Methods
The study protocol was made available for review to ICU physicians from different countries. Once the ICU was enrolled in the research group, data were collected for all admitted patients for 3 consecutive months, in the period between April 2012 and September 2014 using an electronic case report form. Exclusion criteria were (a) age <18 years or >85 years; (b) chronic dialysis; (c) short-term postoperative admission; (d) life expectancy less than 48 h; (e) need for extracorporeal membrane oxygenation (ECMO) within the first 48 h of ICU stay. All types of ICUs were eligible on voluntary basis within the indicated period. Local ethics committees approved the study according to the local regulations.
Data collection
We designed a specific password-protected website for the study. Centres could only access data related to their patients. An automatic data verification system screened each field for missing or out-of-range values and data inconsistencies, generating a visual alert. The online tool included seven sections: (1) Demographics, anthropometrics, and admission diagnoses; (2) Medical history; (3) Admission data; (4) Daily vital signs and laboratory values; (5) RRT; (6) Sepsis and (7) Outcomes/case closure. Data input was on daily basis. Severity of illness was defined by Simplified Acute Physiology Score II (SAPS II) [
12], Sequential Organ Failure Assessment (SOFA) [
13] and Acute Physiology and Chronic Health Evaluation II (APACHE II) [
14] scores at admission and described in a pie chart (Additional file
1: Figure S1). AKI diagnosis was based on creatinine (Cr) levels using the KDIGO classification [
15]. The creatinine value prior to hospitalization (Medical history) was used as baseline. If missing, a reference creatinine was requested utilizing the Modification of Diet in Renal Disease (MDRD) formula assuming a glomerular filtration rate of 75 ml/min/1.73 m
2. Daily data entry included serum creatinine, total fluid intake and output, urine volume, ventilation modality, diuretic therapy and, if present, RRT. As soon as a patient met the criteria for AKI, a full data entry section was activated for 14 days with automatic daily display of AKI stage. Data for SOFA calculation and for epinephrine/norepinephrine dose were additionally included. A dynamic chart displaying fluid balance and clinical management was generated using the data entered daily (see Additional file
1: Figure S2). The RRT section included data on modality, flows (blood, dialysate, reinfusion, pre/post ratio, and net ultrafiltration rate), anticoagulation, duration (for each 24-h period), reason for initiation, type of vascular access and malfunctions. RRT fields were automatically adjusted according to treatment modality. It was mandatory to report the cause of downtime whenever the effective treatment time was less than that prescribed. Data on circuit life span were recorded for each treatment. The case closure section included data on outcome and renal status at discharge. Data on fluid balance were automatically calculated by the CRF from the time of admission, based on daily data entry.
Definitions
Fluid overload (FO) was defined as the ratio between cumulative fluid balance and the initial body weight, in percentage. Maximum fluid overload (MFO) referred to the peak value of FO observed during the entire ICU stay. TMFO represented the number of days between ICU admission and day of MFO. Fluid overload slope (FOSL) was computed as the ratio MFO/TMFO and represented the velocity of fluid accumulation. The term “AKI-RRT” defined AKI patients who received at least one session of RRT while “AKI” referred to AKI patients without RRT, when not differently specified. N-AKI defined patients who never developed AKI. AKI stages were based on the creatinine criteria of the KDIGO classification.
Statistical analysis
Continuous variables were described as mean ± standard deviation (SD) or median and interquartile range (IQR). Percentages were calculated for categorical variables. Bivariate comparisons of patients with AKI and patients without AKI were performed using the Wilcoxon rank-sum test and the chi-square test, as appropriate.
Boxplots were used to illustrate the FO of the different AKI groups (N-AKI, AKI and AKI-RRT) during the first 5 days of ICU stay. The Kruskal-Wallis test was performed to explore whether the three groups differed on individual days. The resulting p values were corrected for the multiple test situations with the Bonferroni formula. Post hoc tests, always comparing two groups, were done using the Wilcoxon rank-sum test and p values were corrected (Bonferroni).
To visualize the trend of fluid accumulation in reference to the development of AKI, the delta between FO at AKI diagnosis and the FO at each day between 3 days before and up to 3 days after diagnosis of AKI was calculated. Patients were censored at the day of AKI recovery. Means ± standard error (StdErr) of FO were plotted for the whole 7-day period.
