Content uploaded by Alice Panchaud
Author content
All content in this area was uploaded by Alice Panchaud on Jun 07, 2023
Content may be subject to copyright.
1
Safety of medicines during breastfeeding – from case report to modeling : a
contribution from the ConcePTION project
Evelina Cardoso1*, Monia Guidi2,3*, Nina Nauwelaerts4, Hedvig Nordeng5,6, Marie Teil7, Karel
Allegaert8,9,10,11, Anne Smits8,9,12, Peggy Gandia13, Andrea Edginton14, Shinya Ito15, Pieter Annaert4#,
Alice Panchaud1,16#
1 Service of Pharmacy, Lausanne University Hospital and University of Lausanne, Lausanne, Switzerland
2 Service of Clinical Pharmacology, Lausanne University Hospital and University of Lausanne, Lausanne,
Switzerland
3 Center for Research and Innovation in Clinical Pharmaceutical Sciences, Lausanne University Hospital
and University of Lausanne, Lausanne, Switzerland
4 Drug Delivery and Disposition Lab, Department of Pharmaceutical and Pharmacological Sciences, KU
Leuven, Leuven, Belgium
5 Pharmacoepidemiology and Drug Safety Research Group, Department of Pharmacy, PharmaTox
Strategic Initiative, Faculty of Mathematics and Natural Sciences, University of Oslo, Oslo, Norway
6 Department of Child Health and Development, Norwegian Institute of Public Health, Oslo, Norway
7 Women of Childbearing Age Program, UCB Pharma, Slough, UK
8 Child and Youth Institute, KU Leuven, Leuven, Belgium
9 Department of Development and Regeneration, KU Leuven, Leuven, Belgium
10 Department of Pharmaceutical and Pharmacological Sciences, KU Leuven, Leuven, Belgium
11 Department of Hospital Pharmacy, Erasmus MC, Rotterdam, Netherlands
12 Neonatal intensive care unit, University Hospitals Leuven, Leuven, Belgium
13 Laboratory of Pharmacokinetics and Toxicology, Purpan Hospital, University Hospital of Toulouse,
Toulouse, France
14 School of Pharmacy, University of Waterloo, Waterloo, Ontario, Canada
15 Division of Clinical Pharmacology and Toxicology, Department of Paediatrics, The Hospital for Sick
Children, Toronto, Ontario, Canada
16 Institute of Primary Health Care (BIHAM), University of Bern, Bern, Switzerland
*Co-first authors equally contributed to the work
#Co-senior authors equally contributed to the work
Corresponding author: Alice Panchaud (alice.panchaudmonnat@unibe.ch)
Journal: Expert Opinion Drug Metabolism and Toxicology
2
Abstract:
Introduction. Despite many research efforts, current data on the safety of medicines during
breastfeeding are either fragmented or lacking, resulting in restrictive labeling of most medicines. In
the absence of pharmacoepidemiologic safety studies, risk estimation for breastfed infants is mainly
derived from pharmacokinetic (PK) information on medicine. This manuscript provides a description
and a comparison of the different methodological approaches that can yield reliable information on
medicine transfer into human milk and the resulting infant exposure.
Area Covered. Currently, most information on medicine transfer in human milk relies on case reports
or traditional PK studies, which generate data that can hardly be generalized to the population. Some
methodological approaches, such as population PK (popPK) and physiologically based PK (PBPK)
modeling, can be used to provide a more complete characterization of infant medicine exposure
through human milk and simulate the most extreme situations while decreasing the burden of
sampling in breastfeeding women.
Expert opinion. PBPK and popPK modeling are promising approaches to fill the gap in knowledge of
medicine safety in breastfeeding, as illustrated with our escitalopram example.
Keywords: Breastfeeding; Breastfed infants; Drug exposure; Human milk; Pharmacokinetics;
Physiologically-based pharmacokinetics
Article highlights:
• Infant exposure to medicines through human milk is an obvious driver for the assessment of
medicine safety during breastfeeding.
• Population pharmacokinetic (popPK) and physiologically based pharmacokinetic (PBPK)
modeling are two methodological approaches to assess medicine exposure in the infant
through breastfeeding.
• The popPK approach provides an attractive sampling design for breastfeeding women who
cannot be enrolled in traditional intensive sampling PK studies.
• The PBPK approach can predict the medicine pharmacokinetics at a very early stage of the
pharmaceutical development, without human sampling.
• There is a need to further adapt popPK and PBPK approaches specifically for lactation studies.
3
1. Introduction
Many women require the use of medicines during the postpartum period, and more than 50 % take at
least one medicine [1]. Breastfeeding is frequently considered as a major challenge for both health-
care providers and mothers when maternal medicines are required. Indeed, this situation raises the
question of the compatibility of maternal medicine intake with infant safety during breastfeeding.
Based on limited evidence, it seems that only a few medicines are associated with a potential harm to
the breastfed infant (e.g. sedation, poor weight gain, and irritability gain) and thus considered as
contraindicated in breastfeeding mothers (e.g. some radioactive compounds) [2]. Most commonly
used medicines are considered relatively safe for the breastfed infant with usually rare and mild
adverse drug reactions (ADRs), and breastfeeding discontinuation is most often not necessary [3].
Several specialized information sources (e.g. LactMed and Brigg’s drugs in pregnancy and lactation)
summarize the available substance-specific safety information during breastfeeding and can thus
provide adequate advice to health-care providers and parents [4-6].
Breastfeeding women have been frequently excluded from pharmaceutical clinical trials as a
precautionary principle, when adequate information to assess the risk to the infant was lacking. For
years, this trend has led to knowledge gaps, particularly at the time of medicine marketing, with poor
information on infant safety in the labeling of most approved medicines [7]. Corresponding labeling
was restrictive and usually limited to two recommendations, namely « do not breastfeed with this
drug » or « use this drug with caution while breastfeeding », without providing data on which to base
them [8]. Currently, most of the safety data during breastfeeding arises from small sample size studies
including breastfeeding women in a limited subpopulation or from individual cases obtained through
many post-marketing research efforts. For a large majority of medicines, the available data are
reassuring and therefore discordant with the labeling [8]. This can easily lead to confusion among
health-care providers and mothers on the medicine use during breastfeeding, which might potentially
lead to inadequate decision-making [9]. On the one hand, mothers might be discouraged from starting
or continuing breastfeeding in order to take a medicine, at the expense of the multiple health benefits
of breastfeeding for both themselves and their breastfed infant [10,11]. On the other hand, some
mothers choose to discontinue the medicine or to postpone its initiation during the breastfeeding
period, regardless of the risk associated with the untreated disease for both themselves and their
infant.
For some years now, several efforts and initiatives have been made to provide more accurate and
helpful information on the effects of medicines used during breastfeeding [12-14]. First, the
recruitment of breastfeeding women into clinical trials is encouraged by regulatory agencies, such as
the European Medicines Agency (EMA), and by several working groups, such as the task force on
4
Research Specific to Pregnant Women and Lactating Women (PRGLAC) [15,16]. Several
recommendations are formulated to obtain data from breastfeeding women in an early stage of the
development of medicine for use by women of childbearing potential [17]. In 2015, new labeling
standards were implemented by the Food and Drug administration (FDA), i.e. the Pregnancy and
Lactation Labeling Rule (PLLR), which proposed major revisions to prescription drug labeling [18,19].
The PLLR recommends, on the one hand, to provide relevant information, such as the amount of drug
transferred into human milk and the specific potential effects on breastfed infants, for decision-making
when treating breastfeeding women and, on the other hand, to continuously include clinically relevant
information from well-conducted studies published in the medical literature in the labeling. Moreover,
guidance has been developed to facilitate the conduct of lactation studies, and recommendations have
been proposed to overcome barriers to the participation of breastfeeding women in research [15,20].
Following these recommendations would allow the generation of reliable information on the safety of
medicine intake during breastfeeding. Yet, their implementation so far remains poor.
The pharmacodynamic (PD) response in infants exposed to a medicine through human milk is an
evident factor to consider for safety evaluation. This information is, however, hardly available as
pharmacoepidemiologic safety studies designed to assess the effects of medicine on infants exposed
through human milk are still very rare [21]. Even though pharmacoepidemiology is deemed crucial in
defining ADR prevalence, this approach will not be further developed as it is outside the scope of this
article. Thus, the risk estimation for breastfed infants comes mostly from pharmacokinetic (PK)
information on medicine as the most obvious driver of the presence and extent of any PD effect is the
infant’s medicine exposure through human milk. The most reliable and direct method to assess infant
exposure to maternal medicine is its quantification (i.e. concentration measurement) in the infant
plasma. This information, combined with milk and plasma samples from the mother (mother-infant
pair study), provides complete information on the medicine transfer from mother to infant of a given
compound. However, sample collection in infants is rarely performed because of ethical concerns,
logistic limitations, and parental reluctance. Two study designs can be proposed to mitigate these
limitations and to estimate the breastfed infant exposure through the medicine excretion into human
milk: the milk only and the milk-plasma studies in lactating women. Whereas these study designs are
well described in the 2019 FDA guidance and in the literature, some methodological approaches, such
as population pharmacokinetic (popPK) and physiologically based pharmacokinetic (PBPK) modeling,
are still rarely reported and applied in the field of medicine and breastfeeding [20,22,23].
The aim of this paper is to describe and compare different methodological approaches that can yield
reliable information about the transfer of medicine into human milk and the subsequent infant
exposure. First, Section 2 summarizes the key elements of medicine transfer into human milk and its
5
disposition in the breastfed infant, and discusses the associated PK variability. Then, Section 3 reviews
the available methodological approaches to assess medicine exposure in infants through breastfeeding
with a concrete example to illustrate the potential of popPK and PBPK modeling (Section 3.3). Section
4 summarizes the relevant data presented in the manuscript. Finally, our expert opinion (Section 5)
proposes several strategies that could provide robust and generalizable data while reducing the
burden of studies on breastfeeding women.
2. Principles of medicine transfer from the mother to the infant through human milk
Almost all medicines taken by the breastfeeding woman transfer to some extent into human milk.
However, according to available evidence, only a few medicines reach the breastfed infant systemic
circulation in relevant clinical amounts as, for the majority of them, transfer is limited by several
barriers and dilution processes.
