Population pharmacokinetics of human antithrombin concentrate in paediatric patients

Abstract

Aims

Antithrombin is increasingly used in paediatric patients, yet there are few age‐specific pharmacokinetic data to guide dosing. We aimed to describe the pharmacokinetic profile of human (plasma‐derived) antithrombin concentrate in paediatric patients.

Methods

A 5‐year retrospective review was performed of patients <19 years of age admitted to our institution who received antithrombin concentrate, were not on mechanical circulatory support and had baseline (predose) and postdose plasma antithrombin activity levels available for analysis. Demographic and laboratory variables, antithrombin dosing information and data on the use of continuous infusion unfractionated heparin were collected. Population pharmacokinetic analysis was performed with bootstrap analysis. The model developed was tested against a validation dataset from a cohort of similar patients, and a predictive value was calculated.

Results

A total 184 patients met the study criteria {46.7% male, median age [years] 0.35 [interquartile range (IQR) 0.07–3.9]}. A median of two antithrombin doses (IQR 1–4) were given to patients (at a dose of 46.3 ± 13.6 units kg^–1), with median of three (IQR 2–7) postdose levels per patient. Continuous infusion unfractionated heparin was administered in 87.5% of patients, at a mean dose of 34.1 ± 22.7 units kg^–1 h^–1. A one‐compartment exponential error model best fit the data, and significant covariates included allometrically scaled weight on clearance and volume of distribution, unfractionated heparin dose on clearance, and baseline antithrombin activity level on volume of distribution. The model resulted in a median −1.75% prediction error (IQR –11.75% to 6.5%) when applied to the validation dataset (n = 30).

Conclusions

Antithrombin pharmacokinetics are significantly influenced by the concurrent use of unfractionated heparin and baseline antithrombin activity.

What is Already Known about this Subject

• Human (plasma‐derived) antithrombin concentrate is administered to paediatric patients using a variety of dosing strategies, often extrapolated from experience with adults.

• There are currently no substantive data describing antithrombin pharmacokinetics in paediatric patients.

What this Study Adds

• A reliable pharmacokinetic model of human antithrombin concentrate in paediatric patients has been developed and validated.

• Concurrent dosing with unfractionated heparin and predose antithrombin activity level are significant pharmacokinetic covariates that appear to influence antithrombin clearance and volume of distribution, respectively.

Introduction

Antithrombin (AT) is a serine protease inhibitor that regulates the intrinsic, extrinsic and common clotting pathways through inactivation of multiple coagulation factors, including factor Xa and thrombin 1. AT has two main functional domains which bind to proteases and to heparin. The protease binding site results in an inactivation of thrombin. The rate of such inactivation is relatively slow but it is catalysed greatly by heparin binding, which induces a conformational change in AT 2. Anticoagulation with heparin thus depends on the adequate presence of AT in the plasma. Infants and children have been reported to have lower endogenous concentrations of AT III and thus often require higher doses of unfractionated heparin (UFH) to achieve therapeutic anticoagulation 3.

As anticoagulation needs have expanded, the use of AT concentrate in paediatric patients has increased over the past several years, in a wide variety of disease states 4, 5, 6, 7, 8, 9, 10, 11, 12, 13. Supplementation of AT has been reported in paediatric patients with cancer, intracranial haemorrhage, sepsis and mechanical circulatory support. However, the dosing of AT in paediatric patients has been based on adult experience, and a wide range of strategies have been employed to attempt to achieve therapeutic outcomes 4, 10, 14. The current dosing of AT in paediatric patients has been extrapolated from adult guidelines. The published dosing recommendations from the manufacturer are as follows: [desired AT level (%) – baseline AT level (%)] × body weight (kg)] divided by 1.4 = units of AT required 15. As this formula was based on data from adults, paediatric dosing has been often reported in units kg^–1 per dose for initial doses, and subsequent doses adjusted based on patient response 5, 6, 10, 16, 17. Formal assessments of AT in paediatric patients have focused on attainment of goal concentrations but have never evaluated its pharmacokinetics or covariates which could potentially be important for optimizing dosing strategies. The characterization of AT pharmacokinetics in this patient population could improve empirical and subsequent dosing and monitoring in paediatric patients.