A figure with boxplots for the three groups (N-AKI, AKI and AKI-RRT) illustrated the MFO during the ICU stay. Similarly, for the AKI-RRT group, a boxplot described the fluid status at different time points. Survivors and non-survivors were also plotted. The horizontal axis described the median day for the corresponding boxplot event.
To characterize the FO prior to death or ICU discharge, means ± StdErr of all patients who stayed in the ICU for at least 5 days were plotted.
A Kaplan-Meier analysis was performed to predict the time to death for the three AKI groups (N-AKI, AKI and AKI-RRT) separately. The difference between the three groups was tested by a log-rank test. This analysis was restricted to the first 30 days of follow-up; patients who stayed in the ICU for more than 30 days were censored at this time.
An unadjusted logistic regression model was used to illustrate the predicted probabilities of MFO on ICU mortality. For this model, follow-up was restricted to the median time in the ICU (12 days).
Additionally, the previous model was adjusted for AKI status (yes/no during the first 12 days of follow-up) and APACHE II score (at baseline). The predicted probabilities for ICU death were plotted for different APACHE II scores. For both models, the odds ratios (OR) and corresponding 95 % confidence intervals (CI) were reported.
Cox proportional hazard regression analysis was applied to evaluate the time to death. The main predictor was FOSL. Hazard ratios (HR) and corresponding 95 % confidence intervals were reported using an unadjusted model as well as a model adjusted for: age, sex, SAPS II, sepsis (yes/no, at admission), mechanical ventilation (yes/no, at admission), diabetes (yes/no, at admission), cardiovascular disease (yes/no, at admission), and hypertension (yes/no, at admission). Only significant variables were shown in the selected models.
P values less than 0.05 were considered to be significant. The analysis was conducted with the statistical software SAS, version 9.4 (SAS Institute Inc., Cary, NC, USA).
Discussion
Fluid therapy is an integral component of the management of critically ill patients. However, wide variation in clinical practice has been observed, in particular related to type of fluid, rate of administration and methods for assessing fluid responsiveness [
16]. Our study shows that ICU patients tended to accumulate fluid in various degrees from the day of admission onwards. In AKI patients, more fluid accumulated between the 3 days prior to the diagnosis of AKI and 3 days later. Despite the fact that all groups accumulated fluid during the ICU stay, AKI-RRT patients showed the highest degree of FO. In this group, FO peaked several days after RRT initiation, which indicates the challenges of removing fluid even with extracorporeal support.
Our results are complementary to the findings of other studies in the literature [
17], but also provide new insights into a very complex clinical problem.
Among the overall population as well as the subgroups (N-AKI, AKI and AKI-RRT), non-survivors accumulated more fluid. Moreover, there was a progressive increase in fluid accumulation during the 4 days prior to death. Patients without AKI had a survival benefit (univariate model). The AKI-RRT group had the lowest survival rate during the whole period while AKI patients had an intermediate chance of survival, as previously described in the literature [
1,
3,
4,
8].
Furthermore, MFO was an independent risk factor for mortality [OR 1.044 (CI 1.023–1.065)], and the predicted probability of death increased exponentially. AKI patients, including those receiving RRT, were more likely to die than N-AKI. Moreover, the higher the MFO, the wider the difference in mortality between the AKI and N-AKI cohort, which suggests that FO aggravates the patients’ underlying condition [
18,
19]. This finding is novel and has not been reported in the literature before. At variance, the PICARD group and others [
8,
20‐
22] concluded that the risk of mortality was proportional to fluid accumulation above a particular cutoff value. Our model predicted the probability of death as a continuum, with an exponential relation to the MFO increase. This suggests that “fluid overload” should be defined as any degree of positive fluid balance rather than a value above an arbitrary cutoff.
We also demonstrated another difference between N-AKI and AKI patients. It seemed that, while AKI independently worsened patients’ outcome, the level of maximum fluid overload was more harmful for patients with AKI than the N-AKI cohort. In addition, severity of illness as defined by APACHE II score was independently associated with mortality. Of interest, the difference in mortality risk between the AKI versus N-AKI cohort tended to increase with higher APACHE II scores.
