Extracorporeal membrane oxygenation (ECMO) is a life-support technique used in critical care medicine to provide temporary support to the lungs and heart of patients with severe respiratory or cardiac failure (Murphy et al., 2015). Veno-venous (VV) ECMO provides both oxygenation and carbon dioxide removal by draining deoxygenated blood from a vein, oxygenating it externally, and returning it to the patient’s circulation through another vein. Veno-arterial (VA) ECMO, on the other hand, supports both oxygenation and cardiac output by draining deoxygenated blood from a vein and returning oxygenated blood to the patient’s circulation through an artery (Abrams et al., 2014; Gajkowski et al., 2022). The use of ECMO has increased rapidly in recent years due to advances in machinery and the increasing demand. The Extracorporeal Life Support Organization (ELSO) is a non-profit organization founded in 1989 and has been registering patient data on the use and outcomes of ECMO ever since. By 1 July 2016, a total of 78,387 patients had been registered, increasing to 196,108 by 2022 (Thiagarajan et al., 2017; Extracorporeal Life Support Organization, 2023). In 2021, during the height of the coronavirus disease 2019 (COVID-19) pandemic, a total of 21,895 patients were enrolled across 591 centers.

Anticoagulation therapy is a crucial requirement for patients undergoing ECMO due to their heightened risk of thrombotic complications from coagulation pathway activation and blood exposure to foreign surfaces during treatment (Murphy et al., 2015). Anticoagulants, including unfractionated heparin, argatroban, and bivalirudin, are utilized for antithrombotic therapy during ECMO. Therapeutic monitoring of these anticoagulants involves various parameters such as activated clotting time, activated partial thromboplastin time (aPTT), anti-factor Xa level, antithrombin level, and viscoelastic hemostatic assays (Levy et al., 2022; McMichael et al., 2022). Ensuring effective antithrombotic therapy to prevent thrombotic events while minimizing the risk of bleeding poses a significant clinical challenge (Cavayas et al., 2018; Nguyen et al., 2022). ELSO registry data from 2010 to 2017 were analyzed in a study, revealing that 3,044 (40.2%) of 7,570 adult patients undergoing VV-ECMO experienced bleeding or thrombus-related complications. Among these cases, 37% (1,127) had only bleeding events, 41.7% (1,270) had only thrombotic events, and 21.2% (647) experienced both (Nunez et al., 2022). Another analysis of the same registry showed that among 11,984 adult patients on VA-ECMO, 8,457 adverse events related to hemocompatibility were observed. Of these events, 62.1% (5,252) were classified as bleeding events, while 37.9% (3,205) were categorized as thrombotic events (Chung et al., 2020).

Unfractionated heparin (UFH) is widely used as the primary anticoagulant during ECMO or continuous renal replacement therapy (CRRT), given its reliable anticoagulation efficacy and extensive clinical experience (Zarbock et al., 2020; Sadeghipour et al., 2021). Nevertheless, clinical management with UFH remains challenging due to potential adverse effects such as heparin-induced thrombocytopenia, thrombotic complications, bleeding, and heparin resistance, each of which complicates patient management and necessitates careful monitoring and timely intervention. Consequently, clinicians have sought alternative anticoagulants, especially for patients prone to bleeding or those unable to tolerate UFH.

