Prediction of filter clotting using longitudinal trends of circuit pressure parameters during continuous renal replacement therapy (CRRT): an exploratory study Authors: Yiting Hu, Zhen Hu, Chuan Li, Xi Yang, Yongchun Ge, Dehua Gong Published: 2026 ABSTRACT The fluctuation of circuit pressure parameters may reflect the severity of circuit clotting, but it is unclear yet which one has the highest precision for the prediction of filter clotting. Increased filter pressure drop (FPD) and transmembrane pressure (TMP) are associated with filter coagulation during continuous renal replacement therapy (CRRT). They are affected by blood flow velocity and ultrafiltration rate, which can lead to false coagulation alarms. To eliminate these effects, we proposed two parameters: blood flow resistance (BFR) and transmembrane flow Resistance (TFR), and explored their role in predicting filter clotting. We conducted a retrospective analysis of patients who underwent CRRT at Jinling Hospital between March and June 2025, and experienced at least one filter change due to clotting during CRRT. CRRT modes included continuous veno-venous hemofiltration (CVVH) and hemodialysis (CVVHD). The records of continuously monitored circuit pressure parameters were extracted. The calculated parameters, including filter pressure drop (FPD), blood flow rate-adjusted FPD(blood flow resistance, BFR), transmembrane pressure (TMP), and ultrafiltration rate-adjusted TMP(transmembrane flow resistance, TFR), were further analyzed for the association of their changes with filter clotting. To verify the necessity of adjusting FPD by blood flow rate and TMP by ultrafiltration rate, a prospective study was conducted to investigate the effect of variations in blood flow rate and ultrafiltration rate on 4 pressure parameters by intentionally stepwise modulating flow rate during the start period of the CRRT session. A total of 96 CRRT circuits (76 CVVHD and 20 CVVH) from 51 patients were included. In CVVHD mode, the efficacy of ΔBFR and ΔFPD was significantly higher than that of ΔTFR and ΔTMP (all p < 0.001) for the prediction of 1-hour later clotting. AUC of ΔBFR was bigger than that of ΔFPD, although non-significant. In CVVH mode, the AUC of ΔTFR and ΔTMP was bigger than that of ΔBFR and ΔFPD, although non-significantly (all p > 0.05). A Combination of ΔBFR and ΔTFR by the parallel rule (positive if ΔBFR > 0.075 mmHg/(ml/min) or ΔTFR > 0.115 mmHg/(ml/h)) had yielded a sensitivity of 77.1% and a specificity of 62.9% for the prediction of 1-hour later filter clotting irrespective of CRRT mode. The prospective part of the study showed that FPD was blood flow rate-dependent but not BFR, and TMP was ultrafiltration rate-dependent but not TFR. BFR (Blood flow rate-adjusted FPD) and TFR (ultrafiltration rate-adjusted TMP) are potential predictors of filter clotting, warranting further studies to verify their value.Supplementary InformationThe online version contains supplementary material available at 10.1038/s41598-026-48433-2. Introduction Continuous renal replacement therapy (CRRT) has become a predominant modality of renal support in critically ill patients, valued for its ability to achieve gentle, sustained removal of toxins and excess fluid1. Typically administered for more than 24 h, CRRT is often limited by premature circuit failure due to progressive clotting, despite the recommended maximum lifespan of the extracorporeal circuit being up to 72 h. Thrombus formation can occur throughout the circuit, with the filter being the most common site of coagulation. Early filter clotting is notoriously difficult to detect; alarms are triggered only after clotting has induced a substantial increase in resistance, at which point the filter can no longer be salvaged by flushing or anticoagulation adjustment. Such events lead to unplanned treatment discontinuation, resulting in reduced therapeutic efficacy, blood loss2, increased workload for medical staff, and elevated treatment costs3,4. Proactive intervention before the development of severe coagulation can effectively reduce the incidence of CRRT-related adverse events and mitigate associated losses. Therefore, accurate prediction of filter thrombosis is essential to inform timely adjustments to treatment strategies. Currently, two primary methods are used to assess filter clotting status during clinical treatment. One is a visual assessment of the degree of filter thrombosis, quantified as the percentage of clotted fibers5. This approach provides only a static snapshot of