Acute kidney injury (AKI) is a well-recognized complication of critical illness with an important impact on morbidity, mortality and health resource utilization 12345. Renal replacement therapy (RRT) is often required and represents a substantial escalation in the complexity and cost of care for critically ill patients with AKI 4. Despite its extensive use in clinical practice, there is uncertainty about the optimal time and indications for initiation of RRT in the ICU 6. Clearly, the process involved in deciding when to initiate RRT in critically ill adult patients is complex and can be influenced by numerous factors, including patient-specific and clinician-specific factors and those related to organizational/logistical issues (Table 1). Indeed, studies have shown marked variation of practice between clinicians, and across institutions and countries 78.

An evaluation of timing of RRT initiation has been the focus of a number of clinical studies. These have recently been summarized in a systematic review and meta-analysis 6910111213. Most of these studies have been small, retrospective or secondary analyses, and have arbitrarily dichotomized the study population into 'early' or 'late' RRT initiation based on biochemical criteria, urine output criteria, or by 'door-to-dialysis' time 14. The meta-analysis by Seabra and colleagues 12 also included five randomized trials. A pooled analysis from these trials showed a non-statistically significant trend towards reduced mortality with earlier initiation of RRT (relative risk 0.64; 95% confidence interval (CI), 0.40 to 1.05, P = 0.08). However, this pooled analysis only included data from 270 patients, thus limiting its statistical power. Accordingly, this limits the inferences about timing of RRT initiation and prohibits a simple translation of such data easily to the bedside to guide clinical management. While large prospective studies are urgently needed, the currently available data would indicate a potential benefit associated with earlier initiation of RRT for those patients where RRT is likely to be needed in terms of both survival and recovery of kidney function 1215.

Currently, there exists no broad consensus to guide clinicians on this important issue. In fact, RRT initiation has been repeatedly identified as a research priority 161718. Accordingly, we have developed an opinion-based clinical algorithm to aid in the decision on when to consider initiation of RRT in critically ill patients (Figure 1). The algorithm gives a more quantitative characterization of 'timing' and incorporates several patient-specific factors, based on clinical evidence when available, that may influence when to initiate RRT. We adapt the terminology proposed by the Acute Kidney Injury Network (AKIN): 'illness trajectory' refers to the pace of clinical evolution of the patient, and AKI 'trend' refers to the rate of clinical and/or biochemical changes (including urea and creatinine) 16. The objective of this algorithm is to provide a starting point to guide clinicians on when to consider use of RRT in adult critically ill patients. This algorithm will not specifically address additional key issues in the delivery of renal support, such as RRT modality (continuous versus intermittent), mode (convection versus diffusion) and dose delivery 17181920. Importantly, the algorithm is also intended to provide a starting point for further prospective evaluation to understand the ideal time/circumstances for when to initiate RRT that could, in due course, promote higher quality of patient care and improved clinical outcomes.

The first priority after a patient is admitted to ICU is determination of whether there are absolute indications and/or emergent need for RRT. A summary of proposed absolute indications for RRT initiation, based on consensus, is presented in Table 216. It is important, however, to recognize that RRT initiation in these circumstances can largely be viewed as 'rescue therapy' where delays may have deleterious consequences for the patient. Moreover, these indications are largely adapted from 'classic conventional' indications for RRT in end-stage kidney disease, wherein the main objective is alleviation of uremic complications.

Since AKI is common in critical illness, in the absence of absolute indications for RRT, the next logical step is to determine whether patients have AKI. In a multi-center multinational study, Uchino and colleagues 4 found AKI occurred in 5 to 6% of all ICU admissions, with 70% of these eventually receiving RRT. Recent data indicate the incidence of AKI is rising 212223. Historically, however, establishing incidence estimates of AKI has been problematic due to the lack of a standardized definition. Fortunately, a consensus-driven classification scheme for AKI, the RIFLE criteria (and modified AKIN criteria), has been recently proposed, which represents a noteworthy advance for clinical practice and research in AKI 2425. The RIFLE criteria have been validated and proven robust for clinically relevant outcomes in patients with AKI across numerous studies 3526272829. Epidemiologic studies of AKI, when defined by the RIFLE criteria, have shown that 11 to 67% of ICU patients may develop AKI during their illness course 29.

