The effectiveness and cost-effectiveness of erythropoiesis-stimulating agents (epoetin and darbepoetin) for treating cancer treatment-induced anaemia (including review of technology appraisal no. 142): a systematic review and economic model

Random allocation

The method of random allocation was clearly stated and sufficient in nine trials,17,48,53,62,67,69,72,74,77 whereas 14 trials50–52,63–66,68,70,71,73,75,76,78–80 did not specify the method used.

Concealment of allocation

The method of concealment of allocation was not clearly reported in any of the included trials. Fourteen trials17,50–52,63–65,67,68,70–72,74,78–80 did not report any information on allocation concealment, whereas eight trials48,53,62,66,69,73,75,76 provided some information. A centralised system for randomisation was reported in seven trials53,62,66,69,73,75,76 and authors of one trial48 stated that only the person administering study medication was unblinded. It is therefore possible that the allocation sequence was concealed in these eight trials. However, as no specific details on allocation concealment were reported, this remains unclear.

Treatment allocation masked from participants

Participants were blinded to treatment allocation in 12 trials. 17,48,50,53,63,66,69–71,73,74,77,79 Ten trials51,52,64,65,67,68,72,75,76,78,80 did not blind participants from treatment allocation and one trial62 did not report any information about blinding participants to treatment allocation.

Treatment allocation masked from clinicians

The 12 trials17,48,50,53,63,66,69,70,71,73,74,77,79 that blinded participants to treatment allocation also masked treatment allocation from clinicians. Eight trials51,64,65,68,72,75,76,78,80 did not blind clinicians to treatment allocation and three trials52,62,67 did not report any information about blinding of clinicians to treatment allocation; these three trials compared erythropoietin groups with standard care.

Reporting of losses to follow-up, withdrawals and dropouts

Losses to follow-up, withdrawals and dropouts were fully reported in nine trials48,52,62,66,68,70,72,76,77 and partially reported in 12 trials. 17,50,51,53,63,65,67,71,79,73–75,78,80 Among the 12 trials in which this information was partially reported, five trials17,50,71,74,78–80 reported withdrawals and dropouts until the end of the trials but did not provide any data on the follow-up period. Two trials64,69 did not report any information on losses to follow-up, withdrawals and dropouts.

Intention-to-treat analysis or < 10% of participants lost

Intention-to-treat (ITT) analysis or < 10% of participants lost was reported in 14 studies17,48,50,51,53,62,63–65,69,71,74,76,78–80 for all measured outcomes. ITT analysis or < 10% of participants lost was reported in seven studies52,66,67,70,73,75,77 for the primary outcome and most of the secondary outcomes. Only two trials68,72 did not use ITT analysis or reported ≥ 10% of participants lost.

Haemoglobin change

The mean Hb change was measured as a change in Hb level (g/dl) from baseline until the end of the treatment period. In total, 20 trials17,48,50,51,53,62–72,74,76–79 measured Hb change, of which 1617,48,50,51,53,63–67,69,70,74,77–79 were included in the meta-analysis. Two studies62,76 [the study by Moebus and colleagues62 was included as an abstract33 in the Cochrane review by Tonia and colleagues11) reported only the median change in Hb (g/dl) without any measure of variance and in two studies,68,72 no point estimates were reported and the results were presented graphically. These four studies were excluded from the analyses.

Overall, the analysis included 16 trials with 3170 participants. 17,48,50,51,53,63–67,69,70,74,77–79 Four trials were newly identified in the update searches. 48,74,77,78 As some trials with multiple experimental arms were split into subsets,48,63 the number of trials displayed is 18.

The random-effects meta-analysis demonstrated a statistically significant difference in Hb change in favour of treatment [weighted mean difference (WMD) 1.59 g/dl, 95% CI 1.33 to 1.84 g/dl; Figure 2]. Although all individual studies indicated a beneficial effect of ESAs with regard to Hb change and varied only in magnitude, there was statistically significant heterogeneity between the trials [I² = 75.9%, p < 0.001; χ² = 70.52, degrees of freedom (df) = 17; p < 0.01]. To assess whether publication bias was likely, a funnel plot was constructed (see Appendix 12). The funnel plot analysis did not show statistically significant asymmetry (p = 0.133). In addition, a meta-regression using publication year as a covariate (to assess the effect of publication year on Hb change) showed that the effects of ESA on Hb change were independent of any effect of publication year (p = 0.180); the meta-regression plot is presented in Appendix 12. The fixed-effects meta-analysis undertaken as a sensitivity analysis also showed a statistically significant difference in Hb change in favour of treatment (WMD 1.49 g/dl, 95% CI 1.37 to 1.60 g/dl; I² = 75.9%; p < 0.001); the forest plot of this analysis is provided in Appendix 12.

FIGURE 2.

Forest plot: Hb change overall (g/dl). Random-effects meta-analysis (DerSimonian–Laird). SD, standard deviation.

To identify sources of heterogeneity, subgroup analyses were conducted (Table 16). In addition, meta-regression models that included random effect and subgroup as covariates (to assess the effects of subgroups on Hb change) were performed; the F-statistics from these analyses are reported in Table 16. All covariates showing a significant effect (p < 0.05) in a univariate analysis were considered further in model selection.

Univariate analyses identified significant differences for one of the subgroups, ESA therapy [short-acting (erythropoietin) vs. long-acting (darbepoetin)] (p = 0.023; Figure 3). For subgroup analysis by erythropoietin treatment type, the short-acting ESA treatment (WMD 1.74 g/dl, 95% CI 1.49 to 2.00 g/dl; I² = 62.7%; p = 0.001) appeared to have a greater benefit than the long-acting ESA treatment (WMD 1.06 g/dl, 95% CI 0.61 to 1.52 g/dl; I² = 71.4%; p = 0.015). The results were also investigated visually. One small study69 (n = 35) appeared to differ from most of the other included trials; this study reported the highest mean difference between the ESA group and the control group. Excluding this study from the meta-analysis did not change the overall conclusions (data not reported). We therefore included all 18 trials in the analysis of Hb change.

FIGURE 3.

Forest plot: Hb change by treatment drug (g/dl). Note: random-effects meta-analysis (DerSimonian–Laird). SD, standard deviation.

