An indirect comparison between bosutinib, nilotinib and dasatinib in first-line chronic phase chronic myeloid leukemia

Abstract

Objective

Bosutinib, nilotinib and dasatinib are approved for the treatment of patients with newly diagnosed chronic-phase chronic myeloid leukemia (CP-CML). In the absence of head-to-head comparisons between second-generation tyrosine kinase inhibitors (TKIs), the objective of this study was to indirectly compare the efficacy of bosutinib with nilotinib and dasatinib in first-line (1L) CP-CML.

Methods

Cross-trial heterogeneity in terms of patient baseline characteristics and imatinib dose escalation are difficult to adjust for in network meta-analyses and anchored matching-adjusted indirect treatment comparisons (MAICs). Therefore, an unanchored MAIC was performed using patient level data from bosutinib (BFORE trial) and published aggregated data from nilotinib (ENESTnd) and dasatinib (DASISION) trials. After matching, cytogenetic and molecular responses, and disease progression, after a minimum follow-up of 24 months were compared between nilotinib versus bosutinb and dasatinib versus bosutinib.

Results

The comparison of nilotinib versus bosutinib resulted in no statistically significant differences for MMR at and by 24 months, MR4 by 24 months, MR4.5 at and by 24 months, CCyR by 24 months, and disease progression, however, a decreased odds of MR4 at 24 months in favor of bosutinib versus nilotinib was observed. The comparison of dasatinib versus bosutinib by 24 months resulted in no statistically significant differences for MMR, disease progression, and CCyR, however a decreased odds of MR4.5 in favor of bosutinib versus dasatinib was observed.

Conclusions

Overall, in these analyses bosutinib demonstrates equivalent efficacy to nilotinib and dasatinib in the treatment of patients with newly diagnosed CP-CML.

Keywords:

• Chronic myeloid leukemia
• first line
• bosutinib
• dasatinib
• nilotinib

Introduction

Recent studies have estimated the five-year survival rate of patients with chronic myeloid leukemia (CML) at 90% and have shown that the life expectancy of patients with CML approaches that of the general population^1,2. This high figure is largely due to the emergence of tyrosine kinase inhibitor (TKI) therapies, ranging from imatinib, to the second-generation TKIs nilotinib, dasatinib, and bosutinib.

The latter three TKIs have also shown improved clinical efficacy against imatinib as first-line (1 L) treatment of chronic-phase CML (CP-CML). In the ENESTnd³ trial, a head-to-head comparison was conducted between nilotinib and imatinib in patients with newly diagnosed Philadelphia chromosome-positive (Ph+) CP-CML. A similar comparison was conducted between dasatinib and imatinib in the DASISION⁴ trial and between bosutinib and imatinib in the BFORE⁵ trial. The ENESTnd³, DASISION⁴, and BFORE⁵ studies have shown a higher overall efficacy for the second generation TKI when compared against imatinib in terms of cytogenetic and molecular responses. However, to date, no head-to-head trials have been conducted between the three second generation TKIs (nilotinib, dasatinib, and bosutinib) in the 1 L setting.

A previously conducted network meta-analysis (NMA) including nilotinib, dasatinib and bosutinib showed for CCyR at 12 months that bosutinib, nilotinib, and dasatinib had similar probabilities of being the first, best treatments and for MMR at 12 months, nilotinib had the highest likelihood of being the preferred treatment⁶.

A limitation to this study is that for results up to 12 months, outcomes of patients from the BFORE study were used but for outcomes beyond this time point, the NMA included outcomes of patients from a previous bosutinib study, the BELA trial⁷. Another NMA reported that nilotinib tended to have the highest probability of being ranked first for several endpoints⁸.

Nonetheless, two apparent sources of heterogeneity make a naïve/indirect comparison (via an NMA) between bosutinib and nilotinib and/or dasatinib unreliable. These sources concern the heterogeneity in baseline characteristics and the differences in the rate of imatinib dose escalation in earlier pivotal clinical trials (e.g. both the ENESTnd and DASISION trials started in 2007) relative to more recent ones (e.g. the BFORE trial started in 2014).

In the BFORE trial, 28⁹ and 31% of patients had an imatinib dose escalation at 12 and 24 months, respectively, compared with 16³ and 29%¹⁰, respectively, in ENESTnd, and 20%¹¹ at 24 months in DASISION (12-month data not reported). Differences in imatinib dose-escalation regimens are related to both the trial design and conduct of the trial, and, therefore, cannot be adjusted for using NMAs or matching-adjusted indirect comparisons (MAICs). NMAs and anchored MAICs are also not able to correct for cross-trial heterogeneity in patient disease- and baseline characteristics, however, unanchored MAICs, despite their inherent limitations (impact of having access to only marginal covariate distribution, choice of scale for indirect comparisons, choice of target population and sampling variation in the target population), can adjust for the latter (heterogeneity in disease- and baseline characteristics) by using individual patient data (IPD) from trials of one treatment (for example, bosutinib) to match baseline summary statistics reported from trials of another treatment (for example, nilotinib and dasatinib).

