Utility of Central Venous Oxygen Saturation Gradient in Predicting Mortality in Dialysis with Catheter Access

Introduction

Fluid and hemodynamic management is a modifiable cardiovascular risk factor in patients on maintenance dialysis. Fluid overload is associated with inflammation, left ventricular hypertrophy, and mortality [1–3]. To avoid fluid overload, a high ultrafiltration rate (UFR) is sometimes required. UFR exceeding the plasma refilling rate from interstitium into vascular space results in blood volume decline and intradialytic hypotension (IDH) [4], which is known to be associated with vascular thrombosis, inadequacy of dialysis, and ischemia of vital organ [5–7]. Overzealous ultrafiltration (UF) is associated with impaired myocardial blood flow and cardiac dysfunction [8]. It is imperative to maintain normohydration status for long-term survival.

Central venous oxygen saturation (ScvO2), a proxy measure of upper-body blood flow, is known to be a marker of mortality in cardiovascular surgery, cardiogenic shock, septic shock, and other critical settings [9]. It has been demonstrated low ScvO2 levels and high ScvO2 variability are associated with poor prognosis [10,11], and might be more specific to pathophysiologic events preceding IDH than blood volume change [12]. Intradialytic oxygen supplement and adequate UF volume can increase ScvO2 [13]. ScvO2 is well-correlated with arterial oxygen saturation (SaO2) and tissue oxygenation [14]. Given that hypoxemia is associated with cardiac arrhythmia, cardiovascular event, and mortality, ScvO2 might be a potential marker for prediction of adverse outcomes [14].

Fluid removal by adequate UF can achieve better SaO2, whereas overzealous UF results in high ScvO2 variability and IDH [10], and it also causes myocardial stunning and end-organ damage [15]. Several blood volume monitors, including biomarker, clinical, and instrumental assessments, have been introduced to avoid overzealous UF [7], but some of them are inaccurate or do not occur in real time [7]. Intradialytic monitoring of blood volume, including relative blood volume (RBV) and absolute blood volume (ABV), has been measured by hematocrit-based and UF jump-based methods to prevent IDH [15,16]. To counterbalance the hypotension, plasma refilling, sympathetic nerve system activation, and vasoactive hormone have been implemented [4]. The counter-regulatory mechanism is mainly activated according to ABV reduction [17]. ABV change was shown to be better correlated with UFR than is RBV change [18]. It is not reduced interstitial volume but intravascular underfilling that compromises adequate circulation [18]. Since the balance between UFR and capillary refilling is the determinant of IDH, ABV change might be better than RBV change as a real-time fluid monitor. As fluid depletion and overload both resulted in increased mortality, real-time intradialytic blood volume monitor is crucial. In the present study, we hypothesized that the intradialytic slope of ScvO2, affected by blood volume change, is a predictor of mortality. Therefore, this retrospective study used hospital record data of 144 dialysis patients with central venous catheter access and aimed to evaluate the ScvO2 gradient, blood volume, and patient mortality

Material and Methods

ETHICS STATEMENT:

This study was conducted in adherence with the Declaration of Helsinki (2000) of the World Medical Association, and the study protocol was reviewed and approved by the institutional review board of Taoyuan General Hospital (TYGH.112057). Written informed consent was obtained from each patient.

PARTICIPANTS AND DATA SOURCE:

