A Prediction Model of Sucient Filter Lifespan in Anticoagulation-free CRRT Patients

Background: Anticoagulation-free continuous renal replacement therapy (CRRT) was recommended by the current clinical guideline for patients with increased bleeding risk and contraindications of citrate and resulted in heterogeneous lter lifespan. There was no prediction model to identify the patients would have sucient lter lifespan when they have to accept CRRT without the use of any anticoagulation. The purpose of our present study is to develop a clinical prediction model of sucient lter lifespan in anticoagulation-free CRRT patients. Method: Patients who underwent anticoagulation-free CRRT in our center between June 2013 and June 2019 were retrospectively included. The primary outcome was sucient lter lifespan ( ≥ 24 hours). The nal model was established by using multivariable logistic regression analysis. And, the prediction model was validated in an external cohort. Results: A total of 170 patients were included in the development cohort. Sucient lter lifespan were observed in 80 patients. The probability of sucient lter lifespan could be calculated using the following regression formula: P (%) = exp (Z)/1 + exp (Z), where Z = 0.49896-(0.08552*BMI)+(0.44107*T)+(0.03373*MAP)-(0.03389*WBC)+(1.51579* [vasopressor=1])-(0.01132*PLT)+(0.00422*ALP)-(2.66910*pH)-(0.00214*UA)+(0.05992*BUN)+(0.00400*Db)– (0.00014*D-dimer)+(0.02818*APTT). The area under the curve (AUC) of the stepwise model and internal validation model was 0.82 (95%CI [0.76-0.88]) and 0.8 (95%CI [0.74-0.87]), respectively. At the optimal cut-off value of -0.1052, the positive predictive value and the negative predictive value ALP, alkaline phosphatase; APTT, activated partial thromboplastin time; BMI, body mass index; BUN, blood urea nitrogen; CI, condence interval; Db, direct bilirubin; Hb, hemoglobin; Hct, hematocrit; MAP, mean arterial pressure; OR, odds ratio; PLT, platelet count; SOFA, sequential organ failure assessment; T, temperature; Tb, total bilirubin; UA, uric acid; WBC, white blood cell. ROC, receiver operating characteristic curve; SOFA, sequential organ failure assessment; SD, standard deviation; T, temperature; Tb, total bilirubin; TMP, transmembrane pressure; UA, uric acid; UFH, unfractionated heparin; UFR, ultraltration rate; VIF, variance ination factor; WBC, white blood cell.


Background
Renal replacement therapy (RRT) is applied in 12%-15% of intensive care unit (ICU) patients and 75% of the RRT modalities were continuous renal replacement therapy (CRRT) [1]. Anticoagulation is a key intervention to maintain the patency of the extracorporeal circuit [2].
Unfractionated heparin (UFH) is the most widely used anticoagulant for CRRT worldwide [3]. However, critically ill patients are commonly complicated with impaired coagulation or increased bleeding risk [4]. A platelet count of < 50 × 10 9 /L was seen in 12% -15% of the critically ill patients and a prolonged global coagulation time (i.e. prothrombin time [PT] or activated partial thromboplastin time [APTT]) in 14% -28% [5]. The reported bleeding incidence in patients underwent heparin anticoagulated CRRT ranged from 4%-25% [6,7]. Therefore, the increased bleeding risk limits the applicability of heparin in critically ill patients [7]. Regional citrate anticoagulation (RCA) are gaining increasing popularity due to its advantage over heparin in terms of prolonged lter lifespan and reduced bleeding risk [8]. The KDIGO guideline recommended RCA as the preferred anticoagulation strategy in patients without citrate contraindications, including severe liver failure and shock with muscle hypoperfusion [4]. The reported incidences of liver dysfunction and shock with hypoperfusion in ICU patients were 2%-5% [9,10] and 40% [11,12], respectively. Therefore, a signi cant number of ICU patients had the contraindications of both heparin and citrate.
For these patients, CRRT was suggested to be proceed without the use of any anticoagulant [4]. In clinical practice, approximately 33%-50% patients did not receive any anticoagulants during CRRT [13][14][15]. The averaged or median lter lifespan of anticoagulation-free CRRT ranged from 10 [16] to 40 [17] hours, which were associated with signi cant heterogeneity. For anticoagulation-free CRRT, 60% of the lters were replaced because of lter failure before the accomplishment of a treatment regiment (commonly 24 hours) [18]. Several parameters, including platelet (PLT), international normalized ratio (INR), and APTT were reported to be related to the lter lifespan in anticoagulation-free CRRT patients. However, the speci c cut-off points have not been determined for these parameters to indicate the possibility of su cient lter lifespan for anticoagulation-free CRRT [4]. To the best of our knowledge, there was no effective model to predict su cient lter lifespan in anticoagulation-free CRRT patients as well [19]. The development of an effective prediction model for su cient lter lifespan would be helpful for the individual choice of anticoagulation strategy for CRRT patients.
Therefore, the aim of our present study is to develop a clinical prediction model to predict su cient lter lifespan for a treatment regiment (commonly 24 hours) in anticoagulation-free CRRT patients.

