**4. Association between haemoglobin variability and mortality**

An emerging body of evidence suggests that haemoglobin variability is associated with increased risk of all-cause death in both non-dialysis CKD and ESKD patients. Boudville and colleagues found an association between haemoglobin variability and death not only in non-dialysis CKD patients treated with ESAs (n=1,823), but also in those who were not on ESAs (n=3,143) (Boudville et al., 2009). For each 1 g/L increase in the residual standard deviation, HR (95%CI) for patients on ESA throughout the study and those who were not receiving ESA were 1.02 (1.01 to 1.04) and 1.03 (1.02 to 1.05), respectively. The analysis of the pooled data from these 2 groups showed similar results. Compared to patients with haemoglobin levels consistently within the target range, those with low amplitude fluctuation with low haemoglobin values (HR 1.62, 95%CI 1.36 to 1.94) and high amplitude fluctuation (HR 1.57, 95%CI 1.24 to 1.98) were at increased risk of all-cause mortality.

Association Between Haemoglobin Variability

consistent findings.

and Clinical Outcomes in Chronic Kidney Disease 35

Not all studies have demonstrated a positive association between haemoglobin variability and death in CKD. Eckardt and colleagues studied the effect of haemoglobin variability on mortality using 5 definitions: within-patient standard deviation, residual standard deviation, time-in-target haemoglobin, amplitudes of fluctuation, and AUC (Eckardt et al., 2010). In the adjusted Cox regression model, haemoglobin variability was not a statistically significant factor in all 5 methods, except for the group of patients with low amplitude fluctuations with low haemoglobin levels (HR 1.74, 95%CI 1.00 to 3.04). However, 95% confidence intervals were very wide and the lower 95% confidence interval was 1.00. Thus,

Brunelli and colleagues included a retrospective cohort of 6,644 incident patients who commenced haemodialysis between 2004 and 2005 from the Fresenius Medical Centre database (Brunelli et. al., 2008). In contrast to their 1996 cohort study results (Yang et al., 2007), the association between haemoglobin variability and mortality was not statistically significant. The unadjusted and adjusted HR (95%CI) for were 0.96 (0.81 to 1.14) and 1.11 (0.92 to 1.33), respectively. The discrepancy in these analyses may be explained by the addition of a large number of variables in the Cox regression model for the 2004-2005 cohort as more data were available. Consequently, when the analysis was restricted using the same limited variables as in the 1996 cohort study; the association achieved statistical significance with a HR of 1.22 (1.01 to 1.48). Although the investigators have attempted to adjust for known variables, the possibility of residual confounding could not be excluded. The complexity of these statistical models makes interpretation of the results difficult, particularly when the different statistical methods or approaches did not generate robust or

Weinhandl and colleagues reported the association between haemoglobin variability and all-cause mortality using 3 Cox proportional hazards regression models (Weinhandl et al., 2011). In the case-mix-adjusted model, the HR (95%CI) for the contemporary prevalent, historical prevalent and incident groups for 1 g/dL haemoglobin variability were 1.27 (1.24 to 1.31), 1.32 (1.27 to 1.38) and 1.08 (1.03 to 1.13), respectively. In the comorbid conditionadjusted model, haemoglobin variability was associated with increased risk of death in both the prevalent groups with a HR of 1.07 (1.04 to 1.10) in the contemporary prevalent group and 1.10 (1.06 to 1.15) in the historical prevalent group. However, there was no statistically significant association in the incident group (HR 1.03, 95%CI 0.98 to 1.09). In the expanded comorbid condition-adjusted model, the statistically significant association was limited only to the historical prevalent group (HR 1.07, 95%CI 1.03 to 1.12). Haemoglobin variability was not associated with increased risk of death in either the contemporary prevalent (HR 1.02, 95%CI 0.99 to 1.05) or incident groups (HR 1.01, 95%CI 0.95 to 1.06). These findings suggest that the association between haemoglobin variability and mortality was weak in this study

Since the introduction of ESAs, most of the clinical trials with ESA therapy have focused on haemoglobin targets in CKD patients. There is a shortage of clinical trials studying the optimal strategy for haemoglobin monitoring in patients treated with ESAs and

although this association was statistically significant, it was weak.

and was sensitive to adjustment for concurrent disease severity.

**5. Management of haemoglobin variability** 

interventions to reduce haemoglobin variability.

Minutolo and colleagues reported that longer time with haemoglobin within the target range of 11 to 13 g/dL was associated with decreased risk of renal death (defined as a composite endpoint of all-cause death on dialysis or after kidney transplantation) (Minutolo et al., 2009).

