Renal Replacement Therapy in Patients With Acute Renal Failure:  A Systematic Review FREE

Acute renal failure (ARF) is increasingly common and is associated with high costs and adverse clinical outcomes, including excess mortality, increased length of hospital stay, and the requirement for chronic dialysis in survivors.¹ Diverse options are currently available for prescribing acute renal replacement, including intermittent, continuous, and extended-duration hemodialysis and hemofiltration and combinations thereof. Despite advances in dialysis technology, many questions remain about how best to provide renal replacement to patients with ARF.

This review will critically evaluate current evidence for the optimal dialytic management of ARF, present an evidence-based approach to this clinically important problem, and identify key areas for future research.

This study was conducted and reported in accordance with published guidelines.^2- 3

An expert librarian conducted a comprehensive search to identify prospective cohort studies of renal replacement therapies (RRTs) in patients with ARF. Only articles published as full manuscripts in English were considered. MEDLINE (1966-October 2007), EMBASE (1988-October 2007), All EBM Reviews (October 2007), and a variety of gray-literature sources (n = 36) were searched (clinical trial registries, health technology assessment agencies, and manufacturer Web sites; for detailed search strategies, see the Alberta Kidney Disease Network Web site at http://www.akdn.info). Each citation or abstract was screened by a subject specialist and a methodologist. Any study considered relevant by 1 or 2 reviewers was retrieved for further review. The reference lists of included trials and relevant reviews were also reviewed for pertinent trials. We also contacted manufacturers of renal replacement therapy products and the authors of included studies for information about further studies.

Each potentially relevant study was independently assessed by 2 reviewers for inclusion in the review using predetermined eligibility criteria and a preprinted form. Studies meeting the following criteria were eligible: (1) Study design: randomized controlled trials (RCTs) or prospective cohort studies with 10 or more allocated participants in at least 2 intervention groups. Data collected prospectively for the purposes of registries or quality assurance databases were categorized as retrospective⁴ and therefore excluded. (2) Population: adults with ARF. (3) Comparison: RRT modality, time of RRT initiation, dose, schedule, flux, membrane type, buffer, and anticoagulant. (4) Outcomes: mortality, length of stay, chronic dialysis dependence, blood pressure, and hypotension (filter failure and bleeding complications for anticoagulant comparison). Disagreements regarding eligibility arose with 10% of the articles (κ = 0.79); disagreements were resolved by a third party through consensus.

We assessed characteristics of study quality for all RCTs (eg, method of allocation concealment⁵ and Jadad score⁶). Prospective cohort studies were assessed using 4 items from the Downs and Black checklist⁷: by whom and when treatment groups were accrued, description of withdrawals/dropouts, and whether there was adjustment for potential confounders. Additionally, we extracted data on funding sources⁸ and intention-to-treat design, given their potential to introduce bias. Two reviewers assessed each included study independently. Disagreements (observed in 17% to 33% of cases) were resolved by consensus with the aid of a third party.

We used a standard method to record the following properties of each study: study characteristics (country, design, sample size, setting, duration of follow-up); participants (age, sex, presence of liver failure or systemic inflammatory response syndrome), illness severity (levels of serum creatinine and urea, Acute Physiology and Chronic Health Evaluation score, need for mechanical ventilation, use of vasopressors or inotropes); renal replacement regimens (technique, device and manufacturer, membrane material and flux, dose, schedule, buffer, prescribed blood flow, prescribed dialysate flow, anticoagulation regimen); outcomes (timing of outcome, in all patients or in survivors only); and results. We considered the following outcomes: mortality (in–intensive care unit [ICU], in-hospital, at any other time point); length of stay (ICU, hospital); chronic dialysis dependence, blood pressure (mean arterial pressure [MAP], hypotension); filter life; and bleeding complications. One reviewer extracted the data and a second checked for accuracy.

