Evaluation of Clinical and Machine Data of Critically Ill Adult COVID Patients with Acute Kidney Injury Exposed to Enhanced Hemoadsorption during CRRT

Acute kidney injury (AKI) affects up to two-thirds of critically ill adult COVID patients [1‒3]. COVID patients with AKI have higher mortality rates and increased requirements for intensive care unit (ICU) care and mechanical ventilation [4‒6]. Among critically ill adult COVID patients, 5–23% require continuous renal replacement therapy (CRRT) [7]. Due to the high mortality of critically ill COVID patients with increased inflammatory markers, newer treatments have been sought out to attempt to ameliorate the inflammatory milieu in COVID patients with multi-organ failure [8].

In this context, the Food and Drug Administration (FDA) authorized the emergency use of the oXiris^® filter to reduce pro-inflammatory cytokine levels in critically ill adult COVID patients presenting with respiratory failure or severe/life-threatening disease [9, 10]. The oXiris filter is a hollow fiber filter with an internal surface permanently grafted with heparin which allows two procedures to be carried out simultaneously: CRRT and the clearance of endotoxins and inflammatory mediators [10]. The oXiris filter unselectively removes inflammatory mediators and endotoxins by adsorption due to means of ionic interactions at the membrane surface [9].

In COVID patients, the use of oXiris has been associated with a decrease in inflammatory cytokines (i.e., IL-6 and TNF-α) [11‒14], lower Sequential Organ Failure Assessment (SOFA) scores, and improvement in respiratory status [13, 14]. However, these studies have seldom used contemporary control groups. Therefore, we evaluated clinical and CRRT machine data of critically ill adult COVID patients with AKI exposed versus not exposed to enhanced hemoadsorption with oXiris during CRRT. We hypothesized that patients exposed to oXiris will have better clinical outcomes than those exposed to a standard CRRT filter. We further hypothesized that CRRT machine parameters will differentiate oXiris (vs. standard filter) properties of enhanced hemoadsorption but will also exhibit heterogeneous patient-level trajectories that can assist with the evaluation of patient selection and heterogeneity of treatment effect in prospective studies.

We conducted a retrospective cohort study including critically ill adult (≥18 years) COVID patients with AKI that required CRRT at the University of Kentucky ICUs between March 2020 and April 2021. Kidney transplant recipients and patients with a prior diagnosis of end-stage kidney disease were excluded from the study. We evaluated clinical and CRRT machine data of patients exposed versus not exposed to oXiris during CRRT. The study was approved by the University of Kentucky Institutional Review Board (IRB # 43159) and patient informed consent was waived given the retrospective and observational nature of the study.

Patients received treatment according to COVID-19 management guidelines used at the time of the index admission. Patient selection for CRRT initiation was based on standard practice for supporting critically ill patients with AKI in the ICU. The choice of using the oXiris filter was based on the clinical judgment of the intensivists and nephrologists in charge of the patient, according to FDA indications of use [9]. The oXiris exposure was defined as receiving CRRT with the oXiris filter for more than 12 cumulative hours and more than one-third of the time within the first 72 h of CRRT initiation (Fig. 1). The non-oXiris group only received HF1400^® as the default CRRT filter for more than 12 cumulative hours within the first 72 h of CRRT initiation. The filters were used with Prismaflex or PrisMax CRRT devices as per FDA requirements, and the hemofilters were replaced every 72 h or after a circuit failure as per standard operational procedures. The CRRT modality used was continuous veno-venous hemodiafiltration with regional citrate (Anticoagulant Citrate Dextrose Solution, Solution A) as the most frequent form of anticoagulation [15].

We obtained patient demographics, comorbidities, laboratory data, acute illness parameters, CRRT indications, daily CRRT parameters (type of filter, type of anticoagulation, etc.), and COVID-specific treatments from electronic health records. The data were extracted manually and stored in a REDCap survey. Diagnosis of SARS-CoV-2 infection was defined as a positive real-time reverse transcriptase polymerase chain reaction result from nasal/oral swabs and AKI was defined by KDIGO criteria.

The CRRT machine parameters – filter pressure, return pressure, effluent pressure, filter life, and filter-related and clotting alarms – were obtained systematically through automated examination of the CRRT machine data cards.

Transmembrane pressure (TMP) and pressure drop were obtained as follows:

• 1.

  TMP: [(Filter pressure + Return pressure)/2] – Effluent pressure

• 2.

  Pressure Drop: Filter pressure – Return pressure

Filter-related and clotting alarm definitions are presented in online supplementary Item S1 (for all online suppl. material, see https://doi.org/10.1159/000535773).

