Survival by Dialysis Modality in Critically Ill Patients with Acute Kidney Injury

Materials and Methods

Results

Of the 398 patients who required dialysis for severe AKI, 206 started on CRRT and 192 started on IHD. Table 1 shows demographic, historical, clinical, and selected laboratory values by initial dialysis modality. Modality assignment differed significantly by site: Initial assignment to CRRT ranged from 27% at Vanderbilt University to 61% at CCF and University of California San Diego. In general, patients who were assigned to CRRT had more organ system failure and more significant physiologic disturbances, including hypotension and tachycardia. Fewer than half of all patients had a pulmonary artery (PA) catheter in place on the day of dialysis initiation, although PA catheter use was more common, as expected, among patients who were started on CRRT (96 [47%] of 206 versus 43 [22%] of 192; P < 0.0001). Among patients with a PA catheter, there were no significant differences in mean PA systolic (P = 0.17) or diastolic (P = 0.89) pressure or pulmonary capillary wedge pressure (P = 0.54) by initial dialysis modality.

Discussion

A relatively large fraction of critically ill patients with severe AKI require dialysis during an ICU stay. The traditional approach to management is IHD, usually delivered three times per week for several hours per session, not unlike maintenance hemodialysis that is used in patients with ESRD. In the acute setting, peritoneal dialysis has fallen out of favor as a result of infectious and other (e.g., respiratory and metabolic) complications. In the past two decades, CRRT, including continuous venovenous hemodialysis (CVVHD), continuous venovenous hemofiltration, and continuous venovenous hemodiafiltration, have gained in popularity and usage, particularly for the treatment of hemodynamically unstable patients. Although several small studies have suggested improved physiologic parameters in response to CRRT relative to IHD (11–16), these data are not definitive. Evaluation of outcomes that are associated with modality choice are hampered by residual confounding and selection bias, because most programs tend to use CRRT for more severely, acutely ill patients (18,22,30). For example, Swartz et al. (18) compared mortality rates by modality assignment in 349 patients with dialysis-requiring AKI during 1995 through 1996 at the University of Michigan. The odds for death with assignment to CRRT was roughly twice that of IHD; however, after exclusion of patients with hypotension (systolic BP <90 mmHg), severe hyperbilirubinemia (>15 mg/dl), and a very short period of renal replacement (<48 h), the risk for death that was associated with CRRT no longer was significantly higher (RR 1.09; 95% CI 0.67 to 1.80). In a re-analysis of this cohort, Martin et al. (31) found that the CRRT-treated patients had an unexpected higher mortality, particularly in patients who were categorized as low risk by the Cleveland Clinic score. In a follow-up study during 2000 through 2001 (n = 383), Swartz et al. (32) found an increase in the risk for death with CRRT on unadjusted analysis and no significant difference after adjustment for comorbidity and severity of illness. Subgroup analyses were conducted, some of which suggested favorable trends with CRRT, although none was statistically significant and there was no consideration of multiple comparisons. Chang et al. (33) described 148 South Korean ICU patients with dialysis-requiring AKI. As expected, patients who were treated with continuous venovenous hemodiafiltration were more severely ill and had significantly lower survival rates (21 versus 46%; P = 0.002). On subgroup analysis, patients with APACHE III scores >103 and more than three organ failures had nominally higher survival with CRRT. Other observational studies have demonstrated conflicting findings, examining mortality and renal recovery after dialysis-requiring AKI.

Several randomized clinical trials have compared CRRT and IHD in severe AKI, although none has been adequately powered. In the largest randomized clinical trial, Mehta et al. (19) compared CRRT and IHD in 166 critically ill patients with severe AKI and found a significantly higher ICU mortality rate in patients who were randomly assigned to CRRT (60 versus 42%; P = 0.02). However, despite randomization, patients who were assigned to CRRT were significantly more likely to have liver failure and had a higher overall severity of illness, as determined by APACHE III score. Adjustment for these factors attenuated the increased risk that was attributed to CRRT (OR 1.6; 95% CI 0.7 to 3.3). More recently, two (underpowered) randomized clinical trials that compared CRRT and IHD failed to show a significant difference in survival by dialysis modality (21,34). Meta-analyses also have been conducted and have concluded that there is no significant difference in survival by modality, although study quality and heterogeneity were not optimal (17,20).