To improve the dosing of AT we sought to characterize the pharmacokinetics of human‐derived AT in paediatric patients and to identify clinically relevant covariates that alter antithrombin pharmacokinetics in these patients.

Methods

A retrospective cohort study design was chosen, and institutional review board approval from Baylor College of Medicine and affiliated institutions was obtained. Patients who received human (plasma‐derived) AT concentrate (Thrombate®, Grifols Inc., Research Triangle Park, North Carolina, United States) were identified by querying the Texas Children's Hospital pharmacy database from 1 January 2011 to 31 December 2015. Patients were included if they met the following criteria: less than 19 years of age, inpatients during the study period, received human AT concentrate, had a baseline (predose) AT activity level and had one or more postdose AT activity levels. Patients were excluded if they were undergoing mechanical circulatory support during AT administration or monitoring.

Data collection included: patient age, gender, height, weight, gestational age, indication for AT supplementation, AT dose and activity levels, concomitant dose of continuous infusion UFH, serum albumin level, serum creatinine level, haemoglobin level, haematocrit, platelet count and postmenstrual age (PMA) in weeks (gestational age plus day of life, calculated at the time of AT administration).

Samples for AT activity levels were collected in sodium citrate tubes and measured on the STA‐R analyser using the STA – Stachrom AT kit (Diagnostica Stago, Asnières, France) at our institution. Plasma was incubated with a known excess of thrombin in the presence of heparin. The assay uses a 1:20 dilution of sample to buffer. The coefficient of variation for the test was 9.35%. AT activity levels were reported as the percentage of normal, which is equivalent to international units per decilitre (units dl^–1). The normal values of AT in paediatric patients are not well defined, and range from 85–130% in paediatric patients, with decreases in patients under 3 months of age 18. In our clinical experience, a goal of 120% is often recommended, although this varies for each clinical scenario.

Descriptive analysis

The patient population was described using percentages, mean, standard deviation, median and range, for normally and non‐normally distributed data as appropriate. Graphical representation of the initial data was also utilized. Statistical analyses were performed using Stata IC v.12 (StataCorp, College Station, TX, USA), and Excel 2013 (Microsoft Corp, Redmond, WA, USA).

Data handling

AT levels were often monitored in the patient population as surveillance to determine the need for AT supplementation. To include levels appropriate for analysis, those included in the dataset consisted of baseline levels, the first level after the dose of AT, and any subsequent levels that were decreased in value after the first level. Baseline activity levels were included as a covariate, and were included in the error model to account for the endogenous production of AT, according to the B3 method by Dansirikul et al. 19.

Model building

Population pharmacokinetic analysis was performed using NONMEM v.7.2 (Icon, PLC; Dublin, Ireland) and PDx‐Pop 5.1 (Icon, PLC) using first‐order conditional estimation with interaction (FOCE‐I). Interindividual variability (IIV) was modelled exponentially.

The visual inspection of AT activity levels plotted against time after dose (TAD) were used to help to determine the base model. One‐ and two‐compartment models, along with additive, proportional, exponential and combined proportional/additive error models, were employed and evaluated. After selection of the base model, weight and allometrically scaled weight, with an exponent of 0.75 on clearance (CL) and 1 on volume of distribution (VD), were added to the base model and evaluated for goodness of fit.

Covariate analysis

Initial covariate analysis occurred by plotting empirical estimates of pharmacokinetic parameters against clinically or pharmacologically relevant covariates and visually identifying potential relationships. The covariate with greatest clinical significance was selected whenever two covariates were strongly correlated (e.g. height and weight). Covariates were selected based upon the investigators’ clinical experience and review of the pharmacology and relevant literature associated with AT. The primary rationale for the inclusion/exclusion of covariates was based upon stepwise addition/elimination of covariates from the model with allometrically scaled weight and change in the objective function value (OFV).

