Physiologic indirect response modeling to describe buprenorphine pharmacodynamics in newborns treated for neonatal opioid withdrawal syndrome

Abstract

Background:

Buprenorphine has been shown to be effective in treating infants with neonatal opioid withdrawal syndrome (NOWS). However, an evidence-based buprenorphine dosing strategy has not been established in the treatment of NOWS due to a lack of exposure-response data. The aim of this study was to develop an integrated pharmacokinetic and pharmacodynamic (PK/PD) model to predict buprenorphine treatment outcomes in newborns with NOWS.

Methods:

Clinical data were obtained from 19 newborns, median (range) gestational age 37 (34-41) weeks, enrolled in a pilot PK study of buprenorphine. Sparse blood sampling, comprising 3 specimens obtained around the second dose of buprenorphine, was performed using heel sticks with dried blood spot technology. Standardized NOWS severity scores (Finnegan scores) were collected every 3-4 hours based on symptoms by bedside nursing staff. Mean Finnegan scores were used as a PD marker in the exposure-response modeling. The blood concentration-Finnegan score relationship was described using a physiologic indirect response model with inclusion of natural disease remission.

Results:

A total of 52 buprenorphine blood concentrations and 780 mean Finnegan scores were available for the PK/PD modeling and exposure-response analysis. A one-compartment model with first-order absorption adequately described the PK data. The buprenorphine IC₅₀ for the inhibition of disease progression was 0.77 ng/mL (95%CI: 0.32-1.2). The inclusion of natural disease remission described as a function of postnatal age significantly improved the model fit.

Conclusion:

A buprenorphine PK/PD model was successfully developed. The model could facilitate model-informed optimization of the buprenorphine dosing regimen in the treatment of NOWS.

1. Introduction

The opioid epidemic in the US has resulted in a more than five-fold increase in neonatal opioid withdrawal syndrome (NOWS) incidence during 2000-2014, reaching 14.4 per 1000 births in 2014 [1]. Newborns experience a range in severity of symptoms of withdrawal after birth, which is referred to as NOWS. Newborns with NOWS often experience a more complicated and prolonged hospitalization [2]. Withdrawal symptoms commonly occur within the first 24-72 hours of life. The clinical manifestations include low birth weight, poor weight gain, sleep disturbances, respiratory distress, feeding difficulties, irritability, fever, and tremors [3].

Pharmacologic treatment is required when nonpharmacologic measures are insufficient in improving symptoms in newborns with NOWS [3]. Oral morphine and methadone are first-line opioids used in infants who require pharmacological intervention [3]. In a randomized controlled trial (RCT), buprenorphine was associated with a shorter length of treatment than morphine [4], and methadone in a retrospective cohort [5]. However, these findings have yet to be replicated in a large multi-centered RCT with evaluation of long-term outcomes. Buprenorphine is a semi-synthetic opioid that exhibits pharmacological activity as a partial mu-opioid receptor (MOR) agonist and kappa-opioid receptor (KOR) antagonist [6]. Sublingual buprenorphine is approved by the US Food and Drug Administration (FDA) for the treatment of opioid dependence in adults. For the pharmacologic treatment of NOWS, it has been demonstrated that the use of a standardized pharmacologic protocol is important to improve treatment outcomes [3]. Indeed, the use of a refined dosing protocol for methadone developed using pharmacokinetic (PK) model-based simulation was associated with better clinical outcomes compared to the conventional protocol [7, 8]. However, such a standardized evidence-based dosing protocol has not been established for the buprenorphine in NOWS.