In a study involving 34,963 haemodialysis patients who were enrolled in the Fresenius Medical Care database in 1996, Yang and colleagues reported that the risk of all-cause mortality increased proportionately with haemoglobin variability (Yang et al., 2007). The hazard ratio and 95% confidence intervals (CI) per 0.50 g/dL, 0.75 g/dL, 1.00 g/dL, and 1.50 g/dL increases in haemoglobin variability were 1.15 (1.10 to 1.20), 1.24 (1.16 to 1.32), 1.33 (1.22 to 1.45), and 1.53 (1.35 to 1.75), respectively.

Gilbertson and colleagues found that out of 6 categories of haemoglobin variability (categorized as low <11g/dL, intermediate 11 – 12.5g/dL & high >12.5g/dL and further divided into low-low, intermediate-intermediate, high-high, low-intermediate, intermediate-high, low-high), patients in the low-high and low-intermediate groups experienced an increased risk of death compared with those in the intermediateintermediate group (Gilbertson et al., 2008). The HR and 95%CI for the low-high and lowintermediate groups were 1.19 (1.10 to 1.28) and 1.44 (1.33 to 1.56), respectively. Although this categorization broadly identifies patients with stable hemoglobin and either low or high amplitude fluctuations in hemoglobin, it assumes a unidirectional and linear change in hemoglobin.

Lau and colleagues measured haemoglobin variability as rate of haemoglobin change: average positive-only (positive haemoglobin deflections) and average negative-only (negative haemoglobin deflections) (Lau et al., 2010). While negative haemoglobin deflections were not associated with mortality risk (HR 1.07, 95% CI 0.94 to 1.21 per g/L/week), rapid rise in haemoglobin was associated increased risk (HR 1.23, 95% CI 1.03 to 1.48 per g/L/week).

Regidor and colleagues analysed a cohort of 58,058 prevalent haemodialysis patients from the DaVita dialysis organisation (Regidor et al., 2006). Compared to patients whose haemoglobin remain unchanged during the first 6 mo of the 2-year cohort study period, the risks of all-cause and cardiovascular mortality were significantly higher in patients with reduction in haemoglobin by more than 1.50 g/dL/quarter. In the fully-adjusted model (adjusted for demographic characteristics, comorbidities, smoking, dialysis dose, nutritional status, iron studies, doses of ESAs and iron), there was no association observed with increment in haemoglobin and mortality. However, decrease in hemoglobin was associated with increased mortality. While this study described an association between change in hemoglobin per quarter and mortality, it did not specifically study an association between hemoglobin variability and mortality.

Pisoni and colleagues reported mortality outcomes using haemoglobin variability at the facility-level using the DOPPS database (Pisoni et al., 2011). In the adjusted model, the HR for every 0.5 g/dL higher facility-level haemoglobin standard deviation was 1.08 (95% CI 1.02 to 1.15). Compared to the reference category of the lowest quartile of facility-level haemoglobin standard deviation, the HR (95% CI) for the 2nd, 3rd and 4th quartiles were 1.08 (1.10 to 1.34), 1.15 (1.35 to 1.69) and 1.19 (1.04 to 1.37), respectively. As previously mentioned, the facility-level haemoglobin standard deviation correlated well with withinpatient haemoglobin standard deviation.

Minutolo and colleagues reported that longer time with haemoglobin within the target range of 11 to 13 g/dL was associated with decreased risk of renal death (defined as a composite endpoint of all-cause death on dialysis or after kidney transplantation) (Minutolo et al., 2009). In a study involving 34,963 haemodialysis patients who were enrolled in the Fresenius Medical Care database in 1996, Yang and colleagues reported that the risk of all-cause mortality increased proportionately with haemoglobin variability (Yang et al., 2007). The hazard ratio and 95% confidence intervals (CI) per 0.50 g/dL, 0.75 g/dL, 1.00 g/dL, and 1.50 g/dL increases in haemoglobin variability were 1.15 (1.10 to 1.20), 1.24 (1.16 to 1.32), 1.33

Gilbertson and colleagues found that out of 6 categories of haemoglobin variability (categorized as low <11g/dL, intermediate 11 – 12.5g/dL & high >12.5g/dL and further divided into low-low, intermediate-intermediate, high-high, low-intermediate, intermediate-high, low-high), patients in the low-high and low-intermediate groups experienced an increased risk of death compared with those in the intermediateintermediate group (Gilbertson et al., 2008). The HR and 95%CI for the low-high and lowintermediate groups were 1.19 (1.10 to 1.28) and 1.44 (1.33 to 1.56), respectively. Although this categorization broadly identifies patients with stable hemoglobin and either low or high amplitude fluctuations in hemoglobin, it assumes a unidirectional and linear change in

Lau and colleagues measured haemoglobin variability as rate of haemoglobin change: average positive-only (positive haemoglobin deflections) and average negative-only (negative haemoglobin deflections) (Lau et al., 2010). While negative haemoglobin deflections were not associated with mortality risk (HR 1.07, 95% CI 0.94 to 1.21 per g/L/week), rapid rise in haemoglobin was associated increased risk (HR 1.23, 95% CI 1.03

