A systematic review of population pharmacokinetic analyses of digoxin in the paediatric population

Abstract

This is a PROSPERO registered systematic review (CRD42018105207), conducted to summarize the available knowledge regarding the population pharmacokinetics of digoxin in paediatrics and to identify the sources of variability in its disposition. PubMed, ISI Web of Science, SCOPUS and Science Direct databases were searched from inception to January 2019. All paediatric population pharmacokinetic studies of digoxin that utilized the nonlinear mixed‐effect modelling approach were incorporated in this review, and data were synthesized descriptively. After application of the inclusion–exclusion criteria 8 studies were included. Most studies described digoxin pharmacokinetics as a 1‐compartment model with only 1 study describing its pharmacokinetics as 2‐compartments. Age was an important predictor of clearance in studies involving neonates or infants, other predictors of clearance were weight, height, serum creatinine, coadministration of spironolactone and presence of congestive heart failure. Congestive heart failure was also associated with an increased volume of distribution in 1 study. The estimated value of apparent clearance in a typical individual standardized by mean weight ranged between 0.24 and 0.56 L/h/kg, the interindividual variability in clearance ranged between 7.0 and 35.1%. Half of the studies evaluated the performance of their developed models via external evaluation. In conclusion, substantial predictors of digoxin pharmacokinetics in the paediatric population in addition to model characteristics and evaluation techniques are presented. For clinicians, clearance could be predicted using age especially in neonates or infants, weight, height, serum creatinine, coadministration of medications and disease status. For future researchers, designing pharmacokinetic studies that allow 2‐compartment modelling and linking pharmacokinetics with pharmacodynamics is recommended.

What is already known about this subject

• Digoxin is a drug known to have a narrow therapeutic index and high pharmacokinetic variability.

• Few population pharmacokinetics models were developed in the paediatric population, but no systematic literature reviews concerning pharmacokinetic variability of digoxin has been reported in this population.

What this study adds

• This review systematically summarizes the knowledge pertaining to population pharmacokinetics of digoxin in the paediatric population in a quantitative and a qualitative manner.

• Age was an important predictor of clearance in studies involving neonates or infants, other predictors of clearance in paediatrics were weight, height, serum creatinine, coadministration of spironolactone and presence of congestive heart failure.

• Designing pharmacokinetic studies that allow 2‐compartment modelling and linking pharmacokinetics with pharmacodynamics is recommended.

1. INTRODUCTION

Digoxin is a cardiac glycoside that is considered to be the oldest drug used in cardiovascular medicine. ¹ In the myocardium, the drug exhibits both positive inotropic and negative chronotropic effects, positive inotropic effects are related to inhibition of the sodium/potassium ATPase pump in myocardiocytes, an action that results in a transitory surge of intracellular sodium, which subsequently enhances calcium influx through the sodium–calcium exchanger, leading to a more forceful contraction in the course of excitation–contraction coupling. Elevated intracellular calcium levels prolong phase IV and phase 0 of the cardiac action potential, subsequently reducing the heart rate. ² , ³ In general, these molecular events are also responsible for its toxicity, which includes for example generation of delayed afterdepolarizations that can, in turn, provoke polymorphic ventricular tachycardia. ²

In paediatrics, digoxin is mainly indicated for children with symptomatic heart failure, as based on adult data it is not recommended for children with asymptomatic left ventricular dysfunction, because no survival benefit was observed in adults with heart failure and low ejection fraction. ⁴ It is also indicated for controlling tachyarrhythmias, including fetal tachyarrhythmia. ³ The position of digoxin in the paediatric cardiologist's armamentarium continues to be ambiguous. ⁵ The prescription of digoxin by paediatric cardiologist has declined recently ⁶ probably due to the digoxin in heart failure (DIG) trial in adults, which exhibited that digoxin lowered hospital admissions for heart failure by 28% with no effect on mortality, ⁷ and the general alternation in paradigm in heart failure pathophysiology that resulted in transference from inotropic agents to neurohormonal modulating agents, which made the benefits achieved by the DIG trial obtainable by other means. ² , ⁸ Other possible reasons are concerns regarding drug–drug interactions and toxicity. Possible interactions include interactions with amiodarone, quinidine and antihypertensive agents. The description and aetiology of digoxin toxicity is limited in paediatrics as opposed to adults where it is well documented. ⁹ Possible toxicities reported in infants and older children include atrioventricular block, nausea, vomiting, anorexia and, very rarely, ectopy. ⁹ , ¹⁰

Nevertheless, digoxin is still used in the paediatric setting. Hsieh et al. ¹¹ reviewed the medication administration data relating to 450 386 infants discharged from 305 neonatal intensive care units in the USA during 5 years and reported that digoxin was from the top 100 medications used in this population. Furthermore, relatively recent studies indicate that digoxin might be useful in reducing interstage mortality in infants with single ventricle congenital heart disease. ⁶ , ¹² Its use was also associated with absence of significant toxicities, when used in the single ventricular reconstruction trial, ⁶ or when used prophylactically in infants with supraventricular tachycardia. ¹³

Due to its narrow therapeutic index and high interindividual variability (IIV), there is an immense need to optimize its use through appropriate pharmacokinetic methods. Routinely, therapeutic drug monitoring (TDM) is employed to guide its use. Using routine TDM in addition to appropriate pharmacokinetic methods that take into consideration different sources of variability in the drug's pharmacokinetics including possible patient‐specific factors such as body weight, renal function or age are important for individualizing therapy in different patient subpopulations. An important subpopulation to consider is the paediatric population. Determining the appropriate dose in the paediatric population is difficult due to the ethical and logistical constraints relating to conducting traditional pharmacokinetic studies in this fragile population. ¹⁴ For this reason, there are many drugs whose doses in the clinical practice are calculated through extrapolation from adult doses on basis of weight or body surface area, the use of digoxin in neonates and infants, for instance, has been historically based on such methods. ¹⁵ However, such a simplified approach may be useful to initiate therapy but is inadequate for maintaining long‐term therapy as developmental changes need to be taken into account when designing an appropriate individualized dosage regimen. ¹⁶ Rapid developmental changes that occur during childhood can have a profound impact on the pharmacokinetics of drugs and hence influencing the degree of exposure over time, which has an important reflection on the safety, efficacy or toxicity of a selected dose. ¹⁷ For instance, developmental differences in the rate of gastric emptying and in the expression and activity of intestinal enzymes can profoundly affect the rate and extent of absorption. Age‐related changes in body composition and changes in the type and quantity of circulating proteins can alter the distribution of certain types of drugs. Drug elimination can depend on certain hepatic enzymes; lack of maturation of these metabolizing enzymes can result in toxicity. Furthermore, the serum concentration of drugs that are extensively eliminated through the kidney is expected to be influenced by developmental changes in the renal function. ¹⁶