Buprenorphine pharmacokinetics (PK) and pharmacodynamics (PD) have been studied intensively in the treatment of opioid dependence in adults [6]. However, buprenorphine clinical pharmacology data remain limited in newborns. The first population PK modeling study for sublingual buprenorphine in neonates with NOWS was published by Ng et al. [9], who describe the developmental changes in clearance and volume of distribution. More recently, Moore et al. [10] reported the first population PK and PD modeling study of buprenorphine in the treatment of NOWS using data obtained from 28 neonates in the Blinded Buprenorphine OR Neonatal morphine solution (BBORN) trial [4]. In this study, authors described the PK/PD relationship with a specific focus on time to NOWS symptoms stabilization as a PD endpoint [10]. Their analysis showed that neonates with lower buprenorphine clearance had a shorter time to stabilization of NOWS symptoms. This is also the first model describing the PK of the primary metabolite norbuprenorphine in neonates. In the present PK/PD study, we developed a buprenorphine PK/PD model using a standardized symptoms score (Finnegan score) to quantitatively describe and predict the clinical outcomes of buprenorphine in the treatment of NOWS.

2. Methods

2.1. Patients

The data were obtained from 19 neonates as part of a prospective pilot pharmacologic study of buprenorphine in the treatment of NOWS [11]. In the pilot study, neonates (>34 weeks of gestational age) with chronic in utero exposure to opioids as determined by maternal history or positive drug test were enrolled. Fourteen of the infants were delivered to mothers on buprenorphine prescribed at a medication-assisted program, two infants were exposed to methadone, one to heroin and two to other opioids. Seven infants were exposed to tobacco in utero. Patients with congenital malformations, major chromosomal anomalies, acute illness, or other confounding medical conditions necessitating therapy with opioids were excluded from the study. The need for pharmacologic treatment of NOWS was determined by elevated Finnegan scores [12] as dictated by the local NOWS treatment protocol. In our institution all babies being observed for NOWS room-in with their mothers initially unless a higher level of care is needed, but the institution does not provide options for mothers to room-in while their infant is receiving pharmacotherapy for NOWS. The study was approved by the Institutional Review Boards of the participating hospitals. Informed consent was obtained from mothers once the care providers decided to treat their infants with buprenorphine.

2.2. Buprenorphine administration

Sublingual buprenorphine was administered and weaned according to the institutional protocol in the Neonatal Intensive Care Unit (NICU) at the University of Cincinnati Medical Center (UCMC) (Supplemental Text). Buprenorphine treatment was initiated when infants had a Finnegan score of ≥8 on three consecutive occasions or greater than 12 on two consecutive scorings within a 24-hour period despite optimized non-pharmacologic interventions (subdued ambient lighting, reduced environmental noise level, minimal stimulation, high calorie feeds). Buprenorphine solution was administered under the tongue. The buprenorphine starting dose was 4.4 μg/kg every 8 hours by trained nursing staff. After every two doses, the dose was increased by 0.8 μg/kg until the average daily Finnegan score was ≤8. The maximum dose per protocol was 13 μg/kg/dose. Once the Finnegan score was 8 or below, patients continued on the same dose for 24 hours (three consecutive doses) before a tapering regimen was initiated. Buprenorphine was weaned each day by 0.8 μg/kg until a dose level of 4.4 μg/kg was reached. This was followed by 2.6 μg/kg every 8 hours for three doses, 1.7 μg/kg every 8 hours for three doses, 1.7 μg/kg every 12 hours for two doses and finally 1.7 μg/kg for one dose. Weaning was halted when the 24-hour average Finnegan score was 8-12. Dose escalation was resumed when the average daily Finnegan score was more than 12. Weaning was also adjusted based on the stability of vital signs (respiratory and heart rates) and weight gain.