After matching, the resulting treatment outcomes are compared across balanced trial populations.

The objective of the current research was to compare the efficacy of bosutinib with nilotinib and dasatinib in newly diagnosed CP-CML via unanchored MAICs^12,13.

Methods

Data sources

For nilotinib, published aggregate data was used from the ENESTnd trial (disease- and baseline characteristics data from Saglio et al., 2010, 24-months outcome data from Kantarjian et al., 2011, and the Summary of Product Characteristics for nilotinib) evaluating the efficacy of nilotinib (300 mg twice daily and 400 mg twice daily) versus imatinib (400 mg once daily) for newly diagnosed Ph + CML patients^3,10,14,15. Only the nilotinib 300 mg twice daily arm was used in our analysis as this is the approved dose for newly diagnosed patients.

For dasatinib, published aggregate data was used from the DASISION trial (disease- and baseline characteristics data from Kantarjian et al., 2010, 24-months outcome data from Kantarjian et al, 2012, the Summary of Product Characteristics for dasatinib, and clinicaltrials.gov) evaluating the efficacy of dasatinib (100 mg once daily) versus imatinib (400 mg once daily) in newly diagnosed Ph + CP-CML patients^4,11,16,17.

For bosutinib, individual patient data (IPD) was available from the BFORE trial evaluating the efficacy of bosutinib (400 mg once daily) versus imatinib (400 mg once daily) in adult patients with newly diagnosed CP-CML from an unlocked database with a data cutoff of July 12, 2017 (24 months after last patient randomized). Although the trial included both Ph + and Philadelphia negative (Ph-) patients, our analyses focused only on the Ph + patients, in order to more closely match the DASISION and ENESTnd study populations. An overview of the 3 pivotal trials is shown in Table 1 and Table A1.

Table 1. Overview of the pivotal trials included in the MAIC.

Characteristic	BFORE (bosutinib, n = 249)†	ENESTnd (nilotinib, n = 282)	DASISION (dasatinib, n = 259)
Daily drug dose	400 mg	600 mg*	100 mg
Study design	Phase 3, randomized, open-label, multinational^5	Phase 3, randomized, open-label, multicentre^3	Phase 3, randomized, open-label, multinational^4
Key inclusion criteria	Adults^5 Molecular diagnosis (detection of BCR-ABL rearrangement) of CP-CML within 6 months of randomization^5 Adequate hepatic and renal function^5 ECOG performance status of 0 or 1^5	Adults, within 6 months after the diagnosis of Ph + CP-CML by conventional cytogenetic analysis of bone marrow^3 ECOG of 0, 1, or 2^3 Adequate organ function^3	Adults with Ph + CP-CML diagnosed by means of bone marrow cytogenetic studies performed within 3 months before study entry^4 ECOG performance status of 0–2^4 Adequate hepatic and renal function^4
Primary endpoint	Major molecular response at 12 months^5	Major molecular response at 12 months^3	Confirmed complete cytogenetic response by 12 months^4
Secondary efficacy endpoints	Complete cytogenetic response by 12 months^5 Major molecular response by 18 months^5 Duration of major molecular response^5 Duration of complete cytogenetic response^5 Event-free survival^5 Overall survival^5	Durable major molecular response at 24 months^10 Time to and duration of major molecular response^10 Complete cytogenetic response by 1 month^3 (also time to and duration of this response)^29 Progression to accelerated phase or blast crisis^3 A ≥4 and ≥4.5 log reduction in BCR-ABL transcripts at 6, 12, and beyond 12 months (also time to and duration of this reduction)^29 Rate of hematologic response^Citation29 Progression-free survival ≤5 years of follow-up^29 Event-free survival ≤5 years of follow-up^29 Overall survival ≤5 years of follow-up^29	Time in confirmed complete cytogenetic response overall^11 Major molecular response at any time^4 Time to confirmed complete cytogenetic response overall^4 Time to a major molecular response overall^4 Progression-free survival^4 Overall survival^4

*Based on a 300 mg twice daily dose.

†The molecular diagnosis (detection of BCR-ABL rearrangement) inclusion criterion from the BFORE study pertains to all the N = 268 randomized patients. However, only Ph + patients from the BFORE study were included in the current analysis.

Abbreviations. CP-CML, chronic-phase chronic myeloid leukemia; ECOG, Eastern Cooperative Oncology Group; Ph+, Philadelphia chromosome positive.