First, the medicine transfer in milk is directly related to the available amount of medicine in maternal
blood. This level depends, besides the maternal dosing regimen, on the PK parameters of the medicine,
such as bioavailability (F), volume of distribution (Vd), clearance (CL), and half-life (t½ which depends
on both CL and Vd) [24]. For most medicines, subsequent transfer from the maternal circulation into
milk occurs through passive diffusion across the biological membranes of the mammary epithelial cells
that form the barrier between maternal blood and milk. Some medicines are, however, actively
transported into human milk by membrane transporters (or carrier proteins), such as the Breast Cancer
Resistance Protein (BCRP; ABCG2 gene). Medicines that are BCRP substrates (e.g. nitrofurantoin,
acyclovir, and cimetidine) thus show increased passage to the milk compared to predictions based only
on passive diffusion mechanisms [25]. The ability of a medicine to enter milk depends on several
factors, such as physicochemical properties of the molecule (e.g. molecular weight, lipophilicity,
degree of ionization or plasma protein-binding) and biochemical characteristics of human milk (e.g.
lipid concentration and pH). The medicines that show the highest milk partitioning are lipophilic, have
a low molecular weight (< 500 Da) and a relatively low degree of plasma protein binding compared to
milk [26].
Several parameters exist to quantify the amount of maternal medicine received by the breastfed
infant. A frequently reported parameter is the Milk-to-Plasma ratio (M/P), i.e. the relative
concentration ratio in milk over maternal plasma, estimated according to different methods reviewed
in [27]. In clinical lactation studies, the M/P can be derived based on milk and maternal plasma samples
collected simultaneously. It is possible to calculate this ratio with single time point concentrations or
with 24 h area under the concentration-time curves (AUCs). Although it may be difficult to obtain
enough time points to compute AUC (see section 3.2.1), this approach is preferable compared to single
6
time point M/P as the latter might change with the sampling time giving rise to a wide range of M/P
values for the same medicine [17,28]. Worth emphasizing, an M/P greater than 1 does not categorize
the medicine as unsafe for the breastfed infant as this parameter reflects the relative efficiency of a
compound to transfer into milk but not the infant systemic exposure to the medicine. Also, the M/P is
more commonly lower than 1, except for some medicines that tend to be concentrated in human milk,
such as actively transported medicines (e.g. cimetidine) or those subject to “ion trapping” mechanisms
due to a lower pH of human milk compared with plasma (e.g. beta-blockers) [29,30]. Although its direct
clinical utility is limited, M/P may inform PBPK modeling in lactating women.
The daily infant dose defines the dose that the infant receives through human milk per day and can be
estimated using the human milk medicine concentration (Cmilk) and the volume of milk (Vmilk) ingested
by the infant at each feeding over 24 h as follows:
𝐷𝑎𝑖𝑙𝑦&𝑖𝑛𝑓𝑎𝑛𝑡&𝑑𝑜𝑠𝑒&
[
𝑚𝑔 𝑘𝑔/𝑑𝑎𝑦
⁄]
=&
∑
𝐶!"#$"
[
𝑚𝑔 𝑚𝐿
⁄]
×&𝑉!"#$"
[
𝑚𝐿 𝑘𝑔/𝑑𝑎𝑦
⁄]
%
"&'
(1)
where i denotes the feeding time and n the number of feedings per day. Because of the practical
difficulty of gathering such information in real-life, the daily infant dose can be estimated assuming
the same Cmilk for each feeding (i.e. the one available at a certain feeding) as follows:
𝐷𝑎𝑖𝑙𝑦&𝑖𝑛𝑓𝑎𝑛𝑡&𝑑𝑜𝑠𝑒&
[
𝑚𝑔 𝑘𝑔/𝑑𝑎𝑦
⁄]
=&𝐶!"#$
[
𝑚𝑔 𝑚𝐿
⁄]
×&𝑉!"#$
[
𝑚𝐿 𝑘𝑔/𝑑𝑎𝑦
⁄] (2)
With Vmilk being the daily ingested milk volume. A daily volume of 150 mL per kilogram of infant body
weight for an exclusively breastfed infant is typically assumed in the PK literature [20,31]. However,
Yeung and al. recently proposed an equation to estimate a more representative daily milk intake for
term and preterm infants as a function of infant’s age [32].
When human milk measurements are completely lacking, Cmilk can be estimated as a product of the
literature M/P value and maternal plasma concentration (Cmaternal) measured or available in the
literature at a given time after medicine administration:
𝐶!"#$&
[
𝑚𝑔 𝑚𝐿
⁄]
=&𝑀 𝑃
⁄
&×&𝐶!()*+%(#
[
𝑚𝑔 𝑚𝐿
⁄] (3)
Therefore, daily infant dose can also be calculated using the equation:
𝐷𝑎𝑖𝑙𝑦&𝑖𝑛𝑓𝑎𝑛𝑡&𝑑𝑜𝑠𝑒&
[
𝑚𝑔 𝑘𝑔/𝑑𝑎𝑦
⁄]
=&𝑀 𝑃
⁄
&×&𝐶!()*+%(#
[
𝑚𝑔 𝑚𝐿
⁄]
×&𝑉!"#$
[
𝑚𝐿 𝑘𝑔/𝑑𝑎𝑦
⁄]
&
(4)
Another parameter that reflects the level of the ingested dose by the breastfed infant is the relative
infant dose (RID), expressed as:
𝑅𝐼𝐷&[%] =& ,("#-."%/(%) .012*.
3
!4 $450(-
6 7
,("#-.!()*+%(#.012*.
3
!4 $450(-
6 7
&× 100
(5)
7
This weight-normalized parameter relates the infant dose to the dose administered to the mother, and
is widely used to estimate the infant risk. Even though highly debatable, some experts consider an RID
of less than 10% indicative of a generally safe medicine for breastfeeding [33].
For medicine approved in pediatrics, it is more informative to compare the daily infant dose with the
therapeutic dose for an infant of the same age:
𝑅𝐼𝐷&)8*+(9*:)"; [%] =& ,("#-."%/(%).012*.
3
!4 $450(-
6 7
,("#-.9*0"()+";.012*.
3
!4 $450(-
6 7
&× 100
(6)
This weight-normalized parameter is a surrogate endpoint of the likelihood of PD effects.
Finally, the systemic exposure in the infant depends on the PK processes (i.e. absorption, distribution,
metabolism, and excretion: ADME) of the medicine in the infant. For instance, due to immaturity of
hepatic and renal functions, (pre)term neonates present a lower capacity to metabolize and excrete
the medicine and consequently have a higher risk of accumulation compared to term neonates and
even lower compared to infants [34,35].
2.1 Variability expected in the mother and the infant
Successful validation of bioanalysis methods for quantification of medicines in milk and plasma implies
that bias and imprecision are constrained to a maximum of 15% across the determination range
(except that 20% is tolerated at the quantification limit) [36]. Furthermore, several non-analytical
sources of variability can be observed during all steps of the medicine transfer through human milk.
First, a difference in plasma concentrations can be observed among mothers (inter-individual
variability) and within the same mother (intra-individual variability) due to intrinsic factors, such as
differences in body mass index, renal or hepatic impairment or genetic polymorphisms, or to extrinsic
factors, such as drug-drug interactions or food effects [37]. For example, decreased maternal CL leads
to increased medicine exposure for the breastfed infant even if in normal circumstances the exposure
was expected to be small.
Moreover, changes in milk composition over time within a given feeding may affect the medicine
transfer by passive diffusion. Indeed, a lipophilic medicine is more concentrated in the milk at the end
(hindmilk) than at the beginning (foremilk) of the feeding due to a difference in fat content [28]. The
milk composition also changes, as the breastfeeding stage progresses from colostrum, to transitional
and then to mature milk.
Finally, important variability in breastfed infants is also expected. A difference in ingested milk volume
or in the type of feed (e.g. exclusive or mixed feeding) implies different daily dose ingested by the
infant. Variability in infant exposure is expected to be due to changes in absorption processes and
8
gradual maturation of hepatic and renal functions during the first months of life [38]. In addition,
variability in genes that code metabolizing enzymes and transporters has also an important impact on
the infant exposure [39]. It is worth emphasizing that these described variabilities are not addressed
in the calculation of the M/P or the infant exposure markers.
3. Available approaches to assess medicine exposure in the infant through breastfeeding
This section reviews the available methodological approaches to study the transfer of medicine into
human milk and the subsequent infant exposure. Table 1 summarizes the identified advantages and
disadvantages.
Table 1. A summary of the main advantages and disadvantages of the described approaches to
assess medicine exposure in the infant through breastfeeding.
Approaches
Advantages
Disadvantages
Case reports/
case series
Opportunistic approach for sentinel
data often triggered by potential
ADR (high and fast applicability)
Low evidence level pyramid
Low cost
Difficult to evaluate causality between
effect-medicine
Absence of patient recruitment
processes
Poor representativeness with lack of
population estimates and variations in
medicine concentrations
Usually, short follow-up time
Challenges associated with
interpretation of case reports on the
same medicine
Easy to report to the scientific
community
Lack of prospectively collected
information
Publication bias in the literature
Traditional PK
Easy methodological approach
Often high selection of participants
limiting relevant insights regarding the
sources of variability
Small number of participants in
comparison to population approach
Multiple samples per individual
(typically 6-9 samples)
Easy calculations
Practical constraints due to fixed
sampling times
Most often not possible to quantify the
interindividual variability
Difficult to identify sources of PK
variability
PopPK
Few samples per individual (e.g. 1-
3)
More complex methodological
approach (although user-friendly
software nowadays available)
Open source platform (R-package)
available as well as academic
licences for commercial platforms
(e.g. Monolix®)
Commercial software not easily
available for small companies
9
Possibility to pool data from
different studies, with different
designs
Reliability and predictive performances
of popPK model are highly dependent
on the availability/ quality of the input
data
Variable sampling times
Large number of participants
Identification of covariates
explaining PK variability
Possibility to predict medicine
exposure by simulations
PBPK
Do not strictly require clinical data:
bottom-up approach, based on first
principles and established
pathophysiology
Complex methodological approach
Open source platform (PK-Sim®)
available as well as academic
licences for commercial platforms
(e.g. Simcyp®, GastroPlus®)
Commercial platforms not easily
available for small companies and
contract research organization (CRO)
Several compound-specific models
can be used as a starting point for
extrapolation to new compounds
(i.e. the virtual subjects description)
A thorough quantitative
understanding of the medicine ADME
processes is required
Possibility to predict medicine
exposure by simulations
The predictive performance of the
PBPK model is highly dependent on the
availability/ quality of the input data
ADME: absorption, distribution, metabolism, excretion; ADR: adverse drug reaction; PBPK: physiologically-
based pharmacokinetics; PK: pharmacokinetics; popPK: population pharmacokinetics.