existing clot formation and cannot predict impending filter failure; moreover, the need for frequent rinsing often results in unnecessary treatment interruptions. The other method involves monitoring pressure parameters from the machine. Transmembrane pressure (TMP) and filter pressure drop (FPD) have been shown to correlate well with filter coagulation. In hollow-fiber filters, TMP represents the pressure gradient across the membrane, and a rapid increase in TMP serves as a critical indicator of membrane pore occlusion6,7. FPD, defined as the pressure differential between the filter inlet and outlet, rises rapidly in response to intraluminal thrombosis that physically obstructs the hollow-fiber channels8. However, elevated pressure readings are not specific to filter clotting. Factors such as blood flow velocity and ultrafiltration rate can induce similar changes. As defined by the fundamental relationship: TMP = Quf/Kuf (Kuf: filter’s filtration coefficient)9, MP is influenced by the prescribed ultrafiltration rate(Quf), which is adjusted according to clinical needs during treatment. Similarly, according to the Hagen–Poiseuille equation, FPD is influenced by the blood flow rate (Qb)10. To date, no standardized method exists for accurately predicting filter clotting. Nonetheless, pressure parameters remain commonly employed in clinical practice because they can be readily obtained in real time. However, few studies have accounted for the impact of flow velocity on these parameters, which may lead to misinterpretation in clinical decision‑making. To address this, we derived two flow‑independent resistance parameters: transmembrane flow resistance (TFR = TMP/Quf) and blood flow resistance (BFR = FPD/Qb). Theoretically, TFR and BFR provide a more accurate assessment of filter clogging by eliminating the confounding effects of flow-rate variation. Moreover, both parameters can be derived from real-time data recorded by the machine. Thus, the objective of this study was to evaluate the predictive value of these two novel metrics for filter change, thereby laying the foundation for future integration into automated systems. Patients and methods The study consisted of two parts: a retrospective part and a prospective part. The retrospective part of the study included patients who received CRRT at Jinling Hospital (affiliated with Nanjing University Medical School) between March and June 2025. The inclusion criterion was that patients experienced at least one filter change due to clotting. The clinical baseline data and CRRT-related data were extracted from the hospital’s electronic medical record system. CRRT was provided to each patient using an Aquarius renal replacement system (Baxter) and filters, including UT1100 (Nipro) and AV600S (Fresenius) hemofilters, with continuous veno-venous hemodialysis (CVVHD) and post-dilution continuous veno-venous hemofiltration (CVVH) modes. Vascular access was established via central venous catheterization of the femoral or jugular vein using an 11.5 Fr catheter (Tuoren, China). The initial blood flow rate was set at 140–160 ml/min. During treatment, the circuit was regularly rinsed with normal saline. The severity of filter clotting was evaluated by a specialized CRRT nurse based on a semi-quantitative method: grade I, no signs of clotting; grade II, few dark red fibers; grade III, less than 50% dark red fibers; grade Ⅳ, with > 50% dark red fibers or overall redness11. A filter was replaced when judged Grade Ⅳ clotting. CRRT-related data included real-time records of circuit pressure at different sites, blood flow rate, and ultrafiltration rate, as well as CRRT dose, mode, and filter lifespan. During treatment, the CRRT machine automatically monitors and records the pressure of different sites of the circuit to ensure the safety of extracorporeal circulation. After treatment, pressure parameters were retrieved from the machine. Baseline values were defined as the average of the parameters over a 10-minute period following the initial 30 min of stable CRRT operation with a new filter. Subsequent recordings were made every 2 h. During the 4 h preceding coagulation, recordings were taken every half hour to accurately capture the trend until coagulation occurred. Data recorded during machine recirculation while patients were out for examinations or surgery were also excluded because they did not reflect the patients’ actual treatment status. The directly detected pressures include Access Pressure (AP), Pre-Filter Pressure (PFP), Effluent