Patient selection
The study was approved by the University of Maryland, Baltimore Institutional Review Board (IRB) according to international and national laws (approval number HP-00082080). We conducted a prospective observational study in adult patients generated from a convenience sample of patients admitted to our medical, surgical, trauma or neurotrauma ICUs. Patients were screened for eligibility between 8 and 4 pm on Monday–Friday. Patients were considered eligible if they could be scanned within 24-h of ICU admission. Those with end-stage renal disease on outpatient hemodialysis, a transplanted kidney or liver or those who were transitioning to comfort care were excluded. After consent was obtained, patients underwent a focused rapid echocardiographic evaluation (FREE) [11] which includes HV, PV and RV assessments.
Sonography
All exams were performed by one of four critical care fellows with POCUS and transthoracic echocardiography training. Participating fellows underwent a 1-month clinical rotation in which they were trained in the FREE under the guidance of a Registered Diagnostic Cardiac Sonographer (RDCS) and a RDCS critical care attending. All Doppler findings were obtained during the end-expiratory phase of the patient’s respiratory cycle and with concurrent multi-lead ECG tracings.
Blinding of data and outcomes
Given the observational design, the images, interpretations or recommendations for care were not discussed with or made available to the treating clinicians. Pre-determined clinical data were abstracted from the electronic medical record immediately prior to the performance of the study by the fellow performing the study (described below).
In order to address potential bias, Doppler waveforms were reviewed by two members of the research team (RS and SM) blinded to clinical outcomes. Patients in whom US images were missing or of insufficient quality to determine Doppler waveforms were excluded from the analysis concerning that specific waveform pattern.
Ultrasound image acquisition and interpretation of waveform morphology
Hepatic Vein Doppler (HVD)
To obtain the HV PW, a phased array transducer is used with cardiac pre-sets. ECG leads are placed to assist in the interpretation of HVD in sinus rhythm as well as atrial fibrillation. The middle hepatic vein is identified from a mid-subcostal or lateral views (Fig. 1a). It is interrogated 2–4 cm from its junction to the IVC. Color flow Doppler (CFD) is used to identify high flow parallel to the transducer, and the PW is obtained. The waveform is recorded, and the scale optimized. All Doppler findings were obtained during the end-expiratory phase of the patients respiratory cycle.
The normal waveform is triphasic with three components, a retrograde A wave, a large antegrade S wave and an antegrade D wave (Fig. 2). The A wave occurs during atrial systole when right atrial pressure increases above the mean systemic filling pressure (MSFP), leading to retrograde flow away from the heart. Some describe a V wave as a correction back to baseline, but this is largely noncontributory. Once the right atrium relaxes and thus when the blood flow is normally the greatest, the S wave occurs during right ventricular contraction. The D wave occurs during right ventricular relaxation when tricuspid opening leads to a second drop in RAP. With normal RAP, the S is greater than D (S > D).
With an increase in RAP, the S wave decreases as RAP equilibrates with MSFP earlier. Due to this early equilibration, more RA filling occurs after the tricuspid valve opens leading to a relatively larger D wave. This is what is known as S-D reversal. As venous congestion worsens, the S wave becomes retrograde and often becomes continuous with the A wave leading to a biphasic pattern. Flow changes from normal to elevated; S > D, S < D, S above the baseline fused with A.
The HV was considered abnormal when the maximum negative velocity of the S wave was less than the maximum negative velocity of the D wave.
Portal vein Doppler
From a lateral costal or subcostal window, the portal vein is identified in the coronal plane using a phased-array transducer. Portal veins appear smaller and more hyperechoic than hepatic veins (Fig. 1b). CFD helps identify measurable blood flow. A PW Doppler gate is positioned in the middle of the vessel and the waveform is obtained, and the Doppler scale adjusted.
The normal waveform is a continuous monophasic flow above baseline with minor variations (Fig. 2). As venous congestion worsens, the portal flow becomes pulsatile, with decreases in flow during right ventricular systole. In extreme cases, flow can be interrupted or even have a retrograde component giving a to-and-fro appearance.
The PV were considered abnormal if the pulsatilty index was greater than 30%. The portal pulsatility index was defined as: (VMax − VMin/VMax) * 100%. Here, VMax is the maximal velocity and VMin is the minimal velocity during the cardiac cycle. We also examined PPI as a continuous variable to assess whether small changes in portal vein flow was a clinically important marker of venous congestion.