3.1 Case reports and case series
Case reports and case series are methodological approaches commonly used in clinical research to
describe unusual, novel, or complex medical observations. In clinical pharmacology, case reports
contribute to identify signals for new safety issues and allow generating hypotheses that may lead to
larger-scale studies. In the literature, they represent a high volume of publications as they benefit from
several facilitators such as their low cost, the usual absence of patient recruitment processes and short
follow-up time, allowing for good reactivity to report to the scientific community. However, case
reports are not at the top of the evidence-level pyramid because they cannot easily provide causality
between an observation/adverse reaction and a medicine. They also have very poor
representativeness as they report the situation of a single or a few individuals. Case reports should
thus be interpreted with caution [3,40].
In the area of drug safety in breastfeeding, case reports are particularly appropriate for describing ADR
or the absence of ADRs in breastfed infants whose mothers take a medicine. More importantly, clinical
observations in this specific area often provide the first or only concentrations of medicines in maternal
and/or infant plasma, and/or in human milk. These measurements often allow the first estimation of
10
RID, which remains case-specific and can hardly be generalized to the population. Case reports with
drug concentrations in maternal or infant plasma, or human milk, are generally pilot approaches that
provide valuable sentinel data for larger-scale studies.!!
Unfortunately, due to a large heterogeneity in the data collection method, such as in the medicine
dosage or duration, the time elapsed after the dose intake, the sampling periods, with or without first
in utero exposure, and the clinical assessment of the infant at different ages, an important variability
in results of case reports is observed [41]. Moreover, it is difficult to explain the sources of this
variability, primarily because there is a lack of prospectively collected information.
In order to improve the reporting of case reports in terms of completeness and transparency, Anderson
et al. have suggested a list of relevant information (“core outcome set”) [41]. Among the proposed
elements, Anderson et al. have pointed out the importance of clearly mentioning the dose and the
route of administration of the maternal medicine, the maternal therapy duration before blood sample
collection, as well as the moment of the milk sample collection (i.e. foremilk, midfeeding milk,
hindmilk, or an aliquot from a complete emptying of breast(s)). The overarching aim of this initiative is
to make these case reports more relevant to be extrapolated to other individual settings.
3.2 Pharmacometric approaches
PK describes the relationship between the dose of a medicine and the manner in which its
concentration changes over time (i.e. PK profile) in several matrices, such as plasma, saliva, or human
milk. ADME processes drive the kinetics of a medicine and the most important PK parameters include
the CL, the Vd, the AUC, the t½, the time to peak concentration (Tmax) and the peak concentration (Cmax)
[42]. When present, active metabolites should be included in the characterization of medicine
behavior, as their pharmacological potency (therapeutic and/or toxic) and PK can contribute
significantly to the overall treatments’ safety and efficacy. Differences in the ADME processes among
individuals result in PK variability, which is also an important aspect to explore.
Several strategies exist to gain more insight into the medicine ADME. Traditional methods rely on
simple mathematical equations, such as non-compartmental analyses (NCA), and are more adequate
to study homogeneous populations, in which the sources of variability are very limited [43].
Conversely, mammillary compartmental methods divide the body into interconnected compartments
and use complex systems of differential equations to describe medicine PK. Three main compartmental
PK strategies can be discerned: the top-down (i.e. popPK), the middle out and the bottom-up (i.e. PBPK)
modeling approaches. PopPK modeling provides a simplified description of the observed data, and it
is thus a “data-driven” methodology, in which the body is divided into concentration homogeneous
areas (i.e. compartments) that exchange between them. On the other hand, PBPK modeling employs
11
a mechanistic description of human anatomy and physiology (e.g. organ volume and blood flow) and
medicine-related parameters (e.g. molecular weight, lipophilicity, and intrinsic ADME properties, such
as CL at the enzyme level) [44]. The middle-out approach combines both strategies by starting from a
mechanistic description and then further optimizing the model based on observed clinical data.
Data-driven (i.e. top-down in a modeling process) approaches require data collected from the
population of interest. An opportunistic sampling approach is used and consists of identifying
breastfeeding women that take a pharmacological treatment in routine clinical care and invite them
for enrollment in a PK study. Importantly, these studies also inform bottom-up modeling strategies as
they allow for their refinement and verification of the predictive performance (see section 3.2.3).
3.2.1 Traditional PK approach
Traditional PK analyses require intensive (or rich) sampling from each individual, reflected in multiple
samples collected at fixed usually pre-established time intervals. Consequently, a full concentration-
time profile is built for each individual, and individual PK parameters are usually calculated using simple
mathematical equations or models. The usual summary statistics then allows describing average PK
parameters and associated standard errors [45].
Such methods are useful to analyze data collected in a small and relatively homogeneous group of
highly selected individuals, with the subjects that might inflate the overall interindividual variability
systematically excluded. Indeed, this approach does not allow determination of the underlined
interindividual variability, whose contribution is hidden in the PK parameters standard deviations
together with other sources of variability, such as analytical and procedural errors. The results of
traditional PK studies are not mainly obtained in the “real-world” population due to the highly selection
of participants. Moreover, the small sample size is another limiting factor for the assessment of the
“real-world” PK variability. Interestingly, traditional PK studies remain the reference for
pharmaceutical phase I clinical trials and, despite the cited limitations, are usually conducted in post-
marketing phase to evaluate medicine safety during breastfeeding. The lack of information on
variability provided by traditional PK studies is an important limitation because it does not allow the
identification of possible risk factors explaining extreme concentrations that may in fact be the
situations that pose clinical problems [46].
3.2.2 PopPK modeling
PopPK modeling is a top-down approach with individual data (i.e. concentrations, dosage
administration history, and individual characteristics, such as age, weight, phenotype) collected in
several subjects and pooled together to describe the typical PK profile (i.e. the average concentration-
12
time profile in the population of interest) together with the observed variabilities through combined
mathematical and statistical models. In addition, it helps identifying the demographic, environmental,
or genetic factors responsible for the reported variabilities [47]. Parametric and non-parametric
software programs (e.g. NONMEM, Monolix, Phoenix, and the R package Pmetrics) exist for popPK
analyses [48-51]. Figure 1 provides a schematic representation of the popPK general principles and
main applications. In general, user-oriented methodological aspects are discussed elsewhere [52,53].
Figure 1. Population pharmacokinetic approaches allow characterizing medicine pharmacokinetics and associated variability
pooling together samples collected in several individuals receiving the medicine of interest. The methodology helps
identifying factors that are responsible for the observed variability and retrieving individual PK profiles that are used for
therapy individualization. The continuous line represents the typical PK profile, and the two black dashed lines the extreme
percentiles (usually 5% and 95% percentiles). The colored dashed lines represent the individual PK profiles. Figure created
with permission from Biorender.com.
The main advantages of the popPK approach are that few samples per individual (sparse sampling) are
enough for adequate model development, and that it allows studying data collected in heterogeneous,
homogeneous or combined settings. Therefore, it is possible to combine data from different studies
into a single analysis, whether from routine clinical care (opportunistic sampling) or from clinical
research. Moreover, this approach allows for sample collection at variable time following medicine
administration, as long as the exact sampling times relative to the dose administration are
documented. Simultaneous plasma and milk, but also milk-only collections, are conceivable for the
application of popPK approaches in breastfeeding studies.
13
The popPK approach assumes that the body is divided into interconnected compartments, i.e.
kinetically homogenous body parts that do not necessarily correspond to true organs or physiological
compartments, that exchange between them. In the field of medicine and breastfeeding, the milk is
treated as a peripheral compartment linked to maternal plasma, i.e. the central compartment, with an
immediate medicine exchange between plasma and human milk (Figure 2). In studies with
simultaneous plasma and milk sample collections, classic microconstant rates between any peripheral
and central compartments might be used to describe the equilibrium between milk and maternal
plasma compartments, which alternatively might be well captured assuming a proportionality
between their corresponding medicine concentrations (equation 3) [54]. Mutual description of
concentration-time profiles of parent medicine and active metabolites is possible following the same
strategy for the metabolite(s) quantified in plasma and milk.
Figure 2. Compartmental structure of a popPK model. k12 and ki12: drug absorption rate constant in mother and infant,
respectively; ki12,m: metabolite absorption rate constant in infant; k20 and ki20: elimination rate constant from the parent drug
plasma compartment function of drug clearance and volume of distribution in mother and infant, respectively; k23 and ki23:
metabolic rate constant from the drug to the metabolite in mother and infant, respectively; k30 and ki30: elimination rate
constant from the metabolite plasma compartment function of metabolite clearance and volume of distribution in mother
and infant, respectively. The transfer of the drug and metabolite from the plasma to the milk can be describe with
microconstants or with the milk-to-plasma ratio.
PopPK model development consists of two principal steps: structural, which defines the number of
compartments (i.e. central and peripheral compartments) as well as the type of absorption and
elimination, and statistical, characterizing the observed interindividual and residual variabilities. The
model development is followed by the covariate analysis, which allows detecting the available
14
clinically, biologically and demographically plausible factors associated with such variabilities. A
combination of well-established statistical and graphical techniques in popPK as well as clinical
considerations helps in identifying the model best capturing simultaneously the active compounds PK
and associated variabilities in both milk and plasma in relationship with influential covariates. Being a
data-driven approach, data availability and quality are fundamental to build a reliable and informative
popPK model [55].
Once developed, the final popPK model needs to be validated by assessing its prediction performances
and reliability throughout statistical and visual internal as well as external validation methods [56,57].
The latter is rarely performed because of the difficulty to find an externally independent dataset and
it becomes practically impossible in the breastfeeding context where even less data are available. A
valuable alternative is data splitting, which, however, implies having a sufficiently large initial dataset
[58].
The validated model can then be used for both Bayesian forecasting and model-based simulations,
which are fundamental applications of population approaches. The first method consists of predicting
the individual PK parameters of a subject given some concentrations and the final model so as to
predict the concentration-time profile of the subject in the future for medicine dosage adjustments if
needed [59]. In the case of well-established medicine plasma concentration-response relationship,
Bayesian forecasting thus allows treatment efficacy while preventing undesired secondary effects for
lactating mothers. Model-based simulations, including the observed variability, are instead used to
predict medicine exposure and, if deemed necessary, to investigate alternative dosage regimens and
scenarios for a subgroup of individuals with specific characteristics, as described elsewhere [54,60]. In
breastfeeding, the simulation of maternal concentration profiles in human milk for a large number of
virtual mother-infant pairs allows predicting the variability in medicine transfer into milk and therefore
the exposure for breastfed infants.
To illustrate this approach, Weisskopf et al. studied escitalopram and its main metabolite plasma-milk
transfers in lactating women through a popPK approach and identified mothers’ predicted
polymorphisms (i.e. cytochrome (CYP) 2C19 and 2D6) and milk composition (i.e. fat content during the
same breastfeeding and sampling moment from delivery) as the main factors affecting both
compounds PK [54]. In this study, model-based simulations allowed estimating both RID and adult dose
equivalent over 6 months as a function of the retained covariates using the daily ingested medicine
amount through breastfeeding as the medicine dose, being the active metabolite amount negligible in
milk. The obtained results provided reassurance on the use of escitalopram by breastfeeding mothers.
