Dose response of nafamostat mesylate for anticoagulation monitoring in pediatric critical care during continuous venovenous hemodiafiltration

Abstract

Nafamostat mesylate (NM) is increasingly used as an alternative anticoagulant during continuous venovenous hemodiafiltration (CVVHDF) in pediatric intensive care, especially when unfractionated heparin or regional citrate is contraindicated. However, evidence regarding optimal monitoring of its anticoagulant effects remains limited. This study aimed to investigate the association between NM infusion rate and coagulation parameters—specifically activated clotting time (ACT) and activated partial thromboplastin time (aPTT)—in critically ill pediatric patients receiving CVVHDF, and to examine how hepatic function and transfusion status may influence this relationship. In this retrospective study of 99 patients, we analyzed 340 matched data pairs of NM infusion orders and coagulation test results. Linear mixed-effects models revealed a significant positive correlation between NM infusion rate and aPTT (β = 8.54, P < 0.001), while no significant association was found with ACT. Subgroup analyses stratified by liver function and transfusion status consistently supported the association between NM and aPTT. Non-linear regression further suggested a dose–response pattern for aPTT, but not for ACT. These findings indicate that aPTT may be a more sensitive and reliable parameter than ACT for monitoring NM’s anticoagulant effects in pediatric CVVHDF. The results challenge current ACT-based monitoring practices and highlight the need for prospective validation to refine anticoagulation strategies using NM in critically ill children.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-025-28308-8.

Introduction

Continuous venovenous hemodiafiltration (CVVHDF) is the preferred modality for renal support in the pediatric intensive care unit (PICU) due to its ability to facilitate gentle fluid removal and solute stabilization in hemodynamically unstable childre^1–3. During CVVHDF, the interaction of blood with the extracorporeal circuit activates the intrinsic and extrinsic coagulation pathways and platelets, leading to clotting and potential filter occlusion⁴. Clotting in the extracorporeal circuits can result in blood loss, impaired solute clearance, and reduced ultrafiltration, underscoring the importance of selecting an appropriate anticoagulation modality.

Unfractionated heparin (UFH) and regional citrate anticoagulation (RCA) are commonly used during CVVHDF. UFH is familiar and widely available but carries risks such as bleeding, a narrow therapeutic index, and heparin-induced thrombocytopenia. RCA is recommended over UFH by the 2012 Kidney Disease: Improving Global Outcomes guidelines when no contraindications exist, offering regional anticoagulation without systemic effects^1,5. However, RCA depends on hepatic metabolism of citrate, and in patients with liver dysfunction, citrate accumulation can lead to metabolic complications, including hypocalcemia, alkalosis, and increased mortality^6,7. Nafamostat mesylate (NM) offers a viable alternative, especially when UFH or RCA is unsuitable. Although its exact mechanism is not fully elucidated, NM is a potent serine protease inhibitor that exerts inhibitory effects on various enzymatic systems, including coagulation, fibrinolysis, and the complement cascade^8–10. With a short half-life and plasma esterase metabolism, NM avoids hepatic dependence and reduces metabolic risks^11,12. However, one study reported that the anticoagulant effect of NM may also vary depending on hepatic function, and the current evidence remains limited, particularly in pediatric patients¹³. Additionally, due to its rapid metabolism and localized mechanism of action, the anticoagulant effect of NM is typically monitored using activated clotting time (ACT)¹⁴. An in vitro study reported that ACT was markedly prolonged when NM was administered above a certain concentration, suggesting the possibility of a concentration-dependent change in anticoagulant effect¹⁵. These findings highlight the need for caution in dosing and monitoring NM and underscore the necessity for additional evidence to better establish its safety and efficacy, especially in children. However, as this finding is based solely on an in vitro experiment, its clinical relevance has not yet been validated, and further investigation in actual patient populations is warranted.

Therefore, the aim of this study was to investigate the relationship between the infusion rate of NM and its anticoagulant effect in pediatric patients undergoing CVVHDF, and to evaluate how this relationship may be influenced by hepatic function. In particular, we sought to examine how various coagulation parameters—including ACT and activated partial thromboplastin time (aPTT)—change in response to different NM infusion rates during real-world clinical use.

Materials and methods

Ethics statement

This study was approved by the Institutional Review Board of Seoul National University Hospital (approval number: H-2307-153-1452). Given that the study involved a retrospective analysis of pseudonymized electronic medical records, it was classified as minimal-risk research, and the Institutional Review Board of Seoul National University Hospital accordingly waived the requirement for informed consent from patients or their legal guardians. The study was conducted in full compliance with the principles set forth in the Declaration of Helsinki.