Pressure (EP), and Return Pressure (RP) (Fig. 1). The calculated parameters include TMP, FPD, BFR, and TFR. The variation of these calculated pressure parameters in 4 h prior to filter clotting was analyzed for their capacity to predict filter clotting. Fig. 1Schematic diagram of pressure parameters monitored by the machine in the CRRT circuit. Schematic diagram of pressure parameters monitored by the machine in the CRRT circuit. The calculation formulas are as follows: TMP=(PFP + RP)/2–EP. FPD = PFP-RP. BFR = FPD/Qb. TFR = TMP/Quf. Whereas Qb is the blood flow rate, Quf is the ultrafiltration rate (Quf = prefilter replacement fluid rate + postfilter replacement fluid rate + net ultrafiltration rate). The prospective part of the study enrolled patients undergoing CRRT to observe the effects of intentional variations in blood flow rate and ultrafiltration rate on the above-mentioned pressure parameters. After the start of CRRT, the blood flow rate was sequentially set to 100, 150, and 200 ml/min at 10‑minute intervals, followed by sequential adjustment of the ultrafiltration rate to 400, 500, and 600 ml/h at the same intervals. After the observation period, the blood flow rate and ultrafiltration rate were set according to clinical decisions. This study was previously approved by the Ethical Review Board of the Jinling Hospital (approval number 2025DZKY-074-02) and conducted in accordance with the principles of the Declaration of Helsinki. Informed consent was obtained from each participant or an authorized family member. Statistical analysis Normality was evaluated using the Shapiro-Wilk test. Normally distributed continuous variables are reported as mean (SD); non-normal data as median (IQR). Categorical variables are reported as counts with percentages. Intergroup comparisons of non-normally distributed parameters were performed using the Kruskal-Wallis (KW) test. Prior to analysis, individual missing pressure parameters were excluded. Receiver Operating Characteristic (ROC) curves were systematically constructed to evaluate the predictive performance of multiple pressure parameters for impending filter clotting at different time thresholds. For each time window, the optimal cut-off value was determined by maximizing the Youden index (J = sensitivity + specificity − 1), providing a balanced trade-off between early detection and false alarm rates. In our prospective module, pressure parameters were recorded at varying flow rates. Group comparisons used one-way repeated-measures ANOVA (P < 0.05). All computations and figures were generated with R version 4.5.1 (R Foundation for Statistical Computing, Vienna, Austria) and GraphPad Prism version 10.0 (GraphPad Software, San Diego, CA, USA). Result Baseline clinical characteristics A total of 96 CRRT circuits (76 CVVHD and 20 CVVH) from 51 patients were included in our retrospective study. The baseline clinical characteristics are shown in Table 1. The median filter lifespan in CVVHD was 20.8 (12.1–27.6) hours, whereas in CVVH it was 16.8 (9.5–23.0) hours (Fig. 2). There was no difference in filter longevity between different catheter placement sites (CVVHD: P = 0.53; CVVH: P = 0.77) and between different anticoagulation modalities (CVVHD: P = 0.16; CVVH: P = 0.39). Table 1Baseline clinical characteristics. RCA, regional citrate anticoagulation; LMWH, low molecular weight heparin.PatientsCVVHDCVVHp value456Age52 (40–60)72(61–79)0.003Gender (Male/Female)32/136/00.127DiagnosisSevere acute pancreatitis38 (84.4%)2 (33.3%)0.006Multiple lesions2 (4.4%)2 (33.3%)Intestinal fistula5 (11.1%)2 (33.3%)Hemoglobin (g/L)67 (60–79)71(68–106)0.199Platelet (×109/L)146(115–232)113(40–161)0.085HCT0.20 (0.18–0.23)0.22(0.21–0.33)0.084PT (s)13.6 (12.6–14.6)15.9(12.5–18.9)0.146APTT (s)32.9(30.3–35.5)34.6(29.2–54.9)0.521INR1.19(1.09–1.26)1.41(1.09–1.69)0.126Total CRRT circuits7620Prescribed CRRT Dose2 L/h67(88.2%)10(50.0%)< 0.0014 L/h9(11.8%)10(50.0%)AnticoagulationRCA15(19.7%)0< 0.001Nafamostat mesylate14(18.4%)0LMWH2(2.6%)6(30.0%)RCA+LMWH31(40.8%)0RCA+Nafamostat5(6.6%)8(40.0%)No anticoagulation9(11.8%)6(30.0%)Vascular accessRight femoral53(69.7%)18(90.0%)0.141Left femoral14(18.4%)2(10.0%)Internal jugular9(11.8%)0Filter Lifespan (h)20.8 (12.1–27.6)16.8(9.5–23.0)0.244 Table 1Baseline clinical characteristics. RCA, regional citrate anticoagulation; LMWH, low molecular weight heparin.PatientsCVVHDCVVHp value456Age52 (40–60)72(61–79)0.003Gender (Male/Female)32/136/00.127DiagnosisSevere acute pancreatitis38 (84.4%)2 (33.3%)0.006Multiple lesions2 (4.4%)2 (33.3%)Intestinal fistula5 (11.1%)2 (33.3%)Hemoglobin (g/L)67 (60–79)71(68–106)0.199Platelet (×109/L)146(115–232)113(40–161)0.085HCT0.20 (0.18–0.23)0.22(0.21–0.33)0.084PT (s)13.6 (12.6–14.6)15.9(12.5–18.9)0.146APTT (s)32.9(30.3–35.5)34.6(29.2–54.9)0.521INR1.19(1.09–1.26)1.41(1.09–1.69)0.126Total CRRT circuits7620Prescribed CRRT Dose2 L/h67(88.2%)10(50.0%)< 0.0014 L/h9(11.8%)10(50.0%)AnticoagulationRCA15(19.7%)0< 0.001Nafamostat mesylate14(18.4%)0LMWH2(2.6%)6(30.0%)RCA+LMWH31(40.8%)0RCA+Nafamostat5(6.6%)8(40.0%)No anticoagulation9(11.8%)6(30.0%)Vascular accessRight femoral53(69.7%)18(90.0%)0.141Left femoral14(18.4%)2(10.0%)Internal jugular9(11.8%)0Filter Lifespan (h)20.8 (12.1–27.6)16.8(9.5–23.0)0.244 Baseline clinical characteristics. RCA, regional citrate anticoagulation; LMWH, low molecular weight heparin. Fig. 2Filter survival curves with 2 CRRT modes, CVVHD and CVVH. Filter survival curves with 2 CRRT modes, CVVHD and CVVH. The fluctuation of pressure parameters prior to filter clotting The profiles of continuously measured and calculated pressure parameters during the final 10-hour period prior to filter clotting were systematically illustrated in Fig. 3(CVVHD) and Fig. 4 (CVVH). In CVVHD mode, just prior to filter change, BFR and FPD increased by 0.13 (0.09–0.28) mmHg/(ml/min) and 19.50(13.25–45.25) mmHg from baseline, respectively; TFR and TMP increased by 0.05(0.02–0.10) mmHg/(ml/h) and 21.50 (4.75–37.50) mmHg, respectively. In CVVH mode, just prior to filter change, BFR and FPD increased 0.09(0–0.18) mmHg/(ml/min) and 14 (0–27.00) mmHg from baseline, respectively; TFR and TMP increased by 0.16 (0.08–0.28) mmHg/(ml/h) and 46(27.50–85.25) mmHg from baseline. Fig. 3Longitudinal trends of pressure parameters (BFR, FPD, PFP, TFR, TMP) in CVVHD circuits during the 10 hours before filter change. The red line represents the locally weighted regression (LOESS) smoothed trend (span = 0.75), and the shaded area indicates the 95% confidence interval. Longitudinal trends of pressure parameters (BFR, FPD, PFP, TFR, TMP) in CVVHD circuits during the 10 hours before filter change. The red line represents the locally weighted regression (LOESS) smoothed trend (span = 0.75), and the shaded area indicates the 95% confidence interval. Fig. 4Longitudinal trends of pressure parameters (BFR, FPD, PFP, TFR, TMP) in CVVH circuits during the 10 hours before filter change. The red line represents the locally weighted regression (LOESS) smoothed trend (span = 0.75), and the shaded area indicates the 95% confidence interval. Longitudinal trends of pressure parameters (BFR, FPD, PFP, TFR, TMP) in CVVH circuits during the 10 hours before filter change. The red line represents the locally weighted regression (LOESS) smoothed trend (span = 0.75), and the shaded area indicates the 95% confidence interval. The rise from baseline of 4 calculated pressure parameters (ΔBFR, ΔFPD, ΔTFR, and ΔTMP) was tested for predicting filter clotting at 4, 3, 2, and 1 h later using ROC Analysis. In CVVHD mode, DeLong’s test showed that the predictive efficacy of ΔBFR and ΔFPD was significantly higher than that of ΔTFR and ΔTMP (all p < 0.001) at 4 time points (Fig. 5). Although the AUC of ΔBFR was bigger than that of ΔFPD, there was no significant difference between them (4 h ΔAUC = 0.003, p = 0.418; 3 h ΔAUC = 0.005, p = 0.137; 2 h ΔAUC = 0.005, p = 0.211; 1 h ΔAUC = 0.004, p = 0.379). The optimal cutoffs of all parameters in 1-hour predictive windows are shown in Table 2. In CVVH mode, the AUC of ΔTFR and ΔTMP was bigger than that of ΔBFR and ΔFPD, although non-significantly (all p > 0.05) (Fig. 6). A power analysis using the observed AUC values (ΔTFR vs. ΔBFR) indicated that the study had only 48% power to detect the observed difference at a two‑sided alpha of 0.05. Thus, the comparative findings for this subgroup are substantially underpowered and should be interpreted as exploratory. Fig. 5ROC curve of pressure parameters’ predictive performance for filter clotting across 4- to 1-hour prediction windows in CVVHD(The dots indicate the cut-off points and the Youden index (J = sensitivity + specificity − 1)). ROC curve of pressure parameters’ predictive performance for filter clotting across 4- to 1-hour prediction windows in CVVHD(The dots indicate the cut-off points and the Youden index (J = sensitivity + specificity − 1)). Table 2Optimal cut-off points and AUC values of pressure parameters to predict filter clotting across 1-hour time windows.ParameterAUC (95% CI)CutoffSensitivitySpecificityYouden Index1 