Nafamostat mesylate (NM) is a serine protease inhibitor used as an anticoagulant for ECMO or CRRT in the Republic of Korea, Japan, and China (Han et al., 2011; Baek et al., 2012; Choi et al., 2015; Kamijo et al., 2020; Lee et al., 2022). NM possesses a markedly short systemic half-life of approximately 8 min (Morikawa et al., 1983) and primarily exerts its anticoagulant activity within the ECMO circuit rather than in the systemic circulation. This pharmacokinetic (PK) profile allows targeted anticoagulation within the extracorporeal system, potentially reducing systemic bleeding risks compared to UFH, which has a significantly longer half-life of about 60 min and produces systemic anticoagulation effects. Previous clinical observations indicated that NM might lower the requirements for transfusions and reduce bleeding-related complications while maintaining similar anticoagulant efficacy as UFH within the ECMO circuit (Kotani et al., 2002; Han et al., 2011; Han et al., 2018). Furthermore, NM’s unique profile permits clinicians to selectively minimize systemic anticoagulation, and some studies have suggested that combination therapy using NM alongside low-dose UFH could prevent ECMO circuit thrombosis effectively (Yamagishi et al., 2004; Doi et al., 2020). These promising aspects of NM spurred broader evaluations of its impact on clinical outcomes like bleeding, thrombosis, and filter lifespan. A systematic review of 11 retrospective studies on patients receiving NM during ECMO revealed contrasting outcomes regarding bleeding and thrombotic events (Sanfilippo et al., 2022). Similarly, a retrospective study involving 243 patients on CRRT demonstrated that NM infusion at a rate of 10 mg/h effectively prolonged filter lifespan in high-risk bleeding patients without increasing the need for RBC transfusions or causing significant bleeding events (Baek et al., 2012). Furthermore, a prospective study involving 55 patients receiving CRRT in a high-risk bleeding state found that, among them, 31 patients received NM (NM group), while 24 did not receive anticoagulant therapy (NA group) (Choi et al., 2015). The NM group showed a significantly longer filter lifespan of 31.7 ± 24.1 days compared to the NA group, which had a filter lifespan of 19.5 ± 14.9 days (p = 0.035), while there were no differences observed between the two groups in terms of transfusion frequency and occurrence of bleeding events.

While NM is used for regional anticoagulation during ECMO and CRRT, a notable research gap exists regarding its detailed PK and pharmacodynamics (PD) in this setting. This gap hinders the development of evidence-based dosing strategies informed by PK/PD principles. The aim of this study is to develop PK/PD models using samples from both ECMO circuits and central venous catheters in ECMO patients treated with NM. This involves investigating the relationship between NM concentrations and changes in aPTT. Monitoring aPTT in patients is crucial for minimizing risks of bleeding and preventing thrombosis, while in the ECMO circuit, it specifically aids in preventing thrombus formation. This dual monitoring approach underscores the importance of precise aPTT management in optimizing therapeutic outcomes for ECMO patients.

Our previous study demonstrated the effectiveness and safety of NM as a regional anticoagulant in patients undergoing VA ECMO (Lee et al., 2022). We administered either NM or unfractionated heparin (UFH) and specifically compared the aPTT of blood samples obtained from the patient’s central vein with blood samples drawn from the ECMO circuit. The results revealed no statistically significant difference between the median aPTT of the patient sample (72.84 s) and the median aPTT of the ECMO sample (72.95 s) when UFH was administered. However, upon switching to NM, a significant difference was observed. The median aPTT of the ECMO sample increased to 73.13 s, while the median aPTT of the patient sample decreased to 68.42 s, with a p-value of 0.031. Moreover, when addressing bleeding adverse events, we switched from UFH to NM, resulting in significant improvement in bleeding symptoms for four patients with cannulation site bleeding, one patient with gingival bleeding, and one patient with hematochezia. Based on our previous study, we recognized the necessity for a quantitative analysis of the PK and PD of NM to enhance our understanding of its administration in ECMO patients.

To the best of our knowledge, no clinical studies have yet developed population PK/PD models for NM in patients. We developed and compared two PK/PD models, one utilizing central venous samples from patients and the other utilizing samples from the ECMO circuit. The PK profiles of NM in both sample types were well described by two-compartment models. In our study, the patient model exhibited a t_1/2α of 0.54 min and a t_1/2β of 19.7 min. On the other hand, the ECMO model demonstrated a t_1/2α of 1.2 min and a t_1/2β of 51.4 min. These results were comparable to those of other studies in Asian populations. In a Phase 1 study conducted in Japan, the observed t_1/2α and t_1/2β were 1.1 min and 23.1 min, respectively (Abe et al., 1984). A study involving healthy adults in China reported t_1/2α ranging from 3.65 to 3.78 min and t_1/2β ranging from 112.42 to 129.19 min (Cao et al., 2008). However, a study conducted with dialysis patients revealed a half-life of 8 min for NM (Morikawa et al., 1983). The advantage of remarkably short half-life of NM (approximately 8 min), especially when compared to UFH (60–90 min), argatroban (45 min), and bivalirudin (25 min), has prompted multiple studies to affirm its suitability as an anticoagulant for ECMO or CRRT patients at increased risk of bleeding (Baek et al., 2012; Lee et al., 2014; Park et al., 2015; Lang et al., 2022; Sanfilippo et al., 2022). According to our two-compartment model, following completion of dosing, the concentration declines rapidly due to the extremely short t_1/2α. As a result, even if t_1/2β is prolonged, the concentration declines to very low levels during the elimination phase. However, the steady-state concentration achieved through continuous infusion may significantly differ from the steady-state concentration determined solely by a single half-life of 8 min, depending on the interplay between t_1/2α and t_1/2β.

