Supplementary Appendix

Transcription

1 Supplementary Appendix This appendix has been provided by the authors to give readers additional information about their work. Supplement to: Gaudry S, Hajage D, Schortgen F, et al. Initiation strategies for renal-replacement therapy in the intensive care unit. N Engl J Med 2016;375: DOI: /NEJMoa

2 TABLE OF CONTENTS OF SUPPLEMENTAL ELECTRONIC MATERIAL Table des matières LIST OF CO-INVESTIGATORS... 2 CONTRIBUTION TO THE STUDY... 4 STUDY FUNDING:... 5 METHODS... 6 Study oversight... 6 Patient information... 6 Participants... 7 Outcomes... 8 Complications potentially related to AKI or RRT: definitions (refer to Table 2) PROTOCOL VIOLATIONS PATIENT FOLLOW-UP INTERIM ANALYSES TABLES Table S1 Criteria mandating RRT initiation in the delayed RRT strategy group Table S2 Characteristics of the patients at baseline Table S3 Distribution of criteria which mandated RRT initiation in the delayed strategy group Table S4 Patient characteristics at the time of RRT initiation Table S5 Characteristics of RRT sessions delivered during the 28 days following randomization Table S6 Baseline predictors of RRT initiation in the delayed group Table S7 Medical treatment of AKI-related metabolic complication before the first RRT session Table S8 Etiologies of hemorrhage FIGURES Figure S1 Flow-chart: Enrollment, Randomization and Follow-up of the study participants Figure S2 Blood urea nitrogen and serum creatinine level changes overtime Figure S3 Diuresis recovery Figure S4 Proportion of patients with spontaneous creatinine decrease UNPLANNED ANALYSES

3 LIST OF CO-INVESTIGATORS The AKIKI (Artificial Kidney Initiation in Kidney Injury) investigators: CH René Dubos, Pontoise and CH de Beaumont, Beaumont sur Oise: Abiramy Thiagarajah CHD La-Roche-sur-Yon : Maud Fiancette CHU Bordeaux: Chloé Gisbert-Mora CHU Louis Mourier, Colombes: Jonathan Messika, Damien Roux CH Sud Francilien, Corbeil Essonnes: Pierrick Cronier CHU Rouen: Fabienne Tamion CHU Lille: Raphaël Favory CHR Orléans: Thierry Boulain CHU La Pitié Salpetrière: Alexandre Demoule CHU Clermont-Ferrand: Mireille Adda CHU Lariboisière: Antoine Goury CHU Marseille: Guillemette Thomas CH Victor Dupouy, Argenteil: Hervé Mentec CHU Amiens: Michel Slama CHU Lyon: Claude Guerin CH Cholet : Elmi Messaï CH Beauvais: Jack Richecoeur, Danièle Combaux CH Saint Denis: Etienne Demontmolin, Daniel Da Silva CHU Saint Antoine: Naïke Bige, Bertrand Guidet CH Versailles: Benjamin Zuber, Guillaume Lacave CH Le Mans: Nicolas Chudeau CHU de Nice: Jean Dellamonica CHU Saint-Priest-en-Jarez: Michaël Darmon 2

4 CH Poissy: Pascal Fangio CH Montreuil: Vincent Das CHU Tenon, Paris: Clarisse Blayau CHU Bichat, Paris: Lila Bouadma 3

5 CONTRIBUTION TO THE STUDY The trial is an investigator-initiated multicenter study led by Didier Dreyfuss and the members of the steering committee (Didier Dreyfuss, Stephane Gaudry, David Hajage Laurent Martin Lefevre, Jean Damien Ricard, Frederique Schortgen, Florence Tubach). All the investigators mentioned as co-authors gathered the data. Data were checked by the clinical research team, and the data base was managed and closed by the Clinical Research Unit Paris Nord, financed by the study funding. The statistical analysis was performed by David Hajage. The paper was written by Stéphane Gaudry, Didier Dreyfuss, David Hajage and Jean-Damien Ricard. The paper was submitted to all the co-authors who made substantial contributions and agreed to submit for publication to The New England Journal of Medicine. 4