In clinical practice, carefully chosen pharmacokinetic models in the target population can be incorporated into Bayesian estimation programs as a priori combined with the current drug concentration to obtain optimal dose regimen. ¹⁸ Thus, the present systematic review aimed to thoroughly summarize the available literature on paediatric population pharmacokinetic models of digoxin and to identify the significant covariates that have an impact on its pharmacokinetics, in addition to portraying the quality of reporting and evaluation of the developed models. This would not only enable the reader to evaluate their quality and specify their usefulness in clinical practice but also aids in pinpointing possible directions of future research.

2. METHODOLOGY

2.1. Registration

The protocol of the present systematic review was registered with the PROSPERO International Prospective Register of Systematic Reviews in 2018 and assigned the identifier CRD42018105207. The full record is available from www.crd.york.ac.uk/Prospero.

2.2. Compliance with the PRISMA principles

In the current analysis, we utilized the Preferred Reporting Items for Systematic Reviews and Meta‐analyses (PRISMA) principles to optimize the review process. The PRISMA statement involves a flow diagram and a checklist encompassing 27 items; those items are considered necessary for transparent reporting of systematic reviews. The PRISMA checklist is provided in the supplementary material.

2.3. Inclusion/exclusion criteria

Pharmacokinetic studies of digoxin were eligible for inclusion if they met the following criteria:

1. Study population: human studies (patients)

2. Age group: neonates, infants, children aged ≤16 years

3. Treatment: receiving digoxin as a treatment drug

4. Analysis method: population pharmacokinetic analysis via a nonlinear mixed‐effect methodology

Studies were excluded if they met at least 1 of the following criteria:

1. Pharmacokinetics in animal studies

2. Pharmacokinetic studies conducted on healthy volunteers

3. Population pharmacokinetic studies on pregnant women or adults

4. Papers focused on laboratory analysis and/or pharmacokinetic modelling, simulation techniques, in‐vitro models or physiological models

5. Pharmacokinetic related studies in a language other than English

6. Review papers (pharmacokinetic)

7. Approaches other than population pharmacokinetics or nonlinear mixed‐effects modelling.

2.4. Search strategy

The following search term was utilized ((“digoxin” OR “dig^*”) AND (“population pharmacokinetic^*” OR “pharmacokinetic model^*” OR “nonlinear mixed effect^*” OR “NONMEM”)) to identify population pharmacokinetic studies of digoxin through the databases PubMed, ISI Web of Science, SCOPUS and Science Direct (all resources excluding book chapters, other [which mainly include indexes], encyclopaedias, conference abstracts, conference information, correspondence, discussions, editorials, short communications, proceedings papers) from inception until the day the search was finalized (January 2019).

2.5. Data extraction

The following data were extracted from the included studies by 2 researchers independently:

1. Study characteristics: number of patients, study location, study type, demographic data (e.g. race, weight, age), digoxin dosage regimen, formulation type, sampling design (e.g. number of samples per patient and sampling time), digoxin assay used and important clinical data (e.g. serum creatinine).

2. Population pharmacokinetic analysis: type of the utilized structural model, type of the statistical models used to describe the IIV and the residual variability, the estimation method used, the structural and statistical parameter estimates, covariates tested and covariates retained in the final model and the criteria for their inclusion, and finally the techniques employed to evaluate the final model.

3. Pharmacokinetic parameters: we calculated the typical value of clearance in each study, in a typical child, by scaling clearance by the mean population weight. For each model we used the mean/median value of continuous covariates included in the final model, while binary covariates were assumed zero, apart from heart failure where clearance was calculated either with or without heart failure in 3 studies. These final models of typical individuals were used to simulate individual pharmacokinetic parameters for 1 000 000 subjects. Simulation was conducted by using R software (version 3.4.3; http://cran.r-project.org). A lognormal distribution was assumed for IIV. The simulated parameters were used to calculate the median, 10^th percentile and 90^th percentile of simulated pharmacokinetic parameters and the results for simulations were graphically summarized. The authors used the tables presented in this review to extract and summarize the data.

2.6. Summery measures

The main summary measures are the typical value of pharmacokinetic parameters in the paediatric population, influential covariates and the characteristics of the included models.

2.7. Quality analysis

To assess the reporting quality of the included studies we employed the checklist developed by Kanji et al. ¹⁹ As this list's scope is pharmacokinetics in general with only 1 item addressing population pharmacokinetics specifically, we incorporated additional items from the literature. Those items were mainly obtained from the limited list of items to be reported in population pharmacokinetic studies created by Dartois et al. ²⁰ which focused on model building strategies. The additional items are presented in the same table as the clinical PK checklist for practical reasons.

In addition to the quality of reporting in each study, we assessed the quality of validation of the developed models. As described by Brendel et al. ²¹ models can be evaluated via 3 general methods of increasing complexity and quality: (i) basic internal methods; (ii) advanced internal methods; and (iii) external methods. We summarized the model evaluation method used in each of the reviewed papers. In addition, we made a subjective judgment about the model evaluation quality by answering the following question “Was the model evaluated?” as described by Brendel et al. ²¹ In short, 3 authors independently assessed how validation was conducted assuming predictive purposes for all models and based on the type of validation, they answered the following question “Was the model evaluated?” with poor when only basic internal methods were executed, good if advanced internal methods with metrics were utilized, excellent when external model evaluation with metrics was conducted or no if no data were found about a specific item. Models were not excluded from the present review based on the quality of their evaluation.