hΔBFR0.751 (0.718–0.784)0.07871.9%66.8%0.388ΔFPD0.747 (0.714–0.780)6.50087.7%49.3%0.370ΔTFR0.633 (0.592–0.674)0.01971.5%50.4%0.219ΔTMP0.627 (0.584–0.669)18.50049.1%74.5%0.236 Table 2Optimal cut-off points and AUC values of pressure parameters to predict filter clotting across 1-hour time windows.ParameterAUC (95% CI)CutoffSensitivitySpecificityYouden Index1 hΔBFR0.751 (0.718–0.784)0.07871.9%66.8%0.388ΔFPD0.747 (0.714–0.780)6.50087.7%49.3%0.370ΔTFR0.633 (0.592–0.674)0.01971.5%50.4%0.219ΔTMP0.627 (0.584–0.669)18.50049.1%74.5%0.236 Optimal cut-off points and AUC values of pressure parameters to predict filter clotting across 1-hour time windows. Fig. 6ROC curve of pressure parameters’ predictive performance for filter clotting across 4- to 1-hour prediction windows in CVVH. ROC curve of pressure parameters’ predictive performance for filter clotting across 4- to 1-hour prediction windows in CVVH. Model exploration combined ΔBFR and ΔTFR For predicting 1-hour-later filter clotting, ΔBFR and ΔTFR were the two parameters with the highest AUC. We further explored the models that combined the two parameters in the pooled data, distinguishing between CVVH and CVVHD modes. The results showed that the AUC of the combined model using the parallel rule (ΔBFR or ΔTFR) was higher than the individual parameters but the difference was not statistically significant (AUC: 0.746 vs. 0.728, p = 0.279) (Fig. 7). The parallel rule (positive if ΔBFR > 0.075 or ΔTFR > 0.115) yielded a sensitivity of 77.1% and a specificity of 62.9%. Table 3. Table 3Prediction performance (AUC) of combined models for filter clotting within 1 h across different treatment modalities (Serial rule: The final result is considered positive only when both TFR and BFR are simultaneously positive; Parallel rule: The result is deemed positive if either TFR or BFR is positive).CVVHD+CVVHCVVHDCVVHΔBFR0.7280.7510.657ΔTFR0.6490.6330.733Serial rule0.6910.7030.660Parallel rule0.7460.7350.777 Table 3Prediction performance (AUC) of combined models for filter clotting within 1 h across different treatment modalities (Serial rule: The final result is considered positive only when both TFR and BFR are simultaneously positive; Parallel rule: The result is deemed positive if either TFR or BFR is positive).CVVHD+CVVHCVVHDCVVHΔBFR0.7280.7510.657ΔTFR0.6490.6330.733Serial rule0.6910.7030.660Parallel rule0.7460.7350.777 Prediction performance (AUC) of combined models for filter clotting within 1 h across different treatment modalities (Serial rule: The final result is considered positive only when both TFR and BFR are simultaneously positive; Parallel rule: The result is deemed positive if either TFR or BFR is positive). Fig. 7ROC Analysis of ΔBFR and ΔTFR: Individual vs. Combined Predictive Value for 1-Hour Filter Clotting (CVVHD+CVVH) (Serial rule: The final result is considered positive only when both TFR and BFR are simultaneously positive; Parallel rule: The result is deemed positive if either TFR or BFR is positive) (The dots indicate the cut-off points and Youden index (J = sensitivity + specificity − 1)). ROC Analysis of ΔBFR and ΔTFR: Individual vs. Combined Predictive Value for 1-Hour Filter Clotting (CVVHD+CVVH) (Serial rule: The final result is considered positive only when both TFR and BFR are simultaneously positive; Parallel rule: The result is deemed positive if either TFR or BFR is positive) (The dots indicate the cut-off points and Youden index (J = sensitivity + specificity − 1)). Prospective observation of the effects of blood flow rate and ultrafiltration rate variation on pressure parameters Theoretically, FPD is not only affected by the severity of filter clotting, but also by the change of blood flow rate, and TMP is not only affected by the severity of filter clotting, but also by the change of ultrafiltration rate; the parameter BFR and TFR should be superior to FPD and TMP for predicting filter clotting, regarding the confounding factors blood flow rate and ultrafiltration rate being adjusted in the former. However, this speculation was not verified in our study, possibly because blood flow rate and ultrafiltration remained little changed during treatment. To verify the speculation, the prospective part of the study intentionally modulated the blood flow rate and ultrafiltration rate during the start period of the CRRT session. As revealed by the results, an increase in blood flow rate simultaneously raised FPD, but affected BFR little (Fig. 8A); similarly, an increase in ultrafiltration rate simultaneously