Regidor and colleagues analysed a cohort of 58,058 prevalent haemodialysis patients from the DaVita dialysis organisation (Regidor et al., 2006). Compared to patients whose haemoglobin remain unchanged during the first 6 mo of the 2-year cohort study period, the risks of all-cause and cardiovascular mortality were significantly higher in patients with reduction in haemoglobin by more than 1.50 g/dL/quarter. In the fully-adjusted model (adjusted for demographic characteristics, comorbidities, smoking, dialysis dose, nutritional status, iron studies, doses of ESAs and iron), there was no association observed with increment in haemoglobin and mortality. However, decrease in hemoglobin was associated with increased mortality. While this study described an association between change in hemoglobin per quarter and mortality, it did not specifically study an association between

Pisoni and colleagues reported mortality outcomes using haemoglobin variability at the facility-level using the DOPPS database (Pisoni et al., 2011). In the adjusted model, the HR for every 0.5 g/dL higher facility-level haemoglobin standard deviation was 1.08 (95% CI 1.02 to 1.15). Compared to the reference category of the lowest quartile of facility-level haemoglobin standard deviation, the HR (95% CI) for the 2nd, 3rd and 4th quartiles were 1.08 (1.10 to 1.34), 1.15 (1.35 to 1.69) and 1.19 (1.04 to 1.37), respectively. As previously mentioned, the facility-level haemoglobin standard deviation correlated well with within-

(1.22 to 1.45), and 1.53 (1.35 to 1.75), respectively.

hemoglobin.

to 1.48 per g/L/week).

hemoglobin variability and mortality.

patient haemoglobin standard deviation.

Not all studies have demonstrated a positive association between haemoglobin variability and death in CKD. Eckardt and colleagues studied the effect of haemoglobin variability on mortality using 5 definitions: within-patient standard deviation, residual standard deviation, time-in-target haemoglobin, amplitudes of fluctuation, and AUC (Eckardt et al., 2010). In the adjusted Cox regression model, haemoglobin variability was not a statistically significant factor in all 5 methods, except for the group of patients with low amplitude fluctuations with low haemoglobin levels (HR 1.74, 95%CI 1.00 to 3.04). However, 95% confidence intervals were very wide and the lower 95% confidence interval was 1.00. Thus, although this association was statistically significant, it was weak.

Brunelli and colleagues included a retrospective cohort of 6,644 incident patients who commenced haemodialysis between 2004 and 2005 from the Fresenius Medical Centre database (Brunelli et. al., 2008). In contrast to their 1996 cohort study results (Yang et al., 2007), the association between haemoglobin variability and mortality was not statistically significant. The unadjusted and adjusted HR (95%CI) for were 0.96 (0.81 to 1.14) and 1.11 (0.92 to 1.33), respectively. The discrepancy in these analyses may be explained by the addition of a large number of variables in the Cox regression model for the 2004-2005 cohort as more data were available. Consequently, when the analysis was restricted using the same limited variables as in the 1996 cohort study; the association achieved statistical significance with a HR of 1.22 (1.01 to 1.48). Although the investigators have attempted to adjust for known variables, the possibility of residual confounding could not be excluded. The complexity of these statistical models makes interpretation of the results difficult, particularly when the different statistical methods or approaches did not generate robust or consistent findings.

Weinhandl and colleagues reported the association between haemoglobin variability and all-cause mortality using 3 Cox proportional hazards regression models (Weinhandl et al., 2011). In the case-mix-adjusted model, the HR (95%CI) for the contemporary prevalent, historical prevalent and incident groups for 1 g/dL haemoglobin variability were 1.27 (1.24 to 1.31), 1.32 (1.27 to 1.38) and 1.08 (1.03 to 1.13), respectively. In the comorbid conditionadjusted model, haemoglobin variability was associated with increased risk of death in both the prevalent groups with a HR of 1.07 (1.04 to 1.10) in the contemporary prevalent group and 1.10 (1.06 to 1.15) in the historical prevalent group. However, there was no statistically significant association in the incident group (HR 1.03, 95%CI 0.98 to 1.09). In the expanded comorbid condition-adjusted model, the statistically significant association was limited only to the historical prevalent group (HR 1.07, 95%CI 1.03 to 1.12). Haemoglobin variability was not associated with increased risk of death in either the contemporary prevalent (HR 1.02, 95%CI 0.99 to 1.05) or incident groups (HR 1.01, 95%CI 0.95 to 1.06). These findings suggest that the association between haemoglobin variability and mortality was weak in this study and was sensitive to adjustment for concurrent disease severity.