Relationships between continuous covariates and pharmacokinetic parameters were modelled using linear, linear proportional and power models, with the covariate normalized to the population median, and categorical covariates were modelled exponentially.

Model selection

A stepwise approach to model building was used, with a reduction of the OFV by >3.84 considered to be statistically significant (P < 0.05) as each covariate was added, and then a stepwise backward elimination for each covariate was performed with an increase of OFV from the full model of at least 10.83 (P < 0.001) as the criterion for covariate retention (see Appendix). Scatter plots of dependent variables (DVs) vs. individual predicted (IPRE) AT activity levels, as well as DV vs. predicted (PRED), and PRED vs. conditional weighted residuals (CWRES), were created to evaluate the model goodness of fit.

Model evaluation

Bootstrap simulations were performed (n = 1000) on the final model, with calculation of 95% confidence intervals. A simulation based upon median patient values from the dataset was performed with 10 000 replications. A validation dataset was developed for patients meeting the same study criteria during the 6 months after the retrospective review period (1 January 2016 to 30 June 2016). The developed model was evaluated for absolute and relative prediction error for the first AT level determined after an AT dose was given for each patient in the validation cohort (Figure 1).

Figure 1.

Formulae for calculating prediction errors

Results

A total of 184 patients met the study criteria (Table 1). The breakdown of age groups was: neonates ≤30 days (n = 50), infants 31 days to 2 years (n = 85), children 3–12 years (n = 30) and adolescents 13–18 years (n = 19). Baseline laboratory values were summarized (Table 2). The majority of patients (87.5%) received AT for decreased AT activity levels during continuous infusion UFH, which was administered at a mean dose of 34.1 ± 22.7 units kg^–1 h^–1. Other scenarios included coagulopathy (8.2%) and enoxaparin therapy (4.3%). Patients received a median of two AT doses [interquartile range (IQR) 1–4] at a mean of 46.3 ± 13.6 units kg^–1 of AT per dose, infused over 15 min. Patients had a median of three (IQR 2–7) AT activity levels analysed, at a median 20.8 h (IQR 12.4–32.1) after a dose, and the mean AT activity level after a dose was 85.9 ± 20.7 units dl^–1.

Table 1.

Demographic variables

Category (n = 184	Value
Male (%)	46.7
Age (years) (median, IQR)	0.35 (0.07–3.9)
Gestational age (weeks, days) (median, IQR)	39 (37 1/7–40 0/7)
Postconceptional age (weeks) (median, IQR)	55 (40–244)
Height (cm) (median, IQR)	58 (50–95)
Weight (kg) (median, IQR)	5.2 (3.4–14.8)

IQR, interquartile range

Table 2.

Baseline laboratory values for antithrombin patients

Category (n = 184)	Value
Baseline antithrombin activity level (units dl –1 ) (mean, standard deviation)	59 ± 17
Albumin (g dl –1 ) (mean, standard deviation)	2.8 ± 0.6
Haematocrit (%) (mean, standard deviation)	33.2 ± 5.5
Haemoglobin (mg dl –1 ) (mean, standard deviation)	11.2 ± 1.9
Platelet count (mean, standard deviation)	234 ± 122
Serum creatinine (mg dl –1 ) (mean, standard deviation)	0.49 ± 0.45

Pharmacokinetic modelling

A one‐compartment exponential error model best fit the data, with a reduction of −303.239 in the OFV when allometrically scaled weight was added to CL and VD, which was a greater reduction than weight alone. Based upon principles of theoretical allometry, allometrically scaled weight was used for the primary model upon which covariates were applied (see Appendix). Variables with a significant reduction in OFV when applied to CL included baseline activity level, UFH dose, serum albumin and age in years. Covariates that resulted in significant reductions in OFV when applied to VD included age in years, baseline activity level, platelet count and UFH dose (see Appendix).