We analyzed data using Review Manager version 4.2.7 (The Cochrane Collaboration, Oxford, England) and Stata release 10.0 (StataCorp, College Station, Texas). Due to the differences expected between studies, we decided a priori to combine results using a random-effects model.⁹ Furthermore, we pooled results for RCTs but not for prospective cohort studies, given the susceptibility to bias in observational studies.¹⁰ For dichotomous outcomes, we used the relative risk (RR) to pool outcomes. Change in MAP was pooled using the weighted mean difference. A correlation of 0.5 was substituted when change-from-baseline correlations were not available in the published reports.¹¹ Length-of-stay data were generally associated with highly right-skewed distributions and therefore were not pooled. Measures of central tendency, spread, tests, and corresponding P values were extracted for each group.

Statistical heterogeneity was quantified for pooled results using the I² statistic,^12- 13 which approximates the percentage of total variation (within and between studies) due to between-study variation. We did not assess publication bias,¹⁴ because each pooled estimate included fewer than 10 trials.

Of 173 studies reviewed in detail, 30 RCTs^15- 44 and 8 prospective cohort studies^45- 52 comparing dialytic strategies in ARF were eligible for inclusion (Figure 1, Table 1, and Table 2). Pooled results for the effect of these strategies on clinically important outcomes are presented in Figure 2, Figure 3, and Figure 4. Additional forest plots that present the results in more detail are available at the Alberta Kidney Disease Network Web site (http://www.akdn.info) or by contacting the authors.

Changes in urine output and levels of serum creatinine traditionally have been used to identify and define ARF. Oliguria is an insensitive indicator of ARF, and nonoliguric ARF may be overlooked if urine output remains adequate. Serum creatinine level is also insensitive to acute changes in renal function⁵³ and may be affected by factors such as age, sex, race, muscle mass, and medications.^54- 55 Therefore, cystatin C may better reflect true glomerular filtration rate in this setting.⁵⁶ Despite these caveats, changes in levels of serum creatinine remain the gold standard for detection of ARF.

To date, there have been more than 35 published definitions of ARF,⁵⁷ which has hampered comparisons between studies. A group that attempted to produce a consensus definition (the Risk, Injury, Failure, Loss, and End-stage Kidney [RIFLE] criteria⁵⁷) stratified patients into 3 categories of severity (risk, injury, and failure) and 2 outcomes (receipt of RRT for >4 weeks and permanent kidney failure), and proposed that the term “acute renal failure” be replaced by “acute kidney injury.” More recently, these criteria have been simplified to focus only on change in levels of serum creatinine (>0.3 mg/dL [25 μmol/L] or a 50% increase) and urine output (<0.5 mL/kg per hour for longer than 6 hours).⁵⁸ While such changes in levels of serum creatinine have been repeatedly linked to adverse outcomes,⁵⁹ the urine output criteria have not been similarly evaluated. The applicability of the RIFLE and Acute Kidney Injury Network criteria to non–critically ill patients is unknown. At present, there is no universally accepted nomenclature for this condition, and we have used the term “acute renal failure” in this review.

Several recent studies suggest that the incidence of in-hospital ARF has increased by more than 10% per year over the last decade,^60- 65 probably due to the increasing average age and comorbidity of hospitalized patients. Using a large national database, Waikar et al⁶⁴ found that the incidence of ARF in the United States was 21 per 1000 hospital discharges in 2002. The epidemiology of ARF has been best characterized in critically ill populations,^60- 62 in whom the incidence varies between 15% and 80% by RIFLE criteria.⁵⁹ The incidence of acute dialysis-dependent renal failure is considerably lower (3.4% to 4.9%), and associated in-hospital mortality generally is between 60% and 70%.^66- 70 While the requirement for renal replacement clearly portends a poor prognosis in patients with ARF, long-term outcomes have not been well studied. The majority of survivors are able to discontinue acute dialysis,^70- 71 although recent data suggest that up to 30% of such patients will require chronic dialysis within 3 years.^72- 73

In summary, acute dialysis-dependent renal failure is associated with a substantial increase in short-term mortality and may cause or accelerate chronic kidney disease, including kidney failure. A critical review of the evidence on which current dialytic support for patients with ARF is based may identify opportunities to improve these long-term outcomes.