2.3. PK sampling and buprenorphine assay

Blood samples were collected using heel sticks with Dried Blood Spot (DBS) sampling to minimize phlebotomy losses. Three timed blood samples were collected from each infant: prior to the second buprenorphine dose, 60-120 minutes post-second dose, and just prior to the third. An additional fourth blood sample was obtained just prior to a buprenorphine dose when a patient would not respond to treatment as anticipated. All DBS samples were air-dried at room temperature for 24 hours and then stored at −70°C until analysis. Buprenorphine concentrations in capillary dried blood spot samples were quantified using an established high-performance liquid chromatography- tandem mass spectrometry (LC-MS/MS) platform for opioids [13] by iC42 Clinical Research and Development (University of Colorado, Aurora, CO, USA). The only difference was that the ion transitions of buprenorphine ([M+H]+, m/z= 468.3 -> 414.4 (quantifier) and m/z= 468.3 -> 396.0 (qualifier) and the internal standard buprenorphine-D4 (m/z= 472.3 -> 400.0) were monitored. Both reference materials were purchased from Cerilliant (Round Rock, TX, USA). The HPLC system was from Agilent (series 1200, Santa Clara, CA, USA) and the MS/MS system was a Sciex API5000 (Concord, ON, Canada). The instrument was controlled and data analyzed using Analyst 1.6.1 (Sciex). Buprenorphine quantification was based on the analyte/internal standard ratios and calibration curves were generated using 1/X weighting and a quadratic fit. The lower limit of quantification was 0.2 ng/mL and the range of reliable response was 0.2 – 400 ng/mL. During study sample analysis, imprecision ranged from 1.2% (coefficient of variance, 40 ng/mL quality control sample) to 13.5% (0.8 ng/mL quality control samples). Accuracy ranged from 97.8- 110%. There was no significant matrix interferences, carry-over or matrix effects (ion suppression/ ion enhancement). Extracted samples were stable in the autosampler at +4°C for at least 48 hours.

2.4. Neonatal abstinence scoring

Finnegan scores were assigned using standardized, validated assessment tool [12] every 3-4 hours by the clinical care team based on the clinical manifestations such as respiratory rate, hyperthermia, poor feeding, and sleep disorders, as part of routine clinical care of infants with NOWS. All nurses who evaluated Finnegan scores were completed scoring training and achieved 90% reliability to minimize inter-rater variability. A mean Finnegan score calculated summarizing the scores for each12 hour period was used for the PK/PD modeling.

2.5. Population PK/PD modeling

Population PK/PD analysis was performed by nonlinear mixed effect modeling using NONMEM (version 7.2, ICON, Ellicott City, MD) with Perl speaks NONMEM (PsN) version 3.6.2 [14] and Pirana version 2.7.1 (Certara, Princeton, NJ). A schematic structure of the integrated PK/PD model for sublingual buprenorphine in NOWS is shown in Figure 1. A one-compartment model with first-order absorption was used to describe the buprenorphine concentration-time data. A two-compartment model was also tested but did not provide reasonable estimates for inter-compartmental clearance and volume of distribution in the peripheral compartment due to the sparse PK sampling. Inter-individual variability (IIV) was assessed using an exponential variability model as follows:

$$P_i=P_{pop}\times \text{exp}({\eta }_{\mathrm{i}})$$

where P_i is the estimated parameter value for individual i, P_pop is the typical population value of the PK parameters such as clearance and volume of distribution, η_i is an inter-individual random effect for individual i with the mean of zero and variance of ω². A proportional error model was used to describe the residual error. A combined proportional and additive residual error model was also tested but did not improve the model fit. All PK models were parameterized in terms of values of clearance (CL), volume of distribution (V), and absorption rate constant (K_a). Due to limited PK data to describe the absorption, the previously identified K_a estimate for sublingual buprenorphine in neonates was included as a fixed parameter in the modeling process [9]. Allometrically-scaled body weight was used to account for the effect of differences in body size on PK parameters as described previously [15]. Model selection was based on goodness-of-fit diagnostic plots, comparisons of the minimum objective function value (OFV), and evaluation of the estimates of population fixed and random effect parameters. A drop in OFV of more than 3.84 (P<0.05) was considered statistically significant. The first-order conditional estimation with interaction method (FOCE-I) was applied for all runs.

Figure 1.

Schematic structure of the buprenorphine PK/PD model.

A relationship between buprenorphine blood concentration and Finnegan score was described using the physiologic indirect response model as described previously by Jusko et al. [16] with the inclusion of natural disease remission as follows:

$$\frac{dR}{dt}=K_{in}\times \left(1-\frac{C_b}{I{C}_{50}+C_b}\right)\times e^{-K_{remi}\times PNA}-K_{out}\times R$$

where R is response representing mean Finnegan score. Cb is buprenorphine blood concentration. IC₅₀ is the buprenorphine blood concentration at 50% of maximum effect. K_in and K_out represents the rate constant for input and output of response (Finnegan score), respectively. K_remi represents the rate constant to describe the spontaneous natural disease remission. The spontaneous remission of NOWS symptoms was described as a function of postnatal age (PNA) using an exponential power model. The final model was evaluated using nonparametric bootstrap analysis (n=1000) [21] and the prediction-corrected visual predictive check (pcVPC, n=1000) [22].