Haematological response

This binary outcome was defined as the proportion of participants with an increase in Hb level of ≥ 2 g/dl or as an increase in haematocrit of ≥ 6 percentage points, unrelated to transfusion. Eight trials defined haematological response as the proportion of participants with an increase in Hb level of ≥ 2 g/dl,17,48,50,65,66,70,77,79 one study defined haematological response as an increase in haematocrit of ≥ 6 percentage points63 and one trial reported haematological response using both definitions;53 for consistency, haematological response as defined by an increase in Hb level was used in the analyses. Two studies69,73 described haematological response as an increase in Hb level of ≥ 2 g/dl or as a Hb level > 12 g/dl and were therefore excluded from the analyses.

Although both the previous HTA review2 and the study by Tonia and colleagues11 used the same definition of haematological response, only the study by Tonia and colleagues11 excluded both the trial by Kurz and colleagues69 and the trial by Vansteenkiste and colleagues73 from the analyses. The previous HTA review2 argued that most of the data in the trial by Vansteenkiste and colleagues73 would have been derived from an increase in Hb of 2 g/dl (considering baseline Hb values) and included it in the analyses. Vansteenkiste and colleagues73 reported mean baseline Hb levels of 10.28 g/dl [standard deviation (SD) 1.08 g/dl] and 9.93 g/dl (SD 1.01 g/dl) in the treatment and control groups respectively. Kurz and colleagues69 reported mean baseline Hb levels of 9.88 g/dl (SD 0.89 g/dl) and 9.85 g/dl (SD 0.60 g/dl) in the treatment and control groups respectively. For consistency with the previous HTA review,2 sensitivity analyses including the trials by Vansteenkiste and colleagues73 and Kurz and colleagues69 were performed.

Overall, the analysis of haematological response included 10 trials with 2228 participants. 17,48,50,53,63,65,66,70,77,79 Two trials were newly identified in the update searches. 48,77 As some trials with multiple experimental arms were split into subsets,48,63 the number of trials displayed is 12.

Haematological response was observed in 759 out of 1213 participants in the ESA-treated groups, compared with 182 out of 1015 participants in the control groups. The random-effects meta-analysis showed a statistically significant difference in haematological response in favour of treatment (RR 3.29, 95% CI 2.84 to 3.81; Figure 4). Heterogeneity between the trials was not significant (I² = 6.4%, p = 0.383; χ² = 11.75, df = 11, p = 0.383), with all individual studies indicating a beneficial effect of ESAs with regard to haematological response. To test whether publication bias was present in the meta-analysis, funnel plot asymmetry was investigated (see Appendix 12). The funnel plot analysis did not suggest statistically significant asymmetry (p = 0.275). A meta-regression using publication year as a covariate to assess the effect of publication year on haematological response suggested that earlier published studies tended to report higher effects than later published studies (p = 0.044). The earlier studies also tended to be smaller trials (see the meta-regression plot in Appendix 12).

FIGURE 4.

Forest plot: haematological response overall. Note: random-effects meta-analysis (DerSimonian–Laird). Events, treatment = number of events/number of participants in the treatment group; events, control = number of events/number of participants in the control group.

The fixed-effects meta-analysis undertaken as a sensitivity analysis also showed a statistically significant difference in haematological response in favour of treatment (RR 3.41, 95% CI 2.96 to 3.92; I² = 6.4%; p = 0.383); the forest plot of the analysis is provided in Appendix 12. Including the trials by Kurz and colleagues69 and Vansteenkiste and colleagues73 in the meta-analysis did not affect the overall conclusions (RR 3.21, 95% CI 2.81 to 3.68; I² = 8.2%, p = 0.363; the forest plot of the analysis is provided in Appendix 12). Similarly to the Hb change outcome, the trial by Kurz and colleagues69 (n = 35) appeared to differ from most of the other included trials. This study reported the highest RR for haematological response, with wide CIs (RR 14.63, 95% CI 0.94 to 226.68).

Prespecified subgroup analysis was performed (Table 17). None of the studies with available haematological response data included ovarian cancer patients. Therefore, the planned ovarian cancer subgroup analysis was not completed. In addition, meta-regression models including random effect and subgroups as covariates to assess the effects of a subgroup on haematological response were performed; the F-statistics from these analyses are reported in Table 17. All covariates showing a significant effect (p < 0.05) in a univariate analysis were further considered in a model selection.

One study70 provided separate results for the subgroups malignancy type (solid and haematological malignancy) and baseline Hb level [≤ 10.5 g/dl and > 10.5 g/dl (but ≤ 12 g/dl)]. In addition to results from the ITT population, results for these subgroups were also used in the PenTAG meta-analyses: using baseline Hb levels the effect estimate was RR 3.29 (95% CI 2.81 to 3.85, I² = 13.4%; p = 0.310; see Appendix 12) and using malignancy type the effect estimate was RR 3.28 (95% CI 2.84 to 3.78, I² = 13.4%; p = 0.403; see Appendix 12). In addition, the trial by Vansteenkiste and colleagues73 also reported subgroup results for participants with baseline Hb levels < 10.0 g/dl and for participants with baseline Hb levels ≥ 10.0 g/dl (but ≤ 11 g/dl) (reported in Vansteenkiste and colleagues84). Including the trials by Kurz and colleagues69 and Vansteenkiste and colleagues73 in the meta-analyses with subgroup results had no impact on the overall conclusions.