Study design
Our present study is a retrospective cohort study and was performed in accordance with the Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis (TRIPOD) statement [20]. This study was performed in accordance with the Declaration of Helsinki and was approved by the ethic committee of our hospital.
The informed consent was waived regarding the retrospective nature of our present study.

Development cohort
Critically ill patients who underwent CRRT between June 2013 and June 2019 in our center (a tertiary teaching hospital with more than 2000 CRRT patients per year) were retrospectively screened. The inclusion criteria included: 1) adult patients (≥ 18 years); and 2) received anticoagulation-free CRRT. Patients were excluded if they ful lled any of the following criteria: 1) received systemic anticoagulation within 24 hours prior to or during CRRT for indications other than CRRT; 2) received other extracorporeal therapies (i.e. Plasmapheresis, plasma exchange, hemoperfusion, or extracorporeal membrane oxygenation [ECMO]) during CRRT; 3) switched to systemic heparin or regional citrate anticoagulation after the start of a anticoagulation-free CRRT regiment; 4) patients underwent CRRT via arteriovenous stula; 5) patients with missing data of important parameters (i.e. liver function, coagulation parameters, or lter lifespan); 6) the rst circuit was replaced within 24 hours due to selective reasons (i.e. imaging procedures, transport, low blood pressure, resuscitation discharge, surgery, or death); or 7) the center venous catheter function was insu cient for the targeted blood ow.

Validation cohort
A separate external validation cohort was retrospectively recruited to assess the generalizability of the prediction model. The patients who underwent CRRT between January 2018 and December 2019 in West China Hospital of Sichuan University were screened according to the inclusion and exclusion criteria employed in the development cohort.
CRRT protocol in the development cohort Continuous veno-venous hemo ltration (CVVH) was the modality routinely adopted during the study period. The machines used for CRRT were the Prisma ex System (Gambro) and DIAPACT (Braun), which were equipped with Multi ow-100 (0.9 m 2 , AN69 membrane) and AV600 (polysulfone, 1.4 m 2 ; Fresenius) hollow-ber lters, respectively. The vascular access was built by inserting a 13.5F double lumen catheter into the femoral vein or jugular vein. Blood ow was maintained at 200 ml/min. Commercially available replacement uid was infused at a rate of 2 L/h with a ratio of pre to post-dilution of 1:1. In cases of large body size (weight > 100 kg), the prescribed dose was set at 20-25 ml/kg/hour according to the recommendations of the 2012 KDIGO guideline [4]. A bicarbonate buffered uid was infused pre-lter separately at an appropriate rate depending on the acid-base status of the patients. The ultra ltration rate (UFR) was adjusted according to the hemodynamic parameters and the goal of treatment.
CRRT protocol in the validation cohort Continuous venovenous hemodia ltration (CVVHDF) was the sole CRRT modality used in the validation cohort.
CVVHDF was performed using the Prisma ex System (Gambro) machine equipped with several types of lters including ST150, M150, and M100. The parameters of CVVHDF were following: blood ow rate 200 ml/min; dialysate ow rate 1 L/hour; and, replacement uid ow rate 1 L/hour with pre-, post-or hybrid dilution models.

Outcomes
Su cient lter lifespan was de ned as lter lifespan ≥ 24 hours. Filter lifespan was de ned as the time interval (hours) between initiation and cessation of an individual circuit. Filter failure was con rmed by 1) transmembrane pressure (TMP) > 300 mmHg, 2) visible clots, or 3) inability to operate the blood pump [24,25].

Factors included in the analysis
Thirty-seven factors (Additional le 1: Table S1) were included in the analysis based on their clinical signi cance and previous reports [26]. Data including demographic characteristics, admission diagnosis, pre-existing disease, illness severity, indications of CRRT, risk of bleeding, contraindications for citrate, laboratory tests, mechanical ventilation (MV), vasoactive agents use, transfusion requirement, and CRRT protocol characteristics were collected from the patient medical records. In order to control confounding bias, only the rst circuit in the rst session was analyzed for patients who received several sessions of CRRT during one admission.

Missing data
The missing data were addressed by mean imputation using SPSS (IBM Corporation) software version 24. All the imputations were performed before the univariable and multivariable analyses.