Although it is generally accepted that patients with ARF require renal replacement when reduced glomerular filtration rate leads to established or incipient complications of solute imbalance or extracellular fluid volume overload, the specific indications for initiation of RRT are controversial. The traditional indications for chronic dialysis (such as encephalopathy, pericarditis, and coagulopathy) are late complications of ARF and are rarely seen in contemporary patients with ARF. Critically ill patients should be thoroughly evaluated for nonrenal causes of these conditions when present. Similarly, metabolic acidosis, hyperkalemia, and hypervolemia can often be medically managed. Arguably, the only absolute indications for RRT in critically ill patients are metabolic acidosis, hypervolemia, and hyperkalemia that do not respond to other forms of therapy. Although the early initiation of renal replacement might be beneficial in theory, data guiding the optimal timing of dialysis in patients with ARF are scarce.^46,74- 75

To date only 2 randomized trials^24,26 have addressed this question, both in critically ill patients. In one trial²⁶ (n = 106), “early” dialysis was started after 6 hours of urine output of less than 30 mL/h but did not affect mortality or dialysis dependence in survivors. The other trial²⁴ (n = 28) found a large reduction in mortality among patients with an earlier initiation (RR, 0.17; 95% confidence interval [CI], 0.05-0.61) but had several markers of poor quality and used unusual definitions of “early” and “late” initiation of dialysis, which would be difficult to implement in clinical practice. The single eligible prospective cohort study reported that the risk of death in critically ill patients was significantly lower among patients starting RRT with blood urea nitrogen levels of 76 mg/dL (27 mmol/L) or less (adjusted hazard ratio, 0.54; 95% CI, 0.34-0.86).⁴⁶ Thus the available literature is inconclusive as to the optimal indications for and timing of renal replacement in ARF.

Renal replacement for ARF may include peritoneal dialysis, intermittent hemodialysis, or continuous renal replacement therapies (CRRTs). Although acute peritoneal dialysis is an important method for treatment of adults with severe ARF in developing countries, it is infrequently used in high-income countries and is not further considered in this review. Intermittent hemodialysis is performed using venovenous access for a few hours at variable intervals (typically 4 hours, 3 to 4 times per week).⁷⁶ Sustained, low-efficiency dialysis (SLED) or extended daily dialysis are submodalities of intermittent hemodialysis in which the duration of dialysis is extended (6-12 hours), allowing for more gradual removal of solutes and fluid.⁷⁷

CRRT is performed continuously (ie, approximately 24 hours per day) through arteriovenous or venovenous vascular access, using much slower blood flow rates as compared with intermittent hemodialysis, and is typically only delivered in an intensive care setting. The most commonly applied submodalities of CRRT are continuous venovenous hemofiltration, continuous venovenous hemodialysis, and continuous venovenous hemodiafiltration (Box). CRRT provides slower solute clearance per unit of time compared with intermittent therapies, but over 24 hours, the total clearance may exceed that provided by intermittent hemodialysis, especially for larger solutes such as cytokines.^78- 79 While fluid is removed more slowly using CRRT, it requires continuous anticoagulation (thus creating the potential for bleeding) and involves continuous exposure to an extracorporeal circuit (which might lead to depletion of nutrients, subtherapeutic levels of antimicrobial agents, or infection).