Outcomes assessed

Outcomes assessed in the BFORE study included the primary endpoint major molecular response (MMR) at 12 months corresponding to a ≥ 3 log reduction from standardized baseline [i.e. ≤0.1% BCR-ABL1 IS], and secondary and exploratory endpoints including complete cytogenetic response (CCyR) by 12 months, deep molecular responses, MR4 corresponding to a ≥ 4 log reduction in BCR-ABL transcripts [i.e. ≤0.01% BCR-ABL1 IS], and MR4.5 corresponding to a ≥ 4.5 log reduction [i.e. ≤0.0032% BCR-ABL1 IS], and progression to accelerated phase (AP) or blast phase (BP) (on-treatment disease transformation) at any time. Endpoints assessed in the ENESTnd trial included the primary endpoint MMR at 12 months and several secondary endpoints, such as, CCyR by 12 months, MR4 and MR4.5 rates beyond 12 months, and disease transformation. The primary endpoint of the DASISION trial was confirmed CCyR by 12 months, secondary and exploratory endpoints included CCyR and MMR by 12 months and disease transformation. For a complete list of all primary and secondary efficacy endpoints included in the 3 trials, refer to Table 1.

Since several efficacy endpoints were assessed in all three clinical trials the MAIC analyses assessed and compared MMR, MR4, MR4.5 at and by 24 months, CCyR by 24 months, and on-treatment disease transformation at any time.

Molecular data for all three studies was centrally reviewed by Molecular MD and assessed according to the international scale. Cytogenetic data was performed locally for all three studies, however CCyR was imputed from MMR in BFORE if no valid cytogenetic assessment was available, whereas CCyR for the other two studies was based solely on bone marrow cytogenetic data.

Overall survival (OS), progression-free survival (PFS) and event-free survival (EFS) were not compared as the focus of our analysis was on short-term endpoints since only 24-month data was available for the BFORE study and progressions events and deaths are expected to be infrequent by this timepoint. In addition, endpoint definitions for PFS and EFS differed across studies.

Statistical methods

Using logistic regressions, variables accounting for patient baseline characteristics were tested for effect modification and prognostic factors for each outcome assessed. Variables accounting for either effect modifiers or prognostic factors that were statistically significant at the 20% level (p<.20) were included in the MAIC^18,19. Table 2 shows a list of the covariates included in the MAIC for each endpoint.

Table 2. Covariates included in the analyses per endpoint.

Endpoint	Comparison with nilotinib		Comparison with dasatinib
	Effect modifier	Prognostic factor	Effect modifier	Prognostic factor
MMR at 24 months	Gender, race, Sokal score	Race, Sokal score	–	–
MMR by 24 months	Gender	Race, Sokal score, ECOG	Gender	Sokal/Hasford score, race, ECOG, peripheral blood basophils
MR4 by 24 months	White-cell count	–	–	–
MR4 at 24 months	Race, platelet count, white-cell count	Sokal score	–	–
MR4.5 by 24 months	Race	Age, race, Sokal score, platelet count, white-cell count	Peripheral blood basophils, race	Age, race, Sokal/Hasford score, platelet count, white-cell count
MR4.5 at 24 months	Race	Age, Sokal score, platelet count, white-cell count	–	–
CCyR by 24 months	Gender, race, white-cell count, ECOG	Race, ECOG	Gender, ECOG, white-cell count, race	ECOG, race, peripheral blood basophils
Disease transformation at any time	Gender	Race, Sokal score, white-cell count, ECOG	Gender	Sokal/Hasford score, race, ECOG, white-cell count, peripheral blood basophils

Abbreviations. MMR, major molecular response; MR, deep molecular response; CCyR, complete cytogenetic response; ECOG, Eastern Cooperative Oncology Group.

The ENESTnd and BFORE studies used the Sokal score to measure patient disease risk, whereas the DASISION study used the Hasford scale. Both scales have the same categories (low, intermediate and high) and for the purpose of our analyses were assumed to be comparable.

Unanchored MAICs comparing bosutinib with nilotinib and dasatinib were performed to match the summary statistics in the BFORE trial to the reported summary statistics in the ENESTnd (nilotinib) and the DASISION (dasatinib) trials^20,21. After the matching, population sample size was estimated using the effective sample size (ESS), and comparative efficacy of the TKIs was assessed using odds ratios (OR) and their corresponding 95% confidence intervals (CIs). CIs excluding 1 were considered statistically significant. No adjustments for multiple comparisons were made.