3. RESULTS

3.1. Study identification

Totals of 62, 158, 344 and 403 studies were identified from PubMed, Web of Science, Scopus and Science Direct searches, respectively. Following duplicate removal, title and/or abstract screening ensued, and 68 studies were eligible for full‐text screening. Of these, 60 studies were excluded; reasons for exclusion are presented in the PRISMA diagram (Figure S1).

3.2. Characteristics of included studies

The 8 studies investigated in the present review were published between 2001 and 2014, they encompassed the following paediatric subpopulations: 1 study was conducted in neonates, ²² 3 included neonates and infants, ¹⁵ , ²³ , ²⁴ 1 in infants and young children (<5 years), ²⁵ 2 included infants/children aged <15 years, ²⁶ , ²⁷ and 1 study included neonates, infants and older children. ²⁸ The number of recruited patients (excluding patients recruited for validation) in the included studies was 718 (range 30–172), while the number of total samples included in the analysis was 1385 (range 30–448). Digoxin was administered for management of heart failure or rhythm disturbances in various dosage forms; exclusive intravenous administration of digoxin was reported in 1 study, ²⁷ while the remaining studies utilized either a combination of oral and intravenous formulations ²⁴ , ²⁸ or exclusively oral formulations; the administered dose ranged from 7.5 to 10.1 μg/kg/d for studies involving only oral administration. The general characteristics of the included studies are summarized in Table 1.

Table 1.

General characteristics of the included studies

Reference	Study site	Sample size (I/V)	Population characteristics						Digoxin prescription details			Sampling strategy		Digoxin level (ng/mL)
			Age	Weight (kg)	Sex (M/F)	Serum creatinine (mg/dL)	Diseases (np) or {ns}	Drugs (np) or {ns}	Formulation	Total daily dose (μg/kg/d)	Frequency	Total number of samples	Sampling time (h postdose)
Gong, 2014 15	China	I: 107	I: 125.6 (98.8) d	I: 5.5 (1.9)	I: (66/41)	I:0.3 (0.14)	I: CHF (49)	Spironolactone (I: 87, V: 13)	Elixir, 0.005%	I: 7.5 (1.2)	Twice daily	I: 125	≥6	I:1.1 (0.2)
		V:24	V:117.8 (120.2) d	V:5.2 (0.4)	V: (16/8)	V:0.3 (0.04)	V: CHF (12)	Enalapril (I: 39, V: 2)		V: 6.6 (2.1)		V: 24		V:1 (0.2)
Yukawa, 2011 25	Japan	I: 117	I:0.76 (0.83) (0.08–4.43) y	I: 5.6 (2.8) (1.8–15.1)	I: (58/59)	I: 0.31 (0.08) (0.17–0.57)	I: CHF {189}	I: Spironolactone {241}	Powder/elixir	I: 10.1 (5.3)* (2.6–28.6)	Twice daily	I: 245	12	I: 0.85 (0.40) (0.31–2.73)
		V:51	V: 1.5 (1.32) (0.25–4.99) y	V:8.1(3.2) (2.87–16.2)	V: (25/26)	V: 0.28 (0.1) (0.2–0.6)	V: CHF {24}	V: {42}		V:9.8 (4.3)* (2.5–27.2)		V: 55		V: 0.64 (0.31) (0.29–1.78)
Preechagoon, 2009 28	Thailand	I:130	I: 6.7 (4.7) y	I: 19.5 (12.5)	I: {117/147}	I: 0.55 (0.17)	I: CHF {257} arrhythmia {7}	I: Spironolactone {13} ACEI: {101} diuretics:{185} antacids: {8}	Elixir or tablets (98.2% of patients):	I: 6.60 (2.60)	Variable	I: 264	Various times after administration	I: 0.71 (0.50)
		V:57	V: 6.9 (4.8) y	V: 19.4 (12.2)	V: {22/35}	V: 0.57 (0.17)	V: CHF {56} arrhythmia {1}	V: Spironolactone {0} ACEI: {22} diuretics:{41} antacids: {1}	Intravenous formulation (1.8%)	V: 6.56 (2.66)		V: 57		V: 0.6 (0.35)
Yukawa, 2007 22	Japan	I: 71	I: 18.4 (6.8) (2–29) d	I: 2.84 (0.47) (1.60–4.05)	I: (41/30)	I:0.36 (0.20 (0.1–1.4)	NR	NR	Powder	9.76 (1.37) (4.78–15.2)	Twice daily	I: 129	12	I: 0.95 (0.44) (0.19–2.1)
		V: 9	V: 21 (7.1) (2‐29) d	V: 2.83 (0.42) (2.16–3.67)	V: (5/4)	V:0.47 (0.1) (0.28–0.62)				V: 8.67 (1.32) (6.83–11.0)		V:13 1		V: 0.99 (0.46) (0.6–2.15)
Desoky, 2005 27	Egypt	I: 40	5.94 (0.33–15.00) y	17.86 (4.25–42.00)	(19/21)	0.63 (0.2–1.3)	CHF (40)	Prednisone (7)	Intravenous	10	Twice daily	80	At 4 h and at trough (n = 30)	0.6 (0.4) (0.3–1.4)
		V: not done						Furosemide (10)					At 0.5 h and at 6 h (n = 10)
Martín‐Suárez, 2002 24	NR	I: 51	3.6 (2.9) (0.2–12) mo	4.2 (1.1) (2.3–7.2)	NR	NR	NR	Patients receiving interacting drugs were excluded	Elixir/intravenous	9.16 (3.09) (2.53–20.47)	Variable	64	1) 12–24 h if given once daily	1.25 (0.41) (0.6–2.2)
		V: not done											2) 8–12 h when given more than once daily
Desoky, 2002 26	Egypt	I: 30	8.88 (3.01) (2‐14) y	23.9 (5.8) (12‐37)	(15/15)	0.63 (0.16) (0.4–1.1)	CHF (30)	None received interacting drugs	Elixir/tablet	10 (2) (5.0–20)	Once daily	30	24	0.83 (0.38) (0.4–1.94)
		V: not done
Suematsu, 2001 23	Japan	I: 172	I: 86.4 (79.0) (8–362) d	I: 3.66 (1.31) (1.49–9.65)	I: (96/76)	NR	I: CHF {265}	#I: Spironolactone {378}	Powder	I: 9.4 (1.92) (3.82–15.72)	Twice daily	I: 448	12	I: 0.84 (0.39) (0.24–2.10)
		V:66	V: 92.1 (88.9) (12–362) d	V: 3.99 (1.45) (2.1–8.21)	V: (35/31)		V: CHF {62}	#V: Spironolactone {67}		V: 9.69 (1.81) (4.76–13.3)		V:81		V: 0.85 (0.35) (0.29–1.83)