These outcomes included TMP (mm Hg) and filter pressure drop (mm Hg) which are commonly monitored in CRRT practice to evaluate clotting (TMP and filter pressure drop simultaneously trending up) or clogging (TMP progressively trending up) events. Additionally, filter lifetime (hours) and the frequency of filter/clotting alarms were evaluated. To exclude outliers, only data from filters that lasted ≥6 h were considered, except for the filter “survival” analysis which included all filters that lasted >1 h.

These outcomes included hospital mortality, dialysis dependence at discharge, and the composite of hospital mortality or dialysis dependence at discharge. Ventilator- and ICU-free days were analyzed at 28 days after ICU admission in the whole cohort (i.e., 28 days minus days on ventilator/ICU, assigning zero to patients that died), and total ventilator and ICU days were analyzed separately in survivors only. In addition, we evaluated evidence of bleeding (gastrointestinal or intracranial) and hypercoagulable state (new diagnosis of stroke, deep vein thrombosis, or pulmonary embolism). Finally, adverse filter reactions clinically attributed to the oXiris filter (heparin bonded AN69 filter) were also evaluated.

Medians/means and interquartile ranges/standard deviations were used for continuous variables and frequencies and percentages for categorical variables. Exposure groups were compared using t test or Mann-Whitney U test as appropriate for continuous variables and χ² test for categorical variables.

When evaluating CRRT parameters, continuous data <1st and >99th percentiles were excluded from analyses, except for the analysis of the frequency of alarms per filter. Trajectory plots of TMP and filter pressure drop were developed using the mean values (95% confidence intervals) of each hour. The trajectories were evaluated individually by autoregressive integrated moving average (ARIMA), and the ARIMA model parameters were compared by F test. Filter survival probabilities were compared with Kaplan-Meier curves and by log-rank test. We represented the frequency of alarms per filter (filter-related and clotting alarms) with bar charts. Every bar represented the average number of alarms per filter within a 6 h duration. Comparisons of alarms between the two groups were made with Mann-Whitney U test.

To evaluate clinical outcomes, inverse probability treatment weighting (IPTW) was used to balance potential confounders in the exposed versus nonexposed groups [16]. The steps were as follows: first, the propensity scores were estimated by fitting a logistic regression model to predict the individual’s probability of using the oXiris filter with selected clinical variables as the model covariates. These variables included age, gender, race, body mass index, CRRT indications, use of pressors/inotropes, use of extracorporeal membrane oxygenation, use of mechanical ventilation, and SOFA score at ICU admission. All acute illness parameters used for IPTW predated and were close to CRRT initiation. Missing continuous data were imputed with median values. Second, the weights for individual patients were calculated as 1/propensity score for the exposed group and 1/(1-propensity score) for the unexposed group. Then the standardized mean differences (SMDs) between the two groups for all model covariates were assessed before and after weighting. As illustrated in online supplementary Figure S1, all covariates had an absolute SMD <0.1 except for age, which had an absolute SMD = 0.11 after the adjustment, indicating a negligible imbalance between groups. Thus, only age was further used for adjusting the subsequent weighted regression models. Linear regression was used for continuous outcomes and logistic regression for binary outcomes. Statistical analyses were performed with R Programming, and p values below 0.05 were considered statistically significant.

A total of 14,043 h of CRRT corresponding to 85 critically ill adult patients were analyzed. Among these, 2,736 h corresponded to oXiris exposure (n = 25 patients) and 11,307 h to a standard CRRT filter (n = 60 patients) (Fig. 2). The overall cohort mean age was 59.3 (±11.5) years, 55 (64.7%) were male, and 70 (82.4%) were white. The demographic and clinical characteristics of patients exposed versus nonexposed to oXiris are presented in Table 1. The main indication for CRRT initiation was fluid overload in both the oXiris (76.0%) and non-oXiris (61.7%) groups. Patients in the oXiris group had more sepsis (92.0% vs. 66.7%, p = 0.015) and received more IL-6 inhibitors (24.0% vs. 5.0%, p = 0.009) than those in the standard filter group.