6 STUDY FUNDING: The study was supported by a grant from the French Ministry of Health (Programme Hospitalier de Recherche Clinique National 2012 (AOM-12456)). The sponsor was the Département de la Recherche Clinique de l Assistance Publique - Hôpitaux de Paris (France). 5

7 METHODS Study oversight - Members of the steering committee: Didier Dreyfuss, Stephane Gaudry, David Hajage Laurent Martin Lefevre, Jean Damien Ricard, Frederique Schortgen, Florence Tubach - An independent Data Safety and Monitoring Board (DSMB) blinded to allocation of groups conducted two interim analyses - Members of the independent DSMB: Pr Laurent BROCHARD (Critical Care Saint Michael s Hospital, University of Toronto, Canada), Pr Christian MELOT (Erasme University Hospital, Emergency Department, Université Libre de Bruxelles, Belgium) and Pr Alexandre HERTIG (Hôpital Tenon, Urgences néphrologiques et transplantation rénale, Université Pierre et Marie Curie, Paris, France) Patient information Patients or surrogates were informed both verbally and with a written document about the AKIKI study by the investigators. By French law, written informed consent was not required, as the standard of care encompasses both study interventions. Patients and surrogates were informed that they could refuse participation at any time and their decision was recorded in patient files. Patients who were eligible but incapable of receiving information and for whom a substitute decision maker was not available might be randomized through a process of deferred information, according to French law. They were informed about participation as soon as their clinical status allowed. 6

8 Participants Inclusion criteria Eligible patients were adults (18 years of age or older) admitted to the ICU with AKI compatible with the diagnosis of acute tubular necrosis in a context of ischemic or toxic aggression and receiving invasive mechanical ventilation and/or catecholamine infusion (epinephrine or norepinephrine). To be randomized, patients should have AKI stage 3 of KDIGO classification defined by at least one of the following criteria: serum creatinine concentration of more than 4 mg/dl (354 µmol/liter) or greater than 3 times the baseline creatinine level, anuria (urine output of 100 ml/day or less) for more than 12 hours, oliguria (urine output below 0.3 ml/kg/h or below 500 ml/day) for more than 24 hours. Randomization and initiation of treatment (RRT or conservative treatment) were mandatory within six hours after fulfillment of the final inclusion criteria. The non-inclusion criteria were: Presence of one of the following conditions: blood urea nitrogen of more than 112 md/dl (40 mmol/liter), serum potassium concentration of more than 6 mmol/liter or more than 5.5 mmol/liter persisting despite medical treatment (bicarbonate and/or glucose-insulin infusion), ph below 7.15 in a context of pure metabolic acidosis (PaCO 2 below 35 mmhg) or in a context of mixed acidosis with PaCO 2 of 50 mmhg or more without possibility of increasing alveolar ventilation, acute pulmonary edema due to fluid overload responsible for severe hypoxemia requiring oxygen flow rate of more than 5 L/min to maintain an SpO 2 of more than 95% or an FiO 2 greater than 50% in patients already on invasive or non-invasive mechanical ventilation and despite diuretic therapy. Pre-existing severe chronic renal failure (defined by a creatinine clearance < 30 7

9 ml/min) Patients already included in the study Patients with inclusion criteria already present for more than 5 hours (to avoid delayed inclusions) AKI caused by urinary tract obstruction or renal vessel obstruction or tumour lysis syndrome or thrombotic microangiopathy or acute glomerulopathy Poisoning by a dialyzable agent Child C liver cirrhosis Cardiac arrest without awakening Moribund state (patient likely to die within 24h) Patient having already received RRT for the current episode of AKI Extracorporeal lung or circulatory assistance Patients included in another clinical study of a RRT technique. Renal transplant Treatment limitation (withholding or withdrawal) Outcomes The follow-up duration for each patient was 60 days. The primary outcome was the overall survival measured from the randomization date until death or day 60. For patients discharged alive from ICU, information on the primary endpoint was obtained either directly from the patient or relatives or from the physician in charge when the patient was still hospitalized. Patients alive at day 60 were censored, and patients lost to follow-up before day 60 were censored at their last follow-up assessment. 8