Results

Patients

Before the MAIC, there were some differences in the cross-trial distribution of the variables accounting for baseline characteristics, particularly for age, race, risk score, ECOG performance status, white blood cell count, and platelet count. Patients in the BFORE trial were older (median age 53) compared with patients in the ENESTnd and DASISION trials (median age 47 and 46 years, respectively). The proportion of Asian patients was higher in the ENESTnd and in the DASISION trials (27 and 42% respectively) than in the BFORE study (12%), which, in turn, had a higher proportion of white patients (78%) compared with the other two studies (60% and 51% respectively). Overall, patients in the BFORE trial had a lower disease risk score; the percentage of patients having high risk disease was higher in ENESTnd (28%) than in BFORE (21%), while DASISION had a higher percentage of patients with intermediate risk disease (48%), compared with BFORE (43%). The percentage of patients with a better ECOG performance status was higher in both the ENESTnd and the DASISION studies compared with the BFORE study: 88 and 82% respectively versus 71% had an ECOG performance status of 0, while 12 and 18%, respectively versus 29% had an ECOG performance status of 1 in the ENESTnd, DASISION and BFORE studies, respectively. The median white blood cell count (10^9/L) was lower in both pivotal trials for nilotinib and dasatinib (23 and 25, respectively), compared with the BFORE study (41). The median platelet count (10^9/L) of patients in the pivotal trials for nilotinib and dasatinib (424 and 448, respectively), was higher than in patients in the BFORE study (386).

After the MAIC, the distribution of the variables accounting for baseline characteristics became balanced, making the two study populations identical from a statistical perspective (e.g. bosutinib with nilotinib, and bosutinib with dasatinib). Table 3 shows a detailed description of the summary statistics of the variables accounting for baseline characteristics before and after the reweighting. For bosutinib, the summary statistics for two datasets are shown, the dataset used in the MAIC with nilotinib and the one used in the MAIC with dasatinib. The two MAICs were performed using different variables and due to their reporting, differences in summary statistics occur.

Table 3. Summary statistics of baseline characteristics before and after the MAIC.

Variables	Bosutinib before the MAIC in the comparison with nilotinib/dasatinib N = 248/244	Nilotinib N = 282	Dasatinib N = 259	Bosutinib after the MAIC in the comparison with nilotinib/dasatinib ESS = 142–206/100–118
Median age (range), years	53 (18-84)/53 (18-84)	47 (18–85)	46 (18–84)	47/46
Male, %	58/57	56%	56%	56%/56%
Race, %
Asian	12%/12%	27%	42%	27%/42%
Black	4%/4%	4%	1%	4%/1%
White	78%/78%	60%	51%	60%/51%
Other	6%/6%	9%	6%	9%/6%
Sokal risk score/Hasford risk score*, %
Low	35%/35%	37%	33%	37%/33%
Intermediate	44%/44%	36%	48%	36%/48%
High	21%/21%	28%	19%	27%/19%
ECOG performance status, %
0	71%/71%	88%	82%	88%/82%
1	29%/29%	12%	18%	12%/18%
Median platelet count (range), 10^9/L	386 (72–2195)/385.5 (72–2195)	424 (90–3880)	448 (58–1880)	424/448
Median white blood cell count (range), 10^9/L	41 (1.5–525)/41 (1.5–525)	23 (2–247)	25.1 (2.5–493)	23/25.1
Median peripheral blood basophils (range), %	NR/4 (0–22)	NR	4 (0.27–8)	NR/4%

Peripheral blood blasts were not reported for nilotinib and peripheral blood blasts were not included as a covariate in the dasatinib analysis since the exact baseline value was not always known for patients in the BFORE trial (missing values were confirmed to be <15% [criteria for CP CML diagnosis] as the sum of the remaining differentials were >85% or bone marrow blasts were <15%).

*In the ENESTnd and the BFORE studies disease risk is reported using the Sokal score, while in the DASISION study it is reported using the Hasford score. The two scales have similar categories (low, intermediate and high), but are not entirely comparable. However, for the purpose of the current analysis, the two risk scores were assumed to be comparable.

Abbreviations. ECOG, Eastern Cooperative Oncology Group; NR, not reported; MAIC, matching-adjusted indirect comparison; ESS, effective sample size.

Individual data on patient outcomes of 249 newly diagnosed Ph + CP-CML patients on bosutinib was included in the reweighing and the comparison with the aggregated study-level outcomes of 282 nilotinib CP-CML patients and 259 dasatinib CP-CML patients after a minimum of 24 months of follow-up. The median treatment duration was approximately 25 months for each study. From the 249 bosutinib patients initially considered for the analyses, only 248 and 244 patients were included in the MAIC comparing bosutinib with nilotinib and dasatinib, respectively. The sample size was slightly diminished due to the exclusion of patients with missing values in any of the variables accounting for baseline characteristics that met the MAIC inclusion criteria (statistical significance at the 20% level). Depending on the assessed outcome, the resulting ESS varied between 142 and 206 in the comparison with nilotinib and between 100 and 118 in the comparison with dasatinib.

