Fluid overload is associated with an increased risk for 90-day mortality in critically ill patients with renal replacement therapy: data from the prospective FINNAKI study

We performed a prospective, observational cohort study, the FINNAKI study, from 1 September 2011 to 1 February 2012, in 17 Finnish ICUs. The referral areas of these ICUs cover 85% of the Finnish adult population. The Ethics Committee of the Department of Surgery, Hospital District of Helsinki and Uusimaa approved the study protocol and the use of deferred consent with a signed, informed consent from the patient or a proxy obtained as soon as possible. The Finnish National Institute of Health approved the collection of data from medical records of deceased patients if an informed consent could not be obtained.

We included all adult patients with either emergency ICU admission or elective post-operative ICU admission with an expected ICU stay of more than 24 hours. We excluded patients 1) with end-stage renal disease (ESRD) on chronic dialysis 2) who had already participated in the FINNAKI study and had received RRT during that previous ICU admission 3) not permanently living in Finland or having insufficient language skills for giving an informed consent, and 4) on intermediate care (Figure 1). Of patients receiving RRT, we excluded those who received RRT only for intoxication or plasmapheresis for non-renal indications. For analyses regarding fluid accumulation, we excluded patients with incomplete data on fluid balance. The decision to initiate RRT and the choice of RRT modality were based on the judgment of the treating clinician.

We collected physiological data, laboratory values, Sequential Organ Failure Assessment (SOFA) [20] and Simplified Acute Physiology Score (SAPS) II [21] scores, Acute Physiology and Chronic Health Evaluation (APACHE) III diagnoses, and patients' baseline weight with the help of the Finnish Intensive Care Consortium database using electronic patient records. We collected data on patients' chronic illnesses, baseline creatinine (latest creatinine value a week to a year prior to ICU admission), and treatment prior to ICU admission. The patients' daily fluid balance was calculated in the electronic patient records by summing the amount of fluids given (maintenance and resuscitation fluids, blood products, drug infusions, and enteral and parenteral nutrition), from which losses (urine output, bleeding, output from drains, rectal and nasogastric tube, and surrogate for evaporation (1000 mL for normothermic patients, and an addition per each Celsius degree of fever per hour)) were subtracted. We recorded data on patients' daily fluid balance from ICU admission to day five and obtained data on cumulative fluid balance of patients with RRT initiated after day five. We collected data on the given RRT treatment daily during the first five days and, thereafter, twice a week. We recorded the RRT replacement fluid flow and dilution mode, dialysis fluid flow, and the blood flow once daily. We calculated the CRRT dose (mL/kg/hr) according to previous equations [22] and considered treatment modality, blood flow, dilution mode, hematocrit, replacement and dialysis fluid flow, and the patient's weight. We recorded hospital mortality, length of ICU and hospital stay, and patients' need for RRT at 90 days from admission. We obtained patients' vital status 90 days from ICU admission from the Finnish Population Register Centre.

We calculated the total cumulative fluid balance from ICU admission to RRT initiation (including the day of RRT initiation) and defined the percentage of fluid accumulation by dividing the cumulative balance in liters by patient's baseline weight and multiplying by 100%. We then used the cutoff value of 10% of fluid accumulation as a definition of fluid overload [18, 23]. We assessed presence of sepsis on admission and on days one to five according to the American College of Chest Physicians/Society of Critical Care Medicine definition [24]. We defined renal recovery as RRT independency at 90 days from ICU admission [25].