Study setting

This retrospective cross-sectional observational study was conducted in a 24-bed PICU at a tertiary children’s hospital in Seoul, Republic of Korea. Patients under the age of 19 who were admitted to the PICU and received CVVHDF between January 2003 and December 2022 were included in this study. Patients who did not receive NM during the course of CVVHDF or who were administered systemic UFH concurrently with NM were excluded. In addition, patients who received other extracorporeal devices, such as extracorporeal membrane oxygenation, in conjunction with CVVHDF were also excluded from the analysis.

Data collection and definitions

All data used in this study were obtained from the clinical data warehouse of the institution. Basic demographic data—including age, sex, and department placing the NM order—were collected. Laboratory results such as ACT, aPTT, aspartate aminotransferase (AST), alanine aminotransferase (ALT), alkaline phosphatase (ALP), total bilirubin (TB), direct bilirubin (DB), and gamma-glutamyl transferase (GGT), along with the date and time of each test, were obtained. Data on anticoagulation included NM infusion rates, corresponding order times, and the presence or absence of systemic UFH administration. To calculate the body weight–adjusted NM infusion rate, the body weight recorded on the day of NM infusion was collected. Transfusion records for blood products—including fresh frozen plasma (FFP), red blood cells (RBCs), and platelets—were also collected. To determine filter lifetime, the start and end times of each filter were collected.

In this study, the term ‘NM order time’ was defined as the time at which an infusion order for NM administration was placed. This included both the initial order and any subsequent orders issued to adjust or maintain the infusion rate, typically based on anticoagulation test results such as ACT or aPTT. Each NM order was paired with the corresponding ACT or aPTT result that directly informed the decision to initiate, adjust, or maintain the infusion rate. In routine practice, the coagulation test result and NM order were paired within minutes, and these matched pairs were defined as ‘coagulation-NM pairs’. These pairs served as the fundamental units for analyzing the relationship between coagulation parameters and NM dosing.

In this study, analyses were performed at the level of individual coagulation-NM pairs, where each NM infusion rate was directly linked to a corresponding coagulation test result (ACT or aPTT) that guided dosing decisions. Since NM infusion rates are influenced by coagulation parameters, which can be affected by transfusions and liver function, we conducted subgroup analyses considering these clinical variables. We assessed whether transfusions occurred within a 12-hour window before or after each coagulation-NM pair and analyzed the specific timing of each transfusion relative to the NM order. Pre-test transfusions could alter coagulation results while post-test transfusions might reflect clinical responses to bleeding risk.

For liver function assessment, we considered routine LFTs (AST, ALT, ALP, TB) and extended LFTs when DB and GGT were additionally measured. However, DB and GGT were not routinely obtained, being ordered only when routine LFTs showed abnormalities or liver disease was suspected. Unlike coagulation tests (which were performed at intervals of 2 to 8 h, starting every 2 h during the initial phase of NM infusion and subsequently extended to 4–8 h depending on the stability of results, but never exceeding 8 h), LFTs are usually done once or twice daily. Therefore, each NM order was matched to the closest LFT result within the preceding 12-hour window. Based on marker availability, coagulation-NM pairs were categorized into the ‘non-extended LFT group’ (AST, ALT, ALP, TB only) and the ‘extended LFT group’ (including DB and GGT).

Outcomes

The primary outcome was the correlation between changes in coagulation parameters—specifically ACT and aPTT—and body weight–adjusted NM infusion rates in the overall patient cohort. In this analysis, NM infusion rates were considered in conjunction with their corresponding anticoagulation test results, matched as pairs based on each defined NM order time.

Secondary outcomes involved subgroup analyses based on hepatic function and transfusion status. Hepatic function groups were defined according to the extended and non-extended LFT group classifications described above. Blood product transfusions administered within 12 h before or after each NM order time were evaluated, as previously detailed, with additional analysis of the temporal interval between transfusions and NM orders. Furthermore, we analyzed the distribution of filter lifetime and examined whether filter lifetime varied according to NM infusion rate.

Equipment and protocol

CVVHDF was performed using the Prismaflex^® system (Gambro, Lakewood, Colorado, USA). Filters were selected based on patient body weight. For patients weighing less than 10 kg, the high-flux (HF) 20 filter was used with a priming volume of 60 mL. For those weighing 10 to < 30 kg, the standard (ST) 60 filter (priming volume 93 mL) was utilized, while the ST 100 filter (priming volume 152 mL) was applied for patients over 30 kg. The ultrafiltration coefficients were 15 mL/h × mmHg for the HF 20 and ST 60 filters and 25 mL/h × mmHg for the ST 100 filter; these values served as the default initial settings, although minor adjustments may have been made depending on the patient’s clinical condition. Anticoagulation was performed using NM (Futhan^®, SK chemicals, Seoul, Republic of Korea). The priming volume (extracorporeal volume) was maintained at < 10% of the total blood volume, estimated as 8% of body weight. Blood priming was required if the priming volume exceeded 8 mL/kg of body weight.