Nilotinib compared with bosutinib

Comparing nilotinib versus bosutinib in terms of MMR at and by 24 months after the re-weighting resulted in an OR of 1.01 (CI 0.70–1.45) and an OR of 1.14 (CI 0.77–1.69), respectively; these differences were not statistically significant. In terms of MR4 at 24 months, when nilotinib was compared with bosutinib following the MAIC, a statistically significant OR of 0.57 (CI 0.38–0.84) in favor of bosutinib was identified, however a non-significant OR of 1.02 (CI 0.71–1.47) was observed for MR4 by 24 months. Non-significant differences were found for MR4.5 at 24 months, with an OR of 0.81 (CI 0.48–1.39), and by 24 months, with an OR of 0.96 (CI 0.63–1.47) respectively. Similarly, there was not a significant difference for CCyR by 24 months with an OR of 1.54 (CI 0.93–2.56). Although on-treatment disease transformation at any time was numerically higher for bosutinib after the MAIC (2.6 vs 0.7% with bosutinib and nilotinib, respectively), this finding was not statistically significant with an OR of 0.26 (CI 0.05–1.41).

Table 4 shows the results for the MAIC between nilotinib and bosutinib.

Table 4. Outcomes of the MAIC for the comparison of nilotinib versus bosutinib.

Outcome	Bosutinib before MAIC	Bosutinib after MAIC	Nilotinib (%)	OR (95% CI)
MMR at 24 months	60.9% (n = 248)*	61.5% (ESS = 190)**	61.7	1.01 (0.70–1.45)
MMR by 24 months	66.9% (n = 248)*	68.3% (ESS = 167)**	71	1.14 (0.77–1.69)
MR4 by 24 months	37.9% (n = 248)*	38.4% (ESS = 206)**	39	1.02 (0.71–1.47)
MR4 at 24 months	33.0% (n = 248)*	36.4% (ESS = 156)**	24.5	0.57 (0.38–0.84)
MR4.5 by 24 months	23.4% (n = 248)*	25.7% (ESS = 149)**	25	0.96 (0.63–1.47)
MR4.5 at 24 months	12.5% (n = 248)*	14.8% (ESS = 149)**	12.4	0.81 (0.48–1.39)
CCyR† by 24 months	81.9% (n = 248)*	81.2% (ESS = 151)**	86.9	1.54 (0.93–2.56)
Disease transformation at any time	2.4% (n = 248)*	2.6% (ESS = 142)**	0.7	0.26 (0.05–1.41)

*Based on the dataset used for conducting the MAIC between bosutinib and nilotinib (patients with missing data were excluded).

**Effective sample size (the actual sample size of the bosutinib dataset after the MAIC).

†CCyR was imputed from MMR in the BFORE study if no valid cytogenetic assessment was available.

Abbreviations. MMR, major molecular response; MR, deep molecular response; CCyR, complete cytogenetic response; MAIC, matching-adjusted indirect comparison; ESS, effective sample size.

Dasatinib compared with bosutinib

Comparing dasatinib versus bosutinib in terms of MMR by 24 months after the re-weighting resulted in a non-significant OR of 0.81 (CI 0.54–1.22). In terms of MR4.5 by 24 months, when dasatinib was compared with bosutinib, a statistically significant OR of 0.56 (CI 0.35–0.90) in favor of bosutinib was identified. A non-significant difference was found for CCyR by 24 months, with an OR of 1.27 (CI 0.74–2.17). For on-treatment disease transformation at any time, despite a numerically higher rate for bosutinib after the MAIC (3.6 vs 2.3% with bosutinib and dasatinib respectively), the results showed a non-significant OR of 0.63 (CI 0.20–2.00).

Table 5 shows the results for the MAIC between dasatinib and bosutinib.

Table 5. Outcomes of the MAIC for the comparison of dasatinib versus bosutinib.

Outcome	Bosutinib before MAIC	Bosutinib after MAIC	Dasatinib (%)	OR (95% CI)
MMR by 24 months	66.8% (n = 244)*	68.7% (ESS = 118)**	64	0.81 (0.54–1.22)
MR4.5 by 24 months	23.8% (n = 244)*	26.8% (ESS = 100)**	17	0.56 (0.35–0.90)
CCyR† by 24 months	81.6% (n = 244)*	82.9% (ESS = 110)**	86	1.27 (0.74–2.17)
Disease transformation at any time	2.5% (n = 244)*	3.6% (ESS = 110)**	2.3	0.63 (0.20–2.00)

*Based on the dataset used for conducting the MAIC between bosutinib and dasatinib (patients with missing data were excluded).

**Effective sample size (the actual sample size of the bosutinib dataset after the MAIC).

†CCyR was imputed from MMR in the BFORE study if no valid cytogenetic assessment was available.