Our population PK analysis identified gas flow rate as a significant covariate influencing CL in the ECMO model and V2 in the patient model. Although a direct pharmacological interaction is unlikely, the interplay between ECMO circuit dynamics and patient hemodynamics may mediate this relationship. Gas flow rate governs CO₂ elimination from the ECMO circuit (Schmidt et al., 2013; Strassmann et al., 2019). Enhanced CO₂ clearance alleviates hypercapnia, which can alter hepatic perfusion, as suggested by studies showing hypercapnia affects liver blood flow (Barash et al., 2007; Cox et al., 2019). Consequently, improved hepatic blood flow could plausibly influence NM disposition, as the drug is metabolized by esterases in both the liver (carboxylesterase 2) and blood (Nakae and Tajimi, 2003). ECMO circuit dynamics also contribute significantly. In this study, NM was infused pre-pump, resulting in high initial drug concentrations within the circuit. Its moderate lipophilicity (logP ∼2) suggests potential adsorption onto circuit components, similar to other drugs like fentanyl and midazolam (Shekar et al., 2012; Gomez et al., 2022). As gas flow is often adjusted alongside ECMO blood flow in clinical practice (Strassmann et al., 2019), higher flows decrease the drug’s residence time in the circuit. This could reduce the extent of drug sequestration, impacting the observed CL in the ECMO model, a phenomenon noted with other drugs like vancomycin (Shekar et al., 2012). Systemically, the observed reduction in V2 in the patient model may reflect improved circulatory efficiency tied to enhanced gas exchange and correlated increases in blood flow, leading to less peripheral drug distribution (Ficial et al., 2021). However, the effect of gas flow on systemic CL in the patient model was not statistically significant. This is likely because patient CL is influenced by a complex array of systemic factors associated with critical illness (e.g., hepatic function, inflammation), potentially masking the isolated impact of ECMO gas flow (Tonna et al., 2021). Collectively, these plausible mechanistic links support the inclusion of gas flow rate as a relevant covariate, highlighting the intricate relationship between ECMO settings and drug disposition. Further investigation into these mechanisms is warranted. The absence of statistically significant effects on other PK parameters might be attributable to study limitations, such as sample size or confounding variables.

Among various PD models, the final turnover model provided a robust explanation for the relationship between NM concentration and the corresponding change in aPTT. This model effectively captures how NM inhibits the mechanism responsible for the decrease in aPTT. A turnover model is a mechanistic approach used to describe drug-induced indirect responses by elucidating the dynamic equilibrium between response production and response loss, providing insights into the underlying mechanisms altered by the drug (Dayneka et al., 1993; Gabrielsson et al., 2019). In our final PD model, we demonstrated the increase in aPTT by an increase in NM as a mechanism by which NM inhibits the loss of response, i.e., inhibits the decrease in aPTT. To date, there have been no studies that have modeled the relationship between NM exposure and aPTT level. However, the IC50s of our models were comparable to those of Hitomi et al. (Hitomi et al., 1985). In their study, NM, with a molecular weight of 347.37 g/mol, exhibited an IC50 value of 3.0 × 10⁻⁹ M (1.0421 μg/L) for plasma kallikrein inhibition and an IC50 value of 3.3 × 10⁻⁷ M (114.63 μg/L) for inhibiting human Hagmann factor fragment. Additionally, the concentration of NM required to double the aPTT was 5.0 × 10⁻⁷ M (173.69 μg/L).