ACT was measured at the bedside using a portable Hemochron 401 device (ITC Medical, Edison, NJ) with Celite-activated HRFTCA510 tubes^16,17. In our institution, the anticoagulation monitoring protocol involves routine assessment of both ACT and aPTT. NM infusion was primarily titrated according to ACT values, with a target range of 140 to 180 seconds¹⁸. However, when ACT values remained within the target range but aPTT showed abnormally prolonged results, clinicians occasionally adjusted the NM infusion rate based on the aPTT value. Thus, NM dosing was guided primarily by ACT, while aPTT served as a secondary parameter that could influence dosing decisions under specific circumstances.

Baseline CVVHDF dose parameters at therapy initiation were based on our institution’s protocol, with recommended blood flow rates of 3–5 mL/kg/minute and combined dialysate plus replacement fluid flow rates of 20–40 mL/kg/hour, aiming for a total dose of approximately 25 mL/kg/hour or higher. Transfusion decisions generally followed unit guidelines—FFP for active bleeding or an aPTT > 120 s, and RBC or platelet transfusions for hemoglobin < 7 g/dL or platelet count < 20,000/µL—but were ultimately guided by clinical judgment rather than strict adherence to these thresholds. Both the prescribed CVVHDF settings and transfusion criteria served as initial frameworks; actual practice during the study period may have varied according to individual patient factors.

Statistical analysis

Categorical variables were presented as numbers (%), while continuous variables were expressed as median (interquartile range). To evaluate the relationship between NM infusion rates and coagulation parameters — namely ACT and aPTT — a linear mixed-effects model was applied to account for repeated measurements within individual patients. In this model, the patient identifier was included as a random effect to adjust for within-subject correlation. Univariable analyses were initially performed to identify factors influencing ACT and aPTT, considering NM infusion rates, age, sex, body weight, LFT results, and transfusion status. Variables found to be statistically significant in the univariable analyses were subsequently subjected to backward stepwise selection based on minimizing the akaike information criterion (AIC), and only the variables contributing to the lowest AIC were retained for the final multivariable model.

Given the possibility that the relationship between NM infusion rate and coagulation parameters may not follow a strictly linear pattern, we also explored several non-linear regression methods. These included polynomial regression, spline regression, locally estimated scatterplot smoothing (LOESS), and exponential/logistic curve fitting. These additional approaches were applied to better capture potential non-linear trends in the data and to ensure robustness of findings across different modeling assumptions.

The analysis was conducted based on individual observations, defined as paired data points of coagulation parameters and corresponding NM infusion rates, rather than on a per-patient basis. Consequently, multiple records could be included for a single patient. In addition, the relationship between filter lifetime and NM infusion rate was assessed using Pearson’s correlation coefficient (r), with statistical significance determined via the corresponding t statistic to derive P values. A P value of < 0.05 was considered statistically significant. All statistical analyses were performed using R software (version 4.3.1, R Foundation for Statistical Computing) and Python (version 3.11)^19,20. In Python, the following packages and libraries were utilized: pandas, numpy, matplotlib, seaborn, scipy, statsmodels, and patsy^21–27.

Results

Baseline demographics and characteristics

A total of 432 NM order–coagulation parameter pairs were initially collected. After excluding unmatched records, patients aged ≥ 19 years, and those who received concurrent heparin, 340 cases from 99 patients (41 female, 58 male) were included in the final analysis. Among them, 171 observations from 26 patients were classified as the extended LFT group, and 169 observations from 73 patients were classified as the non-extended LFT group (Fig. 1). Table 1 presents detailed characteristics of the overall population and each subgroup, and Supplementary Table S1 shows the distribution of departments ordering NM. Transfusions administered within a 12-hour window before and after each individual NM order time in the coagulation-NM pairs (including any NM orders issued during treatment) were recorded and categorized by blood product type. The distribution of transfusion timing relative to each NM order time is detailed in Supplementary Table S2.

Fig. 1.

Study flow chart. A flowchart illustrating the patient selection process. NM = nafamostat mesylate, LFT = liver function test.

Table 1.

Baseline characteristics of the study population overall and by LFT group.