raised TMP, but affected TFR little. The results indicate FPD is blood flow rate-dependent and TMP is ultrafiltration rate-dependent; while BFR is blood flow rate-independent and TFR is ultrafiltration rate-independent (Fig. 8B). Fig. 8Comparison of FPD and BFR under different blood flow rates (100, 150, and 200 ml/min) (N = 9) (A) and comparison of TMP and TFR under different ultrafiltration flow rates (400, 500, and 600 ml/h) (N = 8) (B). Comparison of FPD and BFR under different blood flow rates (100, 150, and 200 ml/min) (N = 9) (A) and comparison of TMP and TFR under different ultrafiltration flow rates (400, 500, and 600 ml/h) (N = 8) (B). Discussion Main findings We retrospectively analyzed continuous pressure data from CRRT and observed that all (BFR, FPD, TFR, TMP, and PFP) increased significantly before filter clotting (> 50% fiber occlusion). BFR and TFR demonstrated higher reliability than FPD and TMP because they eliminated the effects of flow-rate fluctuations, although the AUC difference was not statistically significant. And BFR was more excellent and stable than TFR in forecasting clotting in CVVHD (AUC range:0.744–0.751 vs. 0.633–0.640, p > 0.001). In CVVH, TFR numerically outperformed BFR (AUC ranges: 0.707–0.733 vs. 0.653–0.657), reflecting a preliminary trend that did not reach statistical significance. BFR increasing by 0.075 mmHg/(ml/min) or TFR rising by 0.115 mmHg/(ml/h) with a sustained upward trend was a sign for future filter clotting, whether in CVVHD or CVVH. Herein, we proposed a combination of 2 calculated circuit pressure parameters - BFR and TFR, whose rising values during CRRT have the ability to predict filter clotting. The advantages of these parameters are irrespective of CRRT mode, blood flow rate, and ultrafiltration rate. Relationship to previous studies Anticoagulation is essential to prevent extracorporeal circuits from clotting during CRRT12. The median lifespans of the filters in our study (CVVHD: 20.8 (12.1–27.6) hours; CVVH: 16.8 (9.5–23.0) hours) were shorter than the recommended filter-use duration. One contributing factor to the shortened filter lifespan was the prevalence of coagulation disorders among a considerable subset of enrolled patients, which prevented them from receiving standard anticoagulant doses. And we observed no significant difference in filter longevity between femoral and non-femoral access, as previous studies described13. Previous studies have demonstrated that FPD, TMP, and PFP are strongly correlated with filter failure. A retrospective study showed that TMP and PFP were the strongest predictors of filter coagulation in continuous veno-venous hemodiafiltration (CVVHDF) and indicated that each 1 mmHg increase in TMP or PFP was associated with a 1.1% (p < 0.001) and 1.2% (p = 0.029) higher risk of premature clotting7. Unfortunately, this study did not include FPD for further comparisons. Mateos-Dávila, A., argued that PFP served as a key indicator for determining when to return blood to the patient due to circuit clotting, whereas TMP was not a reliable parameter in both diffusive and convective treatments14. This may be because the article did not analyze coagulation in different parts of the circuit in detail. The rise of FPD and TMP can suggest coagulation of hollow fibers and blockage of membrane pores in the filter, but does not indicate clotting in other circuit segments. While coagulation occurring in any of the following circuit components(the filter, venous chamber, or venous line) will consistently lead to a marked rise in PFP15. Some studies recommended that pressure parameters exceeding certain absolute thresholds (e.g., TMP > 150 mmHg6,16 indicated declining filter performance and warranted replacement. However, in clinical practice, filter clotting is more closely linked to dynamic pressure fluctuations than to static values15. A study showed that a rising threshold of 26 mmHg in FPD had high predictive validity for circuit failure induced by filter clotting8, and it was used as a criterion for filter clotting by another subsequent study17. Our cohort exhibited a lower median FPD increase of 19.5 (IQR: 13.25–45.25) mmHg at filter replacement. This discrepancy may be attributed to our more stringent, clinically relevant filter replacement criteria (Grade 4: >50% fiber occlusion), which prioritized both treatment continuity and patient safety, whereas prior studies relied solely on treatment interruptions as an indicator of severe clotting. In our study, we found that BFR and TFR had