The variables with significant increases in OFV after backwards elimination included only UFH dose on CL, and baseline activity level on CL and VD. In the model implemented after the backwards elimination step, the effect of baseline activity level on both CL and VD was included but this resulted in unacceptable estimates of IIV. When the effects of baseline activity level on only CL or VD was included, the model of the effects of baseline activity level on VD, when analysed graphically, did not have appreciable differences from the other models with baseline activity level as a covariate, and had an acceptable IIV, resulting in a final model of the effects of baseline activity level on VD only (Table 3). The overall final model had a good fit when evaluating the DV vs. PRED, and DV vs. IPRED graphs, and no bias was noted in the CWRES vs. PRED graph (Figures 2, 3, 4). Bootstrap analysis with 1000 replications was performed on the final model (Table 4).

Table 3.

Final model

Model	Interindividual variability (%)	Residual variability (%)
$\mathit{CL}={0.917}^{*}{\left(\frac{\mathit{WT}}{70}\right)}^{.75}+{0.129}^{*}\left(\frac{\mathit{UFH}}{173}\right)$	47.5	11.1%
$\mathit{VD}={67.9}^{*}{\left(\frac{\mathit{WT}}{70}\right)}^{*}{\left(\frac{\mathit{\text{BASE}}}{60}\right)}^{-0.389}$	28.5

BASE, baseline antithrombin concentration (units dl^–1); CL, clearance (dl h^–1); UFH, continuous infusion unfractionated heparin (units h^–1); VD, volume of distribution (dl); WT, weight (kg)

Figure 2.

Antithrombin concentrations (DV) versus predicted concentrations (PRED). There is consistency around the line of unity

Figure 3.

Antithrombin concentrations (DV) versus individual predicted concentrations (IPRE). There is consistency around the line of unity

Figure 4.

Conditional weighted residuals (CWRES) versus predicted concentrations (PRED). No apparent bias is noted in the figure

Table 4.

Bootstrap analysis

n = 184	One‐compartment model		1000 Bootstrap
Parameter	Estimate (RSE%)	95% CI	Median (RSE%)	95% CI
CL (dl h –1 )	0.917 (16.8)	0.615, 1.219	0.957 (20.7)	0.664, 1.410
VD (dl)	67.9 (4.9)	61.41, 74.39	67.9 (5.5)	60.8, 75.7
UFH (units h –1 )	0.129 (15.5)	0.083, 0.174	0.123 (27.9)	0.066, 0.165
BASE (units dl –1 )	−0.389 (37.5)	−0.675, −0.103	−0.388 (47.4)	−0.724, −0.119
ω 1 (%)	47.5 (26.5)	31.9, 59.0	47.1 (41.4)	31.4, 60.1
ω 2 (%)	28.2 (31.2)	17.6, 36.2	28.2 (48.8)	13.7, 37.2
Exponential error (%)	11.1 (26.9)	7.1, 13.4	10.9 (32.3)	7.1, 14.5

BASE, baseline antithrombin (units dl^–1); CI, confidence interval; CL, clearance (dl h^–1); RSE%, relative standard error (%); UFH, unfractionated heparin (units h^–1); VD, volume of distribution (dl); ω1 and ω2, Interindividual variability

According to the developed model, the median patient in our cohort (with a weight of 5.2 kg), with a baseline AT level of 60 units dl^–1, receiving an UFH dose of 34 units kg^–1 h^–1, would require a 60 units kg^–1 dose of human AT to achieve a postdose AT level of 120 units dl^–1 an hour after initiation of infusion. The calculated half‐life for the median patient in our dataset was ~22 h. When the calculations are performed again with a UFH dose of 0, the calculated half‐life is extended to ~27 h. Based on this information, a simulation was performed (n =10 000) with median patient values (weight = 5.2 kg, baseline AT 60 units dl^–1) for a patient receiving 34 units kg^–1 h^–1 of UFH and for a patient receiving 0 units kg^–1 h^–1. Values for CL, VD and half‐life were reported, as well as predicted values at 1 h, 12 h and 24 h post‐infusion (Table 5).