10 Secondary outcomes were the receipt of RRT at least once with the delayed strategy, the number of RRT-free days, dialysis catheter-free days, ventilator-free days, vasopressor-free days (i.e., days alive and without the treatment; 0 day was assigned for patients who died) between randomization and day 28, the SOFA score (25) at day 3 and day 7, the vital status at day 28, the length of stay in ICU and hospital, the proportion of patients with treatment limitations, the occurrence of nosocomial infections and the complications potentially related to AKI or RRT such as hemorrhage requiring red blood cell transfusion, thrombocytopenia (less than platelets/mm 3 ), thrombosis of a large venous axis, hypokalemia (below 3 mmol/liter), hypophosphatemia (below 0.6 mmol/liter), hyperkalemia (more than 6.5 mmol/liter), cardiac rhythm disorders. Other pre-specified outcomes included time between randomization and RRT initiation, time between occurrence of at least one of the criteria that mandated RRT in the delayed strategy and actual initiation, the number of RRT sessions and dependence on RRT at day 28 and day 60. Since diuresis was closely monitored, we defined renal function recovery as a diuresis of more than 1000 ml/24h or 2000 ml/24h (without or with diuretics, respectively), in the absence of RRT initiation or resumption for at least 7 days. The number of patients requiring transfusion and the total number of units of red cells transfused were also compared. These outcomes had not been prospectively defined as such at the time of study conception. 9

11 Complications potentially related to AKI or RRT: definitions (refer to Table 2) Hemorrhage was defined as requirement for red blood cell transfusion. Thrombocytopenia was defined as a platelet count lower than 100,000/mm 3. Thrombosis of a large venous axis was diagnosed by Doppler ultrasonography. Hypokalemia was defined as serum potassium concentration below 3 mmol/liter. Hypophosphatemia was defined as a serum phosphate concentration below 0.6 mmol/liter. Hyperkalemia was defined as serum potassium concentration of more than 6.5 mmol/liter. Severe cardiac rhythm disorders included ventricular tachycardia, ventricular fibrillation and torsade de pointe. Moderate cardiac rhythm disorders were defined as a new onset of atrial fibrillation requiring medical treatment or external electric counter shock. 10

12 PROTOCOL VIOLATIONS In the early RRT strategy, 6 patients did not receive any RRT session because of the following reasons: 3 patients were erroneously included, 2 patients died before RRT initiation, and one patient received incorrect strategy after an investigator misread his treatment assignment. In the delayed RRT strategy, 5 patients received RRT before recommended RRT initiation criteria because of misinterpretation of the protocol. 11

13 PATIENT FOLLOW-UP Follow-up at 60 days was available for 614/620 patients. One patient withdrew permission to use his data and five were lost to follow-up before 60 days (min 6, max 56). 12

14 INTERIM ANALYSES Two interim analyses (making a total of three analyses) were planned in this study. The first and second interim analyses of the primary outcome, with a data cutoff of August 29 TH, 2014 (after 90 deaths) and April 17 TH, 2015 (after 185 deaths) respectively, were conducted by the statistician (DH) aware of study-group assignments. The DSMB was blinded to the allocation group, and recommended continuing the study as planned. All investigators and steering committee remained unaware of study outcomes until the end of the inclusion and follow-up period. To maintain an overall type I error rate of 5%, the significance level of each analysis was adjusted, using the O Brien & Fleming approach of group sequential analysis. The following table summarizes the number of deaths and the significance level of testing at each preplanned analysis of the overall survival, at the time of the study conception. Number of deaths Fraction of Significance level information First analysis Second analysis Final analysis The first interim analysis was conducted after the occurrence of 90 deaths. The second interim analysis was conducted after the occurrence of 185 deaths. The significance boundaries were adapted to the timing of the second analysis, as followed: Number of deaths Fraction of Significance level information First analysis Second analysis Final analysis

15 The final analysis was conducted after the occurrence of 302 deaths. The significance boundaries were adapted as followed: Number of deaths Fraction of Significance level information First analysis Second analysis Final analysis The conclusions of the three analyses remain the same whatever the adaptation of the significance boundaries: HR associated with the delayed strategy CI95% P-value First analysis to Second analysis to Final analysis to