The cohort study enrolled 488 adult and incident hemodialysis subjects in Taoyuan General Hospital and Xinwu Branch since July, 2022. Dialysis was commenced in those with creatinine >0.53 mmol/L in combination with severe hyperkalemia, pulmonary edema, or severe metabolic acidosis. Inclusion criteria were: (1) age ≥20 years and (2) dialysis received for ≥3 months. We retrospectively reviewed medical records of eligible patients, including laboratory data, demographic data, underlying disease, dialysis information, and outcomes. All records about diseases were coded using the International Classification of Diseases, 10th Version, Clinical Modification (ICD-10-CM) in the electronic medical records of Taoyuan General Hospital. Care codes, exam codes, and procedure codes were also obtained. Long-term hemodialysis was confirmed by ICD-10-CM and RIGISTRY FOR CATASTROPHIC ILLNESS PATIENTS (code: 58001C, 58019C, 58020C, 58021C, 58022C, 58023C, 58024C, 58025C, 58027C, 58029C and N18.6). We also excluded those with dialysis time <180 min (n=15), on dialysis less than 3 times a week (n=24), cardiothoracic ratio (CTR) <0.5 (n=80), follow-up less than 3 months (n=16), and using arterio-venous fistula or graft as access (n=204). Those with initial ScvO2 >85% or <25% were regarded as laboratory error and were excluded (n=5). The baseline period was defined as 90 days after enrollment. The end of the baseline period was the initiation of cohort time and were followed up until the date of death, kidney transplantation, or administrative censor on July 31th, 2024. Consequently, a total 144 eligible patients were included for analysis (Figure 1).

OUTCOME ASCERTAINMENT:

Demographic data, comorbidities, biochemical data, and date of death were recorded according to electronic medical records. Patients with last record on AMBULATORY CARE EXPENDITURES BY VISITS before the end of the study were diagnosed as death. Participants who were lost to follow-up without death confirmation were censored at the last date of follow-up. Participants who were being followed for more than 2 years were censored and participants who survived over the administration censor date were censored on July 31^th, 2024.

BLOOD VOLUME MONITOR PROTOCOL, SCVO2, AND BLOOD PRESSURE MEASUREMENT:

The real-time blood volume status was assessed by Crit-Line IV monitor (CLM IV) with blood chamber, which measures hematocrit non-invasively via photo-optical technology (Fresenius Medical Care AG & Co, Bad Homburg, Germany). Real-time hematocrit and ScvO2 were measured every 20 seconds during dialysis session and recorded every 20 minutes by well-trained nursing staff. Maximal ScvO2, minimal ScvO2, mean ScvO2, and ScvO2 at start and end after sessions were also recorded. Standard deviations (SD) of ScvO2 and intradialytic slope of ScvO2, calculated by linear regression during dialysis sessions, were also recorded using software after each session. Intradialytic ScvO2 trend was classified into either decreasing (mean slope <0) (negative trend) or increasing (mean slope >0) (positive trend) based on mean slope of ScvO2. To achieve better fluid management, we adjusted UFR according to the following protocol: (1) set UFR according to inter-dialysis weight gain divided by dialysis time, (2) keep RBV slope within 3–6% per hour, (3) up-titrate UFR 0.5 to 1 ml/Kg/hr while RBV slope is less than 3% per hour, (4) down-titrate UFR 0.5 to 1 ml/Kg/hr while RBV slope is more than 6% per hour, (5) cold dialysis, stop UFR, and oxygen supply once IDH occurred, (6) restart with half the UFR of the last time as IDH was corrected. Blood pressure is automatically measured every 30 minutes oscillometrically. To avoid hypoxemia, oxygen supply with nasal cannula was used routinely. IDH was defined as systolic blood pressure (SBP) less than 90 mmHg. Dialysis information, such as UFR, UF volume, dialysis treatment time, pre-dialysis SBP, frequency of IDH, and RBV and ABV slope, were also recorded. ABV was estimated from RBV around an abrupt change of UF rate [18]. We assumed that UF and capillary refilling are the only factors that change the blood volume during hemodialysis.

Where Qr: capillary refilling rate, Qu: UFR; BV, blood volume

Where a is a very short time.

We introduced a parameter, UF jump, as P

As Qr is continuous at t0; thus,

If we have real-time, high frequency hemoglobin data, obtained from CLM IV and UF jump, we can obtain capillary refilling and ABV. As Hbmass (t0) was solved, we can calculate ABV, according to equation (A).