Results were reported as mean (standard deviation), ranges are also presented when reported

1; 2; 3; 4; CHF: congestive heart failure; ACE‐I: angiotensin converting enzyme inhibitor

#Patients receiving other drugs (other than spironolactone) known to interact with digoxin (e.g. verapamil, quinidine, amiodarone, phenytoin) were excluded from the population (10% of those screened).

^*Approximated by dividing the value of dose in μg/d by the average population weight in kg.

3.3. Population pharmacokinetic analysis

In general, the reviewed articles utilized a population pharmacokinetic approach, through the NONMEM software, to describe the pharmacokinetics of digoxin, and characterize the role of various patients' covariates on the pharmacokinetic parameters of digoxin and hence the dose, utilizing data generated through routine TDM. Usually 1 sample per dosing interval was utilized in the analysis, although Desoky et al. ²⁷ utilized 2 samples per dosing interval (with or without the TDM trough); for this reason this study was the only one able to describe the pharmacokinetics of digoxin with a 2‐compartment model. Digoxin concentrations were measured with various types of immunoassays as presented in Table 2.

TABLE 2.

General description of the digoxin assay and the included pharmacokinetic models

Reference	Type of study	Assay	Software	Type of structural model	Estimation method	Evaluation method
Gong, 2014 15	TDM	Micro‐particle enzyme immunoassay using the AxSYM digoxin III assay method	NONMEM (version 7)	1‐compartment model	FOCE‐I	Basic internal
						Advanced internal
						External
Yukawa, 2011 25	TDM	Cloned enzyme immunoassay using the reagent CEDIA digoxin MAb II	NONMEM (version 6)	1‐compartment model	NR	Basic internal
						External
Preechagoon, 2009 28	TDM	Fluorescence polarization immunoassay technology	NONMEM (version 5)	1‐compartment model	NR	Basic internal
						Advanced internal
Yukawa, 2007 22	TDM	Cloned enzyme immunoassay using the reagent CEDIA digoxin plus	NONMEM (version 5)	1‐compartment model	NR	Basic internal
						External
Desoky, 2005 27	Prospective	TDxFLX fluorescence polarization immunoassay	NONMEM (version 5)	2‐compartment model	FOCE	Basic internal
						Advanced internal
Martín‐Suárez, 2002 24	TDM	Immunofluorescent polarization method TDx digoxin II assay	NONMEM (version 4)	1‐compartment model	FO	Basic internal
Desoky, 2002 26	TDM	Fluorescence polarization immunoassay technique using the TDxFLx system	NONMEM (version 5)	1‐compartment model	FO	Basic internal
						Advanced internal
Suematsu, 2001 23	TDM	Cloned enzyme donor immunoassay using the measurement reagent CEDIA plus	NONMEM (version 4)	Steady‐state model	NR	Basic internal
						External

FO; first order estimation method, FOCE‐I; first‐order conditional estimation method with interaction, NR: not reported; TDM: therapeutic drug monitoring.

The estimated pharmacokinetic parameters in the reviewed studies were usually clearance or more frequently apparent clearance (Cl/F), due to lack of information regarding bioavailability (only 1 study estimated relative bioavailability ²⁵ ). As for other pharmacokinetic parameters, due to the nature of the data in the involved studies (TDM data), certain parameters such as the absorption rate constant were not calculated and thus were frequently fixed to literature values. Similarly, the volume of distribution and its associated IIV were estimated only in 3 studies, ¹⁵ , ²⁶ , ²⁸ and intercompartmental clearance was estimated only in 1 study. ²⁷ The values of estimated apparent clearance in a typical individual (with or without heart failure) standardized by mean weight ranged between 0.24 and 0.56 L/h/kg, while the IIV in clearance ranged between 7.0 and 35.1%. Results of simulations portraying clearance in a typical individual are presented in Figure S2. Residual variability was described by proportional error model in 75% of the studies, while IIV in the pharmacokinetic parameters was described by proportional or exponential models in 87.5% of studies and was not determined in 1 study. ¹⁵ Data relating to the fixed and random effects models are presented in Table 4.

TABLE 4.

