The short term clinical efficacy of imatinib, dasatinib and nilotinib has been demonstrated in large, well designed randomised trials [6, 12, 13]. Long term data from the IRIS clinical trial is strongly supportive of the argument that individuals who receive first line imatinib will live many years longer than those who receive interferon based therapy [7, 9]. Due to the causal link between prognosis and response to treatment it is highly likely, given the significant differences observed in complete cytogenetic and major molecular response rates for second generation compared to first generation TKIs, that a similar pattern will emerge when long term data from DASISION and ENESTnd are published.
Despite the positive clinical profile of dasatinib and nilotinib in the treatment of newly diagnosed CML, reimbursement agencies such as NICE and CADTH will seek to explore the cost and benefit trade-offs associated with the use of these products. From a reimbursement perspective, poor or non-existent evidence is not a reason for delaying or avoiding a decision, and so the use of indirect comparison meta-analysis is becoming more common in order to generate the relative efficacy estimates required for a successful health technology assessment in the absence of head to head data.
For all comparisons of dasatinib and nilotinib to imatinib included in the meta-analysis, the derived results are in line with those reported in the actual clinical studies: dasatinib and nilotinib are clinically superior to imatinib in terms of cytogenetic and major molecular response at a range of time points. However, the key value of this paper is in relation to the comparison of dasatinib to nilotinib, since no head to head trial data currently exists. We found no significant difference between the two treatment options in terms of either CCyR or MMR at any of the time points where enough data was available to inform the meta-analysis. Hence, on the basis of the current literature, from the perspective of both clinicians and reimbursement agencies the two products should be viewed as equally efficacious.
To the best of our knowledge, this is the first study to use the combination of a full and thorough systematic review of the literature and subsequent meta-analysis incorporating all available evidence rather than information from single trials (e.g. DASISION and ENESTnd). One other study has been published that attempted to generate relative efficacy estimates for dasatinib and nilotinib via the alteration of the results from one study via a matching algorithm . This approach neglects the body of evidence in other trials – in particular the S0325 Intergroup study  – and its acceptability within the reimbursement agencies or with other meta-analysts is untested. Hence, the results from the study presented in this paper are the most robust estimates currently available to both clinicians and reimbursement decision makers.
As noted previously, indirect comparison meta-analysis respects randomisation in all included trials, uses all available evidence (both indirect and head to head) and, assuming the trial populations are broadly homogeneous and endpoints consistently defined, generates results that are consistent with conventional meta-analyses when the amount of head to head evidence is sufficient to allow both approaches to be used. Further, the International Society for Pharmacoeconomics and Outcomes Research (ISPOR) set up a task force to generate ‘best practice’ guidelines to be used in conducting and reporting an indirect comparison meta-analysis. These guidelines have been published and the meta-analysis presented here was conducted to meet the standard laid out in this document .
Despite the above mentioned strengths of the analysis, there were a number of limitations. The primary concern is that while the predicted 12 month CCyR values are similar to the values in the ENESTnd study (reported: 80%/78%/65% for nilotinib 600 mg/800 mg and imatinib 400 mg: derived 77.7%/75.7%/62.4% respectively), the predicted value for dasatinib (77.1%) is lower than observed in either of the two identified studies (DASISION: 83.4%, S0325: 82.0%). Given that the absolute highest response rates of all included imatinib studies were in these two studies, when the mean baseline imatinib probability and relative efficacy estimates from the meta-analysis are combined, the result was inevitable. Hence the studies included may be heterogeneous, although no subgroup specific analyses were done to explore this topic. Further work via statistical comparisons of inclusion criteria, baseline patient characteristics, etc. is required to test this hypothesis although without access to patient level data from multiple pharmaceutical companies such analyses may not be feasible.
Secondary concerns relate to the immaturity of the database and to the limited number of studies identified. This impacted on the project in two ways: i) through the large amount of uncertainty surrounding the derived results and ii) through the inability to generate meaningful evidence networks for other clinically relevant outcome measures (e.g. partial cytogenetic response, progression free survival or overall survival). The majority of these problems will be lessened over time with the publication of either new studies or more information from existing ones. However, given the target population and scarcity of events it may never be possible to differentiate the survival profiles (both progression free and overall) of the two second generation products without a head to head trial.
Thirdly, only one manufacturer (Bristol-Myers Squibb) had the opportunity to provide additional material to the analysis, which could be perceived as non-transparent and a bias in favour of dasatinib. As mentioned in the results section, only material available in the public domain was used in the analysis and also that BMS provided information on Nilotinib related studies. Hence, the review maintained its transparency and no bias existed.
It is important to note that the results were generated using aggregate level trial data and hence should not be interpreted as answering the questions ‘who should get dasatinib?’ and ‘who should get nilotinib?’ Such questions would require either a meta-regression or re-running of the analysis using subgroup specific data.