Yet, the study could not provide information about the medicine concentrations in the breastfed
15
infants due to lack of data. Actual breastfed infant exposure can in general be easily predicted using a
classical popPK approach, but this is practically unfeasible because of the lack of medicine
concentrations in infants. A methodological alternative would be to adapt the popPK model developed
on observed adult clinical data to the infants and then use it for exposure prediction. However, to
achieve this aim, several assumptions on the infant ADME as well as knowledge about differences in
infant physiology compared to adults are required [61,62]. Ontogeny of CYPs (and therefore also the
extent of intestinal and hepatic first-pass effects) clearly plays a crucial role for CYP-mediated medicine
disposition, and the same holds true for maturation of renal elimination pathways [38].
Importantly, the FDA has proposed popPK modeling as a possible approach for clinical lactation studies
since 2005 [63]. Despite the proven benefit in terms of clinical study design, the applications of popPK-
based methodology remain rare in the field of clinical lactation studies.
3.2.3 PBPK modeling
PBPK models use a bottom-up and mechanistic approach to predict the PK profile of a medicine and/or
its metabolites [64]. Several commercial and open-source PBPK platforms are available [e.g. Simcyp®
(Certara); PK-Sim®/MoBi® (Open Systems Pharmacology) and Gastroplus® (SimulationsPlus)]. The
major advantage of these platforms is that they feature databases with many of the organism-related
(patho)physiological parameters [64]. Several efforts are ongoing to extend the Open Systems
Pharmacology platform to build lactation PBPK models [65,66]. Similarly, the Simcyp® platform was
recently extended with a specific module for lactation [67].
PBPK models are computational models, made up of physiologically relevant compartments
corresponding to the tissues in the body of a specific population [64]. Figure 3 summarizes the
necessary steps for PBPK model development. First, parameters obtained in vitro (e.g. measurement
of solubility, membrane permeability, enzyme-mediated CL, and protein binding), in silico (e.g.
prediction of logP) and/or in vivo (e.g. determination of tissue distribution in animals) experiments are
entered in the PBPK software. In a second step, variability in these parameters is implemented to
obtain the PK profile in the population. Finally, the profile is compared to data from clinical trials to
evaluate the PBPK model predictive performance. If necessary, the PBPK model is further optimized to
fit the available clinical data. The predictive performance of the PBPK model is highly dependent on
the availability and quality of the input data [68]. PBPK models can be used to make a simulation for
an individual, as well as for a population.
16
Figure 3. Physiologically based pharmacokinetic (PBPK) is a mechanistic approach that combines organism-related and
medicine-related parameters to predict the PK profile in an individual. These parameters are obtained from in vitro (e.g.
measurement of solubility, permeability, enzyme-mediated CL), in silico (e.g. prediction of logP) and/or in vivo (e.g.
determination of tissue distribution in preclinical animals) experiments. In a second step, variability on these parameters is
implemented to obtain the PK profile in the population. Finally, the profile is compared to data from (preferably multiple)
clinical studies to evaluate the predictive performance. The continuous line represents the typical PK profile, and the two
black dashed lines the extreme percentiles (usually 5% and 95% percentiles). Figure created with permission from
Biorender.com.
One of the main benefits of PBPK is that this technique can even be applied when only pre-clinical data
are available [64]. When (sparse) clinical data become available, they can be used for evaluation of the
predictive performance of the model, i.e. to calculate the fold error between the observed and
simulated data. This evaluation of the predictive performance can be based on concentration-time
profiles in plasma, but it is also possible to use other types of observed data (e.g. urine, human milk,
or brain). Information about the dose (amount, administration route) and sampling time is essential.
However, also characteristics of the individual (e.g. age and weight) and confounding factors (e.g.
smoking, co-medication, lactation, or pregnancy) can significantly improve the quality of the
predictions. When higher resolution data become available, the clinical data can be used for parameter
identification or retrograde calculations, for instance, of intrinsic hepatic CL based on in vivo CL. Such
a tool for retrograde calculations is, for example, available within Simcyp® but can also be relatively
easily performed manually. The goal of parameter identification is to minimize the residuals between
the simulation and the observed data, by optimizing the values of specific input parameters using an
algorithm.
A typical PBPK model often does not have a breast/milk compartment. Such a compartment should be
added in order to predict the transfer into the milk. Transfer into the milk can be modeled as direct
17
transfer from the blood to the milk and/or via uptake in the breast tissue [66]. PBPK models were first
used in this field to estimate the transfer of environmental chemicals into the milk [69]. More recently,
PBPK models have also been used to predict the transfer of medicines into the human milk [66]. The
capability to predict concentrations in the tissue(s) of interest constitutes one of the main advantages
of PBPK. PBPK models allow simulating the PK profile of a medicine and/or its metabolites, including
the plasma-time and milk-time concentration curves, even when there is no rapid exchange between
both compartments. This point is particularly important as this approach gives a complete overview of
the kinetic behavior of the medicine in the explored matrices and allows the determination of all
previously mentioned parameters (e.g. M/P, RID).
In addition, PBPK modeling has also been used to predict the systemic exposure of infants to maternal
medicine via breastfeeding [70]. Prediction of infant exposure to maternal medicine via breastfeeding
is possible by coupling the maternal PBPK model to the infant PBPK model (Figure 4). First, a reference
(healthy) volunteer PBPK model is built and then modified to represent a breastfeeding woman. The
simulated concentration of the medicine in the milk is then used to calculate the “dose” (equation 1)
that the infant receives via breastfeeding. Alternatively, the dose can be calculated based on
measurements in human milk. In a next step, an infant PBPK model is made by scaling a (healthy)
reference individual PBPK model to an infant. The combination of the infant PBPK model with the
calculated dosing information from the milk allows predicting the exposure in the infant. The main
advantage of the PBPK approach to estimate the exposure to maternal medicine via breastfeeding is
that an infant PBPK model also considers absorption processes in the infant, which is not addressed
with dose-based parameters like the RID. An infant PBPK model allows predicting the plasma
concentration-time profile of the infant and thus calculating multiple parameters. Recently, the upper
AUC ratio (UAR) has been suggested as a new parameter for risk assessment [71]. The UAR is the 95th
percentile of the simulated infant AUCinfinity divided by the median adult AUCinfinity. The advantage of
this parameter over the currently used parameters is that it allows the identification of “outlying”
infants that are at-risk for a high medicine exposure via breastfeeding.
18
Figure 4. General workflow for development and verification of lactation PBPK models Figure created with permission from
Biorender.com
Furthermore, PBPK modeling allows investigating the effect of a specific genotype or the plausibility of
milk secretion mechanisms [64]. As an example, Olangunju et al. showed that the maternal CYP2B6
genotype affects the efavirenz exposure in the milk [72]. Knowledge on the maturation of several
processes (e.g. enzyme ontogeny) is required for pediatric PBPK modeling and despite recent advances
in this field, such data are currently still lacking [73]. PBPK modeling can also be used to simulate and
compare different scenarios.
It is hereby important to realize that PBPK modeling has been accepted by regulators in various fields
(e.g. drug-drug interactions or drug-food interactions) [64]. The same holds true for simulating the PK
profile in special populations (e.g. renal failure, pediatrics, or postpartum women with medical
conditions) and is therefore promising for simulations during lactation [64].
3.3 Illustration of the relevance of popPK and PBPK approaches to assess medicine safety in
breastfeeding
Estimating medicine exposure via human milk is essential to evaluate medicine safety for breastfed
infants. The quantification of the circulating medicine concentration in this vulnerable population can
increasingly be achieved by modeling and simulation approaches, as also recognized by regulatory
agencies. This section illustrates our proposed strategy to predict actual medicine exposure in
breastfed infants by PBPK and popPK modeling using escitalopram, a selective serotonin reuptake
inhibitor used for perinatal and postpartum depression, as example.
3.3.1 Escitalopram doses through breastfeeding
The escitalopram “medicine doses” received by the infants through breastfeeding were calculated
using a previously published popPK model describing escitalopram and its major active metabolite
19
transfer into human milk (see section 3.2.2 and Supplementary Material) [54]. Model refinement
focused only on escitalopram concentrations because of the minor contribution of the active
metabolite on the overall medicine dose in infants (NONMEM code presented in Supplementary
Material) [54,74].
Milk composition and predicted phenotypes for CYP2C19 and 2D6 significantly affected compounds
PK in both milk and plasma with, in particular, poor metabolizers (PM) presenting higher
concentrations than the other individuals. Virtual mother-infant pairs were simulated considering an
adult dose of 10 mg, CYP2C19 and 2D6 combinations for mothers and term male infants of 7 and 56
days (weight retrieved from WHO charts). Such simulations allowed retrieving mother’s steady-state
exposure (AUCSS) and the medicine ingested by the infant at each breastfeeding. Table 2 summarizes
both mothers’ AUCSS and infant daily doses, obtained summing up the simulated ingested amount at
each breastfeeding during 24 h (equation 1), stratified according to mothers’ predicted phenotypes
and infant age. For instance, in the worst-case scenario, i.e. a mother who is a poor metabolizer for
both CYP2C19 and CYP2D6, a 56-day-old infant would ingest an average dose 100 times lower than
that of the mother through human milk. Weighted median exposure calculated according to the
phenotypic prevalence in the adult Caucasian population (CYP2D6 PM 7%, CYP2C19 PM 2%, and 1.4
‰ for double deficiency) was of 389 µg*h/L.
Table 2. Simulated steady-state infant daily escitalopram doses as a function of mothers predicted
phenotypes for CYP2D6 and CYP2C19 together with mother steady-state exposure.
Mother
Infant
Dose [mg]
CYP2C19
PM
CYP2D6
PM
AUCSS [µg*h/L]
Median (PI90%)
Postnatal age
(Days)
Vmilk*
(L/day/kg)
Daily Dose (
µ
g)
Median/mean (PI90%)
10
No
No
356 (221-585)
7
0.13
15/16 (9-26)
No
Yes
678 (419-1112)
29/30 (17-50)
Yes
No
750 (456-1205)
31/33 (18-53)
Yes
Yes
1425 (867-2290)
60/63 (35-102)
No
No
356 (221-585)
56
0.14
26/26 (15-173)
No
Yes
678 (419-1112)
49/51 (28-83)
Yes
No
750 (456-1205)
53/56 (31-91)
Yes
Yes
1425 (867-2290)
101/103 (58-173)
AUCss: area under the curve at steady state; CYP: cytochrome; PI90%: 90% prediction interval; PM: poor
metabolizers; Vmilk: volume of milk; *according to Yeung et al. equation [32]
20
3.3.2 Escitalopram exposure prediction in infants following milk ingestion: PBPK vs popPK
approaches
Simulations of PBPK and popPK models in infants adapted from adult models were then performed to
predict the escitalopram AUCSS in male infants receiving the medicine through breastfeeding. Single
daily infant dose per age group was retrieved weighting the mean infant daily doses in Table 2 by the
corresponding phenotypic prevalence in the adult Caucasian population. PBPK and popPK model
simulations were thus performed in 7 days (n = 1000) and 56 days (n = 1000) old infants under 18 µg
and 30 µg per day, respectively. PopPK simulations were in addition performed considering actual
medicine intakes through human milk at each breastfeeding per infant as a function of all possible
combination of mothers predicted CYP2D6 and CYP2C19 phenotypes (n = 1000 per group).