Continuous Venovenous Hemofiltration (CVVHF)

Ultrafiltrate produced is replaced with a replacement solution

 Ultrafiltration in excess of replacement results in patient volume loss

 Solute removal is through convection

Continuous Venovenous Hemodialysis (CVVHD)

Dialysate solution is delivered across membrane countercurrent to blood flow

 Blood flow rates are 100 to 200 mL/min

 Dialysate flow rates are 1 to 2 L/h

 Fluid replacement is not routinely administered

 Solute removal is by diffusion

Continuous Venovenous Hemodiafiltration (CVVHDF)

Dialysate solution is delivered across membrane countercurrent to blood flow

 Typical dialysate flow rates are 1 to 2 L/h

 Ultrafiltration volumes are optimized to exceed desired weight loss and enhance solute clearance from convection

 Fluid losses are replaced in part or completely with replacement solution

 Solute removal is both diffusive and convective

CRRT vs Intermittent Hemodialysis . The effect of dialytic modality on outcomes in ARF has only been directly studied in critically ill patients. We identified 9 RCTs^{17,19- 20,22- 23,25,28- 29,43} (989 total participants) that compared CRRT with intermittent hemodialysis. Of these, 7 trials^{17,19- 20,23,25,28- 29} (918 participants) reported all-cause mortality at discharge from hospital, discharge from the ICU, or 28 days. When the latest follow-up available from each trial was considered, the RR of death due to CRRT was nonsignificant compared with intermittent hemodialysis (RR, 1.10; 95% CI, 0.99-1.23; I² = 0%). Results were similar for ICU mortality and in-hospital mortality (data not shown). Data from prospective cohort studies were generally consistent with those from trials: all 4 such studies^45,48,50,52 that we identified found significantly higher unadjusted mortality among CRRT recipients (who tended to be more severely ill), but of the 3 that reported adjusted summaries, none suggested a significant difference between modalities. One additional (partially randomized) study⁴⁷ found no significant difference in unadjusted mortality between modalities.

Available RCTs did not suggest that dialytic modality influenced the frequency with which chronic dialysis treatment (implying end-stage renal disease) was required in survivors (RR for CRRT vs intermittent hemodialysis, 0.91; 95% CI, 0.56-1.49; I² = 0% [5 trials, 308 participants]). Results were similar when the composite of death or dialysis dependence in survivors was considered (RR, 1.11; 95% CI, 0.87-1.42; I² = 66% [4 trials, 425 participants]). Data from 4 RCTs (643 participants) were inconclusive as to the effect of dialytic modality on hospital length of stay.^17,19,23,28 Two prospective cohort studies^45,48 suggested substantially lower risk of dialysis dependence in survivors who received CRRT (RR, 0.48; 95% CI, 0.25-0.90 and RR, 0.43; 95% CI, 0.30-0.62), but CRRT recipients in these studies were substantially more likely to die and therefore survivorship bias may have influenced these results. This is supported by the fact that the risk of the composite outcome of dialysis dependence or death among CRRT recipients was significantly⁴⁸ and nonsignificantly⁴⁵ higher, respectively, in these studies, compared with intermittent hemodialysis recipients.

Four trials (274 participants) measured MAP at various points. In 3 RCTs with no heterogeneity (I² = 0%), the pooled change in MAP from baseline was no different in patients treated with CRRT or intermittent hemodialysis (mean decrease in MAP, 2.5 mm Hg smaller with CRRT; 95% CI, 1.0 greater to 6.0 smaller). Results of the fourth RCT were substantially different than those of the other 3 (I² after including this study increased to 99%) and suggest that patients treated with CRRT may have a slightly higher MAP over the course of treatment (increase in MAP, 0.4 mm Hg; 95% CI, 7.2 greater to 8.0 smaller). Additionally, the pooled risk of hypotension did not significantly differ between treatments (RR of hypotension with CRRT, 0.87; 95% CI, 0.68-1.12; I² = 0% [2 trials, 389 participants]). The sole relevant eligible prospective cohort study suggested a lower risk of hypotension among CRRT recipients (RR, 0.68; 95% CI, 0.52-0.87), despite the higher illness severity of patients in this group.⁴⁵

In summary, data from 9 RCTs suggest no difference in survival between CRRT and intermittent hemodialysis, while data from a subset of these RCTs suggest no significant difference in the frequency with which chronic dialysis treatment was required in survivors or in the incidence of hypotension.