Variables	Total	Non–Extended LFT group	Extended LFT group
	(n = 99)	(n = 73)	(n = 26)
Age, years	7.0 (1.0–12.0)	8.0 (1.0–13.0)	6.5 (1.0–12.0)
Female	41 (41.4)	29 (39.7)	12 (46.2)
Body weight, kg	20.4 (8.9–39.9)	21.0 (10.8–37.8)	18.9 (7.5–42.7)
NM infusion rate, mg/kg/h	0.5 (0.3–1.2)	0.5 (0.3–1.2)	0.4 (0.3–1.2)
LFT
AST, unit/L	65.0 (34.0–153.0)	62.5 (31.5–184.0)	82.0 (47.0–146.0)
ALT, unit/L	42.0 (25.0–106.0)	37.0 (21.0–99.0)	70.0 (36.0–131.0)
ALP, unit/L	110.0 (79.0–164.5)	106.0 (76.0–141.0)	161.0 (103.0–221.0)
Total bilirubin, mg/dL	1.6 (0.8–5.5)	1.1 (0.7–2.8)	4.5 (2.1–9.9)
Direct bilirubin, mg/dL	3.2 (1.4–5.3)	NA	3.2 (1.4–5.3)
GGT, unit/L	64.0 (19.0–110.0)	36.5 (14.0–110.0)	68.0 (48.5–122.5)
Anticoagulation test
aPTT, seconds	41.3 (34.8–58.0)	41.0 (34.7–52.3)	45.8 (35.6–63.0)
ACT, seconds	126.0 (107.0–141.5)	124.0 (107.0–133.0)	143.0 (131.0–149.0)
Transfusion ≤12 h pre-NM†
Fresh frozen plasma	11 (11.1)	8 (11.0)	3 (11.5)
Red blood cells	48 (48.5)	38 (52.1)	10 (38.5)
Platelets	37 (37.4)	24 (32.9)	13 (50.0)
Transfusion ≤ 12 h post-NM‡
Fresh frozen plasma	13 (13.1)	7 (11.1)	6 (23.1)
Red blood cells	44 (44.4)	31 (49.2)	13 (50.0)
Platelets	30 (30.3)	18 (28.6)	12 (46.2)

Values are presented as number (%) or median (interquartile range).

Patients were divided into two groups based on liver function test patterns: the Extended LFT group, who underwent additional liver testing with direct bilirubin and/or GGT, and the Non–Extended LFT group, who received only routine total bilirubin testing.

^†Transfusions administered within 12 h before the NM order time (time at which NM was ordered).

^‡Transfusions administered within 12 h after the NM order time.

LFT = liver function test, NM = nafamostat mesylate, AST = aspartate aminotransferase, ALP = alkaline phosphatase, GGT = gamma-glutamyl transferase, NA = not applicable, aPTT = activated partial thromboplastin time, ACT = activated clotting time.

Main outcomes

Coagulation parameters and NM infusion rate (Primary Outcome)

Scatter plots illustrating the relationship between NM infusion rate and both ACT and aPTT are presented in Fig. 2. As the NM infusion rate increased, aPTT showed a significant upward trend, whereas ACT demonstrated a non-significant tendency to decrease (Fig. 2). In the mixed-effects model analysis of coagulation-NM pairs, TB was the only variable that showed a statistically significant association with ACT in both univariable and multivariable analyses (β = 1.771, standard error [SE] = 0.584, P = 0.002), whereas the NM infusion rate was not significantly associated with ACT (β = −4.158, SE = 2.947, P = 0.158; Table 2). In contrast, regression analysis for aPTT showed that NM infusion rate, age, body weight, ALP, and TB were significantly associated with aPTT in the univariable analysis. In the multivariable analysis, NM infusion rate (β = 8.780, SE = 1.346, P < 0.001) and TB (β = 0.563, SE = 0.243, P = 0.021) remained statistically significant factors (Table 3).

Fig. 2.

Mixed-effect linear regression of NM infusion rate and coagulation parameters. Scatter plots and corresponding regression lines showing the relationship between NM infusion rate and coagulation times. Mixed-effects linear regression with a pseudonymized subject identifier as a random effect revealed no significant association between NM infusion rate and ACT, whereas aPTT showed a significant positive association. ACT = activated clotting time, aPTT = activated partial thromboplastin time, NM = nafamostat mesylate.

Table 2.

Mixed-effects regression^† results for ACT.