Our data showed that the velocity of fluid accumulation is also important. Speculating on a possible role of the fluid accumulation velocity, the more rapidly FO occurred, the higher the risk of dying. For any one unit rise of FO
SL, the probability of death increased 27 times. This finding was significant for all three subgroups in a univariate analysis and remained significant for the AKI cohort but not N-AKI patients after adjustment for other risk factors. This observation may suggest that patients without AKI tolerate a positive fluid balance better during the resuscitation and maintenance phase [
23], especially those suffering from sepsis [
24,
25].
Based on these results, it appears that, instead of using an absolute fluid accumulation or FO value as a predictor of poor patient outcome, fluid accumulation velocity may have a more physiological rationale and a better statistical power to serve as a tool for fluid assessment.
Abbreviations
ACE-I, ACE inhibitor; AKI, acute kidney injury; AKI-RRT, patients with AKI treated with RRT; APACHE, Acute Physiology and Chronic Health Evaluation; ARB, angiotensin receptor blocker; BW, body weight; CI, confidence interval; CKD, chronic kidney disease; Cr, creatinine; CRFs, case report forms; CRRT, continuous renal replacement therapy; DoReMi, The Dose Response Multicentre International Collaborative Initiative; ECMO, extracorporeal membrane oxygenation; FO, fluid overload; FOSL, slope of fluid accumulation, HR, hazard ratio; ICU, intensive care unit; IQR, interquartile range; MDRD, Modification of Diet in Renal Disease; MFO, maximum fluid overload; MFO/TMFO, velocity of fluid accumulation; N-AKI, patients without AKI; NSAID, non-steroidal anti-inflammatory drug; OR, odds ratios; RRT, renal replacement therapy; SAPS, Simplified Acute Physiology Score; SD, standard deviation; SOFA, Sequential Organ Failure Assessment; StdErr, standard error; TMFO, number of days between ICU admission and day of MFO
Acknowledgements
The authors are grateful to Inga Bayh1 and Katharina Brand1 for the statistical support, Matteo Recchia2 and Giovanni Aneloni2 for their invaluable assistance with the information technology, and Fabio Grandi1 for helping with the interpretation of the results.
Other investigators
M. Balciunas3, B. S. Oliveira 4, L. Cachafeiro5, E. G. Porcile6, M. Ranieri7, V. Cantaluppi8, V. Schweiger9, L. Montini10, M. Gurjar11, C. Liuzzo12, N. Brienza13 and J. Silvestre14.
1 Fresenius Medical Care,1 Else-Kröner-Straße. 61352 Bad Homburg – Germany.
2 International Renal Research Institute of Vicenza (IRRIV), San Bortolo Hospital, Vicenza, Italy.
3Anaesthesia and Intensive Care, VilniusUniversityHospitalSantariskiuclinics, Nemencinespl 12A-4, Vilnius, LT-10102, Lithuania. mindaugas.balciunas@santa.lt
4Serviço de Medicina Intensiva, Hospital de Santa Maria - Centro Hospitalar Lisboa Norte. Avenida Professor EgasMoniz. 1649-035, Lisboa. Portugal.brn.s.oliveira@gmail.com.
5Intensive Care, Hospital Universitario La Paz/Carlos III. IdiPAZ, Paseo Castellana 261. 28046, Madrid-Spain. luciacachafeiro@yahoo.es.
6UO Anestesia e Rianimazione, IRCSS AOU San Martino IST, Largo R. Benzi 10, Genova, 16145, Italy. elisagvporcile@gmail.com.
7Nephrology, Dialysis and Renal Transplantation Unit and Department of Anesthesiology, San Giovanni Battista-Molinette Hospital. 88 Corso Bramante. 10126. Torino, Italy. marco.ranieri@unito.it.
8Nephrology and Kidney Transplantation Unit, Department of Translational Medicine, University of Eastern Piedmont,"Maggiore della Carità" University Hospital. 18 Corso Mazzini. 28100 NovaraITALYvincenzo.cantaluppi@med.uniupo.it.
9Department of surgery, Verona University, Piazzale Antonio Scuro 10, Verona, 37134, Italy. vittorio.schweiger@univr.it.