16 TABLES Table S1 Table S1. Criteria mandating RRT initiation in the delayed RRT strategy group* Oliguria or anuria for more than 72 hours after randomization Blood urea nitrogen of more than 112 md/dl (40 mmol/liter) Serum potassium concentration of more than 6 mmol/liter Serum potassium concentration of more than 5.5 mmol/liter despite medical treatment (bicarbonate and/or glucose-insulin infusion) ph below 7.15 in a context of pure metabolic acidosis (PaCO 2 below 35 mmhg) or in a context of mixed acidosis with PaCO 2 of 50 mmhg or more without possibility of increasing alveolar ventilation Acute pulmonary edema due to fluid overload responsible for severe hypoxemia requiring oxygen flow rate of more than 5 l/min to maintain an SpO 2 of more than 95% or requiring an FiO 2 greater than 50% in patients already on invasive or non-invasive mechanical ventilation and despite diuretic therapy * RRT denotes renal replacement therapy 15

17 Table S2 Table S2. Characteristics of the patients at baseline* Characteristic Early RRT strategy (N=311) Delayed RRT strategy (N=308) Age yr 64.8± ±13.4 Male sex no. (%) 209 (67) 198 (64) Weight kg 85.4± ±20.9 Main reason for ICU admission no. (%) Medical Surgical, emergency Surgical, scheduled 247 (79) 48 (15) 16 (5) 246 (80) 47 (15) 15 (5) Serum creatinine before ICU admission mg/dl 0.95± ±0.31 Coexisting condition no. (%) Chronic renal failure Hypertension Diabetes mellitus Congestive heart failure Ischemic heart disease 22 (7) 161 (52) 82 (26) 24 (8) 30 (10) 38 (12) 167 (54) 81 (26) 32 (10) 32 (10) Time from admission to randomization days median (IQR) 1 (1-2) 1 (1-2) SAPS III at inclusion 72.6± ±14.2 SOFA at inclusion 10.9± ±3.1 Oliguric/anuric patients no. (%) 202 (65) 191 (62) Physiological characteristics Mean arterial pressure mm Hg Heart rate beats/min 75.5± ± ± ±25.1 Exposure to at least one nephrotoxic agent in past 2 days no. 194 (63) 195 (65) 16

18 (%) 66 (34) 71 (36) Intravenous contrast no. (%) Aminoglycoside no. (%) Vancomycin no. (%) Physiological support no. (%) Invasive mechanical ventilation Vasopressor support (epinephrine or norepinephrine) Sepsis status no. (%) Sepsis Severe sepsis Septic shock 106 (55) 26 (13) 266 (86) 265 (85) 25 (8) 16 (5) 209 (67) 106 (54) 29 (15) 267 (87) 263 (86) 21 (7) 19 (6) 204 (66) ARDS no. (%) 104 (34) 103 (34) Number of patients with oliguria/anuria no. (%) 202 (65) 191 (62) Biological characteristics Serum creatinine - mg/dl Blood urea nitrogen mg/dl Serum potassium mmol/l Serum bicarbonate mmol/l Serum sodium mmol/l Prothrombin ratio (%) 3.25± ±24 4.4± ± ± ± ± ±24 4.4± ± ± ±19.3 * Plus minus values are means ±SD. A total of 620 underwent randomization and one patient secondary refused the use of his data. There were no significant differences between study groups in any of the measured baseline characteristics except for prothrombin ratio (p=0.05). ICU denotes intensive care unit, ARDS acute respiratory distress syndrome and IQR interquartile range. To convert values for creatinine to micromoles per liter, multiply by To convert values for blood urea nitrogen to millimoles per liter, multiply by Serum creatinine concentration before ICU admission was determined by results of a measurement in the 12 months preceding the ICU stay or estimated (22). The Simplified Acute Physiology Score (SAPS) III with higher scores indicating more severe 17