A summary of final models, fixed and random effect models described in the included studies

Reference	Fixed effect parameters	Value of pharmacokinetic parameter in a typical individual*	IIV		RV
			Type	Value	Type	Value
Gong, 2014 15	CL/F(L/h/70) = 10.4 *(BW/70)0.75 *(PNA/125)0.169	Cl/F:0.28	Cl/F: Not specified	CL/F: 35.1%	Proportional	10.1%
	V/F (L/70) = 1100 *(BW/70)	V/F: 15.7	V/F: Not specified	V/F: 58.0%
	Ka = 0.718 h−1 (fixed)
Yukawa, 2011 25	CL/F (L/h) = 0.302 *BW (kg)1.17 * 0.905 CHF *(Conc–0.540)	CL/FCHF: 0.366	Cl/F: Proportional	CL/F: 21.0%	Proportional	23.0%
	F = 1 for elixirs, 0.754 for powders where CHF is 1 for presence of congestive heart failure, 0 otherwise; conc is digoxin concentration but Conc–0.540 Is 1 for digoxin concentration < 1.7 ng/mL	CL/FNoCHF: 0.405
	Vd/F = 7.5 L/kg (fixed)
	Ka = 0.47/h (fixed)
Preechagoon, 2009 28	CL/F (L/h) for infant (0–1 year) = 0.322 * BW	Children 0–1 year: CL/F: 0.32	Cl/F: Proportional	CL/F: 31.48%	Proportional	41.7%
	CL/F (L/h) for children (> 1 year) = (0.138 * BW + 0.0319 * height) * 0.765 CHF (1 in case of CHF)	Children > 1 year: CL/FCHF: 0.24 CL/FNoCHF: 0.31	Vd/F: Proportional	V/F: 35.36%
	Vd/F (L) = 9.27 * BW* 1.75 CHF	‐ for all ages V/FCHF: 16.2 V/FNo CHF: 9.27
	Ka = 5.6/h (fixed) where CHF is 1 for presence of congestive heart failure, 0 otherwise
Yukawa, 2007 22	CL/F (L/h) = 0.0261 *BW0.645 *Conc‐0.724 *GA0.8	CL/F: 0.35	Cl/F: Proportional	CL/F: 7.0%	Proportional	13.1%
	Vd/F = 7.5 L/kg (fixed)
	Ka = 0.47/h (fixed)
Desoky, 2005 27	CL (L/h/kg) = 0.388‐ [0.78*(SCr‐0.6)]	CL: 0.36	Cl: Exponential	CL: 31.2%	Additive	0.12 ng/mL
	V1 (L/kg) = 1.38 (fixed)
	V2 (L/kg) = 9.11 (fixed)		Q: Exponential	Q: 68.99%
	Q (L/h/kg) = 0.48
	Vss (L/kg) = 9.8
Martín‐Suárez, 2002 24	CL(L/h/kg) = 0.237 *(1 + 0.094 * AGE (mo)) bioavailability of digoxin was fixed to 0.8 for the elixir and 1 for the intravenous doses	CL: 0.32	Cl: Proportional	CL: 7.73%	Additive	0.39 ng/mL
Desoky, 2002 26	CL/F (L/h) = 8.61	CL/F:0.36	Cl/F: Exponential	CL/F: 34.4%	Proportional	26.9%
	V/F (L) = 450	V/F:18.8	V/F: Exponential	V/F: 18.1%
	Ka (h−1) = 1.5 (fixed)
Suematsu, 2001 23	CL/F(L/h/kg) = 0.298 *AGE0.099*Scr‐0.153* 0.882 CHF* 0.897 SPI where CHF is 1 in presence of CHF 0 otherwise, SPI = 1 when spironolactone is administered, 0 otherwise	CL/FCHF: 0.56**	Cl/F: Proportional	CL/F: 32.1%	Proportional	28.9%
		CL/FNo CHF: 0.49**

^*Clearance (L/h/kg) and volume of distribution (L/kg) estimates were generally obtained by dividing the estimated value of clearance (L/h) and volume of distribution (L) in a typical individual by the mean weight of each population.

^**Assuming mean serum creatinine as 0.3 mg/dL.

All studies reported apparent clearance/volume of distribution (CL/F, V/F) apart from Desoky et al. ²⁷ and Martín‐Suárez et al. ²⁴ who reported CL and/or V.

IIV: interindividual variability, RV: residual variability, Ka: absorption rate constant, BW: body weight, PNA: postnatal age, SCr: serum creatinine, CHF: congestive heart failure, GA: gestational age, SPI: spironolactone

Various covariates were screened in the reviewed articles. Age and weight were the most commonly screened covariates as they were screened in all studies; sex was screened in 87.5% of studies but was never found significant; serum creatinine level or creatinine clearance were screened in 75% of the studies; and serum creatinine was included in the final models of 2 studies. ²³ , ²⁷ Other tested covariates were disease states, such as the presence of congestive heart failure (CHF), which was screened in 50% of the studies, and the concomitant use of medications such as spironolactone or angiotensin converting enzyme inhibitors in 62.5% of the studies. Forward inclusion and/or backward elimination were the most frequently used methods to test the significance of covariates. The covariate building process in terms of screened covariates and the statistical criteria for detecting significant covariates, in addition to the relation of these covariates with the pharmacokinetic parameters is portrayed in Table 3, while the final model and associated variability is shown in Table 4.

TABLE 3.