Univariate analyses identified significant differences between trials reporting use of iron supplementation and trials not reporting use of iron supplementation (p = 0.006; Figure 5). Trials that did not report whether they used iron supplementation appeared to offer greater benefits (RR 4.94, 95% CI 3.38 to 7.20, I² = 0%; p = 0.752) than trials using iron supplementation (RR 3.05, 95% CI 2.63 to 3.54, I² = 0%; p = 0.669). The meta-regression model with iron subgroups is presented in Appendix 12. However, including the trials by Kurz and colleagues69 and Vansteenkiste and colleagues73 in the meta-regression model with iron supplementation as a covariate provided different results; the difference between trials using iron supplementation and trials not reporting iron supplementation was no longer significant (p = 0.735). As noted earlier, the trial by Kurz and colleagues69 appeared to differ from the other included studies. A sensitivity analysis including the trial by Vansteenkiste and colleagues73 but excluding that by Kurz and colleagues69 again suggested that trials not reporting iron supplementation offer greater benefits (p = 0.037). The studies not reporting whether they used iron supplementation tended to be smaller (see Figure 5). Univariate analyses using the Hb subgroup results identified significant differences based on baseline and inclusion Hb levels (see Table 17). However, these results seemed to be driven mainly by the study by Littlewood and colleagues70 [Hb subgroup < 12 g/dl; RR 25.52, 95% CI 1.66 to 392.3; I² = not applicable (NA); Figure 6] for both the baseline and the inclusion Hb levels. Because of collinearity we did not combine the baseline and inclusion Hb level subgroups in the same model. A model using the Hb baseline subgroup as a covariate suggests that participants with a higher baseline Hb level (< 12 g/dl; only one study was included in this subgroup) favoured treatment significantly more (RR 25.52, 95% CI 1.66 to 392.3, I² = NA) than participants with Hb baseline values of < 11 g/dl (RR 3.76, 95% CI 2.62 to 5.39, I² = 0%; p = 0.583) and participants with Hb baseline values of < 10 g/dl (RR 3.10, 95% CI 2.64 to 3.64, I² = 19.7%; p = 0.244; see Figure 6). The meta-regression with baseline Hb subgroup as a covariate is presented in Appendix 12. Including the trials by Kurz and colleagues69 and Vansteenkiste and colleagues73 in the meta-analyses with Hb subgroup results had no impact on the conclusions. However, it should be highlighted that only one trial (n = 56) contributed to the subgroup with Hb baseline levels of < 12 g/dl.

FIGURE 5.

Forest plot: haematological response by iron supplementation. Note: random-effects meta-analysis (DerSimonian–Laird). Events, treatment = number of events/number of participants in the treatment group; events, control = number of events/number of participants in the control group.

FIGURE 6.

Forest plot: haematological response using Hb subgroups by Hb baseline level. Note: random-effects meta-analysis (DerSimonian–Laird). Events, treatment = number of events/number of participants in the treatment group; events, control = number of events/number of participants in the control group.

Because of the small number of studies in the meta-analysis, these meta-regressions and subgroup analyses have to be interpreted with caution (see Methods of data analysis/synthesis). The Cochrane Handbook for Systematic Reviews of Interventions54 recommends at least 10 studies per subgroup. In addition, sensitivity analyses (e.g. including data from the trials by Kurz and colleagues69 and Vansteenkiste and colleagues73) suggest that there are differences in the impact of covariates.

Red blood cell transfusion requirement

This binary outcome was defined as the proportion of participants requiring a RBCT. Overall, the analysis of RBCT requirement included 22 trials17,48,50–53,62,63–70,73–79 with 4779 participants. Seven trials were newly identified in the update searches. 48,62,74–80 As some trials with multiple experimental arms were split into subsets,48,63 the number of studies displayed is 24.

A RBCT was required by 554 of 2480 participants treated with ESAs compared with 835 of 2299 participants receiving placebo/no treatment. The random-effects meta-analysis showed a statistically significant difference in RBCT requirement in favour of the treatment group (RR 0.63, 95% CI 0.57 to 0.69; Figure 7). The heterogeneity between the trials was not significant (I² = 10.5%, p = 0.315; χ² = 25.71, df = 23, p = 0.315). All but one individual study78,80 indicated a beneficial effect of ESAs with regard to RBCT requirement. To test whether publication bias was present in the sample included in the meta-analysis, funnel plot asymmetry was investigated (see Appendix 12). The funnel plot analysis did not show statistically significant asymmetry (p = 0.234). A meta-regression using publication year as a covariate to assess the effect of publication year on RBCT requirement was not statistically significant (p = 0.208; see meta-regression plots in Appendix 12).

FIGURE 7.

Forest plot: RBCT. Notes: random-effects meta-analysis (DerSimonian–Laird); trials with multiple experimental arms split into subsets in the analysis: Tjulandin and colleagues48 report data for epoetin theta and epoetin beta and Abels and colleagues63 report data for participants on platinum-based chemotherapy and non-platinum-based chemotherapy. Events, treatment = number of events/number of participants in the treatment group; events, control = number of events/number of participants in the control group.

The fixed-effects meta-analysis undertaken as a sensitivity analysis showed a statistically significant difference in RBCT requirement in favour of treatment (RR 0.62, 95% CI 0.51 to 0.67); the forest plot of this analysis is provided in Appendix 12).

Prespecified subgroup analyses were performed (Table 18). In addition, meta-regression models including random effect and subgroups as a covariate to assess the effects of a subgroup on RBCT requirement were performed. The F statistics from these analyses are reported in Table 18. All covariates showing a significant effect (p < 0.05) in a univariate analysis were considered further in model selection.

One study70 reported results for the subgroups malignancy type (solid and haematological malignancy) and baseline Hb level [≤ 10.5 g/dl and > 10.5 g/dl (but ≤ 12 g/dl)]. Vansteenkiste and colleagues73 also reported subgroup results for participants with baseline Hb levels < 10.0 g/dl and ≥ 10.0 g/dl (but ≤ 11 g/dl) (reported in Vansteenkiste and colleagues84). In addition to results from the ITT population, results for these subgroups were included in the PenTAG meta-analyses: using the Hb subgroups the effect estimate was RR 0.61 (95% CI 0.55 to 0.68, I² = 22.4%; p = 0.015) and using the malignancy subgroups the effect estimate was RR 0.62 (95% CI 0.56, 0.68, I² = 15.8%; p = 0.239; see Appendix 12).

Univariate analyses did not identify any significant differences based on the predefined subgroups (see Table 18).

Number of red blood cell units transfused

Overall, 10 trials51,52,63,65,66,69,73,74,77,79 evaluating a total of 1920 participants were included in the analysis of RBC units transfused. As one study63 was split into subsets, the number of trials displayed is 11. Two trials were newly identified;74,77 neither was included in the Cochrane review11 for the analysis of this outcome. All except one study77 reported the mean number of units transfused per average participant (i.e. regardless of whether participants had received a RBCT). For Tjulandin and colleagues77 this was calculated from the data presented in the published paper.