RBV was easier to estimate as followed:

COMORBIDITIES ASCERTAINMENT AND LABORATORY DATA MEASUREMENT:

Laboratory measures were performed by the Department of Laboratory Medicine at Taoyuan General Hospital using standardized and automated methods. Hemoglobin was obtained by direct current detection method on microscopy (XN9000, Sysmex, Japan). Serum creatinine was measured by Kinetic Jaffe’ method (reference range: 0.6–1.3 mg/dL for men and 0.5–1.1 mg/dL for women) (ADVIA 1800, Siemens Healthcare GmbH, Erlangen, Germany). Serum albumin was determined using the bromocresol green assay, whereas total protein was measured by biuret test (ADVIA 1800, Siemens Healthcare GmbH, Erlangen, Germany). Ferritin and intact parathyroid hormone (iPTH) were measured by chemiluminescence method (ADVIA 1800, Siemens Healthcare GmbH, Erlangen, Germany). Diabetes mellitus (DM) was diagnosed according to ICD-10-CM codes: E08–E14; whereas hypertension (HTN) was encoded as ICD-10-CM code: I10–I16. Congestive heart failure (CHF) was diagnosed according to ICD-10-CM code: I09.81, I40–I43, I50, I11, I13.

STATISTICAL ANALYSIS:

Continuous variables with a normal distribution were summarized as mean±SD unless otherwise stated. Variables with a non-normal distribution are expressed as a median (interquartile range (IQR)). Normality was assessed by the Kolmogorov-Smirnov test. Pearson’s chi-square, one-way analysis of variance (ANOVA), or Mann-Whitney U test was used to determine the differences in the demographic data, the laboratory variables, and clinical characteristics between survivors and non-survivors. Univariate and multivariate logistic regression were used to assess the association between parameters and mortality. To avoid missing possible predictors of death, parameters with P<0.02 were enrolled in the multivariate logistic model. To develop a flexible approach to modelling the nonlinear and asymmetric relationship between dialysis mortality and ScvO2 as well as blood volume, we implemented the multivariable fractional polynomials (MFP) method and maintained blood volume and ScvO2 as a continuous variable [19]. Instead of imposing a specific functional form, the MFP method allows the data to determine the best-fitting functional form for blood volume, ScvO2, and other adjustment variables. This method can capture the relationship between mortality and blood volume and ScvO2 in a compact, parsimonious model. The MFP performance and results were then compared against linear-quadratic models (Supplementary data).

Interactions (MFPI) between adjustment variables were tested to address the possibility of differences in the ScvO2-mortality curve. To integrate the effect of ABV change and mean ScvO2, we assessed the association between slope of mean ScvO2 and mortality. Intradialytic slope of ScvO2 was categorized into ascending (slope >0) or descending (slope <0) slope. Cumulative survival curves for mortality with different trend of ScvO2 were performed according to the log-rank test. Calculations and comparisons in hazard ratios for mortality were conducted using Cox proportional hazards model. The fit of the Cox proportional hazard model was tested by Schoenfeld residuals test. All statistical analyses and plots were performed using STATA 16.0 (StataCorp. College Station, Texas, USA). A two-tailed P value of less than 0.05 was considered statistically significant.

Results

DEMOGRAPHIC CHARACTERISTICS AND CLINICAL OUTCOMES:

The cohort of the 144 subjects were followed up for a median of 14.8 months, ranging from 9.0 to 23.0 months. The mean age for the study population was 66.7 years, whereas the dialysis vintage was 4.8 years. The incidence of all-cause mortality was 16.7 per 1000 patient-months. The average dialysis treatments with CLM during baseline period were 6.9±0.3 times. The prevalences of DM, HTN, and CHF were 52.8%, 65.3%, and 20.8%, respectively. The median hourly RBV and ABV slope were −3.1% and −3.9%, whereas the ABV change at end of dialysis was −16.0% (−12.0% to −18.0%).

COMPARISONS BETWEEN SURVIVORS AND NON-SURVIVORS:

Table 1 shows comparisons between survivors and non-survivors. Non-survivors were older (71.2 y vs 65.5 y) and had a higher proportion of IDH (31% vs 12%), as well as lower creatinine (0.74 mmol/L vs 0.85 mmol/L), lower median treatment time per session (3.5 hr vs 3.9 h), lower mean ScvO2 (64.5% vs 68.0%), and descending slope of mean ScvO2 (−0.88% per h vs 0.69% per h). There were no differences among the 2 groups in BMI, length of follow-up, dialysis vintage, levels of albumin, total protein, hemoglobin, iPTH, and ferritin, as well as dialysis information, such as UF volume, pre-dialysis weight, dialysis adequacy(Kt/Vurea), ABV slope, RBV slope, other ScvO2 measurements, the proportions of sex and comorbidities (Table 1).