greater predictive potential than FPD and TMP. This could be explained from the perspective of fluid dynamics. The relationship between trans-filter blood flow and FPD can be approximated by the Hagen-Poiseuille equation(\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{Q}_{b}=\frac{\pi\:{R}_{f}^{4}\varDelta\:p}{8{\eta\:}_{b}L},\varDelta\:p:axial\:pressure\:drop$$\end{document})10. Thrombus formation within the fiber lumen directly reduces the effective blood flow pathway, increasing flow resistance as blood passes through the hollow fiber bundle. FPD must increase compensatorily to maintain the set blood flow rate. This is also the theoretical basis for previous studies employing FPD as an indicator of filter clotting. However, adjusting the blood flow rate during treatment will lead to a change in FPD without a change in blood flow resistance. By eliminating the effects of changes in blood flow rate, BFR is affected only by the resistance encountered when flowing through the hollow tubes of the filter. If the hollow fibers are clogged, the resistance of blood to flow through the filter increases, which is reflected in an increase in BFR. Similarly, the relationship between TMP, ultrafiltration rate, and ultrafiltration coefficient (KUF) was defined as TMP = Quf/KUF. The filtration coefficient is a critical parameter that quantifies the intrinsic hydraulic permeability of a filter. Clogging of membrane pores directly reduces the ultrafiltration coefficient, and TMP must increase compensatorily to maintain the set ultrafiltration rate. So TMP serves as an indicator of impaired membrane function. If the ultrafiltration rate is adjusted, TMP will change without altering membrane permeability. Therefore, TFR (1/KUF) effectively eliminates the effect of changes in ultrafiltration rate and is a specific indicator of filter clotting. Our experimental measurements (Fig. 8) confirmed the theoretical expectation that FPD and TMP increased proportionally with flow rate, whereas BFR and TFR eliminated the effects of flow-rate changes. However, BFR and TFR showed different prediction ability in different modes. A study also mentioned similar results14. Despite the lack of a direct statistical correlation between BFR and filter coagulation, the data revealed a clear and progressive increase in BFR (from 0.2 to 2.3 mmHg/(mL/min)) corresponding to rising levels of visual filter coagulation (from 1% to 99%) in both diffusive and convective modes. A substantial increase in TMP was observed only in convection mode. It showed that the predictive value of TMP for filter clotting differs between diffusive and convective treatments. There have been a number of studies demonstrating a negative correlation between TMP and filter clearance performance, but they have all focused on convective modalities (CVVH and CVVHDF)7,16,18. In this mode, solute and fluid removal are dominated by solvent drag, a process directly driven by TMP. At this point, the increased resistance to transmembrane water flow caused by pore blockage becomes more sensitive than the elevated blood flow resistance resulting from blockage within the longitudinal fiber channels. This might account for the preliminary trend observed in our study, where TFR had a numerically higher AUC than BFR in CVVH. And the lack of statistical significance might be attributable to the limited sample size. Current evidence inadequately characterizes the correlation between TMP change and filter clotting risk in diffusive treatments. In diffusive mode, solute clearance occurs primarily via diffusion, governed by Fick’s law and driven by concentration gradients across the membrane. TMP plays only a minor role in molecular removal. In this situation, the pressure elevation caused by longitudinal fiber occlusion holds greater predictive value. So BFR performed better than TFR in CVVHD. We believe that there may be additional benefits to monitoring both BFR and TFR simultaneously, regardless of the treatment modality. Simultaneous monitoring of BFR and TFR enables a comprehensive assessment of filter clotting status: dynamic variations in BFR effectively reflect increased hemodynamic resistance due to thrombus formation within the longitudinal fiber bundle, whilst dynamic changes in TFR sensitively detect permeability alterations resulting from membrane pore blockage. The combined predictive efficacy remains robust across treatment modalities. Real-time monitoring of BFR and TFR can be directly derived from existing CRRT system parameters without