Techniques for Intermittent Hemodialysis and CRRT . One small trial³⁰ (39 participants) compared hemodiafiltration with hemodialysis in patients treated with intermittent hemodialysis and found no significant differences in clinically relevant outcomes, although statistical power was low. One high-quality trial¹⁶ compared hemodiafiltration with hemofiltration in 206 participants treated with CRRT and found a significant reduction in mortality at 28 days favoring hemodiafiltration over hemofiltration (RR, 0.63; 95% CI, 0.48-0.82). However, participants in the hemodiafiltration group also received a substantially higher dose of RRT than those in the hemofiltration group, making it impossible to determine if submodality (rather than differences in RRT dose) was truly responsible for the better outcomes. In a sensitivity analysis, the pooled results for overall mortality in trials in which the CRRT group used hemodiafiltration exclusively (RR, 1.07; 95% CI, 0.85-1.35; I² = 62% [3 studies, 650 participants]) did not differ from the findings of the main analysis.

SLED vs Other Dialytic Techniques. We identified 2 RCTs^22- 23 that compared SLED (6-11 h/d; 6-7 d/wk) with a continuous modality (continuous venovenous hemofiltration or hemodialysis) with respect to the surrogate outcomes of hemodynamic stability and uremic clearance. No differences were found, although statistical power was low. To our knowledge, no study has compared SLED with conventional intermittent hemodialysis.

In summary, there was no evidence that either CRRT or intermittent hemodialysis was superior for reducing mortality, resource use, or the risk of chronic dialysis dependence in patients with ARF. Although cohort data suggest a lower risk of hypotension among CRRT recipients, this finding was not confirmed by the data reported within the available randomized trials.

Urea clearance is widely used to quantify chronic dialysis dose, and it is tempting to use similar techniques to quantify dose in acute intermittent hemodialysis. However, patients with ARF are frequently catabolic and have highly variable fluid volumes, which violates several of the assumptions on which chronic dialysis dosing is based. Since there is no consensus as to how intermittent hemodialysis dose should be quantified, it is often prescribed in terms of duration (hours per treatment, typically 4 to 6 hours); frequency (typically 3 to 7 treatments per week), and blood flow rate (150 to 400 mL/min). Since urea equilibrates rapidly across the dialysis membrane, urea clearance in CRRT is essentially equivalent to the volume of effluent dialysate (including any ultrafiltered fluid), and therefore CRRT dose is commonly expressed as L/kg per hour of effluent.

CRRT . Two trials with 531 participants compared different doses of hemofiltration. The larger study³⁵ (425 participants) found that doses of 45 mL/kg per hour and 35 mL/kg per hour reduced mortality compared with the dose of 20 mL/kg per hour (RR, 0.72; 95% CI, 0.54-0.94 and RR, 0.73; 95% CI, 0.56-0.96, respectively). No difference was noted in the likelihood of dialysis dependence among survivors, although the composite outcome of mortality and dialysis dependence was significantly less frequent in patients receiving higher hemofiltration volumes. The second trial²⁶ (106 participants) found a nonsignificant trend toward reduced mortality with higher hemofiltration volumes but no difference in the likelihood of dialysis dependence in survivors or in the composite of the 2 outcomes. In pooled analyses of these 2 studies, the risk of mortality (RR, 0.74; 95% CI, 0.63-0.88; I² = 0%) and of the composite of mortality or dialysis dependence (RR, 0.77; 95% CI, 0.67-0.90; I² = 0%) were significantly lower in patients treated with volumes of 35 mL/kg per hour or greater. Only 1 small prospective cohort study addressed this issue and did not find a significant difference between treatment groups, although statistical power was low.⁴⁹