This study extends previous work in this area using observational data by incorporating multiple sites, a larger sample size, multivariable regression analysis, and the propensity score approach to address residual confounding and selection effects. Although data collection in PICARD generally was comprehensive, all data elements were not collected in all patients (e.g., data from PA catheters), largely as a result of differences in clinical practice among and within sites. In addition, we could not control for other aspects of dialysis care (e.g., urea or other solute clearance) and other co-interventions that were not randomly assigned.

Although propensity scores cannot fully adjust for residual confounding and selection bias, the method has been widely used in observational studies that have examined the effectiveness of various interventions in nephrology and critical care. As with other conditions for which there is uncertainty as to the optimal therapeutic approach, there tends to be wide variation in practice by institution and individual physician in the choice of modality for dialysis-requiring AKI. A propensity score can help account for this variation, which may be unrelated to severity of illness or other biologic factors that influence outcomes. We previously used propensity scores to estimate the effects of the timing of consultation (35) and the use of diuretics (36) and dopamine (37) in AKI. Propensity scores also were used in studies that evaluated the effectiveness of albumin administration (38), blood transfusion (39), and right heart catheterization (40) in the critically ill.

It should be emphasized that the multivariable analyses (with or without the propensity score) support the crude (unadjusted) results that demonstrate an increased risk for death among patients who are assigned to CRRT. In the “naïve” multivariable analyses (not adjusted for the propensity score), we adjusted for significant predictors of death at dialysis initiation. Inclusion of other covariates (e.g., mechanical ventilation, clinical criteria for acute lung injury or adult respiratory distress syndrome, systolic BP) in addition to or in place of the core model covariates did not extinguish the association between assignment to CRRT and mortality. Moreover, when considering factors that were associated with assignment to CRRT but were not significant predictors of death (e.g., use of PA catheter, intake-output balance), we observed risk ratios in the same direction and of the same magnitude as in the unadjusted analysis.

There are several important limitations to this study. First, propensity scores can adjust only for the associations among observed covariates and the chosen treatment or strategy. Other unobserved covariates could influence the likelihood of treatment, and there is no guarantee that the correlation among observed and unobserved covariates is sufficiently high to account adequately for this deficiency. Second, the study was conducted at five academic tertiary care medical centers. Therefore, results may not be fully generalizable to other medical centers, particularly those where CRRT is applied less frequently. Third, we collected no information on long-term survival, functional status, or dialysis dependence for patients who survived hospitalization. Future observational studies and clinical trials in AKI should attempt to understand the long-term effects of dialysis-requiring AKI episodes. Finally, although an extensive number of variables were collected and were done so serially during patients’ ICU stays, we could not capture every aspect of intensive care, so residual confounding by severity of illness is likely.

Although the major results described here could reflect residual confounding, the possibility that CRRT might cause harm still should be considered. CRRT requires continuous anticoagulation; may remove water-soluble vitamins, drugs (including antibiotics), and amino acids; and may result in clearance and/or adsorption of a variety of known and unknown modulators of the inflammatory and counterinflammatory response. Moreover, despite technological advances, it remains a complex intensive therapy that requires considerable nursing and physician expertise. Given the high incidence of AKI in the ICU and the morbidity, associated mortality, and costs that are associated with dialysis-requiring AKI, better evidence is needed to guide AKI treatment strategies.

There are numerous examples in which drugs, devices, and technologies have been introduced into medical practice on the basis of a sound rationale, yet subsequent clinical practice and research demonstrate that the use of these technologies is associated with no benefit or even harm. Recent examples include the use of PA catheters in the ICU (41,42), the use of certain antiarrhythmic agents in an attempt to prevent sudden cardiac death (43), and the use of selected inotropes and vasodilators in the treatment of congestive heart failure (44). When the results of the current observational study on dialysis-requiring AKI in the ICU are integrated with other observational studies and clinical trials, there is no evidence of the superiority of CRRT over IHD and some evidence to support possible inferiority.

At this time, the data that are presented here should be considered hypothesis generating and not definitive and should not be used to change clinical practice. However, in the context of increasing CRRT use, it now is imperative that a randomized clinical trial of adequate power be conducted to determine whether mortality rates that are associated with severe dialysis-requiring AKI can be reduced with the use of CRRT, IHD, or a hybrid technique, such as slow low-efficiency dialysis. Such a study will need to include patients who could be treated successfully with either modality (comparing “apples and apples”), potentially excluding patients with severe hypotension and hemodynamic instability, who may be poor candidates for traditional IHD, and should standardize key elements of therapy, including the timing of initiation, dosage of dialysis, and the expertise of personnel delivering the therapy.