The RIFLE criteria were initially developed to standardize the diagnosis, classify the severity and monitor progression of AKI. Importantly, they were also designed to enable better comparison of clinical studies on the epidemiology and attributable outcomes for AKI 24. Currently, there is no consensus on how to define timing of RRT initiation due to the aforementioned limitations in available data. The concept of 'timing' remains poorly defined and inconsistent 16. Previously, timing of RRT has mostly been described by qualitative criteria (early versus late/delayed). The RIFLE/AKIN criteria provide the possibility of a more 'quantitative' characterization of timing. We recognize that these criteria have not been formally evaluated as a tool for guiding clinicians on when to initiate RRT. Yet, data from numerous observational studies have consistently shown that earlier initiation of RRT (however defined) correlated with improved survival. This would appear to provide some justification of 'early', or perhaps a better term could be 'timely', RRT initiation in selected critically ill patients with AKI. However, further investigation, preferably by prospective randomized trials, is undoubtedly warranted. A prospective analysis of the impact of RRT initiation incorporating the RIFLE/AKIN classification schemes on survival and renal recovery is a potential starting point. Can these criteria have bedside utility to aid in clinical decision-making? We believe this is a logical first step in understanding how research evidence may be translated into clinical practice to improve outcomes in patients with AKI.

The RIFLE/AKIN criteria are also able to classify AKI severity and follow trends over time 29; both are vital to consider in the context of RRT initiation. They are also a tool for dynamic evaluation of response to initial (non-RRT) therapy. We emphasize again this algorithm is not intended to direct all aspects of initial resuscitation and supportive therapy, but rather provide an outline for when to consider RRT initiation. Recent comprehensive reviews, based on consensus, have summarized strategies for initial management of AKI 30.

Likewise, several 'dynamic' factors, including the trends in AKI and/or illness severity, warrant consideration. Acute and/or rapid changes in clinical status, such as whether AKI is progressing (and how rapidly), the probability of kidney recovery, whether illness severity is progressing (and how rapidly), and additional measures of acute physiology such as fluid accumulation 32, relative oliguria (that is, urine output >200 ml/12 h, but insufficient to prevent fluid accumulation) and the trajectory of non-kidney organ dysfunction should factor into the decision of when to initiate RRT for those with established mild-moderate AKI.

Certain co-interventions in the ICU will also influence the decision to initiate RRT in patients with mild to moderate AKI 33. For example, co-interventions may contribute to urea or fluid accumulation, or systemic acidemia, therefore placing a greater demand on already compromised kidney function. The use of adjuvant corticosteroids in severe sepsis/septic shock is common and can aggravate protein catabolism and azotemia 34. The increased urea generation coupled with retention of uremic solutes may create a circumstance where RRT initiation may need to be considered in those with mild-moderate AKI.

The concept of early goal-directed therapy as a guide for acute resuscitation in septic shock has represented a significant philosophical shift in the management of these patients 3536. A key component of early goal-directed therapy is the administration of fluid therapy, ideally targeted to physiologic endpoints. In the trial by Rivers and colleagues 36, enrolled patients received an estimated 13 to 14 liters of fluid therapy in the first 72 hours (most within the initial 6 hours). While this trial did not provide data on AKI occurrence, oliguria or fluid balance, septic patients are known to be at high risk for AKI 3738. In this context, coexistent or rapidly evolving AKI results in impaired free water and solute excretion, and contributes to rapid fluid accumulation and metabolic acidosis (especially with high chloride containing solutions). These complex and integrated conditions present a circumstance where earlier RRT may prove beneficial. Diuretic therapy can be a useful adjunct for management of fluid accumulation; however, their use in patients with AKI should not delay RRT initiation with the intent of avoiding RRT. In one study, diuretic use was associated with increased mortality and non-recovery of kidney function, which may have occurred in part due to delayed initiation of RRT 39.

Critically ill patients with acute lung injury/acute respiratory distress syndrome receiving lung-protective ventilation may intentionally develop respiratory acidosis due to permissive hypercapnea 40. Co-existent and/or evolving AKI in these patients will significantly impair capacity for kidney bicarbonate regeneration to buffer systemic acidemia. Earlier RRT may prove beneficial in these patients prior to the development of severe acidemia, worsening acute respiratory distress syndrome and/or volume overload.

This list of potential 'non-kidney' indications for initiation of RRT in critically patients continues to grow 16. In this context, the term extracorporeal blood purification (EBP), rather than RRT, may be more appropriate.

Critically ill patients whose course is complicated by AKI often receive RRT. RRT is an important therapeutic and supportive measure and is commonly used in clinical practice. However, there remains uncertainty about the ideal circumstances in which to initiate RRT and for what indications. The process of deciding when to initiate RRT in critically ill patients is complex and can be influenced by numerous factors. Currently, there exists large variation in clinical practice between clinicians and across institutions and countries, due, in part, to the lack of consensus on this issue. We have proposed a clinically based algorithm to aid in the decision on when to initiate RRT in critically ill patients that incorporates patient-specific factors and based on available clinical evidence. The intent of this algorithm is to provide a starting point to guide clinicians on when to consider initiation of RRT in critically ill patients and provide a foundation for further prospective evaluation and development of broad consensus on this important issue.

AKI: acute kidney injury; AKIN: Acute Kidney Injury Network; CI: confidence interval; EBP: extracorporeal blood purification; RRT: renal replacement therapy.

SMB, DNC, NG and CR have all participated in ADQI workgroups. NG and CR have participated in AKIN workgroups.