Details of the covariate building process

Reference	Tested covariates	Covariate selection method	Statistical level for selecting significant covariates (P‐value)	Selected covariates and their relation to pharmacokinetic parameters
Gong, 2014 15	BW, PNA, SCr, sex, the presence of CHF, and concomitant medications	1) forward stepwise inclusion	1) P < .05	1) apparent clearance: BW, PNA CL/F (L/h/70 kg) = θ1*(BW/70)0.75*(PNA/125)θ3
		2) backward elimination	2) P < .01	2) apparent volume of distribution: BW V/F (L/70kg) = θ2*(BW/70)
Yukawa, 2011 25	CHF, sex, age, SCr, BW, and combination with spironolactone, infant–young children clearance factor (trough serum digoxin concentration > 1.7 ng/mL), relative bioavailability (elixir vs powder)	1) forward inclusion	1) P < .05	1) apparent clearance: BW, CHF, and infant–young children clearance factor (trough serum concentration of digoxin) CL/F (L/h) = θ 1*BW θ2* θ 4 CHF*Concθ5
		2) hypothesis testing using restricted models of the full model	2) P < .05	2) bioavailability was influenced by formulation F = θ 3; Conc = 1 for serum digoxin concentration < 1.7 ng/mL; CHF, 1 for presence of congestive heart failure, 0 otherwise; F is bioavailability, 1 for elixirs, θ3 for powders.
Preechagoon, 2009 28	BW, height, sex, age, use of diuretics, CHF, ACEIs, SCr, creatinine clearance and serum potassium level	1) forward stepwise inclusion	1) P < .01	1) apparent clearance: BW in children <1 year, BW, height, CHF in children older than 1 year CL/F (L/h) for infant (0–1 year) = θ 2 * BW; CL/F (L/h) for children (> 1 year) = (θ 3 * BW + θ 4 * height) * θ 5 CHF (1 in case of CHF)
		2) backward elimination	2) P < .01	2) apparent volume of distribution: Weight and CHF for all ages Vd/F (L) = θ 6 * BW*θ 7 CHF
Yukawa, 2007 22	BW, GA, PNA, postconceptional age, neonate clearance factor (trough serum concentration of digoxin) and sex	Forward stepwise inclusion	P < .01	Apparent clearance: BW, GA and neonate clearance factor CL/F (L/h) = θ 1*BWθ2·Concθ3·GAθ4
Desoky, 2005 27	Age, BW, height, SCr, sex and coadministered medications	1) generalized additive models (GAM)	1) AIC	1) clearance: SCr CL (L/h/kg) = θ 1*[1 + θ 5* (SCr‐0.6)]
		2) univariate analysis using likelihood ratio test	2) P < .01	2) intercompartmental clearance: None Q (L/h/kg) = θ 3
		3)stepwise backward deletion	3) P < .001
Martín‐Suárez, 2002 24	Age, BW	Evaluated different models via:	1) p < .001	Clearance: Age CL (L/h/kg) = θ 1 *(1 + θ 2* AGE (mo))
		1) χ2 test of the difference in the objective function
		2) plots of weighted residuals, minimization of the magnitude of residual variability, and AIC
Desoky, 2002 26	BW, age, sex, creatinine clearance	Forward stepwise inclusion	P < .05	No covariates were found significant CL/F (L/h) = θ 1 Vd/F (L) = θ 2
Suematsu, 2001 23	BW, age, SCr, daily digoxin dose, sex, spironolactone use, CHF	1) forward stepwise inclusion	1) P < .05	Apparent clearance: Age, SCr, CHF and use of spironolactone CL/F (L/h/kg) = θ 1*AGEθ2*SCrθ3* θ 4 CHF* θ 5 SPI
		2) backward elimination	2) P < .05

BW: body weight; PNA: postnatal age; GA: gestational age; SCr: serum creatinine; CHF: congestive heart failure; ACE‐I: angiotensin converting enzyme inhibitor; SPI: spironolactone; AIC: Akaike's information criterion

3.4. Quality assessment of included studies

The reporting quality of the population pharmacokinetic studies was assessed utilizing the ClinPK checklist ¹⁹ as described in the methodology section, results are portrayed in Table S1. In general, the studies provided the required information in the title, abstract and background (compliance rates 100, 90 and 91.7%, respectively). In the methodology section, several items had poor reporting quality, for instance, the inclusion/exclusion criteria were only identified in 37.5% of the studies. Despite identifying the potentially interacting coadministered medications (or lack of) in most of the included studies, none reported administration or lack of administration of potentially interacting foods. Furthermore, 37.5% of the studies did not report the exact formulation administered to patients and none of the studies reported storage conditions for the withdrawn samples.

In terms of the pharmacokinetic analysis, the NONMEM program provides several estimation methods such as the first order, the first order conditional estimation method and the Laplacian method, the exact method of estimation was identified in only 50% of the included studies. When the covariate model was built the rationale for selecting those covariates and including them in the analysis was reported in only 1 study.

3.5. Validation and clinical utility of included studies

All models were evaluated by at least 1 basic model evaluation method (e.g. goodness of fit plots, uncertainty of the parameter estimates), in addition to either advanced internal methods (e.g. data splitting, bootstrap, Monte Carlo simulation) or external methods, exceptions were noted in case of Martín‐Suárez et al. ²⁴ who used only internal methods and Gong et al. ¹⁵ who used all of the 3 methods to evaluate the model. Four studies ¹⁵ , ²² , ²³ , ²⁵ utilized external datasets for validation, the sample size of these datasets ranged from 9 to 66 participants, which accounts for 12.8–43.9% of the model building sample; the number of samples used in the external validation sets ranged between 13 and 81, which accounts for 10.1–22.5% of the model building sample. Of the 8 included studies, only 2 studies provided the reader with a simple equation to use as a priori method to predict the required dose for a specific individual or the predicted concentration following a certain dose. ²³ , ²⁴ Results of the validation study are shown in Table S2.

4. DISCUSSION

Digoxin is a cardiac glycoside that is licensed for use in children with heart failure and supraventricular arrhythmias. ²⁹ Unfortunately, digoxin is a drug with a narrow therapeutic index and variable pharmacokinetics, ³⁰ which makes individualizing its dosage regimens based on patient specific factors of great importance. This is the first review to summarize the literature knowledge on population pharmacokinetics of digoxin in the paediatric population.

In the present study, CHF was the main indication of digoxin use when reported, it was prescribed for this indication in 45.8% of patients in 1 study ¹⁵ and 59.1–100% of samples in 5 other studies. ²³ , ²⁵ , ²⁶ , ²⁷ , ²⁸ Digoxin bioavailability varies by dosage form; it is reported to be 0.7, 0.8,1 for tablets, elixir and soft gelatine capsules, respectively. ³⁰ Absolute bioavailability was not estimated in any of the reviewed studies, as digoxin was in most studies given only orally. Yukawa et al. ²⁵ estimated the bioavailability of the powder formulation relative to the elixir and found that the relative bioavailability of the powder formulation was 0.754.