Table 5.

Simulated pharmacokinetic parameters and predicted concentrations

Category	UFH (34 units kg–1 h–1)	No UFH
CL (ml kg –1 h –1 ) (mean, standard deviation)	5.7 ± 2.8	2.8 ± 1.4
VD (ml kg –1 ) (mean, standard deviation)	101.0 ± 29.3	101.2 ± 29.9
Half‐life (h)	15.5 ± 9.3	31.6 ± 19.7
Predicted concentrations (units dl –1 ) post‐infusion (mean, standard deviation)
1 h	120 ± 19	120 ± 20
12 h	92 ± 14	105 ± 15
24 h	78 ± 13	92 ± 14

CL, clearance; UFH, unfractionated heparin; VD, volume of distribution

A validation cohort from the 6‐month period after the model building period was identified based on the same study criteria as for the model building cohort (n = 30) (Table 6). The first AT activity level was sampled at a median of 8.8 h (IQR 2.6–21.4 h) after a dose and was 90.1 ± 21.2 units ml^–1. The model, when applied to the validation dataset, resulted in a median −1.75% relative prediction error (IQR –11.75% to 6.5%) and median 10.1% absolute prediction error (IQR 4.8–19.6%) for the first AT activity level after a dose.

Table 6.

Validation dataset demographics

Category (n = 30)	Value
Male (%)	33.3
Age (years) (median, IQR)	0.4 (0.18–4.1)
Weight (kg) (median, IQR)	5.2 (3.4–14.8)
Baseline antithrombin activity level (units dl –1 )	60.8 ± 14.3
Continuous infusion UFH (%)	86.7
Continuous infusion UFH dose (units kg –1 h –1 )	25.2 ± 8.6

IQR, interquartile range; UFH, unfractionated heparin

Discussion

This was the first population pharmacokinetic analysis of human AT concentrate in the paediatric population. Our results are consistent with our clinical experience and current published practice 16. In general, our institution has used doses of approximately 50 units kg^–1 as a dosing strategy, with modifications dependent on baseline AT activity levels and goals of therapy, which is similar to the calculated results from the model.

Our report characterizes the pharmacokinetic parameters of AT in paediatric patients, and these data will be useful to guide dosing strategies. The afore‐mentioned dose of 60 units kg^–1 is a suggestion, as variable dosing strategies have been presented in the literature in paediatric patients. Doses over 200 units kg^–1 per dose have been reported in patients undergoing extracorporeal membrane oxygenation 13. Often, at our institution, the dose of AT is rounded to the nearest complete vial, to minimize waste, owing to the short sterility validity period of a vial of AT, resulting in somewhat variable dosing. The goals of therapy will dictate the dose used, and the attainment of a particular AT level is not always the primary endpoint 10.

The identification of UFH dose as a significant covariate was not surprising, based upon our own clinical experience and our prior clinical investigations 16. UFH potentiates the anticoagulant effect of AT, and high doses and prolonged therapy with UFH can lead to lower AT levels 20, 21. The relationship between UFH and AT was linear, demonstrating increased clearance with increasing UFH doses. The values obtained in our dataset were similar to those reported previously 17, 22, 23. Regular monitoring of AT levels should occur when UFH is administered concurrently.

