Acute kidney injury (AKI) requiring renal replacement therapy (RRT) occurs in 5 to 6% of critically ill patients and is associated with high mortality and significant health resource utilization 1 2 3 . Controversy exists as to what constitutes optimal RRT in this setting. There are several modifiable factors in the delivery of RRT which may potentially influence patient outcome, including RRT modality (continuous or intermittent), solute removal mechanisms (convection, diffusion, adsorption or combination), timing of initiation and dose of treatment. The relationship between patient outcome and treatment dose was first introduced in a landmark study where patients randomised to post-dilution continuous veno-venous haemofiltration (CVVH) at a dose of 35 ml/kg/hour or above had improved survival compared with those randomised to 20 ml/kg/hour 4 . Since then, this issue has been explored in other studies with conflicting results 5 6 7 8 9 . The Acute Dialysis Quality Initiative recommends a higher dose in the absence of definitive data, particularly in septic patients 10 11 . However, practice surveys suggest that this threshold dose has not been widely adopted into current intensive care units (ICU) practice 12 13 .

We performed a prospective European multicentre observational cohort study to evaluate the prescription and actual-delivered RRT dose in ICUs and its relationship with patient outcome, such as mortality and duration of mechanical ventilation and ICU stay. Our hypothesis was that a higher RRT dose would be associated with better patient outcomes.

This study was conducted from June 2005 to December 2007 in 757 patients enrolled in 30 ICUs in eight countries. The protocol was approved by the institutional review boards of the five Steering Committee members. Written informed consent was obtained from patients or next of kin when required by a centre's review board. The design of the study was published in 2005 14 , and registered in the Cochrane Renal Group (CRG110600093).

We conducted a European multicentre observational study to describe clinical outcomes associated with RRT dose in critically ill patients with AKI. The key findings of this study are the following. First, despite a prescribed CRRT dose approximating 35 ml/kg/hour, the recommended 'minimum' for critically ill AKI patients according to expert opinion 10 24 , the delivered CRRT dose was markedly lower than this value. Second, it appears that alternate day IRRT for critically ill patients is uncommon in the participating centres. Third, after adjustment for multiple variables, we did not observe a beneficial effect of more-intensive RRT dose on ICU survival. Fourth, ICU stay and ventilation days were shorter in the more-intensive RRT groups.

Our findings on mortality are congruent with an international observational study 25 and two recent randomised clinical trials on standard versus higher-dose CVVHDF including the large multicentre Veterans Affairs/National Institute of Health (VA/NIH) trial in the US 8 9 . There are conflicting results on the effect of RRT dose on patient outcome. Two earlier single-centre randomised clinical trials showed a beneficial effect of an intensive CRRT dose when compared with a less-intensive dose in both CVVH (≥ 35 ml/kg/hour versus ≤ 20 ml/kg/hour) 4 and CVVHDF (≥ 42 ml/kg/hour versus ≤ 25 ml/kg/hour) 7 . In contrast, Bouman and colleagues did not detect any difference in outcome between ultrafiltration rates of 3 to 4 l/hour and 1 to 1.5 l/hour; however, this study suffered from lack of power and an unexpectedly high ICU survival rate among enrolled subjects 5 . Interestingly, in contrast to other studies 5 7 8 9 , we observed a positive effect of more-intensive RRT dose on ICU stay and duration of mechanical ventilation.

Our study has several notable features. It is the first large observational study specifically oriented towards RRT dose involving multiple ICUs which are a mix of academic and non-academic centres. As such, it is likely to be more reflective of actual clinical practice and a broad patient population. Indeed, in the CRRT group, the prevalence of sepsis was 39% and the overall mortality was 54%. This is similar to that described in the literature 3 6 7 9 22 .

Second, it focused on delivered, rather than prescribed, RRT dose. This is not a minor point, as many factors contribute to delivering a RRT dose lower than prescribed, and it is the delivered dose which the patient 'sees' and which is likely to affect the clinical outcome. A prior observational study reported only on prescribed, but not the delivered, CRRT dose 25 . Furthermore, ours is the first dose study in which correction for percentage predilution was performed in the calculation of CRRT dose, resulting in a more accurate estimate. It has been shown that delivered dialysis dose is generally lower than prescribed, ranging from 68 to 89% of prescribed 7 8 26 27 .

Third, we specifically collected information on treatment 'downtime', which is considered an important factor affecting delivered RRT dose.