When Monte Carlo simulations were performed using the final model, the PK profiles of the patient and ECMO models were significantly different, while the PD profiles were not significantly different. In a study conducted in 1972 involving adult patients not on ECMO, maintaining an aPTT within the range of 1.5–2.5 times the normal value was associated with a reduced occurrence of recurrent venous thromboembolic events (Basu et al., 1972). The current clinical recommendations suggest maintaining an aPTT level of 40–80 s during ECMO, which corresponds to 1.5 to 2.5 times the pretherapy baseline level (Sklar et al., 2016; Esper et al., 2017; Koster et al., 2019; Chlebowski et al., 2020; Sanfilippo et al., 2022). However, this recommendation has not been validated in randomized controlled trials or specifically in patients undergoing ECMO therapy (Gajkowski et al., 2022). To attain the target aPTT level, NM was administered in a dose range of 0.14–0.98 mg/kg/h (equivalent to 9.8–68.6 mg/h for a 70 kg weight) (Park et al., 2015), and in another study, at a median dose of 17.7 mg/h (range: 9.8–21.7 mg/h) (Lee et al., 2022). A systematic review of studies involving the use of NM as an anticoagulant in ECMO patients revealed that the mean dose ranged from 0.46 to 0.67 mg/kg/h (equivalent to 32.2–46.9 mg/h for a 70 kg patient) (Sanfilippo et al., 2022). Based on the findings from these studies, we conducted PK/PD simulations using 10 mg/h increments of NM infusion rates ranging from 10 mg/h to 50 mg/h. Consistent with its administration into the drainage pathway upstream of the ECMO pump, steady-state NM concentrations were markedly higher in the ECMO circuit samples compared to the patient systemic samples (Figure 3). While the observed aPTT levels might appear similar under specific conditions (Figure 3), the underlying PD models developed from patient and ECMO samples reveal important differences (Table 3). Specifically, although baseline aPTT and Kin were comparable between the two models, the IC50 was substantially higher in the ECMO model (581 μg/L) than in the patient model (350 μg/L). The Imax was also slightly higher in the ECMO model. This indicates that while the patient’s systemic circulation is inherently more sensitive to anticoagulant effect (lower IC50), it is exposed to much lower drug concentrations due to rapid metabolism and distribution. Conversely, the ECMO circuit requires higher NM concentrations to elicit a similar anticoagulant response due to its higher IC50. This interplay between higher local concentrations within the circuit and distinct local PD characteristics (higher IC50) explains how NM can achieve therapeutic anticoagulation within the ECMO apparatus while limiting excessive systemic aPTT prolongation, aligning with the goal of regional anticoagulation. In some instances, the concentrations of NM in patient and ECMO samples appear comparable, indicating that variability in carboxylesterase-mediated hydrolysis may arise from factors such as genetic variations, regulatory mechanisms at the transcriptional and posttranslational levels, and interactions with other drugs or disease states (Laizure et al., 2013; Wang et al., 2018). Traditional views hold that carboxylesterase activity is consistent among individuals; however, recent studies suggest significant variability due to genetic and environmental influences, analogous to those observed with cytochrome P450 enzymes. This emerging evidence highlights the urgent need for more comprehensive clinical studies to elucidate how carboxylesterase impacts drug metabolism and to determine its influence on the efficacy and safety of treatments (Liu et al., 2024). Similar to the differences observed between arterial and venous blood sampling in PK studies (Schaedeli et al., 2002; Fagiolino et al., 2013), the variation in NM concentrations between ECMO and patient samples can be attributed to their anatomical positions relative to the site of administration and metabolism. While ECMO samples initially reflect prehepatic exposure, the systemic circulation ensures that both compartments are subject to ongoing equilibration and hepatic clearance over time. Therefore, the observed concentration differences are consistent with physiological expectations and provide important insights into the local vs systemic distribution of NM in ECMO patients. Furthermore, additional insights into the impact of drug lipophilicity on absorption in the ECMO circuit have been provided through various studies. It was observed that lipophilic drugs exhibit significant sequestration in the ECMO circuit, with a positive correlation between lipophilicity (log P) and absorption, particularly for drugs like midazolam (0.62% recovery) and fentanyl (0.35% recovery) (Wildschut et al., 2010). Similarly, it was demonstrated that lipophilic drugs such as fentanyl and midazolam experience substantial sequestration within ECMO circuits, leading to necessitating dose adjustments during ECMO therapy (Shekar et al., 2012). With log P of NM reported to be 1.91 or 2.52 (Kim and Kim, 2022), these findings suggest that NM is likely absorbed in the ECMO circuit to a considerable extent, which would reduce the amount available to reach the patient. This may explain the observed discrepancies in NM concentrations between the ECMO circuit and patient samples. Despite these insights, the extent to which a patient’s condition influences the sequestration of NM within the ECMO circuit remains inadequately explored. This sampling site-specific analysis supports the dual clinical goals of NM therapy: preventing clot formation in the ECMO circuit while avoiding systemic over-anticoagulation and bleeding in patients. The evaluation of clinical adverse events supports the PD interpretation of NM as a regional anticoagulant. Notably, patients who experienced bleeding did not demonstrate disproportionately high NM concentrations or prolonged systemic aPTT values. These findings indicate that bleeding events were not directly attributable to NM overexposure or excessive systemic anticoagulation. Rather, they are likely to result from other clinical factors such as procedural trauma or underlying organ pathology (e.g., hemoptysis in lung disease). Furthermore, the absence of any severe bleeding cases strengthens the safety profile of NM, particularly in comparison to systemic anticoagulants like unfractionated heparin. Notably, the significant interindividual variation observed in aPTT levels between ECMO and patient samples within the same patients underscores the potential benefits of personalized treatment strategies that leverage robust models to optimally maintain aPTT levels.