TMP and filter pressure drop trajectories were statistically significantly different among filters of patients exposed versus nonexposed to oXiris. Specifically, TMP increased rapidly (after the first 12 h) and to higher levels in filters of patients exposed to oXiris versus the standard CRRT filter. Filter pressure drop also increased to higher levels but after the first 24 h in filters of patients exposed to oXiris versus the standard CRRT filter (Fig. 3). Notably, there was considerable heterogeneity in the TMP trajectories of filters from patients exposed to oXiris (online suppl. Fig. S2). Despite these observations, filter lifespan was similar among filters of patients exposed to oXiris versus the standard CRRT filter (median of 36.3 [IQR: 13.7–64.2] vs. 33.1 [15.9–66.6] hours, respectively, p = 0.913, KM log-rank p = 0.926) (Fig. 4). Similarly, filter- and clotting-related alarm frequencies were not different among oXiris versus standard CRRT filters (online suppl. Fig. S3). No filter-attributable adverse reactions were reported.

In the whole cohort, hospital mortality (72.0% vs. 63.3%, p = 0.443) and dialysis dependence at discharge (36.0% vs. 50.0%, p = 0.238) were not different among patients exposed versus nonexposed to oXiris (Table 2). When the analysis was restricted to hospital survivors, patients exposed versus nonexposed to oXiris had fewer days in the ICU (median of 19 [11–28.5] vs. 34.5 [15.8–48.8] days, respectively, p = 0.024) (Table 2).

In adjusted models, oXiris versus non-oXiris exposure was not associated with the composite of hospital mortality and dialysis dependence at discharge (OR 2.13, 95% CI: 0.98–4.82, p = 0.06) or its individual components (Table 3). Similarly, oXiris versus non-oXiris exposure was not associated with more ventilator- or ICU-free days at 28 days in the whole cohort. However, when evaluating the group of survivors only, oXiris versus non-oXiris exposure was associated with fewer ventilator (β = −15.02, 95% CI: −29.23 to −0.82, p = 0.04) and ICU (β = −14.74, 95% CI: −28.54 to −0.95, p = 0.04) days (Table 3).

We evaluated critically ill adult COVID patients with AKI requiring CRRT that were treated with versus without the oXiris filter. Our study expands the prior literature by the use of specific oXiris exposure versus non-exposure definitions through accessing CRRT machine data to determine the timing and duration of exposures. We accessed systematically collected and validated CRRT machine data, and examined CRRT performance parameters relevant to the use of oXiris which have not been previously reported in a comprehensive manner. We observed that oXiris filters exhibited rapid TMP elevation and overall higher levels when compared to the standard CRRT filter used in this study (i.e., HF1400). Despite these differences, filter life and filter/clotting alarm profiles were comparable between both filters. Importantly, the use of oXiris was not associated with improvement in clinical outcomes such as hospital mortality or dialysis dependence at discharge.

The observation of rapid elevation (starting at 12 h) and overall higher levels of TMP in oXiris versus HF1400 filters could be related to the enhanced hemoadsorption inherent to the oXiris filter characteristics. TMP is determined by membrane pore permeability and pressure gradients across the filter; therefore, a raising TMP represents membrane pores that are clogging and losing permeability [17]. We hypothesized that the higher non-selective solute absorption of the oXiris (vs. HF1400) filter may be represented in the described TMP trends [11‒14]. There are no prior studies that have evaluated TMP trajectories in COVID patients treated with oXiris, but a randomized controlled trial reported that TMPs were higher when comparing oXiris versus M150 filters at 12 h of therapy in a non-COVID study population [18]. However, the relationship between TMP and oXiris hemoadsorption properties needs to be further studied in dedicated experiments. One should note that the increment in TMP was very heterogeneous across patients exposed to oXiris, which highlights the role of using dynamic CRRT data, specifically filter-related, for patient sub-phenotyping and evaluation of heterogeneity of treatment effect in future prospective studies (online suppl. Fig. S2). Similarly, filter pressure drop also increased to overall higher levels with oXiris when compared to HF1400 filters, but typically after the first 24 h of filter use (Fig. 3). Collectively, these results support further evaluation of CRRT machine data when using CRRT with or without adjuvant novel extracorporeal blood purification technologies.

Another important point of consideration from our study is the observation that after ∼36 h, both TMP and filter pressure drop remained significantly higher when using oXiris (vs. HF1400), which could also suggest loss of filter efficiency and higher propensity to filter clotting, although we did not find a significant difference in filter life between the two filter types. The latter observation could be due to the effective use of regional or systemic anticoagulation, which may have prevented clotting events [10]. While the oXiris heparin-grafted filter has a local anticoagulant, other components of the extracorporeal circuit still require anticoagulation to prevent clotting [19]. For example, the use of oXiris versus M150 has shown shorter filter life when using anticoagulation-free protocols [18]. In our study, the use of any form of regional (citrate) or systemic anticoagulation was 95.3%. Another prospective study evaluating oXiris in critically ill COVID patients showed 18.9% (7 out of 37) treatments with premature clotting in the first 24 h. The latter occurred in half of the treatments without anticoagulation but only in 18.6% of the treatments using regional citrate [14].