The overall mean difference between groups showed a statistically significant benefit for participants receiving ESAs (WMD –0.87, 95% CI –1.28 to –0.46; Figure 8); the ESA group received fewer units of blood per participant than the control group. The heterogeneity between the studies was significant (I² = 59.3%; p = 0.006). All but one study indicated a reduced need for RBCs in participants receiving ESAs compared with control subjects. A funnel plot analysis did not suggest statistically significant asymmetry (p = 0.137; see Appendix 12).

FIGURE 8.

Forest plot: mean number of RBC units transfused. Notes: random-effects meta-analysis (DerSimonian–Laird); trial with multiple experimental arms split into subsets in the analysis: Abels and colleagues63 reported data for participants on platinum-based chemotherapy and non-platinum-based chemotherapy; mean units transfused per average participant (i.e. regardless of whether participants had received a RBCT).

One study73 provided separate results for participants with baseline Hb levels of < 10.0 g/dl, and ≥ 10.0 g/dl (but ≤ 11.0 g/dl). Meta-analysis including these subgroup results was conducted (WMD –0.87, 95% CI –1.24 to –0.50; see Appendix 12). The fixed-effects meta-analysis undertaken as a sensitivity analysis showed a statistically significant difference for number of RBC units transfused in favour of treatment (WMD –0.64, 95% CI –0.79 to –0.48); the forest plot of this analysis is provided in Appendix 12.

To identify sources of heterogeneity, subgroup analyses were conducted (Table 19). In addition, meta-regression models including random effect and a subgroup as a covariate to assess the effects of subgroups on Hb change were performed; the F-statistics from these analyses are reported in Table 19. All covariates showing a significant effect (p < 0.05) in a univariate analysis were considered further in a model selection.

Univariate analyses identify any significant differences based on the predefined subgroups.

Anaemia-related outcomes: overall summary

All studies included in the analyses of anaemia-related outcomes were of moderate or poor quality. The general problem of reporting of trials on this topic was greatly assisted by the recent Cochrane review,11 as the authors had gathered further details from investigators and manufacturers, which were used in the meta-analysis for this review.

In total, 20 studies measured Hb change, of which 16 were included in the meta-analysis. All of the studies indicated a beneficial effect of ESAs with regard to Hb change, which varied only in magnitude. The overall WMD for Hb level increase was 1.59 g/dl. Hb change was not restricted to patients who were transfusion free; therefore, the results may have been confounded by transfusion in some of the patients.

Haematological response was defined as the proportion of participants with an increase in Hb level of ≥ 2 g/dl or an increase in haematocrit of ≥ 6 percentage points, unrelated to transfusion. In total, 10 trials reported this outcome and all were included in the meta-analysis. The analysis showed that participants treated with ESAs were three times more likely to experience a ≥ 2 g/dl increase in Hb than participants in the control group, with 63% (n = 759/1213) of participants who received ESAs having a haematological response compared with 18% (n = 182/1015) of control patients. Estimates of haematological response were considered robust, with no marked heterogeneity or subgroup effects.

The number of patients receiving RBCTs was the third outcome assessed to investigate the effects of ESAs on CIA. In total, 22 trials reported this outcome and all were included in the meta-analysis. Data were reported for the trial period; the RR of receiving a RBCT was 0.63 in favour of ESAs, equating to 22% of participants in the ESA treatment groups receiving RBCT compared with 33% in the control groups. The number of transfusions per patient was also investigated. Only 10 trials reported this outcome and many of these data were obtained by the Cochrane review authors through further questions to the trial authors. There was little difference between the ESA group and the control group with regard to the amount of blood transfused. Estimates of numbers transfused were considered robust, with no marked heterogeneity or subgroup effects.

Effectiveness estimates were consistent with previously reported estimates for anaemia-related outcomes (Table 20). A graphical summary of the study characteristics, quality appraisal and results for these outcomes is presented in Figure 9.

FIGURE 9.

Anaemia-related outcomes: graphical summary. Notes: for some of the included trials, data for some anaemia-related outcomes were not available for all participants; where different, the numbers included are reported in notes 1–8 and 10–12. 1: Hb change: epo alfa n = 63; 2: HaemR and units: epo alfa n = 66 and placebo n = 66; 3: Hb change: rHuEPO n = 28 and control n = 24; 4: Hb change: epo alfa n = 64 and control n = 58; 5: Hb change: darbe alfa n = 17 and placebo n = 6; 6: RBCT: darbe alfa n = 167 and placebo n = 165; 7: HaemR = increase in Hb of ≥ 2 g/dl and/or Hb level > 12 g/dl; 8: RBCT: epo alfa n = 251 and placebo n = 124; 9: Latin square design; 10: Hb change: epo beta n = 138 and placebo n = 142; 11: Hb change: licensed epo alfa n = 34 and control n = 24; 12: Hb change: darbe alfa n = 330 and placebo n = 359. ?, unknown; chemo, chemotherapy; darbe, darbepoetin; epo, epoetin; haem, haematological; haemR, haematological response; non-plat, non-platinum; plat, platinum; QA, quality appraisal; units, units transfused.

Summary

Seven trials reported tumour response, all of which were included in the meta-analysis. All were of moderate or poor quality. The general problem of reporting of trials on this topic was greatly assisted by the recent Cochrane review,11 as the authors had gathered further details from investigators and manufacturers, which were used in the meta-analysis for this review. Analyses suggest that treatment with ESAs in patients with cancer-induced anaemia did not have a significant effect on complete tumour response (RR 1.10, 95% CI 0.86 to 1.41). In total, 18% (n = 177/1003) of participants who received ESAs had a complete tumour response compared with 16% (n = 142/906) of patients in the control groups. There was non-significant heterogeneity between the trials (I² = 37.5%; p = 0.143); however, the direction of the effects of ESAs with regard to tumour response varied across the individual trials. The data from the seven trials suggest that there is no difference between patients treated with ESAs and patients in the control groups with regard to tumour response; however, the data are insufficient to exclude detrimental effects. It should also be noted that this is a difficult area of assessment, especially in a heterogeneous mix of tumour types, and the results should be treated with caution. Data are presented alongside the results from previous analyses in Table 21. (See Figure 13 for a graphical summary of the study characteristics, quality appraisal and results for this outcome.)