19 disease and a higher risk of death The Sepsis-related Organ Failure Assessment SOFA score with higher scores indicating more severe organ failure (25). Sepsis was defined as suspected or confirmed infection, with at least two out of four signs of a systemic inflammatory response. Severe sepsis was defined as sepsis with evidence of organ dysfunction. Septic shock was defined as sepsis-induced hypotension despite fluid resuscitation of at least 30 ml per kilogram of intravenous fluid administered within the period spanning the 4 hours before and 4 hours after initiation of vasopressor therapy. ARDS was defined according to Berlin definition. 18

20 Table S3 Table S3. Distribution of criteria which mandated RRT initiation in the delayed strategy group* (157 patients of 308 in this group actually received RRT) Criteria Oliguria or anuria for more than 72 hours after randomization no. (%) 59 (38) Blood urea nitrogen of more than 112 md/dl (40 mmol/liter) no. (%) 59 (38) Serum potassium concentration of more than 6 mmol/liter or more than (17) mmol/liter despite medical treatment (bicarbonate and/or glucose-insulin infusion) no. (%) ph below 7.15 in a context of pure metabolic acidosis (PaCO 2 <35 mmhg) or 33 (21) in a context of mixed acidosis with PaCO 2 of 50 mmhg or more without possibility of increasing alveolar ventilation no. (%) Acute pulmonary edema due to fluid overload leading to severe hypoxemia 9 (6) requiring oxygen flow rate of more than 5 l/min to maintain SpO 2 of more than 95% or requiring an FiO 2 greater than 50% in patients already on invasive or non-invasive mechanical ventilation and despite diuretic therapy no. (%) Others 5 (3) * Some patients had several indications. RRT denotes renal replacement therapy 19

21 Table S4 Table S4. Patient characteristics at the time of RRT initiation * Characteristic Early RRT Delayed RRT P Value strategy strategy N=305 N=157 Urine output before RRT ml/24h median 150 (50-600) (IQR) Serum creatinine mg/dl 3.27± ±2.33 <0.001 Blood urea nitrogen mg/dl 52±24 90±34 <0.001 Potassium mmol/liter 4.4± ±0.9 <0.001 Bicarbonate mmol/liter 18.9± ±5.6 <0.001 ph 7.30± ±0.15 <0.001 Sodium mmol/liter 137.9± ± Invasive mechanical ventilation no. (%) 264 (87) 138 (88) 0.75 Vasopressor (epinephrine or norepinephrine) support no. (%) 254 (84) 125 (80) 0.30 Epinephrine dose mg/hour 2.8± ± Norepinephrine dose mg/ hour 4.2± ± * Plus minus values are means ±SD. Only 157 patients of the 308 of the delayed strategy actually received RRT. RRT denotes renal replacement therapy and CI confidence interval. To convert values for creatinine to micromoles per liter, multiply by To convert values for blood urea nitrogen to millimoles per liter, multiply by In the early RRT strategy, 6 patients did not receive any RRT session (see protocol violations) 24-hour urine output was not available in many patients in the early RRT strategy group because they were randomized within 24 hours after admission. 20

22 Table S5 Table S5. Characteristics of RRT sessions delivered during the 28 days following randomization* Characteristic Early RRT Delayed RRT P Value strategy strategy N=305 N=157 Dialysis catheter insertion site no. (%) 0.81 Jugular 123 (41) 68 (44) Femoral 167 (55) 81 (52) Sub-clavian 13 (4) 7 (5) First modality no. (%) 0.97 Intermittent RRT 169 (56) 86 (55) Continuous RRT 135 (44) 71 (45) RRT modalities during ICU stay no. (%) 0.62 Intermitent RRT only 142 (47) 73 (47) Continuous RRT only 101 (33) 47 (30) Both modalities (intermittent and 61 (20) 37 (24) continuous) Mean blood urea nitrogen during RRT mg/dl 38 (17) 57 (27) <0.001 Total number of RRT sessions Number of RRT sessions median (IQR) All patients at day 60 3 (2-7) 4 (2-8) 0.15 Patients dead at day 60 3 (2-7) 3 (2-7) 0.80 Patients alive at day 60 3 (1-8) 6 (3-10) Resumption of RRT after initial cessation no. (%) 11 (4) 4 (3) 0.54 * RRT denotes renal replacement therapy, ICU intensive care unit, and CI confidence interval. As explained in Table S3, only 157 patients of the 308 of the delayed strategy actually received RRT. 21

