Background: At ASCO 2007, we reported increased mortality risks when ESAs are administered to anemic cancer patients who are receiving chemotherapy when target hemoglobin levels are beyond the correction of anemia. In February 2007, a meta-analysis of nine randomized clinical trials with 5,143 patients published in the Lancet [vol 369; 381–88] identified a statistically significant risk of all cause mortality (relative risk (RR) of 1.17, 95% confidence interval (CI) 1.01, 1.35) when anemic patients with chronic kidney disease received ESAs targeted to higher hemoglobin concentrations (120–150 g/L). A recent report from the RADAR (Research against Adverse Drug Reactions) group raises concern that survival analyses might differ depending on whether survival was evaluated as a measure of efficacy versus a measure of safety. Herein, we re-analyze the data by evaluating randomized clinical trials according to whether or not survival was prospectively included as a primary or secondary efficacy outcome.

Methods: Risks of death in randomized controlled clinical trials included in the Lancet meta-analysis were evaluated. We classified those studies based on their mortality outcomes, either as an efficacy outcome or as a safety outcome. Effect estimates for RR and 95% CI were derived from Stata (version 9.1, College Station, TX), calculated with random-effects models and pooled by use of the Dersimonian and Laird method.

Results: In studies where survival was measured as an efficacy endpoint, the relative risk of mortality with ESAs targeted to higher hemoglobin levels was 1.27 (1.08, 1.49), a number greater than the relative risk reported in the Lancet meta-analysis.

Conclusions: Randomized controlled trials should be included in meta-analyses that evaluate harms only if the relevant safety measure is prospectively included as a primary or secondary efficacy outcome measure in the study protocol. When survival was included as part of the efficacy analysis, a statistically significant safety signal was present. Randomized trials that included harms as a measure of safety did not present a statistically significant safety signal. Including randomized trials that include harms as a safety measure introduce noise and can mask safety signals.

Studies:Events:RR (95% CI)
Survival included as a primary or secondary efficacy outcome measure: High vs Low Hb Target:  
Besarab 1998 (n=1233) 183/618 vs 150/615 1.21 (1.01, 1.46) 
Gouva 2004 (n=88) 4/43 vs 3/45 1.40 (0.33, 5.87) 
Drueke 2006 (n=602) 31/300 vs 21/302 1.49 (0.87, 2.53) 
Singh 2006 (n=1432) 52/715 vs 36/717 1.45 (0.96, 2.19) 
Subtotal (n=3355) 270/1676 vs 210/1679 1.27 (1.08, 1.49) 
Survival included only as a safety measure: High vs Low Hb Target:  
Foley 2000 (n=146) 4/73 vs 3/73 1.33 (0.31, 5.75) 
Furuland 2003 (n=416) 29/216 vs 27/200 0.99 (0.61, 1.62) 
Roger 2005 (n=154) 0/75 vs 0/79 not estimable 
Levin 2005 (n=152) 1/74 vs 3/78 0.35 (0.04, 3.30) 
Parfrey 2005 (n=696) 12/396 vs 20/300 0.45 (0.23, 0.92) 
Rossert 2006 (n=390) 1/195 vs 6/195 0.17 (0.02, 1.37) 
Subtotal (n=1954) 47/1029 vs 59/925 0.67 (0.37, 1.19) 
Studies:Events:RR (95% CI)
Survival included as a primary or secondary efficacy outcome measure: High vs Low Hb Target:  
Besarab 1998 (n=1233) 183/618 vs 150/615 1.21 (1.01, 1.46) 
Gouva 2004 (n=88) 4/43 vs 3/45 1.40 (0.33, 5.87) 
Drueke 2006 (n=602) 31/300 vs 21/302 1.49 (0.87, 2.53) 
Singh 2006 (n=1432) 52/715 vs 36/717 1.45 (0.96, 2.19) 
Subtotal (n=3355) 270/1676 vs 210/1679 1.27 (1.08, 1.49) 
Survival included only as a safety measure: High vs Low Hb Target:  
Foley 2000 (n=146) 4/73 vs 3/73 1.33 (0.31, 5.75) 
Furuland 2003 (n=416) 29/216 vs 27/200 0.99 (0.61, 1.62) 
Roger 2005 (n=154) 0/75 vs 0/79 not estimable 
Levin 2005 (n=152) 1/74 vs 3/78 0.35 (0.04, 3.30) 
Parfrey 2005 (n=696) 12/396 vs 20/300 0.45 (0.23, 0.92) 
Rossert 2006 (n=390) 1/195 vs 6/195 0.17 (0.02, 1.37) 
Subtotal (n=1954) 47/1029 vs 59/925 0.67 (0.37, 1.19) 

Author notes

Disclosure: No relevant conflicts of interest to declare.

Sign in via your Institution