Intermittent Hemodialysis . Two trials have directly compared higher doses of dialysis with conventional (every second day) intermittent hemodialysis. One trial²⁷ (n = 160) found significantly lower mortality (RR, 0.59; 95% CI, 0.39-0.91) in patients treated daily, as compared with a conventional schedule. However, the dose of dialysis in the conventional-dialysis group was somewhat lower than expected (based on experience with chronic dialysis). Intensive therapy in another trial⁴⁴ (34 participants) was defined by hemodialysis performed as often as necessary to maintain levels of blood urea nitrogen less than 60 mg/dL (<21 mmol/L) and serum creatinine less than 5 mg/dL (<442 μmol/L) (resulting in daily hemodialysis for all but 2 participants in the intensive therapy group). This study found a nonsignificant trend toward lower mortality with conventional hemodialysis. The pooled overall RR for mortality associated with more intensive intermittent hemodialysis was nonsignificant (0.83; 95% CI, 0.40-1.72) and highly heterogeneous (I² = 73%). Given the methodological differences and low power, it is difficult to make firm conclusions; further RCTs are warranted.

Anticoagulation is essential for both intermittent hemodialysis and CRRT, because blood traveling through the extracorporeal circuit can lead to clotting of the filter, contributing to blood loss and reducing dialysis efficiency. On the other hand, excessive anticoagulation may result in bleeding complications.⁸⁰ Unfractionated heparin is the mainstay of anticoagulation for intermittent hemodialysis and CRRT,⁸¹ although the optimal intensity of heparinization is unknown. Low-molecular-weight heparin is infrequently used due to the need to monitor factor Xa levels.⁸² Alternatives to systemic anticoagulation in patients at high risk of bleeding include regular saline flushes or citrate infusion into the dialysis circuit.

In 1 small randomized trial¹⁸ (30 participants), citrate anticoagulation allowed longer filter life (hazard ratio for clotting, 0.37; 95% CI, 0.20-0.70) compared with standard unfractionated heparin and resulted in a lower risk of serious bleeding (RR, 0.06; 95% CI, >0-0.95). These advantages of citrate anticoagulation must be balanced against its increased complexity and potential for metabolic disturbances including hypocalcemia, metabolic alkalosis, and citrate toxicity. Hirudin may be an alternative to heparin in circumstances in which regional citrate anticoagulation is not available. In another small RCT²¹ (26 participants) comparing hirudin with unfractionated heparin, the filter life was nonsignificantly shorter in patients treated with hirudin (P = .06).

Membranes used for intermittent hemodialysis and CRRT are traditionally characterized in terms of flux (permeability to water and larger solutes) and biocompatibility (degree to which complement is activated by exposing the membrane to blood). Biocompatible membranes are typically made of synthetic materials and are more expensive than bioincompatible membranes, which may be further subdivided into unsubstituted cellulose and cellulosic varieties.

Four trials^31,34,40,42 compared the use of high-flux and low-flux membranes in intermittent hemodialysis and did not find any difference in the risk of mortality (RR, 0.91; 95% CI, 0.74-1.11; I² = 16% [287 participants]) or dialysis dependence in survivors (RR, 1.02; 95% CI, 0.75-1.39; I² = 0% [119 participants]). Since the increased permeability to water facilitates hemofiltration, high-flux membranes are generally recommended for CRRT. Two small trials compared different high-flux membranes in CRRT and found no effect on mortality.^15,39

Seven trials^{32,34,36,38,40- 42} (719 participants) compared mortality in patients treated with bioincompatible vs biocompatible membranes. There was a borderline increase in the risk of death associated with bioincompatible membranes (RR, 1.11; 95% CI, 0.94-1.31; I² = 7%) as compared with biocompatible membranes and no difference in the risk of dialysis dependence in survivors or the composite of these outcomes. The risk of death was significantly higher among patients treated with membranes made of unsubstituted cellulose (a subset of bioincompatible membranes) as compared with biocompatible membranes (RR, 1.23; 95% CI, 1.01-1.50; I² = 0% [5 trials, 494 participants]), and the risk of death or dialysis dependence was of borderline significance (RR, 1.07; 95% CI, 0.90-1.27; I² = 0% [4 trials, 442 participants]). These results, from 7 reasonable-quality RCTs, suggest that the use of unsubstituted cellulosic membranes should be avoided in ARF.