Statistical analysis
Normal distribution and non-normal distribution continuous variables were presented as mean ± standard deviation (SD) and median (interquartile range [IQR]) and compared using Student's t test and the Wilcoxon test, respectively. For categorical variables, data were presented as counts (percentage), and were compared using Chi-square test or Fisher's exact test. The lter survival probability at different time points were graphically analyzed using Kaplan-Meier survival curves with Log-rank test.
All the candidate factors were included in a multivariable logistic regression model with the continuous variables retained on the original scale. The collinearity test was performed at rst and variables with variance in ation factor (VIF) greater than 10 were eliminated. Thereafter, a stepwise approach was employed for the nal model selection based upon Akaike Information Criterion (AIC) [27] measures of model quality and performance. We also developed a full model and a multivariable fractional polynomial (mfp) model. Nomograms [28] were constructed based on the results of the models. Discrimination and calibration of the models were assessed by the area under the receiver operating characteristic (ROC) curve (AUC) with a 95% con dence interval (CI) and the calibration curve, respectively [29]. Sensitivity, speci city, positive predictive value (PPV), negative predictive value (NPV), positive likelihood ratio (PLR), and negative likelihood ratio (NLR) were calculated at the optimal cut-off values of the models which were identi ed according to the maximum Youden Index. In addition to external validation, the performance of the models were internally validated by using bootstrapping (BS) (1000 times). Statistical differences in the AUCs were compared using the Delong test [30]. A 2-sided P-value < 0.05 was considered statistically signi cant. All statistical analyses were performed by using the R software (The R Foundation; http://www.r-project.org; version 3.4.3).

Patients characteristics
During the study period, a total of 496 adult patients who underwent anticoagulation-free CRRT were identi ed, and 326 patients were excluded. The remaining 170 patients were included in the development cohort. The most common reason for patient exclusion was systemic anticoagulation within 24 hours before CRRT (34%) followed by important data missing (10%, Fig. 1).

Final model selection
The full model and the mfp model did not show better accuracy than the stepwise model. Furthermore, the complicated formula of the mfp model compromised the convenience for application. Therefore, the stepwise model was the most parsimonious model under the premise of guaranteeing discrimination performance and was selected for the nal model. formula. An optimal cut-off value of Z was identi ed as -0.1052. The nal model could discriminate 61 out of the 80 patients with lter lifespan of ≥ 24 hours and, the PPV and the NPV were 0.77 and 0.79, respectively. Furthermore, the median lter lifespan was signi cantly longer in patients with Z > -0.1052 (26 [24-38.25] hours vs.15.5 [9.75-21.5] hours, P < 0.001, Fig. 2B).

External validation
After the screening, 44 cases were included in the external validation cohort (Additional le 7: Figure S4). The baseline characteristics of these patients are showed in Additional le 8: Table S3. There were no signi cant differences in age, body temperature, male proportion, BMI, MAP, and APACHE II score between development cohort and validation cohort.

Discussion
In our present study, we found out that, in patients with increased bleeding risk who underwent anticoagulation-free CRRT, lower PLT, UA, and D-dimer as well as higher MAP, APTT, ALP, and BUN, and the use of vasopressor were independently related to longer lter lifespan. Our prediction model could effectively discriminate patients with su cient lter lifespan in both the development and validation cohort. A patient with Z > -0.1052 most likely could have su cient lter lifespan for anticoagulation-free CRRT. In addition, the probability of su cient lter lifespan could be calculated by using a nomogram or an Excel calculator.

Risk factors of lter failure
Fealy, N.et al [25] found in a randomized controlled trial (RCT) that longer APTT (hazards ratio [HR] 0.98, P = 0.002) and decreased platelet count (HR 1.19, P = 0.03) were independently associated with a reduced likelihood of circuit clotting. The similar association of APTT [31] and platelet count [32,33] with CRRT lter lifespan were reported in observational studies as well. Zhang Z et al [34] reported that lower pH was signi cantly associated with longer lter lifespan in a multivariable Cox regression model. These evidences demonstrated the reliability of the predictors in our present model. However, the mechanism of lter failure is highly sophisticated [35][36][37][38][39] and could not be solely interpreted or predicted by classical markers of coagulation including PT, APTT, and platelet count [40]. A recent metaanalysis [26] divided non-anticoagulant determinants of lter lifespan into vascular access factors, circuit factors, and patient factors. Therefore, a clinical prediction model, which was commonly based on more comprehensive indicators, could most likely provide more accurate prediction for lter failure and su cient lifespan.
The appropriate lifespan of a single lter has not yet been well de ned [41]. We choose 24 hours as the cut-off value based on the following reasons. First, 24 hours is a therapy span in our routine practice and the treatment target could be reached in majority of patients within 24 hours. Second, an effective treatment time of 20 hours per day was recommended [41], and it could be achieved by using only 1 circuit with a lifespan of ≥ 24 hours, leaving 1-3 hours downtime per day [42]. In addition, most of the centers and studies de ned successful prevention of clotting as no need for circuit change in the rst 24 hours [43].