Summary

In total, 21 trials reported OS. All were of moderate or poor quality. The general problem of reporting of trials on this topic was greatly assisted by the recent Cochrane review,11 as the authors had gathered further details from investigators and manufacturers, which were used in the meta-analysis for this review. Eighteen trials were included in the meta-analysis. Analyses suggest that treatment with ESAs in patients with CIA did not have a significant effect on OS. In total, 35% (n = 818/2317) of participants who received ESAs died and 35% (n = 744/2137) of patients in the control groups died. The risk of death was 0.97 (HR 0.97, 95% CI 0.83 to 1.13). However, there was significant heterogeneity between the trials (I² = 42.4%; p = 0.030), for which no explanation could be provided. In addition, OS was calculated from the longest follow-up available (no minimum was required) and, as such, this variation between the studies (short-term and long-term studies) should be considered when interpreting the results. Overall, data suggest that, if the licensed ESA dose is followed, there are no detrimental effects of ESAs on OS; however, these results are subject to the limitations acknowledged and should be interpreted with caution. Effectiveness estimates are presented alongside previously reported estimates for OS in Table 23. (See Figure 13 for a graphical summary of the study characteristics, quality appraisal and results for this outcome.)

Summary

On-study mortality was assessed in 12 trials of moderate or poor quality. Analyses suggested that treatment with ESAs in patients with CIA did not have a significant effect on on-study mortality. In total, 11% (n = 174/1586) of participants who received ESAs had died within 30 days of the active study period compared with 12% (n = 164/1381) of patients in the control groups. The risk of death was 0.86 (HR 0.86, 95% CI 0.67 to 1.1). There was no significant heterogeneity between the trials (I² = 16.4%, p = 0.274). Overall, data suggested that, if the licensed ESA dosage is followed, there are no detrimental effects of ESAs on on-study mortality. However, these results should be interpreted with caution. Effectiveness estimates are compared with previously reported estimates in Table 25 and a graphical summary of the study characteristics, quality appraisal and results for this outcome is presented in Figure 13.

FIGURE 13.

Tumour response, OS and mortality outcomes: graphical summary. Notes: for some of the included trials, data for some anaemia-related outcomes were not available for all participants; where different, the numbers included are reported in notes 1–4. 1: tumour response reported for epo alfa n = 64 and placebo n = 63; 2: tumour response reported for epo beta n = 154 and placebo n = 162; 3: OS reported for darbe alfa n = 345 and control n = 369 and mortality reported for darbe alfa n = 353 and control n = 376; 4: OS and mortality reported for all randomised participants (n = 320), whereas other outcomes were reported using all randomised participants receiving at least one dose of study treatment (n = 314). ?, unknown; chemo, chemotherapy; darbe, darbepoetin; epo, epoetin; haem, haematological; non-plat, non-platinum; plat, platinum; QA, quality appraisal.

Thromboembolic events

We identified 14 trials17,51,52,62,63,66,70,73–79 that measured thromboembolic events, including 4013 participants. Of these, 2029 participants were treated with ESAs. As one multiarm study63 was split into subsets, the number of studies displayed is 15. Five included studies were newly identified in the update searches. 62,75–78 If thromboembolic events were not reported, data from the Cochrane review by Tonia and colleagues11 were used in the PenTAG analyses. One study52 did not report any thromboembolic events in the treatment or placebo arms and was excluded from the meta-analysis.

Data from Moebus and colleagues62 were used in the PenTAG meta-analyses (whereas the analysis in Tonia and colleagues11 used data from Moebus and colleagues33). The Moebus and colleagues62 trial showed an increased risk for patients treated with ESAs compared with control participants (RR 2.26, 95% CI 1.09 to 4.70), whereas there was no difference between the treatment arm and the control arm in the study by Moebus and colleagues. 33

Thromboembolic events were reported in 103 out of 2029 participants treated with ESAs, compared with 66 out of 1984 participants in the control group. The random-effects meta-analysis showed a RR of 1.46 (95% CI 1.07 to 1.99), favouring the control group (Figure 14). There was no heterogeneity between the trials (I² = 0%, p = 0.733; χ² = 9.52, df = 13, p = 0.733), with 11 studies indicating detrimental effects of ESA treatment and three studies indicating beneficial effects of ESA treatment with regard to thromboembolic events. To test whether publication bias was present in the sample included in the meta-analysis, a funnel plot was constructed (see Appendix 12). The funnel plot analysis did not show statistically significant asymmetry (p = 0.627). In addition, a meta-regression using publication year as a covariate to assess the effect of publication year on thromboembolic events suggested that the effects of ESAs on thromboembolic events were independent of when the trial results were published (p = 0.871); the meta-regression plot is presented in Appendix 12.

FIGURE 14.

Forest plot: thromboembolic events (overall). Notes: random-effects meta-analysis (DerSimonian–Laird); trial with multiple experimental arms split into subsets in the analysis: Abels and colleagues63 reported data for participants on platinum-based chemotherapy and non-platinum-based chemotherapy. Events, treatment = number of events/number of participants in the treatment group; events, control = number of events/number of participants in the control group.

The fixed-effects meta-analysis undertaken as a sensitivity analysis showed similar results, favouring the control participants over those receiving ESAs (RR 1.52, 95% CI 1.13 to 2.05, I² = 0%; p = 0.733); the forest plot of the analysis is included in Appendix 12.

Prespecified subgroup analyses were conducted (Table 26). In addition, meta-regression models including random effect and a subgroup as a covariate to assess the effects of subgroups on thomboembolic events were performed; the F statistics from these analyses are reported in Table 26. All covariates showing a significant effect (p < 0.05) in a univariate analysis were considered further in a model selection.

Univariate analyses did not identify any significant differences based on the predefined subgroups (see Table 26).