Variables	Univariable			Multivariable
	Coefficient b	SE	P	Coefficient b	SE	P
NM infusion rate, mg/kg/h	–4.158	2.947	0.158
Age, years	–0.963	0.696	0.166
Sex
Female	Reference
Male	–10.773	9.137	0.238
Body weight, kg	–0.112	0.18	0.534
Liver function test
AST, unit/L	0.003	0.008	0.719
ALT, unit/L	0.002	0.011	0.818
ALP, unit/L	0.002	0.029	0.941
Total bilirubin, mg/dL	1.771	0.584	0.002	1.771	0.584	0.002
Direct bilirubin, mg/dL	0.726	0.926	0.433
GGT, unit/L	–0.013	0.019	0.499
Transfusion‡
Fresh frozen plasma	18.986	11.269	0.092
Red blood cells	1.335	5.668	0.814
Platelets	–3.365	5.829	0.564

^†The patient identifier served as the random effect in the mixed-effects regression model.

^‡This refers to the transfusion status from 12 h prior to the commencement of NM administration until the time of its administration.

ACT = activated clotting time, SE = standard error, NM = nafamostat mesylate, AST = aspartate aminotransferase, ALT = alanine aminotransferase, ALP = alkaline phosphatase, GGT = gamma-glutamyl transferase.

Table 3.

Mixed-effects regression^† results for aPTT.

Variables	Univariable			Multivariable
	Coefficient b	SE	P	Coefficient b	SE	P
NM infusion rate, mg/kg/h	9.303	1.298	< 0.001	8.78	1.346	< 0.001
Age, years	–0.865	0.422	0.04
Sex
Female	Reference
Male	–0.419	5.324	0.937
Body weight, kg	–0.257	0.11	0.019
Liver function test
AST, unit/L	0.001	0.003	0.814
ALT, unit/L	–0.001	0.006	0.806
ALP, unit/L	–0.030	0.014	0.036	–0.009	0.013	0.492
Total bilirubin, mg/dL	0.813	0.258	0.002	0.563	0.243	0.021
Direct bilirubin, mg/dL	–0.067	0.571	0.907
GGT, unit/L	–0.006	0.017	0.723
Transfusion‡
Fresh frozen plasma	4.444	5.551	0.423
Red blood cells	–2.566	2.705	0.343
Platelets	–0.308	2.896	0.915

^†The patient identifier served as the random effect in the mixed-effects regression model.

^‡This refers to the transfusion status from 12 h prior to the commencement of NM administration until the time of its administration.

aPTT = activated partial thromboplastin time, SE = standard error, NM = nafamostat mesylate, AST = aspartate aminotransferase, ALT = alanine aminotransferase, ALP = alkaline phosphatase, GGT = gamma-glutamyl transferase.

Subgroup analysis by hepatic function

In the non-extended LFT group, which was classified based on liver function status, analysis of the association between NM infusion rate and coagulation parameters showed that aPTT significantly increased with rising NM infusion rates. In contrast, ACT was not significantly associated with NM infusion rate and instead demonstrated a decreasing trend (Supplementary Fig. S1). In the mixed-effects regression analysis for the non-extended LFT group, ACT was significantly associated only with total bilirubin (β = 7.147, SE = 1.528, P < 0.001). Although FFP transfusion showed a borderline association (β = 27.464, SE = 14.378, P = 0.056), no significant relationship was observed with NM infusion rate (β = −6.062, SE = 5.205, P = 0.244) or other variables (Supplementary Table S3). In contrast, in the multivariable analysis of the non-extended LFT group, aPTT showed a statistically significant positive association only with NM infusion rate (β = 5.208, SE = 2.529, P = 0.039). Total bilirubin also demonstrated a positive trend and narrowly reached statistical significance (β = 1.031, SE = 0.524, P = 0.049). No other variables were significantly associated with aPTT (Supplementary Table S4).

In the extended LFT group, both overall and similarly to the non-extended LFT group, aPTT showed a statistically significant increase with higher NM infusion rates, whereas ACT demonstrated a non-significant decreasing trend (Supplementary Fig. S2). In the univariate mixed-effects regression analysis of the extended LFT group, no factors showed a statistically significant association with ACT (Supplementary Table S5). In contrast, aPTT was significantly associated with NM infusion rate both in the univariable (β = 12.714, SE = 1.508, P < 0.001) and multivariable models (β = 12.373, SE = 1.568, P < 0.001). Although ALP showed a significant negative association with aPTT in the univariable analysis (β = − 0.056, SE = 0.019, P = 0.004), this association was no longer significant in the multivariable model (β = − 0.018, SE = 0.017, P = 0.286; Supplementary Table S6).