The baseline, predose, value of AT as a covariate significantly affected VD, and increased baseline values were associated with decreased VD. In critically ill adults, AT has been reported to stay primarily in the plasma, suggesting that other physiological binding sites are unlikely 22. Other investigations have reported relationships between serum albumin and AT concentrations, due to increases in vascular permeability in patients with disseminated intravascular coagulopathy, but we did not find any relationship when the role of serum albumin was investigated 24. We can hypothesize that patients with lower baseline AT activity levels had increased consumption of AT, due to either a physiological process or high doses of UFH, thus resulting in a diminished rise in AT activity levels after a dose, accounting for the larger VD. However, this is a hypothesis, and further investigation is warranted. Our model incorporates the baseline value of AT, and clinicians should be aware of the apparent inverse relationship between VD and baseline AT activity levels.

The biases and concerns associated with retrospective reviews are inherent in this publication, and a controlled pharmacokinetic investigation are likely to elucidate estimates with less error. Additional future investigations include the characterization of AT in patients undergoing mechanical circulatory support. Anticoagulation strategies are often complex in patients undergoing mechanical circulatory support, and AT administration occurs frequently in the paediatric mechanical circulatory support population 7, 9, 12, 13.

Conclusion

AT pharmacokinetics in paediatric patients are significantly influenced by continuous infusion UFH and baseline AT activity levels.

Competing Interests

There are no competing interests to declare.

Category	OFV	Reduction from base model
Base model	7906.342
Total weight	7653.456	−252.886
Allometric weight	7603.049	−303.293

Covariates on clearance (CL)	OFV	Reduction from allometric weight model
*𝚯^M	7603.04	−0.009
*𝚯^(AGEYRS/3.2)	7597.461	−5.588
*(PCA/59)^𝚯	7597.377	−5.672
*(((PCA/59)^𝚯/((PCA/59)^𝚯 + (375)^𝚯))	7598.514	−4.535
*𝚯^(SCR/0.34)	7599.596	−3.453
*𝚯^(HGB/11.4)	7602.004	−1.045
*𝚯^(HCT/34)	7601.93	−1.119
*𝚯^(ALB/2.9)	7582.477	−20.572
*(PLT/242)^𝚯	7595.395	−7.654
+(𝚯*(UFH/173))	7524.701	−78.348
*(1 + 𝚯^((UFH‐173)/173))	7601.847	−1.202
*((1 + UFH)/173)^𝚯	7549.449	−53.6
*𝚯^((1 + UFH)/173)	7601.592	−1.457
*(BASE/60)^𝚯	7499.852	−103.197
+(𝚯*(BASE/60))	7591.612	−11.437

Covariates on volume of distribution (VD)	OFV	Reduction from allometric weight model
*𝚯^M	7602.93	−0.119
*𝚯^(AGEYRS/3.2)	7591.537	−11.512
*(PCA/59)^𝚯	7591.258	−11.791
*(((PCA/59)^𝚯/((PCA/59)^𝚯 + (375)^𝚯))	7594.903	−8.146
*𝚯^(SCR/0.34)	7601.487	−1.562
*𝚯^(HGB/11.4)	7602.401	−0.648
*𝚯^(HCT/34)	7601.729	−1.32
*𝚯^(ALB/2.9)	7602.428	−0.621
*𝚯^(PLT/242)	7596.276	−6.773
*((1 + UFH)/173)^𝚯	7601.691	−1.358
+(𝚯*(UFH/173))	7601.94	−1.109
*(1 + 𝚯^((UFH‐173)/173))	7598.025	−5.024
*(BASE/60)^𝚯	7593.372	−9.677

AGEYRS, age in years; ALB, albumin; BASE, baseline antithrombin; HCT, haematocrit; HGB, haemoglobin; M, Male; OFV, objective function value; PCA, postconceptional age; PLT, platelet; SCR, serum creatinine; UFH, unfractionated heparin

Moffett, B. S. , Diaz, R. , Galati, M. , Mahoney, D. , Teruya, J. , and Yee, D. L. (2017) Population pharmacokinetics of human antithrombin concentrate in paediatric patients. Br J Clin Pharmacol, 83: 2450–2457. doi: 10.1111/bcp.13359.