Hypertension

We identified 10 trials48,51,52,63,66,70,72,73,77,79 that measured hypertension, including 2086 participants. Of these, 1152 participants were treated with ESAs. As two multiarm studies48,63 were split into subsets, the number of studies displayed is 12. Two included studies48,77 were newly identified in the update searches. If hypertension was not reported, we used data from the Cochrane review by Tonia and colleagues11 in the PenTAG analyses.

Hypertension was reported in 62 out of 1152 participants (5%) treated with ESAs compared with 27 out of 934 participants (3%) in the control groups. The random-effects meta-analysis showed a risk ratio of 1.80 (95% CI 1.14 to 2.85; Figure 15), favouring the control. There was no statistical heterogeneity between the trials (I² = 0%; χ² = 7.10, df = 11; p = 0.791); however, the direction of the effects of ESAs with regard to hypertension varied across the individual trials (see Figure 15). To test whether publication bias was present in the sample included in the meta-analysis, a funnel plot was constructed (see Appendix 12). The funnel plot analysis did not show statistically significant asymmetry (p = 0.689). In addition, a meta-regression using publication year as a covariate to assess the effect of publication year on hypertension suggests that the effects of ESAs on hypertension were independent of when the trial results were published (p = 0.735); the meta-regression plot is presented in Appendix 12.

FIGURE 15.

Forest plot: hypertension (overall). Notes: random-effects meta-analysis (DerSimonian–Laird); trials with multiple experimental arms split into subsets in the analysis: Tjulandin and colleagues48 reported data for epoetin theta and epoetin beta and Abels and colleagues63 reported data for participants on platinum-based chemotherapy and non-platinum-based chemotherapy. Events, treatment = number of events/number of participants in the treatment group; events, control = number of events/number of participants in the control group.

The fixed-effects meta-analysis undertaken as a sensitivity analysis showed similar results (RR 1.97, 95% CI 1.27 to 3.07, I² = 0%; p = 0.791); the forest plot of the analysis is included in Appendix 12.

Prespecified subgroup analyses were conducted (Table 27). In addition, meta-regression models including random effect and a subgroup as a covariate to assess the effects of subgroups on hypertension were performed; the F statistics from these analyses are reported in Table 27. All covariates showing a significant effect (p < 0.05) in a univariate analysis were considered further in a model selection.

Univariate analyses did not identify any significant differences based on the predefined subgroups (see Table 27).

Thrombocytopenia/haemorrhage

Data for thrombocytopenia (decrease of platelets in the blood)/haemorrhage were available from seven trials. 52,65–67,70,76,78 Overall, the analysis included all seven studies with 1715 participants. If thrombocytopenia/haemorrhage were not reported, data were obtained from the Cochrane review by Tonia and colleagues. 11

Thrombocytopenia/haemorrhage was reported in 55 out of 877 participants treated with ESAs, compared with 54 out of 838 participants in the control groups. The random-effects meta-analysis showed a RR of 0.93 (95% CI 0.65 to 1.34; Figure 16), which was not statistically significant. There was no statistical heterogeneity between the trials (I² = 0%; χ² = 3.02, df = 6, p = 0.807); however, the direction of the effects of ESAs with regard to hypertension varied across the individual trials. Because there were only seven primary studies included in the meta-analysis, the funnel plot analysis to test whether publication bias was present was not conducted. 54

FIGURE 16.

Forest plot: thrombocytopenia/haemorrhage (overall). Note: random-effects meta-analysis (DerSimonian–Laird). Events, treatment = number of events/number of participants in the treatment group; events, control = number of events/number of participants in the control group.

The fixed-effects meta-analysis undertaken as a sensitivity analysis showed similar non-significant results (RR 0.91, 95% CI 0.63 to 1.30; see Appendix 12).

Prespecified subgroup analyses and meta-regression models with subgroups as covariates were not conducted because only seven trials were included in the meta-analysis.

Seizures

Data on seizures were available from one trial63 including 289 participants. As this trial was split into subsets, the number of studies displayed in the forest plot is two. If seizures were not reported, we used data from the Cochrane review by Tonia and colleagues11 in the PenTAG analyses.

Overall, five seizure events were reported in the ESA-treated group (n = 148) and four in the control group (n = 141), resulting in a RR of 1.19 (RR 1.19; 95% CI 0.33 to 4.38; Figure 17). There was no heterogeneity between the trials (I² = 0%, p = 0.742; χ² = 0.11, df = 5, p = 0.742), although the two included trials indicated effects in opposite directions. Because only two primary studies were included in the meta-analysis, the funnel plot analysis to test whether publication bias was present was not conducted. 54 The fixed-effects meta-analysis undertaken as a sensitivity analysis showed similar non-significant results (RR 1.19, 95% CI 0.33 to 4.35, I² = 0%; p = 0.742; see Appendix 12).

FIGURE 17.

Forest plot: seizures (overall). Notes: random-effects meta-analysis (DerSimonian–Laird); trial with multiple experimental arms split into subsets in the analysis: Abels and colleagues63 reported data for participants on platinum-based and non-platinum-based chemotherapy. Events, treatment = number of events/number of participants in the treatment group; events, control = number of events/number of participants in the control group.

Prespecified subgroup analyses and meta-regression models with subgroups as covariates were not conducted because only two trials were included in the meta-analysis.

Pruritus (pruritus, rash and irritation)

We identified seven trials52,63,67,69,76,77,79 that measured pruritus (pruritus, rash and irritation were considered11) including 904 participants. Of these, 450 participants were treated with ESAs. Two included studies were newly identified in the update searches. 76,77 If pruritus events were not reported, we used data from the Cochrane review by Tonia and colleagues11 in the PenTAG analyses. One study69 did not report any pruritus events in the treatment and placebo arms and was excluded from the meta-analysis.

The random-effects meta-analysis showed a risk ratio of 2.04 (95% CI 1.11 to 3.75; Figure 18), favouring the control. There was no heterogeneity between the trials (I² = 0%, p = 0.872; χ² = 1.83, df = 5, p = 0.872), with all of the individual studies indicating a detrimental effect of treatment with ESAs with regard to the number of pruritus events. Because only six primary studies were included in the meta-analysis, the funnel plot analysis to test whether publication bias was present was not conducted. 54 The fixed-effects meta-analysis undertaken as a sensitivity analysis showed similar results (RR 2.16, 95% CI 1.18 to 3.92, I² = 0%; p = 0.872); the forest plot of the analysis is included in Appendix 12.