Association of NM infusion rate and coagulation parameters according to transfusion status

As part of the secondary analyses, we examined the association between NM infusion rate and coagulation parameters, stratifying patients based on transfusion status within 12 h before and after each NM order time. Mixed-effects regression analyses were conducted separately for each transfusion group—RBCs, platelets, and FFP—as well as for patients who did not receive each respective transfusion. Across all subgroups, aPTT consistently showed a statistically significant positive association with NM infusion rate. Among patients who received transfusions before each NM order time, the β coefficients for aPTT were 9.171 (P < 0.001) for RBCs, 12.552 (P = 0.008) for platelets, and 10.589 (P < 0.001) for FFP. Similar significant associations were observed in patients who did not receive transfusions before NM order time, with β values of 9.939 (P < 0.001), 9.992 (P < 0.001), and 10.403 (P < 0.001) for those without RBC, platelet, and FFP transfusions, respectively. In contrast, ACT exhibited a decreasing trend with increasing NM infusion rate across all subgroups, but none of these associations reached statistical significance. The β coefficients ranged from − 0.673 (P = 0.904) in the overall analysis to − 10.349 (P = 0.262) in the platelet transfusion group (Supplementary Fig. S3). Additionally, analyses for transfusions administered after each NM order time are presented in Supplementary Fig. S4. In post-NM transfusion groups, aPTT generally demonstrated positive trends with NM infusion rate, though not always statistically significant, whereas ACT occasionally showed significant but negative associations.

In addition, transfusions administered within 12 h prior to the coagulation-NM pair were included as input variables in the multivariable regression analyses since they could potentially influence ACT and aPTT values. However, transfusions administered within 12 h after the coagulation-NM pair might have been given based on coagulation parameter results, and thus were excluded from the overall multivariable regression analysis. Instead, univariable regression analyses were conducted to examine the associations between transfusions and coagulation parameters for each blood product. Results showed that transfusions of RBCs and platelets did not demonstrate statistically significant associations with either ACT or aPTT across all groups. For FFP transfusions, statistically significant associations with ACT were observed in the total cohort (β = 25.751, SE = 11.534, P = 0.026) and the non-extended LFT group (β = 247.744, SE = 27.401, P < 0.001), but not in the extended LFT group (β = −1.695, SE = 12.000, P = 0.888). Regarding aPTT, significant associations were found in the total cohort (β = 17.044, SE = 4.729, P < 0.001) and the extended LFT group (β = 19.434, SE = 6.348, P = 0.002), whereas the non-extended LFT group did not show significant results (β = 11.905, SE = 7.509, P = 0.113) (Supplementary Table S7).

Non-linear relationship between NM infusion rate and coagulation times

Non-linear regression analyses were conducted to assess the relationship between NM infusion rate and coagulation parameters, using four modeling approaches: polynomial regression, spline regression, LOESS smoothing, and exponential/logistic fitting. Analyses were performed for the overall cohort (Supplementary Fig. S5), the non-extended LFT group (Supplementary Fig. S6), and the extended LFT group (Supplementary Fig. S7).

In all groups, aPTT increased with NM infusion rate across all non-linear models. For the overall cohort, the polynomial regression for aPTT showed AICs of 3193.784 (degree 2) and 3195.698 (degree 3). Spline regression for aPTT yielded AICs of 3195.698 (df = 3), 3196.933 (df = 4), and 3196.867 (df = 5). LOESS and exponential/logistic models showed consistent increasing trends. In the non-extended LFT group, the lowest AIC for aPTT was observed with spline regression (df = 5, AIC = 1570.565). In the extended LFT group, aPTT modeling resulted in AICs of 1600.101 (polynomial degree 2) and 1602.101 (degree 3); spline regression AICs ranged from 1602.101 to 1604.733. ACT did not show a consistent association with NM infusion rate in any model. Across all figures and regression types, ACT curves were flat or slightly decreasing, with no evidence of statistically meaningful trends.

Filter lifetime and its relationship with NM infusion rate

The filter lifetime was 17.0 (8.0–36.5) hours. Most filters lasted 24 h or less, with progressively fewer remaining functional for longer durations: 20.1% for 24–48 h, 9.4% for 48–72 h, and 8.2% for more than 72 h. A detailed distribution is presented in Supplementary Figure S8. Regarding the relationship between NM infusion rate and filter lifetime, analysis showed no significant correlation between filter duration and NM dose, regardless of whether mean, median, or maximum NM dose per filter episode was considered (Pearson r = 0.0156, P = 0.8644). Plots and further details are provided in the supplementary figures and Supplementary Figure S9.

Discussion

This study was designed to investigate the relationship between NM infusion rate and its anticoagulant effects—assessed by ACT and aPTT—in critically ill pediatric patients undergoing CVVHDF, with additional subgroup analyses conducted based on hepatic function. To our knowledge, this is the first study to systematically explore this relationship in a pediatric CVVHDF population. Through the overall analysis, several noteworthy findings were identified, along with important limitations and areas for future improvement.

