted indirect comparison using published results for ibrutinib (RESONATE) and physicians’ choice [
8] versus the common comparator ofatumumab [
22].
Several limitations should be noted in the interpretation of the results of this study. First, although a wide range of clinically relevant prognostic factors were available to be adjusted for, residual confounding bias cannot be excluded, as is the case in any observational study. In particular, del(17p)/TP53 mutation, which is a well-known risk factor in CLL, could not be included in the model, due to a lack of such information for most patients from the early years of record keeping. Similarly, IGHV mutational status was also lacking as it was not included in the routine standard-of-care analyses in Sweden. An additional limitation of this report is that time periods when the patients have been treated were different and that duration of follow-up was significantly shorter within RESONATE compared to the previous standard-of-care cohort. However, the PFS and OS associated with ibrutinib were maintained even when restricting the analysis to only patients treated in the same time period (2012–2013). Finally, data from different sources should always be compared with caution.
In conclusion, this study describes a statistical approach which can be used to provide a preliminary comparison between previous real-world treatments and new drugs until comparisons from randomised clinical trials become available.
Part of this research was presented at the 2015 annual meeting of the American Society of Hematology (ASH).
Acknowledgements
This study was funded by Janssen-Cilag and Pharmacyclics, Inc. Additional research funding was provided by The Swedish Cancer Society (Ref no. 15 0894), The Cancer Society in Stockholm (Ref no. 144142, 151313), King Gustav V Jubilee Fund (Ref no. 144193), The Cancer and Allergy Foundation (Ref no. 150 420, 150 431), StratCan Karolinska Institutet (Proj code: 2201), AFA Insurance (Ref no. 130054) and The Stockholm County Council (Ref no. 20150070). Secretarial assistance was provided by Ms. Leila Relander. Editorial assistance was provided by Peter Gray, MRes and Nick Rusbridge, PhD of PAREXEL, and Dusha Jeyakumaran of Janssen-Cilag and Pharmacyclics, Inc.