17 Ofsthun et al. reported a similar analysis of 44 550 prevalent haemodialysis patients from the Fresenius Medical PR-171 nmr Care database.18 The relative risk of death for haemoglobin <90 g/L was 2.11 (P < 0.001) compared with a reference haemoglobin level of 110–120 g/L. The relative risk of death decreased to approximately 1.6 and

1.3 as haemoglobin increased to 90–100 g/L and 100–110 g/L, respectively. There was a 16% reduction in mortality for haemoglobin levels between 120 and 130 g/L (RR 0.84, P = 0.007). Fort et al. prospectively studied the effects of time-dependent haemoglobin and ESA dose on mortality in 2310 incident haemodialysis patients from Spain.19 Using a time-dependent multivariate Cox proportional hazard model, the adjusted HR for death was 1.36 (95% CI 1.01–1.86) for a haemoglobin level <100 g/L compared with a level of 111–120 g/L. In contrast, a haemoglobin

level of >130 g/L was associated with a survival benefit (HR 0.69, 95% CI 0.49–0.97). Analysis of the UK Renal Registry data reported similar outcomes with HRs for death for haemoglobin values <100 g/L and >110 g/L being 1.28 (P < 0.001) and 0.64 (P < 0.001), respectively, compared AZD9668 in vitro with a reference haemoglobin level of 100–110 g/L.20 The HRs decreased as achieved haemoglobin increased (Hb 110–120 g/L HR 0.63; Hb 120–130 g/L HR 0.47, and Hb >130 g/L HR 0.44). Zhang et al. conducted a retrospective study of 94 569 prevalent patients who were on haemodialysis in 2000 and 2001.21 The patients were divided into quartiles of ESA dose (1388–7905 U/week, 7905–13 377 U/week, 13 377–22 068 U/week and >22 068 U/week) and five categories of

haematocrit values (<30%, 30–33%, 33–36%, 36–39% and >39%). Mortality rates decreased as haematocrit values increased. Within each haematocrit category, mortality rates were lowest in the lowest quartile of ESA dose and highest in the highest quartile. A US Medicare study reported outcomes of 393 967 prevalent haemodialysis patients from 2002 to 2004.22 In a fully adjusted Cox proportional hazard model, mortality was higher at all haematocrit levels ADP ribosylation factor below 34.5% compared with the reference haematocrit level of 34.5% to 36%. The HR for death increased from 1.17 (95% CI 1.14–1.20) to 3.11 (95% CI 3.01–3.20) when haematocrit decreased from 33–34.5% to <27%. Similarly, mortality increased at all levels of haematocrit >39%. Mortality was comparable for haematocrit levels between 36% and 39%. When patients were grouped into five categories of erythropoietin dose (0 U/week, 0–6000 U/week, 6000–12 000 U/week, 12 000–18 000 U/week and >18 000 U/week), the HR for death progressively increased with increasing dose of erythropoietin for every level of haematocrit.