FIGURE 18.

Forest plot: pruritus (overall). Note: random-effects meta-analysis (DerSimonian–Laird). Events, treatment = number of events/number of participants in the treatment group; events, control = number of events/number of participants in the control group.

The prespecified subgroup analyses and meta-regression models with subgroups as covariates were not conducted because only six trials were included in the meta-analysis.

Safety-related outcomes: summary

All studies were of moderate or poor quality. In addition, there was considerable variability in the reporting of AEs among the included studies. Given the greater access to data in the Cochrane review11 than in the primary papers, relevant data from the Cochrane review11 were used to conduct meta-analyses for the AEs of interest. Overall, the data suggested that there is an increased risk of thromboembolic events and hypertension after treatment with ESAs, consistent with previous estimates (Table 28). Data for seizures are also consistent with previous meta-analyses, showing no effects of ESAs on seizures (see Table 28). Of note is that all AEs are relatively rare compared with other outcomes considered in this report (e.g. RBCT, Hb change and mortality).

The PenTAG analyses suggest that there is an increased risk of pruritus, with a significant difference found between patients treated with ESAs and participants in the control arms (RR 2.04, 95% CI 1.11 to 3.75). In comparison, the Cochrane review11 did not find a significant difference between patients treated with ESAs and participants in the control arms (RR 1.49, 95% CI 0.99 to 2.24). It must be highlighted that both the current review and the Cochrane review11 combined events of skin rash, irritation and pruritus in the meta-analyses. However, the rates of skin rash, irritation and pruritus may differ and the way that this outcome has been defined may be the cause of the marked variation in event rates.

Also, the summary estimate for risk of thrombocytopenia/haemorrhage associated with ESA treatment found in the PenTAG review was a RR of 0.93 (95% CI 0.65 to 1.34), suggesting that treatment with ESAs in patients with CIA did not have an effect on thrombocytopenia/haemorrhage. However, the Cochrane review11 found a RR of 1.21 (95% CI 1.04 to 1.42), suggesting detrimental effects of ESAs with regard to thrombocytopenia/haemorrhage.

It must be emphasised that the current analyses included only studies complying with the licenced ESA dose, whereas the Cochrane review did not apply any restrictions regarding the ESA posology. However, these results should be interpreted with caution (see Chapter 6, Strengths and limitations of the systematic review of studies of effectiveness for more details).

A graphical summary of the study characteristics, quality appraisal and results for the safety outcomes is presented in Figure 19.

FIGURE 19.

Safety-related outcomes: graphical summary. Notes: for some of the included trials, data for some anaemia-related outcomes were not available for all participants; where different, the numbers included are reported in notes 1–4. 1: hypertension reported for epo beta n = 43 and control n = 28; 2: participants who received at least one dose of study treatment even if they did not meet the eligibility criteria were included in the safety population; 3: one participant randomised to the darbe alfa arm did not receive the study treatment and was included in the placebo arm for the safety evaluation. ?, unknown; chemo, chemotherapy; darbe, darbepoetin; epo, epoetin; haem, haematological; non-plat, non-platinum; plat, platinum; QA, quality appraisal; thrombo/haemor, thrombosis/haemorrhage.

Definition of ‘within licence’

For this HTA review, studies were considered eligible for inclusion if they evaluated starting doses of ESAs according to European labelling, irrespective of how they dealt with Hb levels (see Chapter 1, Marketing authorisations: haemoglobin levels).

With respect to European labelling, additional measures of dose efficiency [inclusion Hb level (≤ 11 g/dl and > 11 g/dl) and target Hb level (≤ 13 g/dl and > 13 g/dl)] were also considered in post-hoc analyses. Studies contributing to these subgroups are listed in Table 31. In addition, we also considered measures relating to the administration of ESAs in conjunction with study quality, specifically blinding (double-blind RCTs).

Results from these subgroup analyses are summarised in Table 32.

Post-hoc analyses offer some limited evidence to suggest that compliance with European labelling results in better outcomes: there are no detrimental effects of ESAs on either on-study or overall mortality in patients with chemotherapy-induced anaemia. These effects are consistent with an improved tumour response and a decrease in the number of thromboembolic events. However, these analyses must be interpreted with caution. The number of studies per subgroup is small, some of the effect sizes are not statistically significant and the CIs remain wide. In addition, the analyses may not have the statistical power to detect the effects of adherence to European labelling on outcomes, if such effects exist. It should also be noted that this is a difficult area of assessment, especially in a heterogeneous mix of tumour types. Furthermore, we have not sought to address multiple testing issues that arise when considering subgroups, and so inference is not straightforward.

Study characteristics

Study characteristics are reported in Table 10. A summary of the HRQoL measures included in the studies is provided in Table 33.

A range of questionnaires was used to measure HRQoL and subsequent changes in response to treatment. The scales are summarised in Appendix 13; however, this review focuses on the FACT tool, as it is has been widely used in ESA trials and is considered to have good responsiveness to change and good convergent and discriminant validity. 11 Furthermore, the FACT tool is the only tool in this review for which there are sufficient studies to enable meta-analysis.

The FACT tool, which asks patients to focus on HRQoL issues over the previous 7 days, is part of a collection of HRQoL questionnaires (Figure 20), beginning with a generic questionnaire called the FACT-G. There are now over 50 different scales and symptom indexes, some of which have been modified over time. The FACT scales used in this review are listed in Table 33. Copies of these questionnaires and details of scoring and interpretation are available at: http://www.facit.org/FACITOrg/Questionnaires (accessed September 2015). It should be noted that since 1997 the scale has been known as FACIT.

FIGURE 20.

Overview of the FACT scales used in this review. Fact-An-An, FACT-Anaemia Anaemia subscale.

