Mylotarg has potent anti-leukaemic effect: a systematic review and meta-analysis of anti-CD33 antibody treatment in acute myeloid leukaemia

The search yielded 1974 articles (Fig. 1). A search of online clinical trial databases suggested four pending trials. One trial was not due for completion until January 2020 (AML SG09-09; ClinicalTrials.gov identifier: NCT00893399) and would fit our inclusion criteria. Another, NCRI AML17 (ISRCTN55675535), is due to be completed July 2014. The other two were sponsored by pharmaceutical companies; one was terminated early, and the second did not have available data (further supplementary 5).

Two different anti-CD33 therapies were identified in this systematic review: GO, a humanised monoclonal antibody against CD33 conjugated to a toxin, and lintuzumab, a humanised anti-CD33 monoclonal antibody not conjugated to a toxin. Two trials involving lintuzumab were found [12, 13] but only one provided sufficient data for analysis of overall survival; this study [13] did not demonstrate a significant impact of lintuzumab on overall survival (HR = 0.94, 95 % CI = 0.63–1.11); however, this study was small and underpowered. Due to their intrinsic biological differences, these studies were omitted from the main meta-analyses that only involved GO. Thus, 11 trials involving 13 randomisations with 7138 patients entered the meta-analysis of trials involving GO [12‐26]. A detailed summary of trials included in this meta-analysis can be found in the supplementary data (further supplementary 4), with a brief summary of each study described in Table 1. We also excluded two small trials from the meta-analysis because they did not include the same concomitant chemotherapy drugs in the GO and the comparator arm [19, 21]. Out of the 11 trials, two had a second independent randomisation at a latter treatment stage [17, 25]. Four studies investigated the use of GO at induction and consolidation phases but with no independent randomisation at the second treatment stage; therefore, these studies contributed to four randomisations only [18, 20, 26, 27]. Three further randomisations involved the use of GO at induction [16, 17, 25], two independent randomisations at consolidation [17, 24] and three at a post-consolidation/maintenance stage [22, 23, 25]. One trial used GO as part of a low-intensity regimen [15]. Up-to-date overall survival and relapse data for the NCRI AML15 trial was used from the AML16 publication [16]. For one study [25], CIR and DCR were calculated from an interim publication [28]; data from this interim analysis was used for the induction randomisation of this study.

Overall, most publications failed to describe the randomisation and allocation method; without this information, we are unable to assess whether these trials suffered from selection bias. Otherwise, the risk of bias was low for the published trials (supplementary table 1).

Two trials [26, 27] were only available in conference proceedings form which limited assessment of risk of bias and data extraction. Of these, one trial [27] had available data for all outcomes, although assumptions were made for resistant disease and induction death (see further supplementary 3). The other trial [26] only provided data for OS and complete remission (CR) rates. Although data was provided for early deaths, it was unclear whether these deaths were induction related or due to disease. Authors of two unpublished trials [26, 27] were contacted to obtain full trial reports, but additional data was not available for analysis. In comparison, eight of the nine published trials [15‐18, 20, 22‐25] involving GO provided complete OS data. The exception was the post-consolidation randomisation of one trial [25]. Out of the nine published trials involving GO, one did not provide RFS rates for a separate consolidation randomisation [17] (further supplementary 2).

All induction phase trials had sufficient data for analysis of CR rates. One non-intensive trial was omitted from this analysis because of the different treatment strategy [15]. Overall response (OR) rates (CR plus CR with incomplete count recovery (CRi) or platelet recovery (CRp)) were also calculated; two trial results published as abstract form did not have sufficient data for this analysis [26, 27]. GO does not have a measurable impact on rates of OR, CR nor CRi/p (Fig. 2 and supplementary figure 1). However, the addition of GO to treatment significantly reduced the rate of resistant disease (failure to eliminate disease) by 23 % (HR = 0.77, 95 % CI = 0.67–0.90, p = 0.0009) (Fig. 2). However, this was at the expense of a statistically significant increase in induction deaths (HR = 1.30, 95 % CI = 1.04–1.63, p = 0.02) (Fig. 2).

GO improved CIR by 14 % (HR = 0.86, 95 % CI = 0.79–0.93, p = 0.0002) (Fig. 3). This is principally driven by its use in induction treatment where the use of GO reduced the rate of relapse by 19 % (HR = 0.81, 95 % CI = 0.74–0.90, p = 0.00003; test for heterogeneity between subtotals p = 0.07). GO had no significant influence on the rates of death in CR analysis (HR = 1.11, 95 % CI = 0.91–1.36, p = 0.3) (supplementary figure 2). Thus, the improvements in CIR correlated with an overall improvement in RFS in the overall analysis (HR = 0.90, 95 % CI = 0.84–0.98, p = 0.01) (Fig. 4).

GO appears to be beneficial in RFS when used as part of remission induction therapy in a subgroup analysis based on treatment stage (HR = 0.86, 95 % CI = 0.79–0.94, p = 0.0007). Although the test for heterogeneity between subtotals was borderline (p = 0.1), there is a substantial treatment effect seen in induction subgroup. Alongside the significant result for CIR in the setting of remission induction treatment, we explored this subgroup through predefined analyses. Subgroup analysis did not provide clear evidence that effect size of RFS was dependent on the age of patients (median age of trial entrants greater or less than 60), on the cumulative dose (greater or less than 9 mg/m²) they received and on whether there was treatment confounding (whether or not the comparative arm had the same dose of accompanying chemotherapy) (test for heterogeneity between subtotals = not significant) (supplementary figures 3–5).

GO had no significant effect on OS (HR = 0.96, 95 % CI = 0.90–1.02, p = 0.2) (Fig. 5), nor was there any evidence for a benefit of GO at any particular treatment stage. However, AML is a heterogeneous condition clinically and by genetic groupings. These analyses were limited by the number of trials involved. However, in one subgroup, there was evidence of benefit from the use of GO in terms of OS: this was in patients with favourable risk cytogenetics (HR = 0.46, 95 % CI = 0.29–0.73, p = 0.001) (test for trend between subtotals p = 0.02; test for heterogeneity between subtotals p = 0.02) (Fig. 6).

GO did not improve OS, in subgroup analyses of induction trials based on age, cumulative dose nor diagnosis (primary vs. secondary AML) (supplementary figures 6–8). Available data for FLT3-ITD and NPM1 analysis was limited to only three trials and was unable to detect a significant influence of these subgroups on OS (data not shown). A subgroup analysis based on treatment confounding was performed (supplementary figure 9): there was some evidence of greater benefit for the use of GO in trials without confounding chemotherapy (test for heterogeneity p = 0.09). There is no evidence that CD33 positivity influences OS of patients treated with GO (supplementary figure 10).

We investigated in detail the increased reports of liver toxicity and incidences of severe veno-occlusive disease (VOD) [29], particularly in the setting of haematopoietic stem cell transplantation [30]. This is displayed in the supplementary data (supplementary table 2), alongside measures of treatment-related fatality. Data was available from four trials using induction treatment (with some coupled with consolidation stage), involving a total of 789 patients [18, 20, 25, 26]. Twelve cases of VOD were reported; however, there was insufficient data available to report the incidence in the control arm. From these trials, there were five fatalities associated with VOD or severe hepatic toxicity when GO was used. Three trials [16, 17, 27] did not specify the number of cases of VOD in the GO arm but declared there were no difference incidence by comparison of study arm. As described above, there was a significant difference in death during induction (Fig. 2 which was extracted from endpoints defined directly as induction deaths [16‐18, 20, 25, 27], with one omitted due to data provided only of early deaths [26]).