Relations to the previous studies
To our knowledge, studies developing prediction models for lter lifespan of CRRT are scarce. A prognostic model reported by Fu, X et al [19] included 302 cases and suggested that insu cient blood ow, without anticoagulation, and values of Hct, lactate, and APTT could be used to predict the likelihood of extracorporeal circuit clotting within 24 hours in patients underwent CRRT with all kinds of anticoagulation models. The AUC of this model was 0.79 (95% CI [0.7-0.87]). Despite both of the AUCs of this model and ours were more than 0.75, a threshold that was deemed as useful discrimination [29], our model has several advantages. First, we included only CRRT patients without the use of any anticoagulation, which could offer important information for the choice of anticoagulation strategies, mainly use or no use of anticoagulation. The study by Fu et al. [19] included patients with and without anticoagulation in their model. The use of anticoagulation de nitely played a major role on the prolonged lter lifespan. And, the lter lifespan in patients accepted anticoagulation mainly attributed to the effective anticoagulation, including the anticoagulant types and the su cient dose of anticoagulant. Therefore, the model by Fu et al. [19] could not offer key clue for the use or no use of anticoagulation for clinicians, especially for patients with relative contraindications to anticoagulation. Second, all the included predictors in our model were readily available prior to CRRT initiation, which could be helpful and useful for clinical decision making. Third, in addition to the regression formula, the nomogram and Excel calculator could provide more convenient and accurate prediction. At last, our model has been validated in an external validation cohort, which suggested very good reliability.
Several previous studies also suggested the use of coagulation parameters and bleeding markers to determine the use of anticoagulation-free protocol. However, in the study by Morabito, S. et al [44], 45% of the patients who initially received anticoagulation-free protocol switched to heparin anticoagulation because the lter lifespan was less than 24 hours. In another study by Morabito, S. et al [18], 33 patients switched to RCA-CRRT because of early circuit clotting (< 24 hours) without anticoagulation, and only 40% of the circuits reached a lifespan of more than 24 hours before switch. In our development cohort and validation cohort, of the patients who were predicted to have lter lifespan > 24 hours, 77% and 85% had su cient lifespan during their CRRT treatment. The results of the aforementioned 2 studies [18,44] and our study indicated that the use of a prediction model instead of an assessment based only on coagulation parameters could signi cantly improve the discrimination ability to identify eligible patients for anticoagulant-free CRRT.

Clinical implications
Our model can facilitate the assessment of lter failure risk and the selection of appropriate anticoagulation for CRRT in patients with relative contraindications to anticoagulation. The patient with expected su cient lter lifespan could initially underwent anticoagulation-free CRRT. For those patients with expected insu cient lter lifespan, a less risky anticoagulation strategies could be considered. Future studies with prospective, randomized, and multicenter design are warranted to validate our ndings.

Limitations
First, there were potential confounding factors and biases due to the retrospective nature of our present study. We had strictly prede ned the inclusion and exclusion criteria and enrolled consecutive real-life patients over a 6-year period to reduce the selection bias and confounding factors. Second, the application of the model might be complex as the nal model includes 13 variables. However, most of these predictors were clinically relevant and easily available in routine practice. Furthermore, an Excel calculator was generated and would facilitate the clinical application of our model. prothrombin time activity; PPV, positive predictive value; PLR, positive likelihood ratio; RBC, red blood cell; RCT, randomized controlled trial; RCA, regional citrate anticoagulation; ROC, receiver operating characteristic curve; SOFA, sequential organ failure assessment; SD, standard deviation; T, temperature; Tb, total bilirubin; TMP, transmembrane pressure; UA, uric acid; UFH, unfractionated heparin; UFR, ultra ltration rate; VIF, variance in ation factor; WBC, white blood cell.

Declarations
Ethics approval and consent to participate Approval from the local scienti c and ethics committee of the Xijing Hospital was obtained; they stated that no informed consent of the patient or next of kin was required regarding the retrospective nature of our present study.

Consent for publication
Not applicable Availability of data and material The datasets analyzed during the current study are available with the corresponding author on reasonable request.

Competing interests
The authors declare that there is no con ict of interest.
Funding Figure 1 The participant ow diagram of the development cohort. CRRT, continuous renal replacement therapy; CVVH, continuous venovenous hemo ltration; RCA, regional citrate anticoagulation