Optimal Drug, Optimal Dose, or Both in the Pharmacological Treatment of Neonatal Opioid Withdrawal Syndrome?

Abstract

The drugs and drug dose regimens used to treat opioid withdrawal symptoms in newborns with in utero opioid exposure have been developed in a mostly empiric fashion. Clonidine has long been used as an adjunct but has been explored as part of an opioid-free approach to primary therapy. The tools of pharmacometrics can provide a path for rational exploration of drug exposure and response.

Antenatal opioid exposure results in post-natal neonatal opioid withdrawal syndrome (NOWS) with cardinal symptomatology in the central and autonomic nervous systems and the gastrointestinal tract. The natural history of NOWS is characterized by an initial increase in symptoms as placentally transferred opioids are cleared from the infant's system. The therapeutic focus in treating newborns is on the reduction in symptom severity, with a goal of ensuring normal infant growth and development and maternal–infant bonding. Rates of NOWS have risen 7-fold over the past 20 years, and in parallel, there has been progress in therapeutic approaches to improving neonatal outcomes. Supportive measures with functional-based symptom assessment control symptoms for many infants. Non-pharmacologic measures have been standardized and refined, and their systematic implementation has reduced the number of infants requiring pharmacologic treatment. This has been an advantage since while pharmacologic treatment controls symptoms, it is associated with prolonged hospital stays and can impair maternal infant bonding. Clonidine is currently used as a rescue therapy in NOWS. The parent trial of Bada¹ (NCT03396588) tested the contrarian hypothesis that clonidine can improve outcomes for those requiring pharmacologic treatment compared with the current standard of opioid replacement using oral morphine. NOWS investigations are often difficult to interpret due to suboptimal design, such as before/after quality improvement programs that implement a suite of interventions, making the impact of any single action hard to assess. Retrospective² or single-arm studies with historical controls³ are subject to differences in NOWS incidence or temporal changes in of maternal drug exposure such as the emergence of fentanyl. In contrast, the Bada study was large (for NOWS investigations), randomized, and blinded. All of these helped to mitigate the temporal bias of fentanyl use patterns as well as the difficult to predict impact of the COVID-19 pandemic. Importantly, the clinical data were magnified by the use of rich pharmacokinetic (PK) collection sampling, outlined in the work of Tang.⁴ External validity of the work is strengthened by parameter estimates congruent with both adult IV administration and the prior work of Xie in NOWS patients.⁵ Simulations provided a number of options to consider in future investigations. Critically, the work is welcomed in moving improvements of NOWS therapeutics from empiric dose optimization to a PK data-driven approach.

The primary and secondary endpoints in the trial are important clinical outcomes for clinicians. However as with many clinical trials with a clinical efficacy endpoint, the results provide describe what the investigators saw but less guidance on the why and mechanisms explaining the results. On first glance, the parent trial could be seen as a null result, with no difference in length of treatment or hospital duration (and even a trend toward worse outcomes with clonidine) and a negative impact with a higher fraction of infants requiring adjunct medication (45% vs. 10%). A closer look at the pharmacodynamic marker of Finnegan scores (measured every 3–4 hours rather than a time to event in days seen with length of treatment) reveals an early advantage with morphine. The faster symptomatic control in the morphine group conforms with quick achievement of a steady-state drug exposure in therapeutic range for morphine, compared with one for clonidine, which will have taken a few days. The Finnegan Score and morphine dosing have been integrated since both began wide adoption in the 1970s, and the current morphine approach of frequent dosing without a loading dose evolved empirically. There is an age-dependent maturation of metabolism of oral morphine,⁶ but the relatively short half-life allows for frequent dose adjustments using symptom assessments every 3–4 hours. Moreover, the increased morphine clearance by 2 weeks of life is of less importance with natural history decline in withdrawal symptoms over that time period. The clonidine dose approach used in the study was mirrored on morphine both out of habit and to provide an ability to blind the study appropriately, rather than modeled based on expected exposures. Without the PK samples collected in the trial, the pharmacodynamic observation of delayed efficacy could have been surmised to be based on slower accumulation but not confirmed. Tang's demonstration of slow clonidine accumulation leading to delayed efficacy mirrors that proposed by Sheng⁷ in critical care sedation. The finer grain focus provided by the Finnegan score compared with the “hard clinical” endpoints such as length of treatment or length of stay allow for exploration of potential drug efficacy for novel therapies in NOWS. Without PK, it is possible this would have been considered an equivocal clinical trial. With the work of Tang, we are provided a path forward for dose optimization of a potentially opioid-free primary NOWS treatment approach.