The PBPK model published by Delaney et al. allowed estimating infant AUCSS [75]. These authors
showed the feasibility of PBPK modeling to predict infant exposure, constructing an adult PBPK model
in PK Sim, including metabolism via CYP3A4, CYP2D6 and CYP2C19. The model was then adapted
(physiology and anatomy) to represent an infant. The standard ontogeny functions available in PK-Sim
were used to scale the elimination to the infant.
PopPK models for escitalopram PK characterization in lactating infants have not been published yet.
Therefore, the developed model in mothers, obtained neglecting the phenotype effects, was scaled
from mother to infant to mimic organ maturation and their impact on elimination using an allometric
function:
𝐶𝐿"%/(%) =𝐶𝐿!1)8*+
B
<*"48)!"#$"%
=>
C
91?*+
(7)
with 70 kg being the typical adult weight and power the allometric exponent. To include the maturation
of CYPs on escitalopram CL, an age-dependent allometric exponent (i.e. 1.1 for infants from 0 to 3
months of age) was preferred to the commonly employed fixed value of 0.75 [61]. The same variability
as in the model developed in mothers was assumed (NONMEM code presented in Supplementary
Material).
In order to validate the predictions of both approaches, a literature search was conducted to retrieve
observed escitalopram concentrations in breastfed infants. One study reported that medicine
concentrations in infants were below the limit of detection (<3 µg/L) without providing information
about the time interval between infant sampling and mother medicine intake nor on mother treatment
history [76]. As suggested in Delaney et al., the model-predicted average concentration at steady-state
(Cav,SS) was thus calculated and compared to the observed escitalopram concentration to assess both
PBPK and popPK models predictive performances [75].
21
The PBPK model-predicted median (PI90%) AUCSS values in the infants shown in Figure 5 were 13.6 (5.7-
28.4) µg*h/L and 14.6 (6.8-29.1) µg*h/L for the 7- and 56-day-old infants receiving the phenotype-
weighted average daily medicine dose of 18 and 30 µg, respectively. Using the popPK approach,
median (PI90%) AUCSS values in the infants were 19.9 (9.9-35.5) µg*h/L and 20.1 (10.3-36.5) µg*h/L for
the 7- and 56-day-old infants, respectively (Figure 5). Comparable results were obtained simulating
each medicine intake through breastfeeding over 24 h. Infant exposure appears to be 4% and 5% of
that predicted in mothers using the PBPK and the popPK approaches, respectively. This results in a
Cav,SS of approximately 0.6 µg/L and 0.7 µg/L for both infant age groups, supporting PBPK and popPK
models predictive performances. It should be noted that in silico PI90% infant exposure obtained
through popPK simulations was of 65.1 (28.8-142.7) µg*h/L for PM mothers for both CYP2D6 and
CYP2C19 (prevalence of 1.4‰), which corresponds to a Cav,SS of 2.7 (1.2-6.0) µg/L.
Figure 5. PBPK (blue boxes) and popPK (red boxes) model-predicted infant exposure (AUCSS) at 7 and 56 days of postnatal age
with phenotypes-weighted daily escitalopram doses of 18 µg and 30 µg, respectively, received through breastfeeding. The
horizontal lines in the box plots represent the median, the boxes the interquartile ranges, the extremities of the vertical lines
the minima and the maxima, and the dots the outliers. Figure generated using the R package.
4. Discussion and Conclusion
When one considers that the first case reports on medicine safety in breastfeeding were published in
the literature already several decades ago, it seems legitimate to ask why we continue to work, as
health-care providers, with information that lacks robustness. The main reason is probably linked to
the historical exclusion of breastfeeding women from clinical trials that limits the generation of
knowledge early in the life cycle of a medicine. In addition to the lack of data from pre-marketing
clinical trials, this area seems to be stuck in the assessment methods used to identify signals or provide
initial security information. The generation of knowledge must shift from case reports or small
22
longitudinal studies to methods allowing a complete characterization of medicine transfer into milk
and the infant systemic exposure linked to the amount transferred. This characterization requires a
description of the variability around milk transfer and infant systemic exposure to allow identification
of risk factors. It is becoming urgent to fill the knowledge gap using methods that can simulate the
most extreme situations (e.g. 95% percentile) in using population-based information. It is only this level
of information that will ultimately reassure all stakeholders (patients, health-care providers, drug
manufacturers, and regulatory authorities) on the compatibility or otherwise of the medicines during
breastfeeding.
The two modeling approaches (PBPK and popPK) presented in this paper hold the potential to fill some
of these knowledge gaps. While these approaches require a more complex computational process,
their outputs bring a whole-new dimension to the information generated, adding the notion of
variability and the possibility to explain it using individual characteristics. Indeed, the variability in
medicine transfer into human milk should be carefully considered as it affects infant exposure and
facilitates identification of circumstances at risk of toxicity. Furthermore, both popPK and PBPK allow
model-based simulation to retrieve the expected range of exposure for breastfed infants in situations
(e.g. worst-case scenario, phenotypes) for which clinical data are much more difficult to capture, using
extreme covariate values. The possibility given by these models to describe numerous scenarios with
a wide range of level of exposures to be expected in the clinical setting probably explains why these
modeling approaches have been used for some time now in populations where designing and
conducting clinical studies is more challenging (e.g. pediatrics) [77]. Both approaches have also been
used to inform drug labeling that otherwise would have been silent in some specific situations [78].
5. Expert Opinion
In the field of research and medicine use during breastfeeding, we are currently in a transition period
with ethical perspectives to protect mothers and infants through research instead of from research
[15]. Although data on medicine exposure in infants through human milk are reassuring for the most
commonly used medicines, these data are often generated by ad hoc case reports or case series that
are not very robust or are based only on theoretical principles (e.g. physicochemical properties). To
ensure evidence-based medicine, it is therefore important to perform larger standardized human
lactation studies.
The limited data available mainly from small sample size studies or from individual cases may reflect
the notorious difficulty to enroll breastfeeding women in clinical studies. Several strategies can be
applied to provide robust and generalizable data while reducing the burden of studies on breastfeeding
23
women in order to improve consent rates for the participation of breastfeeding women in clinical
studies.
One of these strategies would be to capitalize on the existence of milk banks. In recent years, donor-
based human milk banks have demonstrated remarkable growth, as clinicians have come to value the
importance of banking human milk not only as a nutritional option but also as a potentially life-saving
therapy in certain clinical situations such as extreme prematurity [79]. These banks are currently
mostly not designed for research use and do not fully meet requirements that are distinct to clinical
use. Women taking medicines are even often excluded from contribution to these banks. However,
milk banks facilities provide a great opportunity to collect human milk for research on drug safety in
breastfeeding with a minimal investment of resources and energy. They have experienced staff and
standardized operating procedures for milk collection, facilities to store human milk, and the potential
to identify women on medicine treatments as they are commonly considered as exclusion criteria.
Thus, the development of an international research infrastructure to collect, store, and analyze
samples of human milk and blood capitalizing on existing milk banks would provide access to a large
number of lactation data with a minimal investment of resources and energy.
The collection through milk bank will also allow the use of another strategy minimizing the burden of
data collection, which is the sparse sampling approach. Intensive sampling in traditional PK studies
presents practical constraints (e.g. sample collection takes time) which can be difficult to perform
during breastfeeding. Moreover, the FDA recommendation [20] regarding the collection of the entire
milk volume from both breasts over 24 h, with the remainder of the collected milk after aliquoting
being refed to the infant under certain circumstances, is often criticized by ethical boards and by
working groups because of the risk of disrupting breastfeeding (i.e. nipple confusion) [17,20]. With the
collection of sparse samples under less strict conditions than with traditional PK, as well as the
possibility to use opportunistic samples, the sampling design of popPK is especially useful for the
characterization of medicine PK in special population, such as breastfeeding women and infants. The
milk matrix collection coupled with the highly flexible sampling strategy might favorably affect the
recruiting rate in breastfeeding studies.
A third strategy would be to favor milk-only study design. It is easily conceivable to send a milk sampling
kit to mothers’ home from a milk bank facility. One of the drawbacks generally considered with this
strategy is the difficulty of having complete and accurate information on the time of medicine intake
and sampling collection, as well as all the specific information for the covariable analysis when they
are not collected in clinical setting (e.g. laboratory results). Coupling this strategy with a milk bank
facility would favor the involvement of experimented staff empowering women to follow clear study
24
procedures for milk collections, reporting the important data in a well-established case report form
and, thus, preventing this limitation.
Last but not least, human lactation studies should be incorporated into the regulatory requirements,
as done for the pediatric population, to speed things up. One way would be to rely on the 5-year
validity of a marketing authorization and the need for renewal. Regulators could make human lactation
studies mandatory at the renewal stage of a marketing authorization for drugs frequently used by
women of childbearing age.
Finally, interestingly, PBPK modeling can be used to predict infant exposure to maternal medicines via
breastfeeding, even when there are no clinical data on infants. Indeed, PBPK models can be used
during the entire medicine development process and can inform the product labeling until clinical
experience accumulates. Initial PBPK predictions are based on pre-clinical data and model refinement
happens along the process, as more data are available. Alternative scenarios can then be predicted in
a study but also in different populations conditioning on the fact that the specific population model is
available. In the context of breastfeeding, this means having a model for breastfeeding women, which
requires a mechanistic understanding of the physiological changes in postpartum period that is
currently available only to a limited extent. Nevertheless, popPK and PBPK are promising methods,
each in their own way, to minimize sample collection in breastfeeding women, easing research in this
particularly challenging field.
The two modeling approaches, PBPK and popPK described in this paper, allow for simulation of
maternal concentration profiles in human milk for a large number of virtual mother-infant pairs, which
can be used to predict the variability in medicine transfer into milk and therefore the exposure for
breastfed infants as illustrated with the escitalopram cases. Unfortunately, the validation of exposure
simulations in breastfed infants with popPK or/and PBPK based on medicine concentrations in infants
is rarely available, let alone when medicines are not indicated in this population (such as escitalopram).