Dialysate for intermittent hemodialysis is produced as needed by the dialysis machine by mixing specially treated water from the municipal supply with electrolytes. In contrast, dialysate/replacement solution for CRRT must be sterile and is either purchased or produced locally in hospital pharmacies. Options for correction of metabolic acidosis include acetate, lactate, and bicarbonate; the latter has become increasingly popular in recent years due to concern that acetate and lactate may not be adequately converted to bicarbonate in the setting of multiple organ failure.⁸³

One trial³³ with a total of 117 participants compared bicarbonate with lactate in patients receiving CRRT. The difference in mortality was not significant (RR, 0.77; 95% CI, 0.48-1.22), but a significantly reduced risk of cardiovascular events was found in patients treated with a bicarbonate rather than a lactate buffer (RR, 0.39; 95% CI, 0.20-0.79), suggesting that the former may be preferable in CRRT. No difference was found in the only RCT comparing buffers in patients with ARF treated with intermittent hemodialysis (n = 29).³⁷ The sole eligible prospective cohort study⁵¹ identified by our search found no significant difference between lactate and acetate buffers on the risk of death (RR, 1.07; 95% CI, 0.75-1.52) in CRRT recipients.

The same equipment and staff that provide chronic dialysis usually can be used to deliver intermittent hemodialysis to patients with ARF. Therefore, the capital and training expenditures associated with establishing an intermittent hemodialysis program are often low. Regardless of the submodality used, CRRT involves specialized equipment and training for staff, generally requires additional supplies (especially replacement fluids), and consumes greater health care resources than intermittent hemodialysis, a consistent finding regardless of country, scenarios of nursing care, and frequency of intermittent hemodialysis (including SLED).^32,36,38,84 The weekly cost difference between intermittent hemodialysis and CRRT is estimated to be $1100 to $3700, depending on whether daily intermittent hemodialysis (or SLED) or thrice-weekly intermittent hemodialysis is compared with CRRT.^32,38

However, the incremental costs of providing RRT in critically ill patients are small compared with the large costs of the index hospitalization and beyond. While renal failure requiring provision of chronic dialysis is infrequent, it has substantial implications given its high costs.⁴¹ Patients who remain dialysis dependent are estimated to incur an additional $60 000 in direct health care costs compared with those who recover renal function,³⁸ not including indirect or patient productivity costs. Thus, relatively costly interventions may be attractive if proven to improve these longer-term outcomes. Prior to widespread adoption of novel treatments for ARF, decision makers might find it valuable to consider the potential impact on resource utilization in the short and long term (including the index hospitalization and chronic dialysis).

Our recommended management strategy for patients with severe ARF is presented below. The evidence for these recommendations was derived primarily from studies of critically ill patients and thus may not be generalizable to other populations. In the absence of evidence demonstrating clear superiority of one strategy over another, our recommendations have been guided by economic and practical considerations.

Starting RRT. The decision to initiate RRT in patients with severe ARF requires consideration of multiple factors, including assessment of intravascular volume, electrolyte and acid-base status, uremia, nutritional requirements, urine output, hemodynamic status, and the evolving clinical course of each patient. Potential advantages of earlier RRT initiation must be set against the hypothetical risks of treatment-induced renal injury, bleeding due to anticoagulation, and mechanical and infectious complications associated with central venous access.

Prescription of RRT. Available RCTs demonstrate no clear difference between continuous or intermittent modalities with respect to clinically relevant outcomes. Furthermore, the most recently published randomized trial of intermittent hemodialysis vs CRRT demonstrated that even the sickest of patients could be safely treated with intermittent hemodialysis.¹⁷ Given the significantly higher cost of CRRT, intermittent hemodialysis may be preferable for patients with ARF who require RRT.