Due to inherent limitations of TDM studies, which usually include lack of samples collected during the absorption phase, the absorption rate constant (Ka) could not be estimated with high precision. For this reason, when reported Ka was fixed to literature values ²² , ²⁵ or allowed to be estimated initially, Ka was then fixed by removing the component of IIV throughout modelling. ¹⁵ , ²⁶ , ²⁸

Due to limitations in sampling, only 1 study characterized digoxin pharmacokinetics as being 2‐compartments, with an estimated steady‐state volume of 9.82 L/kg. ²⁷ Utilizing a 1‐compartment model, 3 studies were able to estimate the apparent volume of distribution (V_d/F) of digoxin and its associated IIV; in neonates and infants, the V_d/F was not as estimated to be 15.7 L/kg ¹⁵ (IIV: 58%), while a value of 18.8 L/kg (IIV: 18%) was reported in older children (2–14 years), ²⁶ these values are higher than what is reported in adults with normal renal function V_d/F (6‐7 L/kg). ³⁰ In infants, this is probably due to age related changes in body compositions and the higher tendency of digoxin to bind to tissues. ³¹ Preechagoon et al. ²⁸ found that V_d/F in both infants and older children was about 75% higher in patients with CHF as opposed to those who do not have CHF (16.2 L/kg vs 9.3 L/kg); the authors proposed that patients who developed signs of oedema are likely to have a rise in V_d/F when CHF gets worse. However, this correlation between CHF and volume of distribution was not confirmed in Chinese neonates and infants. ¹⁵ Discrepancies between the 2 studies could be attributed to the distribution of CHF cases between the 2 studies, as the population of patients at the study of Preechagoon et al. ²⁸ was predominantly CHF with only 2.7% of analysed samples belonged to patients receiving digoxin for arrhythmia. By contrast, Gong et al. ¹⁵ had a better representation of non‐CHF cases in his study (54.2%).

The estimated value of apparent clearance (CL/F) ranged between 0.24 and 0.56 L/h/kg, while the estimated magnitude of the IIV ranged between 7.0 and 35.1%. Several covariates were found to impact digoxin clearance such as age, weight, presence of CHF and serum creatinine. Three studies involving only infants and neonates found a positive linear ²⁴ or exponential ¹⁵ , ²³ correlation between age and clearance, the increase in clearance with age in this specific age group could be attributed to developmental changes in the kidney function, this consistent with previous reports indicating that the half‐life of digoxin shortens over the first postnatal year ³² and that higher average plasma digoxin concentration was detected in full‐term neonates compared to infants and older children, when maintained on equal weight‐based doses. ³³ Apparent clearance also increased with gestational age in a study conducted only in neonates by Yukawa et al. ²² Preechagoon et al. ²⁸ investigated a more diverse of children (0–15 years), they found that the fit of the model improved when the analysis was separated based on age group (<1 year vs >1 year). Weight was incorporated in 4 out of the 8 included models ¹⁵ , ²² , ²⁵ , ²⁸ in 3 of these the correlation was expressed in an exponential form—the exponent was either allowed to be estimated by NONMEM in these cases the exponent was estimated to be 1.17 in a study involving infants and young children (<5 years) ²⁵ or 0.65 in a study involving neonates ²² —or fixed to a value of 0.75. ¹⁵ The latter approach can be utilized by modellers to depict secondary covariate effects from the effect of size. ³⁴

In a small study conducted by Naafs et al., patients diagnosed with CHF had a lower value of digoxin clearance when compared to patients having similar creatinine clearance and taking similar doses of digoxin for management of atrial fibrillation. ³⁵ Several population pharmacokinetic studies in adults and elderly confirmed this correlation, ³⁶ , ³⁷ , ³⁸ , ³⁹ although others did not. ⁴⁰ , ⁴¹ , ⁴² In the present review, CHF was tested in 50% the studies ¹⁵ , ²³ , ²⁵ , ²⁸ and was found to reduce clearance in 75% of these models. ²³ , ²⁵ , ²⁸ The reduction in clearance was 9.5 and 11.8% in 2 studies conducted in neonates and infants where the percentage of CHF relative to the total analysed samples was 59.1 and 77.1%. ²³ , ²⁵ A higher impact of CHF on digoxin clearance was reported in children aged 1–15 years as clearance was reported to be reduced by a factor of 23.5%. Nevertheless, it is important to emphasize that the majority of blood samples were from to patients diagnosed with CHF (257) as opposed to only 7 samples for patients diagnosed with arrhythmia in the entire population of infants and older children. ²⁸

Drugs known to interact with digoxin pharmacokinetics were either not received or excluded if received in 3 studies. ²⁴ , ²⁶ , ²⁷ Suematsu et al. ²³ allowed the use of spironolactone and excluded patients receiving any other drug that could interact with digoxin such as quinidine. Actually, spironolactone was the most commonly investigated drug; in total, the impact of spironolactone on digoxin clearance was investigated in 4 studies. ¹⁵ , ²³ , ²⁵ , ²⁸ Of those, only 1 study found a correlation with apparent clearance, where it decreased by 10.3% if the patient received the drug (84.4% of the patients received the drug). ²³ The lack of correlation may be explained in 2 of the 3 studies by either the very small number of collected samples receiving spironolactone (4.9%) in 1 study ²⁸ or the small number of samples collected for patients not receiving spironolactone (1.6%). ²⁵

Given that digoxin is eliminated primarily via the kidney, it would not be surprising to identify serum creatinine or creatinine clearance as influential covariates on digoxin clearance. Nevertheless, out of the 6 studies ¹⁵ , ²³ , ²⁵ , ²⁶ , ²⁷ , ²⁸ that investigated these variables only 2 included serum creatinine in the final model. ²³ , ²⁷ This does not negate the impact of renal function on digoxin clearance, simply because lack of correlation could be explained by exclusion of cases of severe renal dysfunction, ¹⁵ the narrow range of serum creatinine level in the included participants ²⁵ , ²⁸ and inability of the model to include covariates. ²⁶ Based on Suematsu et al. ²³ the typical value of clearance would drop from 0.557 to 0.463 L/h/kg (16.9% decrease) in a 3‐month‐old infant not receiving spironolactone or suffering from CHF, if the serum creatinine increases from 0.3 to 1 mg/dL.