One of the key findings of this study was the significant association between NM infusion rate and aPTT, in contrast to the lack of correlation with ACT. According to previous literature, NM exerts its anticoagulant effects independently of antithrombin by directly inhibiting multiple coagulation factors, including IIa, Xa, XIIa, kallikrein, and fibrinolytic enzymes, as well as complement and platelet activation^12,28. Since aPTT reflects the intrinsic and common pathways of coagulation—primarily involving factors such as IIa, IXa, Xa, and XIIa—it is considered an appropriate marker for monitoring the anticoagulant effect of NM. Nonetheless, ACT has traditionally been recommended as the primary monitoring tool¹⁴. Recent studies challenge the reliability of ACT for NM monitoring. Research on ECMO patients revealed ACT does not exclusively reflect anticoagulant effects and may be influenced by patient-specific factors or technical variables²⁹. In CRRT settings, post-filter ACT levels showed poor correlation with bleeding complications, while aPTT-guided NM dosing achieved effective anticoagulation at lower doses (0.41–0.93 mg/kg/h) compared to ACT-based protocols requiring higher doses (1.15–2.19 mg/kg/h)^30,31. A VA-ECMO study demonstrated NM’s regional anticoagulation efficacy through differential aPTT measurements between patient circulation (normal range) and ECMO circuit (prolonged values), whereas ACT failed to show this compartmentalized effect²⁹. These findings align with pharmacological data showing NM’s short half-life (8 min) and targeted factor inhibition are better captured by aPTT’s pathway-specific sensitivity^29,31. Based on a previous in vitro study reporting that ACT was markedly prolonged only when NM concentration exceeded a certain threshold—suggesting a non-linear, concentration-dependent anticoagulant effect¹⁵—we also explored potential non-linear relationships in our data. While our primary analyses employed linear regression models, we conducted additional non-linear regression analyses (Supplementary Fig. S5–S7) to further examine this possibility. Although our study was observational in nature and not designed to identify precise pharmacodynamic thresholds—and this analysis was not part of our primary objectives—we included these models to provide complementary insights. In the non-linear analyses, ACT did not follow a consistent or interpretable trend in relation to NM infusion rate, displaying a relatively irregular pattern. By contrast, aPTT demonstrated a more structured response, with certain models suggesting a steeper slope beyond a potential threshold. These results, while intriguing, should be interpreted with caution, as they are not based on controlled experimental validation and are intended to offer exploratory perspective rather than definitive conclusions.

To investigate the relationship between NM infusion rate and coagulation parameters according to hepatic function, we stratified patients into two groups: the non-extended LFT group and the extended LFT group. Although this classification may not provide the most precise distinction, it was considered a necessary and practical approach given the retrospective nature of the study and the limitations of available data. Specifically, a complete set of LFT values—such as AST, ALT, ALP, TB, DB, and GGT—was not uniformly available across all cases. Conducting regression analyses using only partial data for each LFT parameter would have introduced bias and undermined the validity of inter-variable comparisons. Therefore, we defined the extended LFT group as those patients who had undergone comprehensive LFT assessment, allowing more reliable intra-group regression analyses. In our analysis, ACT did not show a significant association with NM infusion rate in any group. Focusing on aPTT, however, multivariable regression in the overall cohort revealed that both NM infusion rate and TB were significantly associated with aPTT (β = 0.550, SE = 0.246, P = 0.025; Table 3). Furthermore, in subgroup analysis of the non-extended LFT group, TB was the only variable significantly associated with ACT despite NM infusion rate showing no such association (Supplementary Table S3). In the same group, aPTT was significantly correlated with both NM infusion rate and TB in the multivariable analysis (β = 1.031, SE = 0.524, P = 0.049; Supplementary Table S4). In contrast, in the extended LFT group, none of the LFT variables showed significant associations with either ACT or aPTT in multivariable models. This may be attributed to the fact that, although this group was selected to represent patients with potentially impaired hepatic function, it did not necessarily include only those with clinically severe liver dysfunction. Therefore, it is possible that the absence of LFT-related associations and the prominence of NM infusion rate in this group reflect the dominant influence of NM dose over mild-to-moderate variations in liver function.

As part of the analysis, a further evaluation was conducted regarding transfusion history. RBC, platelet, and FFP transfusions administered within 12 h prior to the NM order time were reviewed, as these could have influenced coagulation parameters. Interestingly, ACT remained unassociated with NM infusion rate regardless of transfusion status, whereas aPTT consistently showed statistically significant correlations across all subgroups (Supplementary Figure S5). These findings imply that, even in the presence of recent transfusions that may have affected coagulation status, aPTT retains its clinical value as a monitoring indicator of NM’s anticoagulant effect—unlike ACT.