Abbreviations. MMR, major molecular response; MR, deep molecular response; CCyR, complete cytogenetic response; MAIC, matching-adjusted indirect comparison; ESS, effective sample size.

Discussion

The pivotal trials for each of the second generation TKIs demonstrated superiority to imatinib in terms of efficacy when examining short-term endpoints such as MMR. However, to date, there have been no head-to-head trials comparing any of the second generation TKIs to each other. In the absence of such trials, we conducted an indirect comparison (via unanchored MAICs) in order to evaluate the relative efficacy between the second generation TKIs bosutinib, dasatinib, and nilotinib.

Two recently conducted NMAs^6,8 reported that, based on SUCRA values, nilotinib tended to have the highest probability of being ranked first for several endpoints such as MMR in 1 L CP-CML. However, these results are likely to be biased due to the violation of one of the general assumptions underlying NMAs: similarity of trials. The cross-trial differences in dose escalation patterns between the three pivotal clinical studies (ENESTnd, DASISION, and BFORE), results in differences in the trial conduct that cannot be corrected for with traditional NMAs. Since trial design matters and differences in control arms, e.g. dose escalation, cannot be corrected for using NMAs, unanchored MAICs are the preferred method, from a technical perspective, to indirectly compare nilotinib and dasatinib with bosutinib in 1L CP-CML. In that sense, an MAIC published in 2011²² compared the relative efficacy of nilotinib (ENESTnd) with dasatinib (DASISION) by month 12, however it did not include bosutinib in the analyses. Similarly, another MAIC published in 2015¹⁵ compared the molecular responses by 12 and 48 months for the same TKIs but did not include bosutinib either.

The aim of these analyses was to compare the efficacy in terms of response and transformation rates of bosutinib, nilotinib, and dasatinib in newly diagnosed Ph + CP-CML patients by means of unanchored MAICs using the available 24-month data. We chose to conduct unanchored MAICs for several reasons: (1) access to full individual patient level data (IPD) was available for one trial (BFORE for bosutinib) and published, aggregate-level data was available for the other trials (ENESTnd for nilotinib and DASISION for dasatinib), (2) as opposed to a naïve comparison, MAICs can adjust for disease- and patient-level cross-trial heterogeneity, making the corresponding outcomes more reliable to compare, and, (3) the differences in imatinib dose escalation over time due to changing treatment practices could bias the results of an NMA or an anchored MAIC. Because of these factors, the unanchored MAIC methodology was considered the most appropriate and robust approach for making indirect comparisons between bosutinib, dasatinib, and nilotinib.

The results of our MAIC analyses showed that aside from bosutinib being statistically significantly better for MR4 at 24 months compared to nilotinib and statistically significantly better for MR4.5 by 24 months compared to dasatinib, all other comparisons between the treatments were not statistically significantly different. Bosutinib therefore demonstrated comparable effectiveness to nilotinib and dasatinib in the first-line CP-CML setting.

These statistical findings are similar to what has been observed in clinical practice, where the three second generation TKIs are considered similar in terms of efficacy but differ in their safety profiles. Clinicians therefore can tailor their choice of treatment to the individual needs of the specific patient. The results of the current report are similar to those from a previous MAIC confirming a similarity in the efficacy of short-term endpoints for bosutinib, dasatinib and nilotinib in second-line CP-CML²³. These analyses showed that when dasatinib was compared with bosutinib in terms of MCyR, the result was a non-significant OR of 0.78 (95% confidence interval 0.53–1.16). A comparison between nilotinib and bosutinib resulted in a non-significant OR for MCyR of 0.98 (95% confidence interval 0.71–1.35). Thus overall, for both 1 and 2L CP-CML, bosutinb, dasatinib, and nilotinib can be considered largely comparable in terms of efficacy as demonstrated through the use of MAICs.

The six patients in the BFORE trial who experienced disease transformation were counted in the disease progression analysis, however, three cannot be regarded as true progressions, as these patients transformed within 2 weeks after randomization and based solely on basophil count. One subject achieved MMR after transformation and none of the three subjects discontinued treatment due to disease progression to AP/BP CML or death. It was therefore considered that these events do not appear to be true transformation events, since their clinical courses are not consistent with AP/BP. This could explain the higher naïve rate of disease transformation for bosutinib patients compared with the nilotinib population. In addition, for nilotinib, only data collected and reported for progression events after discontinuation of treatment were used in the analysis of progression to accelerated or blast phase on treatment if collected within 14 days of treatment discontinuation. Despite this, after the MAIC there were no statistically significant differences between nilotinib and bosutinib in terms of disease transformation.