3. Results

3.1. Population PK/PD modeling

A population PK model was first developed before performing the PK/PD modeling. A total of 52 buprenorphine concentrations were used for the population PK analysis. A one-compartment model with first-order absorption adequately described the PK data. For the PK/PD modeling, a total of 780 mean Finnegan scores were available. The relationship between buprenorphine blood concentration and the Finnegan score was described using the physiologic indirect response model. Inclusion of natural disease remission described as a function of postnatal age significantly improved the model fit (ΔOFV=193.8, p<0.0001). The PK/PD parameter estimates for the final model are summarized in Table 2. All PK/PD parameters, except the absorption rate constant (K_a), were estimated simultaneously. The buprenorphine IC₅₀ for the ameliorating disease progression was 0.77 ng/mL (95% CI: 0.32-1.2 ng/mL). Goodness-of-fit plots indicated that the developed final model adequately predicts the buprenorphine concentrations and Finnegan scores, respectively (Figures 2 and 3). The non-parametric bootstrap analysis confirmed the homogeneity of variances for the developed final model (Table 2). The pcVPC demonstrated that the simulated concentrations were in reasonable agreement with the observed values (Supplementary Material Figure 1).

Table 2.

Parameter estimates for the buprenorphine PK/PD model

Parameters	Final model			Bootstrap (n=1000)
	Estimates	RSE (%)	Shrinkage	Median	95% CI
					Lower	Upper
CL/F (L/h/70 kg)	32.6	12.0	11.3	32.5	25.5	38.4
V/F (L/70 kg)	464	13.9	41.6	467	364	591
Ka (h−1)	0.416 (Fixed)*	—	—	0.416	0.416	0.416
Kin (h−1)	0.690	20.3	4.2	0.672	0.479	0.899
Kout (h−1)	0.0496	20.8	13.1	0.0476	0.0346	0.0624
Kremi (day−1)	0.0346	18.8	16.7	0.0347	0.0260	0.0479
IC50 (ng/mL)	0.766	29.8	12.4	0.778	0.467	1.17
Inter individual variability (CV%)
IIV for CL/F	38.2	31.4	—	37.9	18.0	59.7
IIV for V/F	39.1	16.3	—	37.6	15.7	49.3
IIV for Kin	62.0	21.1	—	63.2	28.6	79.2
IIV for Kout	51.9	25.5	—	52.5	23.6	68.5
IIV for Kremi	62.6	18.9	—	58.7	34.0	79.3
IIV for IC50	87.1	31.9	—	83.2	23.3	127.6
Residual errors (CV%)
PK prop.	27.7	10.8	16.5	27.3	22.4	32.8
PD prop.	29.1	4.6	3.3	29.1	27.0	31.4

CL/F, oral clearance; V/F, oral volume of distribution for the central compartment; Ka, absorption rate constant; K_in, rate constant for input of response, K_out, rate constant for output of response; K_remi, rate constant describing the spontaneous disease remission; IC50, concentration at the 50% of maximum effect; IIV, inter-individual variability; CV, coefficient of variation; CI, confidential interval; RSE, relative standard error; PK prop., proportional error for buprenorphine concentration, PD prop., proportional error for Finnegan score.

^*Ka was adapted from the previous study published by Ng et al. [15]

Figure 2.

Goodness-of-fit plots for the final buprenorphine PK model. Observed versus population predicted (A) and individual predicted (B) buprenorphine concentrations (line of identity shown for clarity). The conditional weighted residual (CWRES) versus population predicted concentrations (C) and time after the last dose (D).

Figure 3.

Goodness-of-fit plots for the final buprenorphine PD model. Observed versus population predicted (A) and individual predicted (B) Finnegan scores (line of identity shown for clarity). The conditional weighted residual (CWRES) versus population predicted Finnegan score (C) and time after the birth (D).