One could argue that an exposure–response relationship should have been explored before engaging in a large, blinded efficacy trial. The model of Xie et al. would have allowed dose exploration through simulation that would have been able to achieve therapeutic concentrations quicker, which may have led to downstream improvement in the clinical endpoints of treatment time and length of stay. Indeed, Xie et al. identified the maturation of clearance in their paper and suggested increasing the dose by 50% at Week 2 of life. In defense of the investigators, this specific change would have not made a difference in the early days of treatment. There is also the challenge for the clinical trialist of tension between the conservatism of the clinicians who care for the infants on the ward and acceptance of novel but untested dose regimens. This is not unwarranted, given the potential cardiovascular effects of clonidine and the large interpatient variability in maturation of the clearance of clonidine. Lastly, assuming investigators had incorporated a loading dose, the operational complexity of maintaining a blind would complicate clinical trial pharmacy operations (in what was already a challenging study to run). In the end, the trial that was done provided a large number of samples paired with hard clinical outcomes. This provides greater confidence in the simulations for future guidance.

Where to go next? The authors state that the modeled doses be “assessed in a separate prospectively designed clinical study.” This would be approached in an ideal world with many resources and clinical trial capacity. Given the limits on funding, the size of public health impact, and long latency until dissemination of clinical trial results, an alternative approach may be to roll out modeled dose regimens carefully into clinical care. This approach has been employed in the pharmacometric dose optimization based on observational methadone PK data. This drove a change in the standard of care regimen leading to a 20% reduction in length of treatment.⁸ Similar approaches have been taken with buprenorphine regimens that have been explored with modeling,³ but not rigorously tested in a randomized clinical trial before being implemented in the ward. As this could involve exposures higher than previously tested, this could be done in an incremental fashion. Precisely since this would not be a clinical trial, there is not a need for a fixed regimen in a large number of patients. Instead, an iterative approach could gradually refine the doses. This is not revolutionary. Instead, this is an improvement in the prior approach of pure empiricism with clinical outcomes as the only guide. Such an approach could incorporate observations and PK sampling on top of a standard of care change in dose in the systematic generation of generalizable knowledge in a human subject's research framework. The dose alterations could be part of a high-quality local quality improvement project that fits frameworks of Plan-Do-Study-Act cycle. Such approaches are already commonly used to alter NOWS dose regimens,⁹ just typically without the power of pharmacometric guidance.

Some cautions should be taken. The parent study had a consent rate of 30%, a value comparable to many studies in this population. It is the observation of trialists that mothers who decline study participation are less often on a stable outpatient methadone or buprenorphine regimen are more likely to have polysubstance abuse and may have higher rates of undertreated comorbid psychiatric illness or heritable personality traits such as impulsivity. All of these are covariates associated with more severe NOWS symptoms. It is unknown whether disease severity would impact the dose response relationship, but low consent rate impact on NOWS generalizability is a common issue in the field. Long-term neurodevelopmental impacts of various medications are difficult to assess, and balance against endpoints such as time away from family in the hospital. For example, when used as an adjunct therapy phenobarbital is associated a shorter hospitalization than clonidine, but at the expense of longer overall duration of therapy at home.¹⁰ Phenobarbital has been widely used in neonatal seizures but does have concerns about impact on neurodevelopment. The underlying premise for the two arms of the Bada trial is that even a short post-natal treatment with an opioid has clinically important long-term neurodevelopmental impact. Though debated,¹¹ this is by no means established. If there is no safety advantage of clonidine over an opioid, its use in NOWS becomes much less compelling.

The question has often been about what is the “ideal drug,” when instead the question should be what is the “ideal dose of the ideal drug?” Tang provides a convincing example of the value of integrating pharmacokinetics into clinical trials and develop exposure–response relationships that can serve as a basis for optimizing the individual dose. PK collection in a neonatal population is not easy but adds a value magnifier in a way that may not be immediately apparent. Given the power of modeling and simulation, clinicians who care for infants with NOWS should consider ways in which pharmacometric outputs can be used to optimize treatment regimens even in the absence of Level I clinical trial evidence. This will be a clear improvement over the current anchoring to empiric and instinct-derived regimens.

WK is supported by NICHD 1UG1HD107628–01.

The author declared no competing interests for this work.