Study aim
The aim of this study was to assess prevalence and risk factors of HIRRT associated with preload-dependence during the first 7 days of CRRT.
Study design and setting
We conducted a prospective, observational, single-center cohort study between May 9, 2017 and September 1st, 2020 in a 15-bed medical intensive care unit (ICU). The study was approved by an ethics committee (CPP Ile de France IV, ID-RCB 2017-A00483-50) and was registered on ClinicalTrials.gov (NCT 03139123) on May 2nd, 2017. Informed consent for study inclusion was obtained from all individual participants and/or their closest relatives.
Patients
To be eligible, the subjects had to fulfill all the following inclusion criteria: aged 18 years or older, with acute kidney injury KDIGO 3 [14], treated with CRRT for less than 24 h and monitored by mean of a PiCCO® device (Pulsion Medical Systems, Feldkirchen, GERMANY) mandated by acute circulatory failure. Exclusion criteria were pregnancy, lower limb amputation, intracranial hypertension, known obstruction of inferior vena cava, ongoing directives to withhold or withdraw life sustaining treatment, lack of consent by patient or next of kin, lack of affiliation to social security, patient under a legal protective measure, inclusion in another research study and previous inclusion in current study.
Data collection
The following variables were recorded at inclusion: demographic and anthropometric data, time of ICU admission and inclusion, admission category, Sequential Organ Failure Assessment (SOFA) score [15] at ICU admission, and Simplified Acute Physiology Score (SAPS) II at ICU admission [16].
The following variables were recorded at inclusion, every 4 h and at the onset of each HIRRT episode until study completion: heart rate, systolic, diastolic and mean arterial pressures, pulse pressure variation (PPV), central venous pressure, cardiac index assessed by both thermodilution and pulse contour analysis, stroke volume variation (SVV), extravascular lung water index, global end-diastolic volume index, pulmonary vascular permeability index, global ejection fraction, vasopressor administration and dose, inotrope administration, mechanical ventilation use, CRRT settings (blood flow, ultrafiltrate or dialysate rate and temperature, net ultrafiltration rate), and preload-dependency tested as described below.
The following variables were recorded at inclusion and daily until study completion: SOFA score [15], body weight, fluid balance, arterial blood gas, arterial lactate, hemoglobin, fulfillment of sepsis and septic shock criteria [17].
Missing data per variable are reported in Additional file 1.
Study follow-up
Patients were followed during the first 7 days after inclusion or less in case of occurrence of any of the following events: death, end of life care, CRRT cessation or interruption of PiCCO® monitoring.
CRRT management
The indication, technique [continuous veno-venous hemofiltration (CVVH) or continuous veno-venous hemodialysis (CVVHD)] and settings of CRRT were under the responsibility of the clinician in charge of the patients, in accordance with current practice guidelines [14]. CRRT was performed with the Multifiltrate® station and the Ultraflux® AV1000S hemofilter (Fresenius Medical Care, Bad Homburg, GERMANY). CRRT settings were adjusted by the attending physician. The ICU policy was to promote hemodynamic monitoring, using the PiCCO® device whenever severe shock was present in patients being treated with CRRT.
Hemodynamic measurements
HIRRT was defined as mean arterial pressure below 65 mmHg justifying any therapeutic intervention among the following ones: fluid administration, initiation or increase in vasopressor dose, or discontinuation or decrease of net ultrafiltration rate on CRRT. Once hypotension occurred and before any therapeutic intervention, a postural test (PLR in the supine position or Trendelenburg maneuver in the prone position) was performed by trained ICU nurses during 1 min to assess for preload-dependence. PLR was performed from the semi-recumbent position with the trunk at 45° [18] and the Trendelenburg maneuver was performed from a 13° upward bed angulation to a − 13° downward bed angulation in patients in the prone position [13]. Preload-dependence was deemed present if the pulse contour-derived cardiac index increased by at least 10% and 8% during the PLR test and the Trendelenburg maneuver, respectively.
Therapeutic management of HIRRT was at the discretion of the clinician in charge of the patient and was not protocolized. A 1-h period without new hemodynamic assessment was allowed after each HIRRT episode onset to wait for treatment effect.
Hemodynamic measurements including a postural test were systematically performed by trained ICU nurses every 4 h and during each HIRRT episode. Regular training sessions of nurses to hemodynamic measurements were organized to ensure quality of data acquisition. Arterial and central venous blood pressures were continuously monitored, using arterial femoral and jugular vein catheters, respectively, connected to an Intellivue MP40 monitor equipped with the PiCCO® technology module (Philips Healthcare, Andover, MA, USA). Cardiac output was assessed using the PiCCO® device, calibrated with the transpulmonary thermodilution technique at least every 4 h, using a triplicate intravenous infusion of 15 mL cold serum saline. Cardiac output was then continuously monitored using pulse contour analysis with the PiCCO® device. Arterial dynamic elastance was computed as the ratio of PPV over SVV.
End points
Primary end point was the rate of HIRRT associated with preload-dependence, with reference to the total number of HIRRT episodes occurring during the first 7 days after inclusion. Secondary end point was the identification of risk factors for HIRRT associated with preload-dependence.
Statistical analysis
Statistical analyses were performed using R software version 4.0.2 [19] and the following packages: lme4 [20], Lmertest [21], pROC [22], PropCIs [23], MultinomialCI [24] and mice [25]. A p value below 0.05 was chosen for statistical significance. The statistical unit was the hemodynamic measurement. Power of the study was computed using the normal approximation confidence interval method [26]. Assuming a rate of HIRRT associated with preload-dependence between 0.25 and 0.5, we calculated that with a sample size between 72 and 96 HIRRT episodes, the study would provide at worst a ± 10% precision in the 95% confidence interval of the prevalence of HIRRT associated with preload-dependence. We decided to include conservatively at least 100 HIRRT episodes and at least 50 patients to ensure minimal representativity. Analyses were performed on all included patients, including those prematurely withdrawn. Medians and interquartile ranges were reported for continuous variables and counts in each category with corresponding percentages were reported for categorical variables. Ninety-five percent confidence intervals (CI95%) for multinomial proportions were computed using the Sison and Glaz method [27]. To test whether each therapeutic intervention (namely fluid administration, initiation or increase in vasopressor dose, or discontinuation or decrease of net ultrafiltration rate on CRRT) differed between preload-dependent and preload-independent HIRRT episodes, we used 3 logistic regression mixed models with HIRRT type as the dependent variable, each therapeutic intervention as binary independent variable and patient as variable with a random effect, and the Bonferroni correction was used to account for multiple testing. To test which variables could predict occurrence of HIRRT associated with preload-dependence, the whole dataset was restricted to hemodynamic measurements obtained without HIRRT, and a new variable was computed [occurrence of HIRRT associated with preload-dependence in the subsequent measurement (Yes/No)]. Variables were entered into a mixed logistic regression model, using patient as variable with a random effect, and occurrence of HIRRT associated with preload-dependence in the subsequent measurement as the dependent variable. Some continuous variables were entered in the model as dichotomized variables, using ROC curve analysis and computation of optimal cut-off points by maximizing the Youden’s index. Independent variables associated with occurrence of HIRRT with preload-dependence with a p value below 0.2 in univariate analysis were selected for inclusion in a multivariable mixed logistic regression model, using backward stepwise descending selection. Interactions between predictors were assessed on the final model. Missing data in multivariate analyses were handled using multiple imputations and predictive mean matching. Model calibration was assessed by the Hosmer–Lemeshow test and model discrimination by the C-statistic.