3.2. Individual model-based PK/PD predictions

Figure 5 shows the individual model-based predicted PK/PD profiles with observed data in two representative patients who were also included in the model development. Panel A represents the blood concentration- and Finnegan score-time data for a male patient who started at a buprenorphine dose of 4.4 μg/kg 4.5 hours after birth. The buprenorphine dose was step-wisely increased per the protocol by 0.8 μg/kg increments at the same dosing frequency (every 8 hours) based on symptomatology. The detailed dosing data are shown in the Figure legend. The highest dose level reached in this patient was 7.6 μg/kg. On Day 8, the tapering regimen was initiated after an improvement in average daily Finnegan score <8, and buprenorphine dose was gradually decreased over a week (until Day 15). In response to the decreased buprenorphine dose, the average daily Finnegan score rebounded increasing to over 8 on Day 15; however, the score finally improved and normalized again without pharmacological intervention. Panel B represents a female neonate who initiated buprenorphine at 27 hours after birth. The buprenorphine starting dose was 4.4 μg/kg every 8 hours which was increased by 0.8 μg/kg up to 7.6 μg/kg. The tapering regimen was started on Day 6 treatment. However, the weaning was suspended on Day 13, and the dose was increased again up to 4.4 μg/kg since the patient showed worsening Finnegan scores (more than 12). After three consecutive doses of 4.4 μg/kg, tapering was resumed.

Figure 5.

Individual model-based predicted buprenorphine PK/PD profiles in two representative patients. The upper panel shows the buprenorphine blood concentration-time data, and the lower panel shows the NOWS disease (Finnegan) score-time data. Solid lines indicate the model-based individual prediction. Closed circles represent the observed data. The horizonal dashed line indicates the reference Finnegan score (less than 8). Panel A represents a male neonate who started at a buprenorphine dose of 4.4 μg/kg. The dose was increased per the protocol by 0.8 μg/kg increments at the same dosing frequency (every 8 hours) based on symptomatology. The highest dose level reached in this patient was 7.6 μg/kg. On Day 8, the tapering regimen was initiated, and buprenorphine dose was decreased until it reached 4.4 μg/kg. On Day 12, the dose was further decreased to 2.6 μg/kg every 8 hours and eventually decreased to 1.7 μg/kg every 8 hours for 3 doses on Day 13, 1.7 μg/kg every 12 hours for 2 doses on Day 14, and 1.7 μg/kg for 1 dose on Day 15. Panel B represents a female neonate who initiated buprenorphine at 4.4 μg/kg every 8 hours which was increased by 0.8 μg/kg up to 7.6 μg/kg. The tapering regimen was started on Day 6 treatment, when the dose was decreased by 0.8 μg/kg to 4.4 μg/kg every 8 hours. The dose was further decreased to 2.6 μg/kg every 8 hours and then decreased to 1.7 μg/kg every 8 hours. After continuation of this dose for 9 doses, the dose was increased again to 2.6 μg/kg every 8 hours and then further increased to 4.4 μg/kg every 8 hours since the patient showed worsening Finnegan scores (more than 12). After three consecutive doses of 4.4 μg/kg, tapering was restarted at a dose of 2.6 μg/kg every 8 hours for three doses, 1.7 μg/kg every 8 hours for three doses, 1.7 μg/kg every 12 hours for two doses and lastly 1.7 μg/kg for one dose.