Additionally, although ALP was significantly associated with aPTT in univariable analysis—which aligns with its role as a marker of cholestasis and potential hepatic synthetic dysfunction—it lost significance in multivariable models, suggesting this finding may be confounded or sample-specific and warrants confirmation in larger, prospective studies (Table 3)^32,33.

While prior studies have shown that adequate anticoagulation with NM can extend filter life^34,35, our modest correlation likely reflects the multifactorial nature of filter patency—including patient factors (e.g., platelet count, fibrinogen), vascular access quality, treatment settings, and membrane characteristics—which were not fully controlled in this retrospective analysis^36,37. These findings suggest that filter longevity is influenced by a broad range of factors beyond NM dosing alone (Supplementary Fig. S9).

Supplementary Table S7 provides key insights into post-NM transfusions and coagulation parameters. Since FFP transfusions are often given when ACT or aPTT are elevated, it is expected to find significant positive associations in most groups. However, ACT was not significantly related to FFP transfusion in the extended LFT group, and aPTT was not significant in the non-extended LFT group, despite a notable positive trend (β = 11.905). The unexpected negative beta for ACT (β = −1.695) in the non-extended LFT group may reflect limited sample size or confounding factors, needing further study. Conversely, RBC and platelet transfusions seem to be less directly affected by coagulation test results compared to FFP. Although elevated ACT or aPTT values may indicate a bleeding risk prompting FFP transfusion, RBC and platelet transfusions are typically guided by clinical evidence of actual bleeding or confirmed decreases in hemoglobin or platelet counts. This distinction reflects differences in transfusion indications for each blood product rather than a simple reflection of coagulation parameters alone.

This study has several limitations. First, as a retrospective analysis, the extent and timing of laboratory testing were not prospectively controlled. Extended LFTs were not uniformly performed across all cases. In our unit, extended LFTs are typically ordered when basic liver tests are abnormal, but the decision is based on clinical judgment rather than strict criteria. As a result, while extended LFTs were commonly performed in relevant cases, the process was not standardized. To address this, patients were stratified into extended and non-extended LFT groups based on available hepatic data. Although this may not fully reflect liver function status, it provided a practical way to reduce bias from missing data and allow valid subgroup comparisons. Second, while the study focused on pediatric patients undergoing CVVHDF, stratification by CVVHDF dose was not performed. This decision was based on the understanding that CVVHDF dosing is largely weight-dependent, and body weight was included as an analytical covariate in the regression models. Given the relatively small sample size, further stratification by dose may have led to overfitting or underpowered comparisons. Third, the study was conducted at a single tertiary care center using a standardized protocol for ACT measurement and a single type of CVVHDF device (Prismaflex^® system). Although this ensured protocol consistency, the findings may not be fully generalizable to other institutions using different equipment or clinical protocols. External validation in broader, multi-center cohorts is warranted to confirm the applicability of these findings across various clinical settings. Fourth, although we analyzed transfusions within 12 h after each NM order time according to our institution’s standard protocol, transfusion thresholds were not strictly enforced in this retrospective study, which may introduce variability in their impact on coagulation parameters. Similarly, CVVHDF dosing generally followed unit guidelines but was not rigidly standardized. These factors reflect inherent limitations of the retrospective design; however, given the absence of major clinical outcome differences, our findings should still be interpreted in the context of real-world practice.

Conclusion

This study demonstrated a consistent and statistically significant association between NM infusion rate and aPTT, whereas ACT showed no such relationship. These findings suggest that aPTT may serve as a more reliable and sensitive marker for monitoring NM’s anticoagulant effect in pediatric CVVHDF. TB was also associated with both ACT and aPTT in some analyses, suggesting a modulatory role of hepatic function on NM pharmacodynamics. However, NM infusion rate remained the dominant determinant of aPTT across subgroup analyses. Taken together, these results call into question the adequacy of ACT-based monitoring and support consideration of aPTT-guided anticoagulation strategies in this clinical context. Further prospective, multicenter studies are needed to validate these findings and optimize monitoring protocols for NM use in critically ill children receiving CVVHDF.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Study concept and design: B.L. Data collection, analysis and cleaning: S.L. and B.L. Interpretation of data: S.L., J.K., and B.L. Drafing and revising the manuscript: S.L., J.K., and B.L. Critical edition: B.L. Supervise the original draft and article: J.K. and B.L. All authors read and approved the final manuscript.

Funding

This research was conducted without any funding or support.

Data availability

The data used in this study is not publicly available. However, if needed, you may contact the corresponding author with valid reasons, and it could be provided upon request.

Declarations

Competing interests

The authors declare no competing interests.