In the Literature: On Clinical Performance Measures and Outcomes Among Hemodialysis Patients

What Did This Study Show? 

Rocco et al2 studied 15,287 prevalent patients, who were selected from a 5% random sample of patients receiving long-term outpatient hemodialysis in the US from 1999 and 2000. The clinical performance measure targets were hemoglobin value of 11 g/dL (110 g/L) or greater; serum albumin value of 4 g/dL (40 g/L) or greater or 3.7 g/dL (37 g/L) or greater (bromcresol green and bromcresol purple laboratory methods, respectively); use of a fistula for vascular access; and measured single-pool Kt/V urea value of 1.2 or greater. In multivariable analysis, they determined whether achieving these performance measures was associated with better mortality and hospitalization rates during the next 12 months. During the initial period, 6% of patients did not meet any target, 24% met 1 target, 39% met 2 targets, 24% met 3 targets, and 7% met all 4 targets. During the 12-month follow-up, 55% of patients were hospitalized and 20% died. Patients who met all targets were more likely to be male, younger, of white race, of Hispanic ethnicity, have hypertension or glomerulonephritis as the cause of ESRD, have a lower body mass index, and to have received dialysis for more years; they were less likely to have diabetes mellitus and comorbid conditions (both cardiovascular and noncardiovascular). In other words they were healthier and they were survivors. Adjusted hazard ratios and 95% confidence intervals were 4.6 (3.3-6.4), 3.5 (2.6-4.7), 2.6 (1.9-3.5), and 1.9 (1.4-2.6) for 0, 1, 2, or 3 targets met, respectively, compared with the reference group who met all 4 targets. The authors concluded that meeting multiple clinical measure targets was associated with a decrease in hospitalization and mortality rates. Although the authors stress that it was not possible to determine causality they do state that attainment of “individual quality indicators [was] associated with reductions in death and hospitalization.”

As pointed out by the authors, the study has important limitations. It was not possible to determine the roles of severity of illness, other patient factors, or suboptimal care in failure to meet performance indicators. However achievement of targets was associated with younger age, better health, and being a survivor. It is unlikely that the multiple regression modeling can take into account all the clinical differences between achievers of targets and nonachievers. Consequently an equally tenable hypothesis, deriving from these results, is that failure to achieve targets is a marker for patients who will have adverse clinical outcomes, because they are older and sicker, and that achievement of the recommended targets will not lead to reductions in mortality and hospitalization. This conclusion is supported by the hazard associated with hypoalbuminemia, which is a marker for inflammation, malnutrition, and comorbidity. Serum albumin, a strong predictor of adverse outcomes,3 could not be targeted for quality improvement interventions, because no efficacious interventions were available. As a result serum albumin levels had not changed during the 10-year period of observation.

How Does This Study Compare With Other Studies? 

Prospective longitudinal studies of cohorts of patients have been used to identify risk factors, but in dialysis patients these studies are prone to multiple methodologic problems including survivor bias caused by enrolling prevalent rather than incident patients, unmeasured comorbidity, multiple interventions, selection bias for interventions, and inadequate power to take account of multiple known risk factors. In addition, identification of risk in a cohort study does not indicate that reduction in outcomes will occur by treating the risk factor, because the identified risk factor may only be a marker for some other associated factor which is causing death. Consequently the treatment hypothesis produced by a cohort study must be proven (where possible) by an RCT in which the new intervention is compared to conventional therapy in an adequately powered study of groups of patients considered to be similar because they were randomly allocated to either the intervention or control group. In the absence of an RCT proving efficacy of an intervention to improve an outcome, demonstration of an association between achievement of a performance measure and an outcome is subject to the same methodologic problems as in a cohort study demonstrating an association between the risk factor and the outcome. In addition, another limitation, dose-targeting bias, must be considered in observational studies relating achievement of performance measures to outcomes. Dose-targeting bias describes the likelihood that failure to achieve the performance measure is a marker for some other risk factor which is associated with both unresponsiveness to therapy and subsequent adverse outcomes. Dose-targeting bias is a likely explanation for the divergent results of cohort studies which examined the risk of anemia4, 5 and of different quantities of dialysis (as measured by Kt/V),6 and subsequent RCTs which examined the effect of anemia therapy7, 8, 9, 10 and of increasing dose of dialysis.11, 12