4. Discussion

In this study, we developed a population PK/PD model of buprenorphine in newborns with NOWS. As illustrated by the representative patient cases (Figure 5), the observed buprenorphine concentration data and Finnegan scores were well captured by the model-based predictions using the PK/PD model. The current PK/PD analysis indicates that buprenorphine exposure is associated with the therapeutic effect in the treatment of NOWS, which is in line with the previous findings observed in a larger blinded trial (265 buprenorphine concentrations obtained from 28 neonates) recently published by Moore et al. [10]. The IC₅₀ estimate in this analysis (0.766 ng/mL, 95%CI: 0.32-1.2 ng/mL) was comparable with previously reported buprenorphine IC₅₀ values in neonates with NOWS (0.509 ng/mL, 95%CI: −0.27-1.56 ng/mL) and in adults with opioid dependence (0.67 ng/mL, 95%CI: 0.3-1.04 ng/mL), although there are differences in the study design and data analysis, including the model structure and the PK sample matrix across these studies (e.g., whole blood samples collected by DBS were used for buprenorphine measurements in this study vs. serum/plasma samples were used in other studies) [10, 17]. Although the current study did not include control patients (e.g., untreated patients), a NOWS disease progression model successfully characterized the natural disease remission over time.

Providing the appropriate buprenorphine dose in the treatment of newborns with NOWS is particularly challenging due to the substantial between-patient variability in drug response caused by a variety of maternal-infant factors. In addition, rapid developmental changes in neonates have an impact on PK/PD over the course of infancy [18]. In the current clinical practice, the need for buprenorphine treatment and its dose adjustment are determined based on symptoms with disease scoring systems (e.g., Finnegan score) [19]. Such a dosing strategy is considered reactive and typically has a large “trial and error” component. Developing a standardized protocol for a proactive personalized dosing strategy has the potential to avoid inadequate buprenorphine treatment that can cause prolonged hospitalization, longer duration of opioid-use and increased cumulative opioid exposure, with concerns of potential long-term adverse effects on brain development. Indeed, it has been demonstrated that the use of a standard dosing protocol improves treatment outcomes in infants with NOWS treated with morphine or methadone [20-22]. The developed buprenorphine PK/PD model could be useful in optimizing the buprenorphine dosing regimen in the treatment of NOWS.

To date, the pharmacometrics (modeling and simulation) approach has been successfully employed to optimize dosing protocols in various patient populations, including neonates and infants [23-25]. For instance, in a previous study, we developed an optimized methadone dosing protocol for the treatment of NOWS using population PK model-based simulation [8]. This PK-model-informed methadone dosing protocol has now been implemented across the state of Ohio through the Ohio Perinatal Quality Collaborative (OPQC) [26]. Prospective evaluation in a pre-post cohort study, found that the refined dosing protocol provides better outcomes including reduced duration of opioid weaning and shortened length of hospital stay [7]. A similar approach holds promise for developing improved dosing protocols for buprenorphine which may accelerate rapid achievement of drug exposure targets associated with alleviating withdrawal symptoms in infants with NOWS.

Buprenorphine is a unique drug with complex pharmacology. Various enzymes are involved in the buprenorphine metabolism, such as cytochrome P450 (CYP) 3A4, CYP2C8, CYP3A5 and uridine 5'-diphospho-glucuronosyltransferase (UGT) subfamilies including UGT2B7 and UGT1A1 [27-29]. It has been suggested that genetic variants in those metabolic enzymes may be predictors of buprenorphine PK and response [29-31]. For instance, an in vitro study demonstrated that the UGT1A1*28 allele is associated with decreased buprenorphine glucuronidation in pooled human liver microsomes [29]. The same study also indicated that the combination of UGT1A1*1 allele and UGT2B7 promoter (G-842A) mutation is associated with a higher glucuronide metabolic rate compared to other genotype combinations [29]. Besides metabolic enzymes, the effects of variants in genes related to PD, such as the μ-opioid receptor (OPRM1) and catechol-o-methyltransferase (COMT), have also been investigated in patients with NOWS. In a prospective multicenter study, the authors identified genetic polymorphisms in OPRM1 and COMT genes that were significantly associated with better treatment outcomes in NOWS patients treated with either morphine or methadone [32]. In the current study, we could not examine the effect of genetic variants as patient’s genetic data are not available. Further study is warranted to investigate the implications of genetic factors on buprenorphine PK/PD and incorporate this information into the model to further improve its predictive performance.