requiring additional hardware modifications. These critical hemodynamic parameters are readily calculable through algorithmic processing of standard pressure and flow data already captured during routine CRRT operation. This enables safe monitoring at the bedside without incurring additional costs. For future CRRT systems, we propose integrating continuous BFR and TFR monitoring capabilities with automated analytical algorithms. Our study proposes novel pressure parameters for alerting to filter clotting and prompting clinical pre-clotting intervention, which may be beneficial for prolonging filter lifespan and reducing clotting-related complications, such as blood loss and nursing burden, caused by unplanned filter replacement. The results may inspire future prospective trials with a randomized controlled design to compare an alert-triggered protocol versus standard care. Although this was an exploratory study, several limitations should be acknowledged. Firstly, as a retrospective investigation with a modest sample size, this study did not establish causality nor allow quantification of absolute risk. Future prospective studies are needed to validate the predictive value of the examined parameters. Specifically, findings related to CVVH should be interpreted as reflecting trends only, given the sample size constraints. The potential superiority of blood flow rate and filtration rate over conventional parameters, as well as the predictive performance of a combined model incorporating both measures across different treatment modalities, requires further confirmation in large-scale, multicenter prospective studies. Secondly, ROC analysis with DeLong tests was used to evaluate the performance of multiple parameters. However, the inherent correlation in the data was not accounted for, which may have resulted in underestimated standard errors and overestimated significance of differences in the area under the curve. These findings are therefore exploratory and should be interpreted with caution. Thirdly, as this study was conducted using a single device and filter type, validation across different CRRT machines and filter types is necessary before the findings can be generalized. In addition, this study did not construct a multivariate model to adjust for potential confounders, such as anticoagulation modality, vascular access, or laboratory indicators. The presence of such confounders may introduce bias in effect estimates, either overestimating or underestimating the true prognostic value. Future studies incorporating multivariate analysis are required to validate these results. Furthermore, although no direct subgroup comparisons were performed, differences in baseline characteristics, including age and diagnosis, were present across subgroups. Future studies should aim to balance baseline characteristics to enable meaningful subgroup analyses. Finally, although this study included only filter replacements attributable to nurse‑assessed Grade 4 clotting, clinical factors, such as operator preferences, may still have influenced the timing of replacement. Future studies could adopt more objective measures to refine outcome definitions and improve model precision, such as the quantified percentage of occluded fibers, filter clearance performance, or composite endpoints. Conclusion By retrospective analysis of stress parameters during continuous renal replacement therapy, this study confirmed that BFR, FPD, TFR, TMP, and PFP all showed a significant rise before filter clotting. In CVVHD, BFR (calculated as FPD/Qb) emerged as the superior predictor among pressure parameters for impending filter clotting (> 50% fiber occlusion). In CVVH, our analysis indicated a preliminary trend toward better predictive performance of TFR (TMP/Quf) than BFR, but this did not reach statistical significance. By eliminating the influence of flow rate, BFR and TFR may more specifically reflect changes in flow resistance due to filter occlusion than FPD and TMP. The combined model of BFR and TFR may provide better predictive value when analyzed without mode differentiation. In the future, we shall conduct prospective validation to determine whether intervention when parameters exceed thresholds can effectively extend filter lifespan. Real-time BFR and TFR could be calculated from parameters captured by the existing CRRT system. Future iterations of CRRT machines should integrate continuous monitoring and algorithmic analysis of BFR and TFR. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 1