10Intensive Care Unit and Anesthesiology, Policlinico Agostino Gemelli. 8, Largo Agostino Gemelli. 00168 Roma. Italy.luca.montini@rm.unicatt.it.
11Critical Care Medicine, Sanjay Gandhi Postgraduate Institute of Medical Sciences, Department of Critical Care Medicine, SGPGIMS, Lucknow, 220614, India. m.gurjar@rediffmail.com.
12U.O. Anestesia e RianimazioneAgrigento, San Giovanni di Dio.Contrada Consolida 92100. Agrigento, Italy. carmenliuzzo@gmail.com.
13Department of Emergency and Organ Transplantation, University Aldo Moro of Bari, Piazza G. Cesare, Bari, 70124, Italy. nicola.brienza@uniba.it.
14Unidade de Cuidados Intensivos Polivalente, Hospital São Francisco Xavier. Estrada Estrada do Forte do Alto do Duque. 1449-005. Lisboa- Portugal. joanapsilvestre@gmail.com.
Participating centres
1. Servicio de Medicina Intensiva, Hospital Universitario del Vinalopo, Calle Tonico Sansano Mora 14, Elche, Spain; 2. Intensive Care Unit, Hospital Universitario La Paz/Carlos III. IdiPAZ, Paseo Castellana 261, Madrid, Spain; 3. Department of Nephrology, Shanghai Institute of Kidney and Dialysis, Shanghai Key Laboratory of Kidney and Blood Purification, Zhongshan Hospital, Fudan University, Shanghai, China; 4. Medical Intensive Care, University of Poitiers; CHU Poitiers, Poitiers, France; 5. Department of Nephrology, Dialysis and Transplantation, San Bortolo Hospital, Vicenza, Italy; 6.Intensive Care Service, St Antonio Hospital – Porto, Largo. Prof. Abel Salazar, Porto, Portugal; 7. Intensive Care Unit, General University Hospital, Avd Carlos Haya s/n, Malaga, Spain; 8. Department of Nephrology, The First Affiliated Hospital of Nanjing Medical University, Nanjing, Jiangsu, China; 9. Department of Nephrology, Clinica Las Condes, Estoril 450, Las Condes, Santiago, Chile; 10. Department of Critical Care, King’s College London, Guy’s and St Thomas’ Hospital, Westminster Bridge Road, London, UK; 11. Anaesthesia, Intensive Therapy and Pain Management, Vilnius University Hospital Santariskiu Clinics, Nemencinespl 12A-4, Vilnius, Lithuania; 12. Department of Anesthesiology and Critical Care, University Hospital, Lille, France. 13. Serviço de Medicina Intensiva, Hospital de Santa Maria - Centro Hospitalar Lisboa Norte, Lisbon, Portugal; 14. UO Anestesia e Rianimazione, IRCSS AOU San Martino IST, Genoa, Italy; 15. Nephrology, Dialysis and Renal Transplantation Unit and Department of Anesthesiology, San Giovanni Battista-Molinette Hospital, Torino, Italy; 16. Department of Surgery, Verona University, Verona, Italy; 17. Intensive Care Unit and Anesthesiology, Policlinico Agostino Gemelli, Rome, Italy; 18. Critical Care Medicine, Sanjay Gandhi Postgraduate Institute of Medical Sciences, Lucknow, India; 19. U.O. Anestesia e Rianimazione Agrigento, San Giovanni di Dio, Agrigento, Italy; 20. Department of Emergency and Organ Transplantation, University Aldo Moro, Bari, Italy; 21. Unidade de Cuidados Intensivos Polivalente, Hospital São Francisco Xavier, Lisbon, Portugal.
Authors’ contributions
CT, FG, CR, AM, EK, EGP and VC contributed to the study conception and design. FG, GA and MR participated in web-database design and development, and coordination of the study. MO, MSS, DML, JT, RR, AM, MEHG, HJM, DB, EK, MB, BSO, LC, EGP, MR, VC, VS, LM, MG, CL, NB and JS carried out the study enrollment. DM, IB, FG, KB and AL performed the data and statistical analysis. FG, CR, MO, CT and AM drafted the manuscript. EK, DML, MSS, JT, RR, MEHG, HJM, DB, AL and DM revised the manuscript. All authors read and approved the final manuscript.