Intra-renal venous Doppler
From lateral costal window, the kidney is located in the coronal plane and a color flow box placed over the distal renal calyceal junction to cortex. (Fig. 1c) The PW Doppler gate is placed over interlobar vessels with a velocity set between 15 and 20 cm/s. The overall gain can also be used to try to see flow. It may be challenging to obtain in certain patients, and slow fanning of the probe may sometimes reveal better Doppler signals rather than relying solely on the color flow patterns. The waveform is usually considered adequate when both arterial (above the baseline) and venous (below the baseline) are seen clearly for two or more cardiac cycles.
Intra-renal venous Doppler is normally a continuous monophasic flow below the baseline (Fig. 2), which progressively becomes first interrupted with two phases, analogous to the S and the D waves of the hepatic vein flow. Similar to the hepatic vein pattern, as venous congestion worsens, the S wave becomes smaller and the D wave is more pronounced. Eventually the S wave disappears entirely, leaving only a monophasic D wave.
The intra-renal vein Doppler waveforms were considered abnormal if either a biphasic or monophasic renal vein flow pattern was present. This was defined as discontinuous venous flow with either a systolic/diastolic pattern or a diastolic only pattern.
All Doppler waveforms were reviewed by two members of the research team (RS and SM) blinded to clinical outcomes. Final venous waveform patterns were determined in this final review. Any disagreements were adjugated by a discussion between RS and SM. We calculated the inter-rater reliability between our reviewers. Patients in whom US images were missing or of insufficient quality to determine Doppler waveforms were excluded from the analysis concerning that specific waveform pattern.
Outcome measurements and clinical data
Data were collected using manual and electronic data extraction methods from electronic health records. This included data on pre-enrollment renal function, demographic characteristics, diagnoses, SOFA scores, admission serum lactate values, intravenous fluids administered on the day of the FREE study, plasma electrolyte and creatinine values, need for receipt of renal-replacement therapy, CAM ICU score, fluid balance at admission and ICU discharge, need for mechanical ventilation and vital status at hospital discharge. Trial personnel performing data extraction did so using a standardized data extraction form and were unaware of group assignment.
The primary outcome measure is a major adverse kidney event (MAKE-30) [12]. The MAKE-30 is a composite outcome of an elevation of the creatinine level to ≥ 200% of baseline, need for renal replacement therapy or death. All outcomes were collected at the time of hospital discharge, death or within 30 days of admission to the ICU. Baseline creatinine values were determined using most recent creatinine values obtained during the year before hospitalization. If no baseline creatinine value was available and patients had no documented history of renal disease, the baseline creatinine level was estimated to be 1 mg/dL. We compared the rate of MAKE-30 events in patients with and without venous flow abnormalities in the hepatic, portal and intra-renal veins.
Statistical analyses
Power analysis
The estimated rate of MAKE-30 in our cohort is 30%. From a previous internal review of patients undergoing FREE exams, the rate of MAKE-30 events in patients with venous waveform abnormalities is approximately 50%. Therefore, a cohort of 100 patients would identify a 20% absolute difference in MAKE-30 events between patients with and without venous waveform changes with an 80% power.
Regression analysis
Linear and logistic regression analyses were performed to assess the independent association of the HVD, PVD and RVD on the MAKE30. Variables were initially selected for logistic regression by comparing receiver operating characteristic curves (C-statistic) with the Delong test [13]. Secondly, stepwise selection of variables using criterion-based procedures was done to establish the best set of predictors in each regression model. The Bayes information criterion (BIC) was used to select the final model since this criterion usually results in more parsimonious models [14, 15]. For the logistic and linear regression models, multiple variables were tested. Multiple effect modification terms (e.g., age and SOFA) were created to detect statistical interaction.
All regression models were assessed with regression diagnostics. Logistic regression models were assessed for specification error using the “linktest” function in Stata, followed by goodness-of-fit testing using the Hosmer–Lemeshow goodness-of-fit statistic (computed as the Pearson chi-square from a contingency table of observed frequencies and expected frequencies), testing for multicollinearity and Pearson and deviance residual plot analyses to detect influential observations. Linear regression models were evaluated using the same tests for logistic regression. Additionally, outliers, leverage and influence were assessed using studentized residual plots, Cook’s D test, DFITS and DBETA, respectively. All tests were two-tailed, and a P value of < 0.05 was considered statistically significant.
Descriptive statistics were calculated to describe the cohort. Data are presented as means and standard deviations (SD) for normally distributed variables and medians and interquartile ranges (IQR) for non-normally distributed variables. Continuous measurements were compared with the student’s t test or Wilcoxon rank sum test as indicated.