23 Neither Sustained Low-Efficiency Dialysis (SLED) nor CRRT with high ultrafiltration rates were used in any patient For continuous RRT, one session is considered as one day 22

24 Table S6 *The baseline predictors of initiation of RRT were assessed in the delayed group only, using a Cox semi-parametric proportional-hazards model. Clinically relevant baseline predictive factors were: age, serum urea, serum creatinine, serum potassium, serum bicarbonate, arterial ph, presence of a sepsis, treatment with vasopressors, SOFA score, SAPS III score, presence of an ARDS, and exposure to nephrotoxic agents. All these predefined variables were considered as candidate for inclusion into the multivariate model. First, a univariate analysis was performed. The linearity assumption was checked for all continuous variables, and variables which significantly violated the linearity assumption were dichomized according to the median value**. Final model was selected using a backward stepwise selection process (variables with p-value > 0.20 were iteratively removed from the model). ** Only baseline potassium was dichotomized. 23

25 Table S7 Table S7. Medical treatment of AKI-related metabolic complication before the first RRT session for patients who received it or during the whole ICU stay for patients who did not receive it Characteristic Early RRT Delayed RRT P Value strategy strategy n=311 n=308 Diuretics no. (%) 4 (1.3) 112 (36.5) <0.001 Medical treatment of hyperkalemia no. (%) 17 (5.5%) 67 (22.9%) <0.001 Medical treatment of acidosis no. (%) 21 (6.8%) 49 (16.7%) <

26 Table S8 Table S8. Etiologies of hemorrhage* Early RRT strategy (N=311) Delayed RRT strategy (N=308) P Value All hemorrhages no. (%) 27 (9) 36 (12) 0.21 Dialysis catheter-related Hemorrhage 3 (1) 2 (1) 1 no. (%) Digestive tract hemorrhage 24 (8) 25 (8) 0.85 Hemorrhage of other etiologies no. (%) 1 (0.3) 12 (4) * RRT denotes renal replacement therapy In the early RRT strategy group, one hemorrhage was a complication of arterial puncture during coronary angiogram. In the delayed RRT strategy group, the etiologies were the following: 6 abdominal surgery sites bleeding, 3 hemothoraces, 2 nondialysis catheter-related hemorrhage, one polytrauma patient and one hemoptysis. 25

27 FIGURES Figure S1 Flow-chart: Enrollment, Randomization and Follow-up of the study participants. ICU denotes intensive care unit, AKI acute kidney injury, KDIGO kidney disease improving global outcome and RRT renal replacement therapy 5528 Had AKI and received vasoactive agent and/or invasive mechanical ventilation 2098 Were excluded because AKI did not reach stage 3 of KDIGO classification 3430 Had AKI stage 3 of KDIGO classification 2583 Were excluded 663 Had immediate RRT indication 370 Had severe chronic renal failure 348 Had moribund state 331 Had cardiac arrest without awakening 265 Had treatment limitation 209 Had inclusion criteria already present for more than 5 hours 165 Had already received RRT for the current episode of AKI 149 Had Child C liver cirrhosis 144 Had AKI caused by urinary tract obstruction or renal vessel obstruction or tumour lysis syndrome or thrombotic microangiopathy or acute glomerulopathy 81 Had extracorporeal lung or circulatory assistance 74 Had renal transplant 42 Had poisoning by a dialyzable agent 12 Were included in another clinical study of a RRT 227 Were eligible but not enrolled 620 Underwent randomization 312 Were assigned to early RRT strategy 308 Were assigned d to delayed RRT strategy 1 patient refused the use of data 619 Were included in the analysis 26

28 Serum creatinine concentration (mg/dl) Blood urea nitrogen (mg/dl) Figure S2 Blood urea nitrogen and serum creatinine level changes overtime. *** indicates p<0.001 There was no significant difference between groups in either blood urea nitrogen or serum creatinine levels both on admission and at discharge A 200 *** Group Early RRT strategy Delayed RRT strategy 50 0 Admission Randomization At RRT initiation Discharge (alive) B *** Group Early RRT strategy Delayed RRT strategy 1 Admission Randomization At RRT initiation Discharge (alive) 27