Although OS was also considered for these analyses, the currently available 24-month data from the BFORE trial is not mature enough to draw a conclusion about treatment differentiation for this outcome. Additionally, short term response outcomes were the primary endpoints in the pivotal studies of the assessed TKIs and are currently recommended for the efficacy assessment of therapies in CML. Short term remission endpoints are preferred over the classical OS endpoint because potential survival improvements will not be observed for a long time given the substantial improvement in OS achieved after the approval of TKIs²⁴. In addition, specific short-term endpoints such as the achievement of CCyR or MMR have demonstrated to have prognostic information for longer term outcomes^25,26. Therefore, we believe the comparison of short-term efficacy endpoints is more appropriate for this particular analysis.

There are a few limitations to this study. CCyR was imputed from MMR in the BFORE trial, and not in the other two studies. The rationale for this change is that the BFORE study was initiated approximately seven years after the ENESTnd and DASISION trials, and during that time the relevance of disease monitoring through cytogenetic testing decreased, while the focus of response assessment switched to molecular monitoring. Another limitation is that the ENESTnd and BFORE studies used the Sokal score to measure patient disease risk, whereas the DASISION study used the Hasford scale. Although both scales have the same categories (low, intermediate, and high), these are not entirely comparable because the Hasford score identifies more low-risk patients and places fewer patients in the high-risk group than does the Sokal score²⁷.

A reliable comparison between treatments can only be accepted if both safety and efficacy data are assessed. However, most of the adverse events across the three pivotal trials (ENESTnd, DASISION, and BFORE) are not mutually reported. Similarly, potential imbalances in the adverse event reporting and the assessment of relation to study drug across trials preclude an appropriate comparison of the safety profile of the drugs. Additionally, a comparative analysis of the overall treatment discontinuation rates could not be conducted due to a lack of reporting by a specific time point. Twenty-nine percent (29.7%) of patients in BFORE discontinued treatment at any time after 24 months of follow-up, however, only 26.1% discontinued by month 24.

The reported discontinuation rates in ENESTnd and DASISION include patients who discontinued in their third year of follow-up; because patients are enrolled at different points in time in each trial and given the potential differences in the number of discontinuations over time, making such a comparison would be unreliable.

The general MAIC limitations apply to our analyses as well. Access restricted to marginal covariate distribution means that only observed differences between the trials could be corrected for and not the unobserved ones (e.g. disease and baseline characteristics that were not reported in the clinical study publications). Additionally, having access to only marginal covariate distributions (medians/proportions of baseline characteristics) leads to the additional assumption that the covariates in the comparing trials are identically correlated, which is unlikely to occur.

Furthermore, the scale on which MAICs assume additivity of treatment effects is the natural outcome scale (e.g. proportion) and this comes with certain limitations, as opposed to the more stable transformed scales, such as a logit scale. By matching the baseline characteristics of the bosutinib patients with those of the nilotinib/dasatinib patients, it is implicitly assumed that the latter patients (i.e. from the nilotinib/dasatinib trials) represent the target population. However, for a more objective comparison, the target population should be one in national cohorts or registry studies of CP-CML^20,21.

While a randomized clinical trial is always the preferred approach to evaluate different treatment options, in the absence of head-to-head trials, an indirect comparison can support treatment decisions in routine clinical practice. All three second generation TKIs have demonstrated superiority over imatinib, and after an MAIC adjustment they appear to have similar efficacy. In routine clinical practice, although the ultimate objective of CML treatment is still an optimal survival, recent advances have switched the treatment paradigm to seeking to achieve survival coupled with an optimal quality of life and minimum long-term toxicities.

The choice of therapy is therefore determined not only by the efficacy of the drugs but also by additional considerations such as the overall tolerability, the specific safety profile and the cost-effectiveness of each TKI²⁸.

Conclusions

The use of unanchored MAIC analyses have shown that bosutinib might be similarly effective to nilotinib and dasatinib for the treatment of newly diagnosed patients with CP-CML. Due to the strong assumptions underlying unanchored MAICs, these results should be interpreted with caution. However, in the absence (and the prospect thereof) of a head-to-head comparison clearly demonstrating the superiority of one second generation TKI over the others, the choice of TKI should be personalized and other factors such as the safety profile of the TKIs, the individual patient characteristics and comorbidities^25,28, and the impact on patient health-related quality of life should be taken into account. These findings should be complemented by additional analyses comparing safety outcomes and confirmed by further clinical research.

Transparency

Declaration of funding

This study was sponsored by Pfizer. This publication reports the results of a Pfizer-sponsored non-interventional study.