A limitation of this study is the relatively small number of patients (n=19) and the sparse PK data (3-4 samples per patient) to characterize buprenorphine concentration time profiles after sublingual administration, which limits the robustness of the model. Previous publications have indicated that the buprenorphine plasma concentration profile is best described by using a two-compartment model [9, 10]. However, a two-compartment model was not applicable in our analysis due to the sparse sampling strategy. The model diagnostic plots indicate that there still is substantial variability in buprenorphine concentrations and PD endpoints, which impedes the predictive performance of the model. In a preliminary external validation analysis, the observed concentrations in the current study were not well predicted by the previously published model by Moore et al. [10] when the predictions were solely on covariates without Bayesian estimation (Supplemental Figure 2). The predictions were improved with Bayesian estimation. Given such large unexplained inter-patient variability, further studies with a larger number of patients is warranted to validate and update the developed PK/PD model and identify more predictive covariates. The PD data used in this study were observational and not collected in a blinded fashion. Increasingly, NICUs are adopting a strategy of “as-needed” (PRN) dosing of opioids and replacing the Modified Finnegan Neonatal Abstinence Scoring System with the “Eat, Sleep and Console (ESC)” strategy, which focuses on critical functions of a newborn such as eating and sleeping [33]. Recent studies suggest that the ESC approach is promising to reduce the use of postnatal opioids in the treatment of NOWS [34, 35]. Further study is warranted to evaluate such new knowledge in association with the PK and PD of opioids including buprenorphine. In addition, emerging studies demonstrated that various factors such as rooming-in and breastfeeding could have an impact on the treatment outcome of NOWS [36]. These were not evaluated in this study; however, those factors which potentially influence withdrawal symptoms could be good candidates to be assessed as predictive covariates of response (PD) in future studies. Quantitative “bottom-up” approaches such as physiologically-based pharmacokinetics (PBPK) and mechanism-based quantitative systems pharmacology (QSP) modeling could also be attractive approaches to complement the current practice and data gaps given the complex maternal-infant factors associated with disease development and drug response, and limitations in blood sampling and data collection from such a vulnerable neonatal patient population.

5. Conclusion

A buprenorphine PK/PD model was developed in newborns with NOWS that allows to predict the NOWS disease (Finnegan) score based on buprenorphine concentrations. The developed PK/PD model will facilitate model-informed optimization of the buprenorphine dosing regimen in the treatment of NOWS.

Supplementary Material

40262_2020_939_MOESM1_ESM

Supplemental Figure 1. Prediction-corrected visual predictive check (pcVPC) for the final buprenorphine model for PK (A) and PD (B). Closed circles represent observed buprenorphine concentrations (A) or mean Finnegan scores (B). Lines represent the median, 10^th and 90^th percentiles of the simulated data (n=1,000).

40262_2020_939_MOESM2_ESM

Supplemental Figure 2. Correlations between observed concentrations in the current study and (A) population model-based predicted concentrations and (B) Bayesian estimated individual predictions estimated by the previously published model (Moore et al.) [10]. Population model-based and individual predicted concentrations were generated with the published PK parameters using NONMEM.

Figure 4.

The effect of buprenorphine concentration (A) and postnatal age (B) on input rate constant (K_in) of response (Finnegan score). The solid line indicates the mean estimate. The shaded area represents 95% confidence interval. PNA, postnatal age.

Table 1.

Demographic characteristics of patients (n=19)

Characteristics	Median (Range)
Birth weight (kg)	3.3 (1.9-4.6)
Gestational age (weeks)	37.0 (34.0-41.0)
Age at the start of treatment (days)	2.1 (0.2-4.2)
	Count
Sex (Female/Male)	5/14
Race (White)	19
Median length (range) of stay (days)	21 (9-38)
Median length (range) of treatment (days)	17.5 (8-33)

Key points.

• This study provides a buprenorphine population pharmacokinetic-pharmacodynamic (PK/PD) model in newborns with neonatal opioid withdrawal syndrome (NOWS).

• An indirect response model was established to describe NOWS disease improvement and the buprenorphine concentration-response relationship using a standardized NOWS disease score (Finnegan score) as a marker of withdrawal syndrome severity.

• The developed model incorporates natural disease remission described as a function of postnatal age and could inform dosing protocols for pharmacotherapy for NOWS.