Previous studies have shown that the use of oXiris in COVID patients is associated with a decrease in inflammatory cytokines such as IL-6, IL-8, and TNF-α [11‒14], as well as lower SOFA scores and improved respiratory status [13, 14]. Furthermore, a study of critically ill COVID patients with AKI found a decrease in observed mortality compared with expected ICU mortality (56% vs. 65%). However, the latter study lacks a control group and the analysis was not adjusted by acute illness parameters [14]. On the other hand, a randomized control study of COVID patients without AKI found no significant elimination of inflammatory factors [20]. In our study, we did not observe an association between oXiris versus non-oXiris exposure and hospital mortality or dialysis dependence at discharge, as well as with an increment in ventilator- or ICU-free days. Nonetheless, we did notice a statistically significant reduction in total ventilator and ICU days in survivors, although indication bias should be considered when interpreting these results. Therefore, interventional studies are needed to surpass inherent limitations of selection and indication biases of observational studies. Further, one should recognize that treatment strategies focused solely on improving physiological indicators or biomarkers may not necessarily lead to better clinical outcomes.

In the absence of a clear improvement in clinical outcomes, the application of oXiris should be carefully considered due to outstanding questions to support its more standardized use. For example, some salient questions include: What type of non-selective solutes are being absorbed by the oXiris filter? What is the time span of acceptable efficiency of the filter? How to evaluate heterogeneity of treatment effect? How to enrich patient selection? What are clinically useful time windows for oXiris use? While more robust evidence evolves, our study contributes to ongoing observations that could assist in the design of prospective and interventional studies.

There are limitations of this study to acknowledge. First, this is an observational study that is subject to indication bias. It is possible that oXiris was assigned to more severely ill patients by the treating clinicians, and that the study could have underestimated its effect. Although IPTW analysis was performed to minimize this indication bias, the possibility of unmeasured confounding remains. Second, specific cytokine levels and endotoxins were not measured in the study patients, and therefore it was not possible to assess the effect of oXiris on decreasing inflammatory mediators and its interaction with clinical outcomes. Third, the study collected data during a period of time in which COVID therapies were evolving and therefore variably applied to the study patients. Nonetheless, we did not find statistically significant differences in COVID therapy exposures between the study groups. Finally, since this is a single-center study of relatively small sample size, the generalizability of these observations may be limited.

In critically ill adult COVID patients with AKI requiring CRRT, oXiris filters exhibited a rapid and heterogeneous TMP elevation with overall higher levels of TMP when compared to standard CRRT filters. Similarly, filter pressure drop also increased to overall higher levels with oXiris when compared to standard CRRT filters, but typically proceeding the elevation in TMP. Despite these differences, filter life and filter/clotting alarm profiles were comparable between oXiris and standard CRRT filters. The use of oXiris was not associated with improvement in clinical outcomes such as hospital mortality or dialysis dependence at discharge. The correlation of cytokine/endotoxin removal by oXiris with CRRT machine TMP trajectories should be a subject of future study because it could represent a feasible way to evaluate heterogeneity of treatment effect and improve patient selection and prospective study design.

J.A.N. is supported by grants from NIDDK (R01DK128208, U01DK12998, and P30 DK079337).

The University of Kentucky Institutional Review Board approved this study (IRB # 43159) and waived the need for written informed consent given the retrospective nature of data collection. The study was conducted in accordance with the Helsinki Declaration.

J.A.N. has received consulting honoraria from Baxter, Leadiant Biosciences, Outset Medical, and Vifor Pharma. A.J.T. has received consulting honoraria from Baxter.

No funding was received for the development or implementation of the study.

Research idea and study design: V.O.-S. and J.A.N.; data acquisition: V.O.-S., L.J.L., and S.C.; analysis/interpretation: A.C.-O., V.O.-S., L.J.L., T.T., J.C., A.J.T., and J.A.N.; writing of the first draft manuscript: A.C.-O., L.J.-L., T.T., and J.A.N. All authors contributed intellectual content to the manuscript, reviewed the results, edited the manuscript, and approved the final version of the manuscript.

The data generated or analyzed during this study are not publicly available due to ethical reasons. However, de-identified data could be made available by the corresponding author upon reasonable request.