29 Probability of diuresis recovery Probability of being free of RRT during 7 successive days Figure S3 A--Proportions of patients who did not need RRT for at least 7 days overtime (patients who never received RRT or no longer required it) B--Proportions of patients with diuresis of more than 1000 ml/24h or 2000 ml/24h (without or with diuretics, respectively) A Early RRT strategy Delayed RRT strategy p value: < Days B Early RRT strategy Delayed RRT strategy p value: Days

30 Probability of spontaneous creatinine decrease Figure S4 Proportion of patients with spontaneous (in the absence of renal replacement therapy) creatinine decrease p value: 0.01 Early RRT strategy Delayed RRT strategy No. at Risk Days

31 UNPLANNED ANALYSES Sub-populations were defined after all results were known. Only patients from the early strategy formed a group prospectively defined per-protocol. The 2 other sub-populations resulted from splitting of delayed RRT strategy patients in two sub-populations according to whether patients received RRT or not ( Delayed strategy RRT+ or Delayed strategy RRT ). This splitting was not planned per-protocol but was done after results were known in order to show that the main determinant of patient mortality did not result from the fact that they received either no RRT or early RRT or late RRT but from baseline severity criteria. Variables Early RRT strategy Delayed RRT strategy-rrt- Delayed RRT strategy- RRT+ Total SAPS III at baseline Med [IQR] 71 [64-79] 72 [64-79] 73 [ ] 72 [64-81] Moy (std) 72.6 (14.41) (14.26) (13.68) (14.21) P value p value: (Kruskal-Wallis rank sum test) SAPS III at baseline (groups) < (43.2%) 60 (42.3%) 53 (35.3%) 243 (41%) >= (56.8%) 82 (57.7%) 97 (64.7%) 350 (59%) Total 301 (50.8%) 142 (23.9%) 150 (25.3%) 593 (100%) p value: (Pearson s Chi-squared test) SOFA at baseline Med [IQR] 11 [9-13] 10 [8-12] 12 [9-14] 11 [9-13] Moy (std) (3.18) 9.96 (2.89) (3.12) (3.15) p value: < (Kruskal-Wallis rank sum test) The highest severity at baseline was observed in patients who received RRT in the delayed strategy group (Delayed strategy RRT+). The lowest severity was observed in patients who never received RRT (Delayed strategy RRT ). Patients of the early RRT strategy group had intermediate severity at baseline (Early RRT strategy). 30

32 Proportion of survivors The figure below represents Kaplan-Meier curves of the same three subpopulations of patients (defined after results were known) as in the above table. In addition, these subpopulations were stratified according to SAPSIII at randomization. The observed survival stratified by SAPS III at randomization (solid lines) is very similar to the observed survival stratified by RRT/no RRT (dashed lines): Early RRT strategy SAPS III < 70 Early RRT strategy SAPS III >= 70 Delayed RRT strategy SAPS III < 70 Delayed RRT strategy SAPS III >= 70 Delayed RRT strategy RRT Delayed RRT strategy RRT Days There was no statistically significant interaction between RRT strategy and SAPSIII (p = 0.92) 31

33 The two tables below show that survival differences between the 3 sub-populations defined above were explained by differences in baseline severity: they were no longer present after matching on SAPS III at randomization. Analysis Population HR CI 95% p. value Before matching Early strategy 1.00 Delayed strategy, RRT [0.52;0.97] After matching Early strategy 1.00 Delayed strategy, RRT [0.57;1.21] Analysis Population HR CI 95% p. value Before matching Early strategy 1.00 Delayed strategy, RRT [1.08;1.8] After matching Early strategy 1.00 Delayed strategy, RRT [0.9;1.7] The decrease/increase toward 1 of the hazard ratio after matching on SAPS III at baseline is a strong argument in favor of the fact that the observed differences between the three subpopulations were biased by confounding. 32