Declaration of financial/other relationships

BM has disclosed that, at the time of this research, he was an employee of Ingress-health BV, which received financial support from Pfizer in connection with the conduct of this study and development of this manuscript. In his salaried position, he has worked with a variety of companies and organizations and was precluded from receiving payment or honoraria directly from these organizations for services rendered. BH has disclosed that he is an employee of Ingress-health BV and an equity holder of Ingress-health BV, which received financial support from Pfizer in connection with the conduct of this study and development of this manuscript. CM, JCC, EL and AV have disclosed that they are employees and shareholders of Pfizer Inc. Peer reviewers on this manuscript have received an honorarium from CMRO for their review work but have no other relevant financial relationships to disclose.

Author contributions

BM, CM, JCC, EL, AV, BH: conception, design, analysis and interpretation of the data, drafting of the paper and revising it critically for intellectual content. All authors agree to be accountable for all aspects of the work.

Acknowledgements

Not applicable.

Appendix

 

 

Table A1. Specific inclusion/exclusion criteria*.

Characteristic	BFORE (bosutinib, n = 249)	ENESTnd (nilotinib, n = 282)	DASISION (dasatinib, n = 259)
Daily drug dose	400 mg	600 mg*	100 mg
Key specific inclusion criteria	Molecular diagnosis of CP CML of ≤ 6 months (from initial diagnosis) Diagnosis of CP CML with molecular confirmation by detection of BCR-ABL rearrangement at screening (cytogenetic assessment for Philadelphia chromosome was not required for enrollment) Philadelphia chromosome status was to be identified at screening. Both Philadelphia positive (Ph+) and Philadelphia negative (Ph-) patients could be included Eastern Cooperative Oncology Group Performance Status score of 0 or 1 Adequate hepatic, renal and pancreatic function	Chronic myelogenous leukemia in chronic phase patients within the first 6 months of diagnosis. Standard conventional cytogenetic analysis was to be done on bone marrow. FISH could not be used Diagnosis of chronic myelogenous leukemia in chronic phase with confirmation of Philadelphia chromosome of (9:22) translocations Eastern Cooperative Oncology Group Performance Status score of 0, 1, or 2 Adequate end organ function Patients must have had potassium, magnesium, phosphorus and total calcium (corrected for serum albumin ≥ LLN (lower limit of normal) prior to the first dose of study medication	Chronic phase, Philadelphia Chromosome-positive chronic myeloid leukemia (CML) Previously untreated chronic CML Ph + or variants must be demonstrated by bone marrow cytogenetics. Eastern Cooperative Oncology Group Performance Status score of 0–2 Adequate hepatic and renal function
Key specific exclusion criteria	Any prior medical treatment for CML, including tyrosine kinase inhibitors (TKIs), with the exception of hydroxyurea and/or anagrelide treatment, which were permitted for up to 6 months prior to study entry (signature of ICF) if suitably approved for use in the subject's region Any past or current central nervous system (CNS) involvement, including leptomeningeal leukemia. Extramedullary disease only Concomitant use of or need for medications known to prolong the QT interval Uncontrolled hypomagnesemia or uncorrected hypokalemia, due to potential effects on the QT interval History of clinically significant or uncontrolled cardiac disease Known seropositivity to human immunodeficiency virus (HIV), current acute or chronic hepatitis B (hepatitis B surface-antigen positive), hepatitis C, cirrhosis or evidence of decompensated liver disease. Patients with resolved Hepatitis B can be included Recent or ongoing clinically significant GI disorder, e.g. Crohn's Disease, Ulcerative Colitis, or prior total or partial gastrectomy	Patients who are considered pH – because they do not have a confirmed cytogenetic diagnosis of Philadelphia chromosome with (9:22) translocation Previously documented T315I mutation Treatment with a tyrosine kinase inhibitor prior to study entry was not allowed except for no more than 2 weeks in duration of imatinib Any medical treatment for CML prior to study entry for longer than 2 weeks with the exception of hydroxyurea and/or anagrelide Known cytopathologically confirmed CNS infiltration Impaired cardiac function Severe or uncontrolled medical conditions (i.e. uncontrolled diabetes, active or uncontrolled infection) History of significant congenital or acquired bleeding disorder unrelated to cancer Use of therapeutic coumarin derivatives (i.e. warfarin, acenocoumarol, phenprocoumon) Patients receiving treatment with any medications with the potential to prolong the QT interval History of acute pancreatitis within 1 year of study entry or past medical history of chronic pancreatitis Acute or chronic liver, pancreatic or severe renal disease considered unrelated to disease	Any other prior systemic treatments, with anti-CML activity except for anagrelide or hydroxyurea Known pleural effusion at baseline Uncontrolled or significant cardiovascular disease Significant bleeding disorder unrelated to CML Patients receiving drugs that are generally accepted to have a risk of causing Torsades de Pointes
Study locations†	North America, Europe, Asia, Africa, Australia	North America, South America, Europe, Africa, Asia	North America, South America, Europe, Asia, Australia

*Source: Study protocols^29–31.

†Source: clinicaltrials.gov.