Multiple cohort studies have confirmed that anemia is a cardiac and mortality risk factor,4, 5 but multiple RCTs comparing normalization of hemoglobin with erythropoietin to partial correction of anemia failed to demonstrate a clinical benefit, and produced a signal of potential harm in the higher hemoglobin group.7, 8, 9, 10 While the hazard in the high hemoglobin group was opposite to that anticipated from the cohort studies (Fig 1A), the relationships between achieved hemoglobin and primary outcome in both the intervention and control groups were similar to those observed in cohort studies7 (Fig 1B). It is likely that failure to achieve the higher hemoglobin level in either the intervention or the control group was a marker for some other adverse risk factor unresponsive to anemia therapy.

View Large Image Download to PowerPoint

Figure 1. (A) Probability of death or first nonfatal myocardial infarction in dialysis patients treated with erythropoietin randomly allocated to high and low hematocrit levels. (B) Mean mortality rate (+/− SE) as a function of the average achieved hematocrit in the normal and low hematocrit groups in dialysis patients treated with erythropoietin. Both panels reproduced with permission from Besarab et al7; copyright © 1998 Massachusetts Medical Society, all rights reserved.

A very similar observation was made in the Hemodialysis (HEMO) Study, an RCT where higher dose of dialysis did not prevent adverse clinical events compared to a lower dose, but higher achieved Kt/V in both intervention and control groups was associated with better clinical outcomes11, 12 (Fig 2). Again the latter results from these 2 cohorts were consistent with those reported in previous cohort studies.6 Thus failure to achieve the higher Kt/V in either group was likely associated with some adverse risk factor unresponsive to increasing dialysis dose.

View Large Image Download to PowerPoint

Figure 2. (A) Cumulative survival in hemodialysis patients randomly allocated to high dose of dialysis (urea reduction of about 75% or single pool Kt/V of about 1.65) and to standard dose (urea reduction of about 65% or single pool Kt/V of 1.25). Reproduced with permission from Eknoyan et al11; copyright © 2002 Massachusetts Medical Society, all rights reserved. (B) Association of mortality with the most recent 4 month mean equilibrated Kt/V (e Kt/V). Shown are the adjusted relative risks (RR; and 95% confidence intervals[CI]) of mortality within each of the 5 quintiles of the 4-month average e KT/V values of the standard-dose group and the first, second, fourth, and fifth quintiles of the high dose group, with the middle quintile of the high-dose group as reference. Reproduced with permission from Greene et al12; copyright © 2005 American Society of Nephrology.

What Should Clinicians and Researchers Do? 

Of the 4 quality indicators studied by Rocco et al,2 only a higher dialysis dose has RCT evidence to support the expectations that achievement of these targets would be associated with better hard clinical outcomes.13 Thus, a clinical performance measure to achieve Kt/V > 1.2 is justified. One cannot conclude from the Rocco et al analyses that achievement of other clinical performance targets will lead to reduction in mortality and hospitalization. However achievement of some targets should be associated with other clinical benefits. RCTs demonstrate that treatment of anemia to a target hemoglobin value of 11 g/dL (110 g/L) or greater will improve quality of life and limit blood transfusions.14 It is likely that preferential use of fistulas rather than grafts should lead to better vascular access outcomes.15 Consequently quality assurance initiatives to achieve these targets can be supported. However, a serum albumin target of 4 g/dL (40g/L) as a clinical performance measure is illusory at present, as it is not currently modifiable.

Questions on the efficacy of and target levels for interventions should be answered using RCTs. Effect sizes derived from cohort studies should not be used to influence clinical practice, but to create hypotheses to test in RCTs, except in those situations where the intervention is clearly a dominant strategy as in the provision of dialysis in ESRD, cessation of smoking, or provision of defibrillators in inherited arrhythmogenic right ventricular cardiomyopathy.16 As Rocco et al2 suggest, the premise that achievement of multiple clinical performance measures will lead to improvement in mortality and hospitalization is testable, but will require an experimental trial which randomly allocates patients to intervention and control groups.