As obtaining infant plasma samples is particularly challenging, we believe that validation of mechanistic
PBPK as well as popPK models to predict infant drug exposure through breastfeeding would benefit of
the use of new sampling techniques ethically more acceptable, such as dry blood spot (DBS) sampling
coupled to a validated method for DBS/plasma conversion, to increase parental, ethical, and health-
care provider acceptance [80]. There are several major advantages of using DBS, such as minimal
volume of blood required compared to conventional venipuncture; minimal risk of bacterial
contamination and/or hemolysis as compared with traditional methods; noninvasive and economic
collection of blood; conservation of blood spots for long periods with almost no deterioration of most
of the studied analytes [81]. Conventionally, DBS is being used for screening of neonates for congenital
25
and inherited metabolic disorders [82]. However, numerous emerging applications of DBS collection
have been reported, including PK studies and their use to ease therapeutic drug monitoring of several
medicines [81,83]. The ability of this strategy to facilitate the access to the limited amount of infant
samples enabling the validation of infant exposure predictions seems an interesting avenue to explore.
PBPK and popPK with their potential for simulation of infant exposure through human milk are
promising approaches to fill the gap in knowledge of medicine safety in breastfeeding. However, as
with any method transfer, there is a need to further adapt PBPK and popPK approaches specifically for
lactation studies. The maturation of individual ADME processes in infants is mainly driven by age or
size, further affected by non-maturational covariates like genetics, disease state, or environmental
factors [84]. While the integration of the effect of growth in popPK models is relatively easy through a
body weight metric, the incorporation of physiological maturation as a function of age is less clear.
Strategies to scale models from adults to infants are still limited, and only marginally consider age-
dependent factors. Within the PBPK framework, anatomical and physiological databases representing
the age dependence of processes affecting ADME are available across the age continuum, but still have
knowledge gaps. Consequently, there remain scenarios in which data are not available or less certain
in infants, such as the ontogeny of enzymes that are not typical or for many of the transporters. In
these cases, extrapolation from an adult PBPK model down to children and infants is reasonable until
the maturation of these active processes becomes important to overall exposure, and then there can
be considerable uncertainty in the exposure simulation. Lactation may also present confounding
environmental factors such that it is associated with alterations of various physiological factors, such
as rate of gastric emptying, growth, and body composition, or differences in medicine metabolism
compared to infants fed with formula. This information could be included into PBPK models if known.
There is thus an urgent need to further develop popPK and PBPK methods and models that better fit
the specificities of age-dependent maturation factors as encountered in the infant population. A
similar claim can be made for the postpartum, lactating women as both mother and infant display
time-dependent physiology.
Finally, while popPK and PBPK studies are powerful tools to generate a significant amount of relevant
safety information on medicine in breastfeeding, it seems urgent to define what would be the amount
of safety information that would be considered as enough to be reassured by the different
stakeholders, such as health-care providers, patients, and, ultimately, the manufacturer, and that
would allow a change of label. Currently, for an important proportion of medicines, the available safety
and PK data are reassuring and therefore discordant with the labeling. Well-trained health-care
providers have therefore already a fair level of reassurance while breastfeeding women navigate with
uncertainty because of discordant messages. To allow for a shift of paradigm, there is a need for an
26
alignment between the content of the labeling and the level of reassurance that can be derived from
the available safety and PK information. The level of evidence required by the manufacturer is
therefore a decisive element to shape the milk studies landscape and allow a significant change. A
publication by Roque Pereira et al. reported that only a few studies have generated meaningful
evidence in pregnancy labeling, suggesting that the current requirement for post-marketing studies is
ineffective in determining what constitutes adequate information to be included in the product
labeling [85]. This observation is probably also valid for the breastfeeding labeling. Labeling change
could be determined by a very large range of information types, such as PK information (e.g. non-
significant systemic absorption), information stemming from PK longitudinal studies, with rich
sampling procedure enrolling 10-20 breastfeeding women, bottom up mechanistic modeling
approaches, such as PBPK, with or without clinical validation, large sparse sampling popPK studies
enabling information on inter-individual variabilities with or without infant compartment or
pharmacoepidemiology studies assessing infant outcomes with short to long-term follow-up. All these
approaches have the potential to provide a better understanding of the risks for the infant exposed to
a medicine through breastmilk. It is the amount of time and resources that vary greatly between them.
Thus, should PK characteristics of medicines be screened first to identify candidate for PK longitudinal
studies (e.g. little inter-individual variability expected) from those for large sparse sampling popPK
studies? Should the risk for accumulation in the breastfed infant (e.g. long half-life of the parent
compound or of an active metabolite) or the pharmacological profile of the medicine (e.g. cytotoxicity)
ask for pharmacoepidemiology studies as the PK/PD relationship in adults are not transposable to
infants? Should the prevalence of use of a drug in women of childbearing age decide on the bottom-
up approach (e.g. low prevalence of use) instead of a top-down approach (e.g. high prevalence of use)?
Should pregnancy registries be re-designed to further observe infants during the lactation period and
after ? Can real-world data stemming from administrative data sources be harnessed to assess the
safety of medicine exposure through breastmilk (e.g. develop method to identify breastfeeding in
claims data)? All these issues must be addressed and answers agreed upon by all stakeholders if we
are to make a decisive move in the next 5-10 years to open up treatment to breastfeeding women with
a fair level of safety information to enable evidence-based, not fear-based, decision-making.
Declaration of interest
Marie Teil is an employee of UCB Pharma. The authors have no other relevant affiliations or financial
involvement with any organization or entity with a financial interest in or financial conflict with the
subject matter or materials discussed in the manuscript. This includes employment, consultancies,
honoraria, stock ownership or options, expert testimony, grants or patents received or pending, or
royalties.
27
Funding and Acknowledgements
This work has been completed as part of the ConcePTION study. The ConcePTION project has received
funding from the Innovative Medicines Initiative 2 Joint Undertaking under grant agreement No.
821520. This Joint Undertaking receives support from the European Union’s Horizon 2020 research
and innovation program and EFPIA. Nina Nauwelaerts also received a PhD scholarship by Research-
Foundation-Flanders (1S50721N). The research project leading to these results was conducted as part
of the ConcePTION consortium. This paper only reflects the personal views of the stated authors.
References
1. Saha MR, Ryan K, Amir LH. Postpartum women's use of medicines and breastfeeding
practices: a systematic review. Int Breastfeed J. 2015;10:28.
2. Anderson PO, Manoguerra AS, Valdés V. A Review of Adverse Reactions in Infants From
Medications in Breastmilk. Clin Pediatr (Phila). 2016 Mar;55(3):236-44.
3. Anderson PO, Pochop SL, Manoguerra AS. Adverse drug reactions in breastfed infants: less
than imagined. Clin Pediatr (Phila). 2003 May;42(4):325-40.
4. Drugs and Lactation Database (LactMed) 2006-. [cited 2022 September 1]. Available from:
https://www.ncbi.nlm.nih.gov/books/NBK501922/
5. Akus M, Bartick M. Lactation safety recommendations and reliability compared in 10
medication resources. Ann Pharmacother. 2007 Sep;41(9):1352-60.
6. Briggs GG, Forinash AB, Towers CV. Briggs Drugs in Pregnancy and Lactation: A Reference
Guide to Fetal and Neonatal Risk. Lippincott Williams & Wilkins; 2021.
7. Mazer-Amirshahi M, Samiee-Zafarghandy S, Gray G, et al. Trends in pregnancy labeling and
data quality for US-approved pharmaceuticals. Am J Obstet Gynecol. 2014
Dec;211(6):690.e1-11.
8. Anderson PO. Medication Information Sources for Breastfeeding Mothers. Breastfeed Med.
2017 Sep;12(7):396-397.
9. Schirm E, Schwagermann MP, Tobi H, et al. Drug use during breastfeeding. A survey from the
Netherlands. Eur J Clin Nutr. 2004 Feb;58(2):386-90.
10. Del Ciampo LA, Del Ciampo IRL. Breastfeeding and the Benefits of Lactation for Women's
Health. Rev Bras Ginecol Obstet. 2018 Jun;40(6):354-359.
11. Binns C, Lee M, Low WY. The Long-Term Public Health Benefits of Breastfeeding. Asia Pac J
Public Health. 2016 Jan;28(1):7-14.
12. ConcePTION [cited 2022 September 1]. Available from: https://www.imi-conception.eu/
13. Illamola SM, Bucci-Rechtweg C, Costantine MM, et al. Inclusion of pregnant and
breastfeeding women in research - efforts and initiatives. Br J Clin Pharmacol. 2018
Feb;84(2):215-222.
14. Weld ED, Bailey TC, Waitt C. Ethical issues in therapeutic use and research in pregnant and
breastfeeding women. Br J Clin Pharmacol. 2022 Jan;88(1):7-21.
15. PRGLAC - Task Force on Research Specific to Pregnant Women and Lactating Women. 2018
[cited 2023 January 26]. Available from: https://www.nichd.nih.gov/about/advisory/PRGLAC
16. Nooney J, Thor S, de Vries C, et al. Assuring Access to Safe Medicines in Pregnancy and
Breastfeeding. Clin Pharmacol Ther. 2021 Oct;110(4):941-945.
17. Eke AC, Olagunju A, Momper J, et al. Optimizing Pharmacology Studies in Pregnant and
Lactating Women Using Lessons From HIV: A Consensus Statement. Clin Pharmacol Ther.
2021 Jul;110(1):36-48.
28
18. U.S. Food and Drug Administration. Pregnancy and Lactation Labeling (Drugs) Final Rule.
2015 [cited 2022 September 4]. Available from: https://www.fda.gov/drugs/labeling-
information-drug-products/pregnancy-and-lactation-labeling-drugs-final-rule
19. Dinatale M, Roca C, Sahin L, et al. The Importance of Clinical Research in Pregnant Women to
Inform Prescription Drug Labeling. J Clin Pharmacol. 2020 Dec;60 Suppl 2:S18-s25.
20. U.S. Food and Drug Administration. Clinical Lactation Studies: Considerations for Study
Design. Guidance for Industry. 2019 [cited 2022 June 3]. Available from:
https://www.fda.gov/regulatory-information/search-fda-guidance-documents/clinical-
lactation-studies-considerations-study-design
21. Jordan S, Bromley R, Damase-Michel C, et al. Breastfeeding, pregnancy, medicines,
neurodevelopment, and population databases: the information desert. Int Breastfeed J. 2022
Aug 2;17(1):55.