MODEL FITTING:

The best-fitting model for ABV and mortality was identified by MFPI including ABV and squared ABV. The main adjustment model also included age, albumin, creatinine, total protein, mean ScvO2, and CV of ScvO2. The transformed model (Deviance difference=146.15) significantly improved model fit relative to the untransformed model (Deviance difference=100.70) and linear-quadratic model (Deviance difference=123.66, P<0.05). After finding the best fit for the main model, the ScvO2-ABV interaction were also identified (Deviance difference=9.77, P=0.007). After including the above interaction terms, the final model remained fit the untransformed model (Deviance difference=123.34, P<0.05). Table 2 shows the logistic regression for the final model including fractionally transformed ABV and interaction term.

SCVO2-MORTALITY CURVE AND ABV CHANGE-MORTALITY CURVE:

The fitted curve showed the association between ABV change and probability of 2-year mortality (Figure 2). In those below mean ScvO2, increases in ABV reduction were associated with increased mortality, whereas in those above mean ScvO2, the increases in ABV reduction were associated with decreased mortality. The wide confidence interval at the left tail of the high ScvO2 group was due to the low proportion of negative ultrafiltration. The association between mean ScvO2 and 2-year mortality is shown in Figure 3. The correlation between mean ScvO2 and mortality was weak (r=−0.05), whereas the association between ABV change and mean ScvO2 were a reverse U curve (Figure 4). We used the intradialytic slope of mean ScvO2 to integrate the effect of ABV change and mean ScvO2. The intradialytic slope of ScvO2 remained significant predictors of mortality after multivariate adjustment (adjusted odds ratio and 95% CI=0.421(0.226–0.783), P<0.05), and the intradialytic slope of ScvO2 was also negatively associated with mortality (r=−0.11) (Figure 5). We also categorized our subjects into decreasing slope or increasing slope according to the intradialytic trend of ScvO2. In the crude Cox model, those with a descending slope of ScvO2 had significantly higher risks of mortality than those with an ascending slope (HR [95% CI]=3.43 [1.40–8.44]). After adjusting demographic data, comorbidities, and laboratory data other than ABV change, decreasing ScvO2 remained a predictor of mortality (HR [95% CI]=3.98 [1.22–13.03], P<0.05). The cumulative survival curve also demonstrated that patients with increasing ScvO2 also had better prognosis than those with decreasing ScvO2 (log-rank chi-sq=8.43, P=0.004) (Figure 6).

Discussion

LIMITATIONS OF THE STUDY:

First, it was a retrospective observational study within a single medical center; thus, causal inference is impossible. Secondly, the sample size was limited. Thirdly, despite multivariate adjustment, residual confounding factors cannot be excluded. Those with UF intolerance, such as the elderly, CHF, and high comorbidity, might have distorted the association between ScvO2 and death [35]. We lacked data on residual renal function, which is associated with UF volume and blood volume change [36]. Furthermore, objective indicators of fluid status, especially post-dialytic bioimpedance analysis, were not available, but they can help clarify the relationship between fluid status and ABV change. Finally, ABV measurement might be affected by technical- and treatment-related intervention, such as wrong placement of the measurement chamber and intravenous bolus of fluid.

Conclusions

In summary, a negative trend of intradialytic ScvO2 was associated with mortality. Mean ScvO2 and ABV change were independently associated with mortality. The association between mean ScvO2 and ABV change was a reverse U curve. The integrated effect of mean ScvO2 and ABV change, reflected by intradialytic slope of ScvO2, was a predictor of mortality. This study was limited by study design and sample size. Objective fluid indicators are needed to confirm the association and effect modification between ScvO2 and mortality.