In otherwise stable patients, alternate-day dialysis treatments of 4 or more hours using blood flows of 250 mL/min or greater are usually sufficient in patients with or without concomitant critical illness. More frequent hemodialysis may be required in highly catabolic patients or to achieve treatment targets for fluid, electrolyte, or acid-base management, although data identifying how such targets should be set are limited. Despite the lack of data supporting its superiority and its higher cost, some clinicians may prefer to use CRRT in critically ill patients with ARF and severe hemodynamic instability. If CRRT is used, the target dose should be 35 mL/kg per hour (3 L/h in a 70-kg person).

Technical Considerations. Due to its low cost and wide availability, we recommend unfractionated heparin as the usual anticoagulant for dialysis in ARF, titrated to maintain activated partial thromboplastin time between 1 and 1.4 times the upper limit of normal. In patients at high risk of bleeding, regional citrate anticoagulation should be administered, although this technique is not recommended for usual care, given its substantially higher cost. Heparin-free dialysis with saline flushes may be used as an alternative, especially outside a critical care unit.

Dialysis membrane flux (high vs low) and surface area should be determined based on patient characteristics and treatment goals; unsubstituted cellulose membranes should not be used in patients with ARF. Dialysate composition should be determined by the treatment targets (correction of hyperkalemia, metabolic acidosis), and the use of acetate buffer should be avoided. Recently publicized errors in the preparation of CRRT fluids led to patient deaths due to hyperkalemia and massive hemolysis.⁸⁵ These serious adverse events underscore the importance of quality assurance for the preparation of dialysate fluid, regardless of the composition that is selected for use. Given the potential for dialysis to reduce blood levels of vital medications (as well as the extensive list of medications that must be dose-adjusted or avoided in ARF), advice from an expert pharmacist should be sought for all patients with severe ARF.

To our knowledge, this is the most comprehensive review of options for renal replacement in ARF. Despite the wide variety of available techniques, we identified few good-quality data to guide best practice. Uncertainty remains about when RRT should be initiated, how much dialysis should be provided, and for how long treatment should continue. The design and execution of research in this area has been hampered by a number of factors, including (1) lack of consensus with regard to the definition of ARF and indications for dialysis; (2) limited understanding of the epidemiology of ARF; and (3) poorly designed and inadequately powered studies.

Perhaps the most problematic issue is the lack of a common metric for defining ARF. Changes in serum creatinine levels and urine output are relatively late events that provide little information about the etiology of ARF or the extent to which loss of kidney function is reversible, thus making even the most recently proposed definitions somewhat arbitrary. Future research efforts should be directed toward more specifically defining subsets of patients with ARF before considering interventional studies.

Second, the epidemiology of ARF is poorly understood. Relatively few data describe optimal dialytic support for patients with ARF who are not critically ill—patients who arguably may benefit the most from improved management, given their better overall prognosis. Much remains to be learned about how to improve outcomes after hospital discharge, such as renal recovery, the development or acceleration of chronic kidney disease, and the need for chronic dialysis. Since such outcomes might drive the long-term costs and consequences of a renal replacement strategy, they should be addressed by future studies.

Finally, most studies of RRT in ARF have been small and of poor quality (80% of the Jadad scores were <3). Future studies should be adequately powered to detect differences in clinically meaningful outcomes and should also collect data on duration of hospitalization and use of resources. These objectives would be best achieved through collaborative multicenter studies such as the ARF Trial Network (ATN) study—an RCT of intensive vs conventional dosing of both intermittent hemodialysis and CRRT in critically ill US patients with ARF.⁸⁶ The multicenter Randomised Evaluation of Normal vs Augmented Level of Renal Replacement (RENAL) study is similarly comparing augmented with conventional dosing of CRRT in critically ill patients treated for ARF in Australia and New Zealand and is expected to complete enrollment in 2008.⁸⁷ These trials will provide useful guidance with respect to the dosing and frequency of RRT in patients with ARF as well as to establishing infrastructure that could be used to support future studies in this important clinical area.