We report data as count and percentages or medians with interquartile range (IQR, 25^th to 75^th percentiles). We used Fisher's exact test to compare proportions and Mann-Whitney U-test to compare continuous data and calculated 95% confidence intervals (CI) for the main outcome. We studied factors associated with 90-day mortality with backwards logistic regression. We used stepwise elimination approach and a significance level of < 0.05 for entry and > 0.10 for removal. We entered the following variables: age, SAPS II score without age points, non-renal SOFA score on the day of RRT initiation, time from ICU admission to RRT initiation (days), initial RRT modality (continuous or intermittent), lactate (mmol/L), base excess (mmol/L), and plasma creatinine (umol/L) at RRT initiation, cumulative urine output on the day of RRT initiation, colloid use prior to RRT initiation (including data from ICU stay and 48 hours prior to ICU admission), presence of severe sepsis (yes/no) during the ICU admission, and fluid accumulation (%) at RRT initiation. We studied fluid overload (fluid accumulation > 10%) as a categorical variable in a separate model with all other covariates being the same as in the first model. We studied the potential interactions in separate models between degree of fluid accumulation (%) and 1) SAPS II score without age points, 2) day of RRT initiation, and 3) presence of severe sepsis 4) creatinine prior to RRT initiation and 5) urine output on the day of RRT initiation. We tested the goodness-of-fit with the Hosmer-Lemeshow test and calculated the area under the receiver operating characteristic curve (AUC) and correct classification rate. We examined the 90-day survival with Kaplan-Meier curves, and compared survival at 90 days between patients with and without fluid overload at RRT initiation with the log-rank test. We considered a two-sided P value lower than 0.05 as significant and made no corrections for multiple comparisons. In logistic regression, we set the significance level at 0.01. We performed the statistical analysis using the SPSS Statistics version 19.0 (SPSS Inc, Chicago, IL, USA).

The median (IQR) number of reported indications for RRT was 3 (2 to 4). The most common indications were oliguria in 223 (78%), high creatinine in 196 (70%), acidosis in 181 (65%), and fluid accumulation in 116 (43%) patients. RRT was initiated on the first ICU treatment day in 124 (41.9%), on day two in 86 (29.1%), on day three in 37 (12.5%), and on day four or after in 49 (16.6%) patients. When patients were classified according to the day of RRT initiation, the degree of fluid accumulation increased until the day three (Additional file 3, Figure S3.) Of the 215 patients who initially received continuous RRT (CRRT), 111 (51.6%) received continuous veno-venous hemodialysis, 93 (43.3%) continuous veno-venous hemodiafiltration and 11 (5.1%) continuous veno-venous hemofiltration. The median (IQR) prescribed CRRT dose was 35.3 (31.2 to 40.6) mL/kg/hr (data from 205 of 215 patients). The median (IQR) daily duration of CRRT was 19.0 (9.3 to 24.0) hr (data from 680 CRRT treatment days), and CRRT dose adjusted for the treatment duration was 27.9 mL/kg/hr. The median (IQR) number of days patients who received CRRT (n = 230) was 3 (2 to 6) and number of intermittent RRT (n = 163) sessions 2 (1 to 4).

Of the 296 patients, 96 (32.4%; 95% CI 31.9 to 32.9%) died in hospital and 116 (39.2%; 95% CI 38.6 to 39.8%) before day 90. The SAPS II -based standardized mortality ratio (95% CI) was 0.64 (0.52 to 0.78). The hospital mortality of patients with or without fluid overload was 43 of 76 (56.6%) and 49 of 207 (23.7%), P < 0.001. The unadjusted 90-day survival of patients with fluid overload compared to those without is presented in Figure 2. The 90-day mortality rate increased as the degree of fluid accumulation increased (Figure 3). At 90 days, 34 of 168 (18.9%; 95% CI 13.2 to 24.6%) survivors (12 patients lost to follow-up) were dependent on RRT.

In logistic regression for 90-day mortality, no interactions between fluid accumulation percentage and 1) SAPS II score without age points, 2) time from ICU admission to RRT initiation, 3) presence of severe sepsis 4) creatinine prior to RRT initiation, or 5) urine output on the day of RRT initiation were present. The final model with fluid overload studied as categorical variable had Hosmer-Lemeshow chi-square 8.6 (P = 0.375), an AUC of 0.816 (95% CI 0.767 to 0.866), and correct classification rate of 74.4%. The results of the final model are presented in Table 4.