ECMO has been used worldwide, and its frequency of use is increasing due to the COVID-19 pandemic. Despite technological improvement and accumulated clinical experiences, the optimal anticoagulation strategies and monitoring are not well established, and major bleeding remains both the leading cause of mortality in patients with ECMO and the Achilles heel of ECMO. In this study, we demonstrated the efficacy of NM as a regional anticoagulant comparing the PK/PD profiles of NM in both patient and ECMO samples. Additionally, the changes in aPTT level induced by NM were represented by a turnover model, where NM inhibited the decrease in aPTT. However, in real clinical practice, there is a lack of research on the actual correlation between concentration and adverse events of NM use. Considering diverse clinical situations and variables, additional research is needed to optimally adjust and titrate the dose of NM in real practice.

This study has some limitations. First, the limited number of patients hindered the identification of significant covariates, and the sampling number for each patient was insufficient for the development of a robust PD model with good predictive performance. Consequently, large between-subject variability and imprecise parameter estimates were observed. Due to these imprecise estimates, certain parameters had to be fixed to avoid further compromising the stability and reliability of the model. This highlights the need for caution when extrapolating these findings model to broader populations. Second, as shown in the individual fit plots, the model exhibited suboptimal fitting for a small subset of patients. It appeared that a few patients were documented as continuing their medication despite its discontinuation; due to the absence of any justifiable grounds for exclusion or modification of this data, they were retained for the purpose of model development. Third, we were unable to evaluate other PD markers such as activated clotting time (ACT), prothrombin time (PT), anti-factor Xa, and antithrombin activity, in addition to aPTT. Although ACT and PT data were collected, they had significant missing data and could not be used in model development. However, aPTT represents the most frequently recommended test in clinical guidelines and consensus statements. Fourth, although clinically relevant factors such as ECMO configuration (veno-venous vs veno-arterial), hemodilution, and systemic inflammation were considered during the covariate screening process, none of them demonstrated statistically significant associations with PK or PD parameters and were therefore not retained in the final model. This may be due to limited variability in these clinical characteristics or insufficient statistical power related to the sample size. Future studies with larger and more heterogeneous patient populations are needed to more definitively assess the role of these factors.

The PK profiles of NM in both ECMO and patient samples were well described by a two-compartment model. The changes in aPTT level induced by NM were represented by a turnover model, where NM inhibited the decrease in aPTT. Significant interindividual variability was observed in the concentration of NM and its PD effects on aPTT, underscoring the need for models that account for such variability to optimize NM dosing and achieve targeted aPTT levels. By implementing these refined PK/PD models, we might significantly reduce the risks of bleeding and thrombosis in ECMO circuits, thereby enhancing both patient safety and the overall effectiveness of ECMO therapy.