22. Anderson PO. Drugs in Lactation. Pharm Res. 2018 Feb 6;35(3):45.
23. Begg EJ, Duffull SB, Hackett LP, et al. Studying drugs in human milk: time to unify the
approach. J Hum Lact. 2002 Nov;18(4):323-32.
24. Verstegen RHJ, Ito S. Drugs in lactation. J Obstet Gynaecol Res. 2019 Mar;45(3):522-531.
25. Ito N, Ito K, Ikebuchi Y, et al. Prediction of Drug Transfer into Milk Considering Breast Cancer
Resistance Protein (BCRP)-Mediated Transport. Pharm Res. 2015 Aug;32(8):2527-37.
26. Hale TW. Pharmacology Review: Drug Therapy and Breastfeeding: Pharmacokinetics, Risk
Factors, and Effects on Milk Production. NeoReviews. 2004;5(4):e164-e172.
27. Fleishaker JC. Models and methods for predicting drug transfer into human milk. Adv Drug
Deliv Rev. 2003 Apr 29;55(5):643-52.
28. Anderson PO, Sauberan JB. Modeling drug passage into human milk. Clin Pharmacol Ther.
2016 Jul;100(1):42-52.
29. Oo CY, Kuhn RJ, Desai N, et al. Active transport of cimetidine into human milk. Clin Pharmacol
Ther. 1995 Nov;58(5):548-55.
30. Atkinson HC, Begg EJ. Prediction of drug distribution into human milk from physicochemical
characteristics. Clin Pharmacokinet. 1990 Feb;18(2):151-67.
31. Anderson PO, Valdés V. Variation of milk intake over time: clinical and pharmacokinetic
implications. Breastfeed Med. 2015 Apr;10(3):142-4.
32. Yeung CHT, Fong S, Malik PRV, et al. Quantifying breast milk intake by term and preterm
infants for input into paediatric physiologically based pharmacokinetic models. Matern Child
Nutr. 2020 Apr;16(2):e12938.
33. Hale TW. Hale's Medications & Mothers'Milk 2022 [cited 2022 July 7]. Available from:
https://www.halesmeds.com/mnemonics/47700
34. Kearns GL, Abdel-Rahman SM, Alander SW, et al. Developmental pharmacology--drug
disposition, action, and therapy in infants and children. N Engl J Med. 2003 Sep
18;349(12):1157-67.
35. Gow PJ, Ghabrial H, Smallwood RA, et al. Neonatal hepatic drug elimination. Pharmacol
Toxicol. 2001 Jan;88(1):3-15.
36. U.S. Food and Drug Administration. Bioanalytical Method Validation. Guidance for Industry.
2018 [cited 2022 June 3]. Available from: https://www.fda.gov/regulatory-
information/search-fda-guidance-documents/bioanalytical-method-validation-guidance-
industry
37. Lin YS, Thummel KE, Thompson BD, et al. Sources of Interindividual Variability. Methods Mol
Biol. 2021;2342:481-550.
38. van den Anker J, Reed MD, Allegaert K, et al. Developmental Changes in Pharmacokinetics
and Pharmacodynamics. J Clin Pharmacol. 2018 Oct;58 Suppl 10:S10-s25.
39. Yalçin N, Flint RB, van Schaik RHN, et al. The Impact of Pharmacogenetics on
Pharmacokinetics and Pharmacodynamics in Neonates and Infants: A Systematic Review.
Pharmgenomics Pers Med. 2022;15:675-696.
29
40. Nissen T, Wynn R. The clinical case report: a review of its merits and limitations. BMC Res
Notes. 2014 Apr 23;7:264.
41. Anderson PO. Guidelines for Reporting Cases of Medication Use During Lactation. Breastfeed
Med. 2022 Feb;17(2):93-97.
42. Urso R, Blardi P, Giorgi G. A short introduction to pharmacokinetics. Eur Rev Med Pharmacol
Sci. 2002 Mar-Jun;6(2-3):33-44.
43. Gabrielsson J, Weiner D. Non-compartmental analysis. Methods Mol Biol. 2012;929:377-89.
44. Jamei M, Dickinson GL, Rostami-Hodjegan A. A framework for assessing inter-individual
variability in pharmacokinetics using virtual human populations and integrating general
knowledge of physical chemistry, biology, anatomy, physiology and genetics: A tale of
'bottom-up' vs 'top-down' recognition of covariates. Drug Metab Pharmacokinet.
2009;24(1):53-75.
45. Julious SA, Debarnot CA. Why are pharmacokinetic data summarized by arithmetic means? J
Biopharm Stat. 2000 Feb;10(1):55-71.
46. Charles B. Population pharmacokinetics: an overview. Aust Prescr. 2014;37:210-13.
47. Ette EI, Williams PJ. Population pharmacokinetics I: background, concepts, and models. Ann
Pharmacother. 2004 Oct;38(10):1702-6.
48. Beal SL, Sheiner LB, Boeckmann A, et al. NONMEM 7.4 users guides 1989-2018. Available
from: https://nonmem.iconplc.com/nonmem743/guides
49. MONOLIXSuite. Version 2021R2. Lixoft SAS, Antony, France; 2021.
50. Phoenix. Version 8.3. Certara, Princeton, New Jersey; 2020.
51. Fidler M, Wilkins JJ, Hooijmaijers R, et al. Nonlinear Mixed-Effects Model Development and
Simulation Using nlmixr and Related R Open-Source Packages. CPT Pharmacometrics Syst
Pharmacol. 2019 Sep;8(9):621-633.
52. Guidi M, Csajka C, Buclin T. Parametric Approaches in Population Pharmacokinetics. J Clin
Pharmacol. 2022 Feb;62(2):125-141.
53. Goutelle S, Woillard JB, Buclin T, et al. Parametric and Nonparametric Methods in Population
Pharmacokinetics: Experts' Discussion on Use, Strengths, and Limitations. J Clin Pharmacol.
2022 Feb;62(2):158-170.
54. Weisskopf E, Guidi M, Fischer CJ, et al. A population pharmacokinetic model for escitalopram
and its major metabolite in depressive patients during the perinatal period: Prediction of
infant drug exposure through breast milk. Br J Clin Pharmacol. 2020 Aug;86(8):1642-1653.
55. Samara E, Granneman R. Role of Population Pharmacokinetics in Drug Development. Clinical
Pharmacokinetics. 1997 1997/04/01;32(4):294-312.
56. Lindbom L, Pihlgren P, Jonsson EN. PsN-Toolkit--a collection of computer intensive statistical
methods for non-linear mixed effect modeling using NONMEM. Comput Methods Programs
Biomed. 2005 Sep;79(3):241-57.
57. Bergstrand M, Hooker AC, Wallin JE, et al. Prediction-corrected visual predictive checks for
diagnosing nonlinear mixed-effects models. Aaps j. 2011 Jun;13(2):143-51.
58. Brendel K, Dartois C, Comets E, et al. Are population pharmacokinetic and/or
pharmacodynamic models adequately evaluated? A survey of the literature from 2002 to
2004. Clin Pharmacokinet. 2007;46(3):221-34.
59. Keizer RJ, Ter Heine R, Frymoyer A, et al. Model-Informed Precision Dosing at the Bedside:
Scientific Challenges and Opportunities. CPT Pharmacometrics Syst Pharmacol. 2018
Dec;7(12):785-787.
60. Panchaud A, Garcia-Bournissen F, Csajka C, et al. Prediction of infant drug exposure through
breastfeeding: population PK modeling and simulation of fluoxetine exposure. Clin
Pharmacol Ther. 2011 Jun;89(6):830-6.
61. Mahmood I, Tegenge MA. A Comparative Study Between Allometric Scaling and
Physiologically Based Pharmacokinetic Modeling for the Prediction of Drug Clearance From
Neonates to Adolescents. J Clin Pharmacol. 2019 Feb;59(2):189-197.
30
62. Anderson BJ, Holford NH. Mechanistic basis of using body size and maturation to predict
clearance in humans. Drug Metab Pharmacokinet. 2009;24(1):25-36.
63. U.S. Food and Drug Administration. Guidance for Industry. Clinical Lactation Studies – Study
Design, Data Analysis, and Recommendations for Labeling. 2005.
64. Zhuang X, Lu C. PBPK modeling and simulation in drug research and development. Acta
Pharm Sin B. 2016 Sep;6(5):430-440.
65. Job KM, Dallmann A, Parry S, et al. Development of a Generic Physiologically-Based
Pharmacokinetic Model for Lactation and Prediction of Maternal and Infant Exposure to
Ondansetron via Breast Milk. Clin Pharmacol Ther. 2022 May;111(5):1111-1120.
66. Nauwelaerts N, Deferm N, Smits A, et al. A comprehensive review on non-clinical methods to
study transfer of medication into breast milk - A contribution from the ConcePTION project.
Biomed Pharmacother. 2021 Apr;136:111038.
67. Abduljalil K, Pansari A, Ning J, et al. Prediction of drug concentrations in milk during
breastfeeding, integrating predictive algorithms within a physiologically-based
pharmacokinetic model. CPT Pharmacometrics Syst Pharmacol. 2021 Aug;10(8):878-889.
68. Jones HM, Gardner IB, Collard WT, et al. Simulation of human intravenous and oral
pharmacokinetics of 21 diverse compounds using physiologically based pharmacokinetic
modelling. Clin Pharmacokinet. 2011 May;50(5):331-47.
69. Corley RA, Mast TJ, Carney EW, et al. Evaluation of physiologically based models of
pregnancy and lactation for their application in children's health risk assessments. Crit Rev
Toxicol. 2003;33(2):137-211.
70. Smits A, De Cock P, Vermeulen A, et al. Physiologically based pharmacokinetic (PBPK)
modeling and simulation in neonatal drug development: how clinicians can contribute.
Expert Opinion on Drug Metabolism & Toxicology. 2019 2019/01/02;15(1):25-34.
71. Yeung CHT, Ito S, Autmizguine J, et al. Incorporating Breastfeeding-Related Variability with
Physiologically Based Pharmacokinetic Modeling to Predict Infant Exposure to Maternal
Medication Through Breast Milk: a Workflow Applied to Lamotrigine. Aaps j. 2021 May
17;23(4):70.
72. Olagunju A, Rajoli R, Atoyebi S, et al. Physiologically-based pharmacokinetic modelling of
infant exposure to efavirenz through breastfeeding. AAS Open Research. 2018 05/14;1:16.
73. van Groen BD, Nicolaï J, Kuik AC, et al. Ontogeny of Hepatic Transporters and Drug-
Metabolizing Enzymes in Humans and in Nonclinical Species. Pharmacol Rev. 2021
Apr;73(2):597-678.
74. Rao N. The clinical pharmacokinetics of escitalopram. Clin Pharmacokinet. 2007;46(4):281-
90.