Fourth, it is also one of only two studies to look at a continuum of RRT dose so far 25 . Of note, the majority of patients received a CRRT dose which was in between the 'standard/low dose' and the 'high dose' arms evaluated in randomised clinical trials 4 7 8 9 . Whether having several patients in this 'intermediate' zone served to dilute the true clinical effect of RRT dose remains unclear. It has been suggested that only major changes in the application of dose can be reasonably expected to have a discernible clinical effect 28 . For example, the difference between a delivered CRRT dose of 30 and 35 ml/kg/hour may be too subtle, or could be criticised for being within a calculation error. However, we performed a variety of analyses which included expressing dose in increments of 10 ml/kg/hour (arbitrarily considered a 'significant' increment), or as categories based on cut-offs from the literature or on statistical spread (e.g. tertiles, median), and failed to find a significant effect on ICU mortality (Table 5). It is also possible that therapy in the first 24 to 48 hours is more crucial with respect to patient outcome. We therefore performed a post-hoc sensitivity analysis looking at CRRT dose during these periods (Table 5), and this did not significantly alter the results.

It has been suggested that septic patients may be a specific population which could benefit from higher RRT dose 4 11 . In our post-hoc analysis, the effect of RRT dose on mortality was similar in both septic and non-septic patients (Table 6). It is also possible that more-intensive RRT only benefits patients with an intermediate severity of illness, as suggested by Paganini and colleagues 29 . We performed two sensitivity analyses to address this. First, we limited the analysis only to patients with SAPS scores between 45 and 60, in whom the predicted mortality ranges from 35 to 60%. In five studies evaluating the effects of CRRT dose, mean acute physiology and chronic health evaluation (APACHE) II scores ranged from 22 to 26, giving predicted mortality rates of 42 to 57% in this group 4 5 6 7 8 9 . Our results were similar within this subgroup. Second, patients may have a very short duration of RRT for various reasons. For example, they may be gravely ill and die shortly after RRT initiation. Alternatively, they may be less ill and have rapid recovery of renal function allowing early withdrawal of RRT. Therefore, we performed a secondary analysis looking only at patients who had at least 25 hours of RRT. This was adapted from the definition of an 'adequate trial of therapy' in a randomised trial comparing CRRT and IRRT 30 . The results remained qualitatively unchanged.

This study provides further insight into the prescription and delivery of RRT dose in current clinical practice. There is a gap between prescribed and delivered CRRT dose, as has been shown by others 7 8 26 27 . Treatment downtime is a known contributing factor. In contrast to earlier studies, however, we also considered the effect of percentage pre-dilution in calculating the delivered dose. We hypothesise that lack of attention to this when prescribing CRRT may play a heretofore unrecognised role in under-delivery of dose. As modern machines are able to provide replacement fluid in variable proportions of pre/post-dilution, it is important to keep this in mind. We also observed that CRRT patients receiving more-intensive dose had significantly lower body weights. This may represent indiscriminate 'by the litre' prescription, rather than 'individualised' prescription based on body weight 13 . It is also possible that it is simply more difficult to provide higher doses in larger patients with currently available technology. Perhaps the greatest concern arising from the observed gap between prescription and delivery is the potential downstream effect of prescribing 20 ml/kg/hour to patients. There would be a real risk of effectively underdialysing patients 31 .

We acknowledge certain limitations in this study. As with all observational studies, ours may have suffered from 'selection by prognosis' 32 . Indeed, it is quite plausible that, based on existing literature, the treating intensivist or nephrologist would prescribe a higher RRT dose to a sicker patient, who a priori has a higher predicted mortality. Although we adjusted for potential confounders, including propensity score analysis (Table 3 in Additional data file 2), this may still be insufficient because it is not possible to adjust for confounders that are neither measured nor known. We also chose to exclude a number of patients from the analysis, which may have resulted in some selection bias. It was not possible to analyse patients who crossed over between modalities because there is no single equivalent expression of dose clinically validated for both CRRT and IRRT. However, when we compared the analysed group to the overall population, they were similar in terms of demographics, general severity of illness and co-morbidities. Therefore, if selection bias was present, its effect is likely to be minimal. For IRRT patients, we were unable to correlate outcome with the measured Kt/V, as the necessary laboratory parameters for the calculation were not collected as part of routine practice. This is consistent with the findings of the VA/NIH group in their pre-trial survey that assessment of the delivered dose of IRRT was performed infrequently in clinical practice 13 . Nevertheless, we believe we have a reasonable estimate of prescribed IRRT dose based on the operational parameters collected, with a median prescribed Kt/V of 1.2. Although we found an inverse relationship between RRT dose and duration of mechanical ventilation, as well as with ICU stay, we acknowledge that there were no standard criteria for extubation or ICU discharge in this observational study. Furthermore, we only looked at short-term outcomes. Future studies should attempt to better understand the long-term effects of RRT dose. Lastly, this was a voluntary survey conducted in predominantly CRRT-oriented centres. As such, it may not be possible to generalise the results to other medical centres. It is noteworthy, however, that despite being CRRT-oriented centres, the delivered CRRT dose, although higher than reported in prior studies 26 28 , still fell short of the mark. This begs the question as to whether a dose of 35 ml/kg/hour or 45 ml/kg/hour 4 11 , as suggested for septic patients, is routinely achievable in the real world.