In infants and young children Yukawa et al. ²⁵ reported that the apparent clearance varied with neonatal clearance factor (trough serum level of digoxin, Conc^θ), at a concentration lower than 1.7 ng/mL this factor was equal to 1; however, at higher concentrations, the estimated θ was (−0.54), indicating a nonlinear decrease in clearance at digoxin concentrations >1.7 ng/mL. The same research group found a similar relationship between clearance and elderly (age >65 years) clearance factor, although the magnitude of θ was different (θ = −0.18 at a concentration >1.7 ng/mL). ³⁷ In another study of the same research group in neonates, clearance decreased nonlinearly as the trough serum level of digoxin increased, at the whole spectrum of digoxin concentrations (i.e. there was no cut‐off value). ²² More research is required to confirm these nonlinear correlations in neonates, infants and young children. In terms of pharmacodynamics, none of the included studies investigated the correlation between digoxin concentration and achieving certain pharmacodynamic endpoints such as conversion to sinus rhythm.

Complete and transparent reporting of pharmacokinetic studies is quite important to allow the readers of these studies to evaluate their quality and enable them to make conclusions about the applicability of the findings. ¹⁹ The clinPK statement checklist, despite its limitations, can provide researchers with a tool to help them achieve the required level of reporting quality. An important limitation of this checklist is that it does not address all the required reporting needs for complicated analysis such as population pharmacokinetic analysis. For this reason, we added certain items to address covariate model building, as performance between various methods for covariate model building approaches may differ. ⁴³ The final model may also differ depending on the chosen statistical criteria; thus, such criteria should be declared. It is important to highlight that these additions were made solely to allow the reader of this review to have a comprehensive overview of the presented models, based on reviewed literature and the opinions of the authors of this manuscript. In a separate section, we also evaluated the quality of evaluation of the developed models. Since model evaluation depends on whether the purpose of the model is descriptive (aims to summarize data) or predictive, we assumed for the purposes of this analysis that all models were developed for a predictive purpose and the evaluation quality of 4 models was judged to be good or excellent in 75% of the developed models based on the criteria defined by Brendel et al. ²¹ and outlined in the methodology section.

Carefully chosen pharmacokinetic models can be utilized in clinical practice by incorporating them into Bayesian estimation programs as a priori, combined with the current drug concentration to obtain optimal dose regimens ¹⁸ ; an example of utilizing population pharmacokinetic models in Bayesian programs is given in the case of tobramycin: the population model developed by Hennig et al. ⁴⁴ has been incorporated in the Bayesian forecasting program of DoseMe (http://doseme.com.au) software. ⁴⁵ This requires the availability of a population pharmacokinetic model that sufficiently characterizes the pharmacokinetics of the drug in the population of interest in addition to identifying of important covariates. ¹⁸ Prior to utilizing such models into such programs, the model must describe the pharmacokinetic behaviour in the selected population. An important limitation of the present models that hinders their use in such manner, is the use of steady‐state troughs only, in the majority of models, which mandated the simplification of the known 2‐compartment structure of the pharmacokinetic model to a 1‐compartment model as shown in Table 2. This generally enabled them to only estimate 1 model parameter, which is clearance, other important parameters such as volume of distribution and absorption rate constant usually had to be fixed to literature values or estimated with poor precision. Furthermore, when choosing a model, it is important choose a model with adequate predictive performance in addition to best approximating the target population. For instance, a model that takes into account the coadministration of amiodarone is better suited for a centre where the coadministration of this drug is common with digoxin as opposed to a model that excludes all patients receiving amiodarone.

In the present review, we only included pharmacokinetic studies reported in English; this may be considered a selection bias. However, such action did not impact our results as the identified studies in languages other than English were mainly in adults, involving other analysis methods. We also selected to investigate studies that were analysed through the population approach, as in children this approach is preferred over the traditional approach, which requires extensive sampling at specified times, which could both interfere with patient care and raise ethical concerns especially in infants and neonates if the required volume of the blood sample is large. ⁴⁶ , ⁴⁷

A limitation of the present review is the lack of studies in the US population and the presence of only 1 study that represents the European population. Pharmacogenetics can be an important reason for differences in pharmacokinetics between races. In the case of digoxin pharmacogenetics, there is a controversy about the role ABCB1 gene especially for the rs1045642 in digoxin pharmacokinetics. ⁴⁸ To what degree genetic differences are reflected in clinically significant changes in digoxin pharmacokinetics in paediatrics is difficult to determine, as it has not been studied. Nevertheless, the applicability of the models described in this paper to the US or European populations can be easily tested by applying external validation techniques utilizing a dataset from these populations.

For researchers, the present review provides information concerning the utilized model structure, population pharmacokinetic parameters, influential covariates as well as the degree of pharmacokinetic variability and provides guidance to possible directions of future research studies in paediatrics. Future research should focus on characterizing the 2‐compartment pharmacokinetic profile of digoxin in various paediatric populations and possibly linking pharmacokinetics with pharmacodynamics.

COMPETING INTERESTS

There are no competing interests to declare.

CONTRIBUTIONS

Conceptualization: M.H.A. Conducting the review: M.H.A., N.A., M.A., M.S., K.A. Data analysis: M.H.A., N.A., M.A., M.S. Writing the manuscript: M.H.A., N.A., M.A., M.S., K.A. Reviewing and approving the manuscript: M.H.A., N.A., M.A., M.S., K.A.

Supporting information

TABLE S1 Quality assessment of the included studies

TABLE S2 Summary and assessment of the performed model evaluation conducted by the included studies

DATA S1 Supporting Information

FIGURE S1 Prisma flow chart utilized for inclusion of studies in the present review

FIGURE S2 Typical estimates of clearance (median) in L/h/kg obtained from simulations. The markers on the sides represent the 10^th and 90^th percentiles. The asterisk indicates that clearance rather than apparent clearance was estimated

Abdel Jalil MH, Abdullah N, Alsous MM, Saleh M, Abu‐Hammour K. A systematic review of population pharmacokinetic analyses of digoxin in the paediatric population. Br J Clin Pharmacol. 2020;86:1267–1280. 10.1111/bcp.14272