75. Delaney SR, Malik PRV, Stefan C, et al. Predicting Escitalopram Exposure to Breastfeeding
Infants: Integrating Analytical and In Silico Techniques. Clin Pharmacokinet. 2018
Dec;57(12):1603-1611.
76. Rampono J, Hackett LP, Kristensen JH, et al. Transfer of escitalopram and its metabolite
demethylescitalopram into breastmilk. Br J Clin Pharmacol. 2006 Sep;62(3):316-22.
77. Johnson TN, Rostami-Hodjegan A. Resurgence in the use of physiologically based
pharmacokinetic models in pediatric clinical pharmacology: parallel shift in incorporating the
knowledge of biological elements and increased applicability to drug development and
clinical practice. Paediatr Anaesth. 2011 Mar;21(3):291-301.
78. Jamei M. Recent Advances in Development and Application of Physiologically-Based
Pharmacokinetic (PBPK) Models: a Transition from Academic Curiosity to Regulatory
Acceptance. Curr Pharmacol Rep. 2016;2(3):161-169.
79. Geraghty SR, Davidson BS, Warner BB, et al. The development of a research human milk
bank. J Hum Lact. 2005 Feb;21(1):59-66.
80. Ostler MW, Porter JH, Buxton OM. Dried blood spot collection of health biomarkers to
maximize participation in population studies. J Vis Exp. 2014 Jan 28(83):e50973.
31
81. Gupta K, Mahajan R. Applications and Diagnostic Potential of Dried Blood Spots. Int J Appl
Basic Med Res. 2018 Jan-Mar;8(1):1-2.
82. De Jesús VR, Mei JV, Bell CJ, et al. Improving and assuring newborn screening laboratory
quality worldwide: 30-year experience at the Centers for Disease Control and Prevention.
Semin Perinatol. 2010 Apr;34(2):125-33.
83. Demirev PA. Dried blood spots: analysis and applications. Anal Chem. 2013 Jan 15;85(2):779-
89.
84. Allegaert K, Simons SHP, Tibboel D, et al. Non-maturational covariates for dynamic systems
pharmacology models in neonates, infants, and children: Filling the gaps beyond
developmental pharmacology. Eur J Pharm Sci. 2017 Nov 15;109s:S27-s31.
85. Roque Pereira L, Durán CE, Layton D, et al. A Landscape Analysis of Post-Marketing Studies
Registered in the EU PAS Register and ClinicalTrials.gov Focusing on Pregnancy Outcomes or
Breastfeeding Effects: A Contribution from the ConcePTION Project. Drug Safety. 2022
2022/04/01;45(4):333-344.
Annotated bibliography
** Provide a list of recommendations to promote the inclusion of pregnant and lactating women in clinical trials:
15. PRGLAC - Task Force on Research Specific to Pregnant Women and Lactating Women. 2018
[cited 2023 January 26]. Available from: https://www.nichd.nih.gov/about/advisory/PRGLAC
**European project aiming to shift paradigm in information on safety of medicine in breastfeeding :
12. ConcePTION [cited 2022 September 1]. Available from: https://www.imi-conception.eu/
**Important database for decision making in the clinical setting:
4. Drugs and Lactation Database (LactMed) 2006-. [cited 2022 September 1]. Available from:
https://www.ncbi.nlm.nih.gov/books/NBK501922/
** Provide a list of meaningful information when reporting a case of the use of an unreported medicine in
breastfeeding :
41. Anderson PO. Guidelines for Reporting Cases of Medication Use During Lactation. Breastfeed
Med. 2022 Feb;17(2):93-97.
* Example of popPK modeling in breastfeeding studies:
54. Weisskopf E, Guidi M, Fischer CJ, et al. A population pharmacokinetic model for escitalopram
and its major metabolite in depressive patients during the perinatal period: Prediction of
infant drug exposure through breast milk. Br J Clin Pharmacol. 2020 Aug;86(8):1642-1653.
* Example of PBPK modeling in breastfeeding studies
72. Olagunju A, Rajoli R, Atoyebi S, et al. Physiologically-based pharmacokinetic modelling of
infant exposure to efavirenz through breastfeeding. AAS Open Research. 2018 05/14;1:16.
32
Supplementary material
S.1 Escitalopram doses through breastfeeding
Escitalopram doses received by the infants through breastfeeding were retrieved using an adaptation
of the popPK model developed by Weisskopf et al. [54]. A two-compartment model with linear
absorption, elimination and metabolism adequately described parent medicine and its major active
metabolite data in plasma, with milk concentrations well captured using an M/P per compound.
Influential covariates were milk composition, predicted phenotype for CYP2C19 and 2D6, with, in
particular, poor metabolizers (PM) presenting higher concentrations of escitalopram and its active
metabolite than the other individuals. Milk composition was characterized by both sampling moment
with respect to infant birth (i.e., birth and 1 week vs 1 month) and fat content. Infants would thus
receive cumulative doses of escitalopram and its metabolite through breastfeeding. However, because
of the minor contribution of the active metabolite on overall medicine dose in infants [54,74], only
escitalopram doses were considered in our example. To this aim, original model refinement was
performed in NONMEM neglecting milk and plasma metabolite concentrations, verifying parameter
stability focusing only on escitalopram data and repeating covariate assessment on the reduced model.
A classical covariate analysis procedure was followed testing the significant factors that were
associated with metabolite disposition on the parent medicine CL. Escitalopram model structure was
found to be the same as the original model with the two predicted phenotypes affecting
simultaneously medicine CL (NONMEM code for the final model shown in S.3).
Model-based simulations of 5000 virtual mother-infant pairs per group were performed considering
all plausible predicted phenotype combinations for mothers and term male infants of 7 and 56 days
(weight retrieved by WHO charts), and milk composition to calculate steady-state mother’s exposure
33
(AUCSS) and medicine ingested by the infant at each breastfeeding. The values of the milk composition
covariates were chosen as to maximize medicine transfer into milk. A standard dose of 10 mg per day
was assumed for all virtual mothers with AUCSS calculated as the ratio between dose and individual CL.
S.2 Escitalopram exposure prediction in infants following milk ingestion: popPK approach
The popPK model developed in mothers using only escitalopram data in plasma was scaled to infants
neglecting the phenotypes effect and retaining the sampling moment. Indeed, escitalopram CL in
mothers at 1 month of infant life better reflects normal adult CL without pregnancy-induced
alterations. S4 shows the NONMEM code of the infant popPK model.
S.3 NONMEM code for escitalopram popPK model in mothers with final parameter estimates
$PROBLEM PK
$INPUT ID DAT1=DROP TIME AMT SS II DV NDV CMT EVID MOM FAT CYP2C19 CYP2D6
$DATA Data.csv IGNORE=#
$SUBROUTINES ADVAN6 TOL=3
$MODEL
NCOMP=3
COMP=(DEPOT)
COMP=(PLCIT); escitalopram plasma compartment
COMP=(MICIT) ; escitalopram milk compartment
$PK
IF (AMT.GT.0) THEN
TDOS=TIME
TAD=0.0
ENDIF
IF (AMT.EQ.0) TAD=TIME-TDOS
MOMN=0
IF (MOM.EQ.3) MOMN=1
CYPN=0
IF (CYP2C19.EQ.0) CYPN=1
CYPD=0
IF (CYP2D6.EQ.0) CYPD=1
MFAT=3.1
FFAT=(FAT-MFAT)/MFAT
IF(FAT.LT.0) FFAT=0
TVCL=THETA(1)*(1+MOMN*THETA(5))*(1+CYPN*THETA(6))*(1+CYPD*THETA(10))
CL=TVCL* EXP(ETA(1))
TVV2 = THETA(2)
34
V2=TVV2* EXP(ETA(2))
TVK12 = THETA(3)
K12=TVK12* EXP(ETA(3))
TVMPRC = THETA(4)*(1+FFAT*THETA(7))
MPRC=TVMPRC* EXP(ETA(4))
K20=CL/V2
S2 = V2
$DES
DADT(1) = -K12*A(1)
DADT(2) = K12*A(1)-K20*A(2)
$ERROR
PLC=0
MIC=0
IF (CMT.EQ.2) THEN
PLC=1
IPRED2 = A(2)/S2
W2 = THETA(8)*IPRED2
Y2 = IPRED2 + W2*EPS(1)
IRES2 = DV-IPRED2
IWRES2 = IRES2/W2
ENDIF
IF (CMT.EQ.3) THEN
MIC=1
IPRED3 = MPRC*A(2)/S2
W3 = THETA(9)*IPRED3
Y3 = IPRED3 + W3*EPS(1)
IRES3 = DV-IPRED3
IWRES3 = IRES3/W3
ENDIF
IPRED = PLC*IPRED2 + MIC*IPRED3
Y = PLC*Y2 + MIC*Y3
IWRES = PLC*IWRES2 + MIC*IWRES3
$THETA
(35.4); CL (L/h)
(1140); V2 (L)
(0.553); absorption rate constant (h-1)
(2.01); milk-to-plasma ratio
(-0.21); moment of sampling effect on CL
(-0. 523); CYP2C19 effect on CL
(0.276); fat composition effect on CL
(0.313); Prop error model component for plasma compartment
(0.24); Prop error model component for milk compartment
(-0.474); effect on CL
$OMEGA
0.0879; ETACL corresponding to an inter-individual variability of 29.4%
35
$SIGMA
1 FIX
S.4 NONMEM code for escitalopram popPK model in infants with final parameter estimates
$PROBLEM PK
$INPUT ID TIME AMT DV CMT EVID BW FAT MOM DDOSE DAY CYPCD
$DATA DATA_INFANTS.txt IGNORE=@
$SUBROUTINES ADVAN6 TOL=3
$MODEL
NCOMP=3
COMP=(DEPOT)
COMP=(PLCIT)
COMP=(AUC)
$PK
TVCL=THETA(1)*(1+MOM*THETA(4))*(BW/70)**0.75
CL=TVCL* EXP(ETA(1))
TVV2 = THETA(2)*(BW/70)
V2=TVV2
TVK12 = THETA(3)
K12=TVK12
K20=CL/V2
S2 = V2
$DES
DADT(1) = -K12*A(1)
DADT(2) = K12*A(1)-K20*A(2)
DADT(3) = A(2)/S2
$ERROR
AUC=A(3)
IPRED = A(2)/S2
Y = IPRED*EXP(EPS(1))
$THETA
(31.3) ; CL (L/h)
(1150) ; V2 (L)
(0.577) ; absorption rate constant (h-1)
(-0.208); moment of sampling effect on CL
$OMEGA
0.177 ; ETA CL
$SIGMA
0.0961 ; Prop.RE PLC
36