Using both anchor-based and distribution-based methods to analyse FACT-F, FACT-G and FACT-An, data from three samples of patients (n = 50, n = 131 and n = 2402) determined the minimal clinically important difference to be FACT-F = 3.0, FACT-G = 4.0 and FACT-An = 7.0. 112

Trials identified in the previous Health Technology Assessment review

Of the 11 trials identified in the previous HTA review,2 nine indicated that there was a statistically significant difference in HRQoL between patients treated with ESAs and control participants (Figure 21). Of the two studies that did not show ESAs to be effective compared with placebo, one used an unvalidated assessment tool (Health State Utility Scale)69 and the other used the Psychological Distress Inventory (PDI). 67

FIGURE 21.

Health-related quality of life: graphical summary. Notes: 1: Study population included patients not receiving chemotherapy; 2: 87% and 84% of participants were analysed in the treatment and control groups, respectively; 3: 80% and 73% of participants were analysed in the treatment and control groups, respectively; 4: 54% and 57% of participants were analysed in the treatment and control groups, respectively; 5: 75% and 61% of participants were analysed in the treatment and control groups, respectively; 6: 90–98% and 86–97% of participants were analysed in the treatment and control groups, respectively; 7: 81% of participants were analysed in the treatment and control groups. ?, unknown; BSC, best supportive care; BSI, Brief Symptom Inventory; CLAS, Cancer Linear Analog Scale; D&A, depression and anxiety subscale of the BSI; darbe, darbepoetin; EORTC QLQ-C30, European Organisation for Research and Treatment of Cancer Quality of Life Questionnaire C30; epo, epoetin; FACT-An-An, FACT–Anaemia Anaemia subscale; haem, haematological; LASA, Linear Analogue Scale Assessment; NHP, Nottingham Health Profile; non-plat, non-platinum; PDI, Psychological Distress Inventory; plat, platinum; VAS, visual analogue scale.

Thatcher and colleagues52 reported that only the overall HRQoL level revealed a statistically significant improvement favouring epoetin alfa (p < 0.05). Evaluation of World Health Organization (WHO) performance scores revealed similar findings, with no significant between- or within-group differences. 52

Trials identified, 2004 to current

Three trials were identified following the previous HTA review. 2 Of these, one was a follow-up study79 of a study identified previously. 71 Österborg and colleagues79 report a statistically significant increase in favour of epoetin beta; however, the variability between patients was considerable. Ray-Coquard and colleagues75 stated that there were no statistically detectable differences during the study period, although none of these data were reported. Tjulandin and colleagues77 also found no significant differences between the epoetin theta group and the placebo group.

Post-hoc studies identified, 2004 to current

Five studies60,81–84 reported trends that favour ESA. However, Bajetta and colleagues81 and Littlewood and colleagues83 did not analyse this statistically.

Additional results to the primary study73 provided by Vansteenkiste and colleagues84 indicated a significant difference (p = 0.0147) in HRQoL between the darbepoetin alfa group and the placebo group for those with a baseline Hb level of < 10 g/dl. In contrast, no difference was apparent between groups for those with a baseline Hb of ≥ 10 g/dl.

Meta-analysis: Functional Assessment of Cancer Therapy – Fatigue (13 items) score (random effects)

Given the variability of reporting in the published papers, FACT-F data were extracted from the Cochrane review by Tonia and Colleagues11 for use in the PenTAG analyses. Functional Assessment of Cancer Therapy – Fatigue scores were available from seven studies17,50,65,70,71,73,77,79 including 1794 participants. One new primary study was identified. 77

The WMD was 2.54 (95% CI 1.42 to 3.65; Figure 22). There was low heterogeneity between the trials (I² = 14.9%; p = 0.32) (Table 34). Because only seven primary studies were included in the meta-analysis, the funnel plot analysis to test whether publication bias was present was not conducted in accordance with published giudelines. 54 The fixed-effects meta-analysis undertaken as a sensitivity analysis showed similar significant results (see Appendix 13, Figure 108). All of the studies were similar in terms of quality; however, the trial reported by Boogaerts and colleagues65 did not employ blinding for participants. Removing this study from the meta-analysis had a minimal impact on the results (WMD 2.21, 95% CI 1.131 to 3.280; see Appendix 13, Figure 113), but did improve heterogeneity (I² = 0%, p = 0.51).

FIGURE 22.

Health-related quality of life: FACT-F (13 items) – overall. Note: random-effects meta-analysis (DerSimonian–Laird). BSI, Brief Symptom Inventory; CLAS, Cancer Linear Analogue Scale; D&A, depression and anxiety subscale of the BSI; darbe, darbepoetin; EORTC QLQ-C30, European Organisation for Research and Treatment of Cancer Quality of Life Questionnaire C30; epo, epoetin; FACT-An-An, FACT-An Anaemia subscale; haem, haematological; LASA, Linear Analogue Scale Assessment; NHP, Nottingham Health Profile; non-plat, non-platinum; plat, platinum; VAS, visual analogue scale.

Meta-analysis was performed on FACT-G and FACT-An Anaemia subscale (FACT-An-An) data; however, only three studies70,71,77,79 were suitable for inclusion for each scale with high levels of heterogeneity [see Table 34 and Appendix 13, Figures 114 and 115 (FACT-G), and Figures 116 and 117 (FACT-An)]. The results of no statistical difference between the intervention and the control must therefore be treated with caution.

Univariate subgroup analyses were conducted for FACT-F outcomes according to chemotherapy type, malignancy type, intervention (epoetin or darbepoetin) and study duration and showed significant results; however, the number of studies included was small (see Table 34 and Appendix 13, Figures 109–112).

Health-related quality-of-life outcomes: overall summary

Effectiveness estimates are presented alongside previously reported estimates for HRQoL (see Table 34). A graphical summary of the study characteristics, quality appraisal and results for these outcomes is presented in Figures 21 (HRQoL) and 23 (FACT-F).

FIGURE 23.

Forest plot: HRQoL overall. Note: random effects meta-analysis (DerSimonian–Laird).

Overall, the conclusions of the PenTAG review are in agreement with those of the Cochrane review11 in that there is a statistically significant difference between patients treated with ESAs and control participants when combining HRQoL parameters; however, this is probably not clinically important (minimal clinically important difference = 3.0112). As with previous reviews, however, it should be noted that there are several methodological concerns that may result in bias, for example the substantial quantity of missing data, expecting patients to complete repeated questionnaires leading to a shift in patient response and the various modes of administration of the questionnaires.