Nevertheless, our findings of reduced ICU stay and mechanical ventilation days with more-intensive RRT may potentially have a large impact on health resource utilisation, if confirmed by future studies. For example, the average total cost per ICU day has been estimated at €1200 in a sample of European countries 33 . A possible implication would be potential savings of €8000 to 10,200 per ICU admission with more-intensive RRT. In contrast to our results, two other studies showed no significant difference in duration of mechanical ventilation between lower and higher CRRT dose groups 5 7 ; however, this issue was not specifically addressed by others 4 6 8 9 .

This study highlights that the concept of RRT dose or adequacy is more complex than previously thought. This adds fuel to the debate on the optimal RRT dose for critically ill patients with AKI. Clearly there are other dimensions to RRT adequacy other than removal of various solutes, whether expressed as Kt/V, ml/kg/hour or number of RRT sessions per week. These include prophylactic volume control, as well as acid-base and tonicity control, among others. Furthermore, recognising that critical illness is not a static condition, a 'dynamic approach' to RRT dose, rather than fixed dose, may be more appropriate in this setting 34 . This hypothesis is worthy of exploration in future studies. In addition, it is likely that there are multifaceted interactions between RRT dose and other factors (timing of RRT, modality, patient characteristics, etc.) which influence outcome.

We conducted a prospective European multicentre cohort study of AKI patients treated with RRT. This study provides insight in to how RRT is currently practiced in the ICU. We observed that the median CRRT dose is lower than 35 ml/kg/hour and only 22% of patients received this or a higher dose. In contrast, 60% of IRRT patients were treated daily. We evaluated the association between actual delivered RRT dose and clinical outcomes. The data provide no evidence for a survival benefit afforded by more-intensive RRT. However, higher RRT dose appeared to be associated with shorter ICU stay and duration of mechanical ventilation. In conclusion, within the confines of the dose range examined, there was no effect on survival while effects on non-mortality endpoints should be examined by further study.

• In this observational study, the delivered CRRT dose was markedly lower than 35 ml/kg/hour (median = 27).

• Alternate day IRRT for critically ill patients was uncommon in the participating centres.

• After adjustment for multiple variables, there was no beneficial effect of more-intensive RRT dose on ICU survival.

• Shorter ICU stay and duration of mechanical ventilation were observed in the more-intensive RRT groups.

AKI: acute kidney injury; APACHE: acute physiology and chronic health evaluation; CI: confidence interval; CRRT: continuous renal replacement therapy; CVVH: continuous veno-venous haemofiltration; CVVHD: continuous veno-venous haemodialysis; CVVHDF: continuous veno-venous haemodiafiltration; ICU: intensive care unit; IRRT: intermittent renal replacement therapy; IQR: interquartile range; RIFLE: Risk-Injury-Failure-Loss-Endstage renal disease; RRT: renal replacement therapy; SAPS II: simplified acute physiology score; SOFA: sequential organ failure assessment.

The authors declare that they have no competing interests.

SV participated in conception and design, patient enrollment, acquisition of data, analysis and interpretation of data, and critical revision of the manuscript. DNC participated in patient enrollment, acquisition of data, performed statistical analysis, analysis and interpretation of data, drafting and critical revision of the manuscript. RF, DKM, GM, AM, FM, MF, MM, RR and SL participated in conception and design, patient enrollment, acquisition of data and critical revision of the manuscript. CR conceived and designed the study, participated in analysis and interpretation of data, critical revision of the manuscript, and obtaining administrative and technical support.