Buprenorphine Compared with Methadone in Pregnancy: A Systematic Review and Meta-Analysis

Abstract Introduction Illicit opioid use in pregnancy is associated with adverse maternal, neonatal, and childhood outcomes. Opioid substitution is recommended, but whether methadone or buprenorphine is the optimal agent remains unclear. Methods We searched EMBASE, PubMed, Web of Science, Scopus, Open Gray, CINAHL and the Cochrane Central Registry of Controlled Trials (CENTRAL) from inception to April 2020 for randomized controlled trials (RCTs) and cohort studies comparing methadone and buprenorphine treatment for opioid-using mothers. Included studies assessed maternal and or neonatal outcomes. We used random-effects meta-analyses to estimate summary measures for outcomes and report these separately for RCTs and cohort studies. Results Of 408 abstracts screened, 20 papers were included (4 RCTs, 16 cohort, 223 and 7028 participants respectively). All RCTs (4/4) had a high risk of bias and median (IQR) Newcastle Ottawa Scale for cohort studies was 7.5 (6–9). In both RCTs and cohort studies, buprenorphine was associated with; greater offspring birth weight (weighted mean difference [WMD] 343 g (95% CI: 40–645 g) in RCT and 184 g (95% CI: 121–247 g) in cohort studies); body length at birth (WMD 2.28 cm (95% CI: 1.06–3.49 cm) in RCTs and 0.65 cm (95% CI: 0.31–0.98 cm) in cohort studies); and reduced risk of prematurity (risk ratio [RR] 0.41 (95% CI: 0.18–0.93) in RCTs and 0.63 [95% CI: 0.53–0.75] in cohort studies) when compared to methadone. All other clinical outcomes were comparable. Conclusions Compared to methadone, buprenorphine was consistently associated with improved birthweight and gestational age, however given potential biases, results should be interpreted with caution.


Introduction
Opioid use is common worldwide and is a growing public health challenge. In the United States of America, estimates of opioid use for non-medical reasons in 2019 was 10.3 million people (3.7% of the adult population), with approximately 745,000 people (0.3% of the adult population) consuming heroin every year (Substance Abuse & Mental Health Services Administration, 2020). The widespread adverse effects of illicit opioid use on maternal and child health are widely recognized (Stover & Davis, 2015). A multi-faceted public health response to opioid use in pregnancy is required to archive the World Health Organization's sustainability development goals of improving maternal and offspring health (International Expert Group on Drug Policy Metrics, 2018).
Pregnancy is recognized as an opportunity to change lifestyle behaviors (Cooper et al., 2017), and whilst abstinence from opioids during pregnancy is ideal, withdrawal from opioids during pregnancy is not recommended (National Institute for Health & Clinical Excellence, 2006;World Health Organisation, 2014). Opioid pharmacotherapy programs were established in the 1960s to integrate controlled opioid therapy with obstetric care, and social health interventions (Dole & Nyswander, 1965) and have led to reductions in overdoses (Schwartz et al., 2013), reduced recidivism, and reduced blood born virus transmission (American Society of Addicition Medicine, 2015). Opioid use in pregnancy carries a risk of neonatal abstinence syndrome (NAS), a collection of gastrointestinal, neurological, and behavioral symptoms following the abrupt cessation of opioids after delivery (American Society of Addicition Medicine, 2015). The balance between the risk of NAS and uncontrolled illicit opioid use favors opioid agonist therapies in pregnancy (National Institute for Health & Clinical Excellence, 2006;US Department for Health & Human Services, 2018;World Health Organisation, 2014).
Methadone is commonly used as an opioid agonist medication. Methadone therapy aims to provide stability of opioid levels, prevent withdrawal cycles, and improve engagement with obstetric care and neonatal outcomes, but is limited by stringent observation protocols and risk of overdose (Finnegan et al., 1977;National Institute for Health & Clinical Excellence, 2006). Buprenorphine is a more recently developed opioid agonist therapy and is an alternative to methadone. Buprenorphine is a partial opioid receptor agonist which has a ceiling effect on respiratory depression (limiting harm following overdose), a more flexible dosing design, and may have a more favorable neonatal opioid withdrawal profile when compared to methadone (White & Lopatko, 2007).
Previous meta-analyses have investigated differences between opioid agonist therapies, but have either included only RCTs (Minozzi et al., 2020), or RCTs plus cohort studies from over 5 years ago (Brogly et al., 2014;Zedler et al., 2016). Since 2015, five large cohort studies (~3000 patients) have been published (Bier et al., 2015;Brogly et al., 2018;Meyer et al., 2015;Nechanská et al., 2018;Tolia et al., 2018). Inclusion of these cohort studies will enable further triangulation of evidence given the limited evidence available from RCTs and previous smaller observational cohorts. The objective of this review was to systematically review all the published evidence to determine the optimal opioid substitution therapy in pregnancy.

Methods
The PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) checklist was followed for reporting this systematic review and meta-analysis (Supplementary Table 1) (Moher et al., 2009). The protocol was registered with PROSPERO (CRD42020173882). We defined the research question in accordance with the PICO format (population, intervention, comparator, and outcomes) (Supplementary Table 2). The population of interest was patients taking opioid agonists whilst pregnant and their offspring. The intervention was buprenorphine drug therapy (with or without naloxone), and the comparator was methadone drug therapy. Maternal outcomes were side-effects associated with treatment, maintenance on treatment, illicit drug use, death, and mode of delivery. Offspring outcomes were stillbirth, birthweight, growth (total body length at birth and head circumference at birth, small for gestational age), prematurity, opioid withdrawal treatment, hospital admission duration, death, congenital anomalies, and childhood development. Full details of outcomes are shown in Supplementary Table 3. We conducted a systematic search of EMBASE, PubMed, Web of Science, Scopus, Open Gray, CINAHL and the Cochrane central Registry of Controlled Trials (CENTRAL) from inception for April 2020 using a search strategy led by a senior university librarian, as shown in Supplementary  Table 4. Eligible studies were full-text RCTs and observational cohort studies comparing methadone and buprenorphine and reporting maternal and or neonatal outcomes. We included cohort studies in accordance with Cochrane guidance (Reeves et al., 2022), to provide evidence of the effects (benefit or harm) of interventions for which only a small number of randomized trials are available (or are likely to be available). We excluded case-reports, case-series, case-control studies, and editorials. We limited our results to human studies. Non-English studies were translated using Google Translate.
Screening of titles was conducted using the Covidence software platform (v2619). One reviewer (MK) conducted the search and removal of duplications. Two reviewers (MK and YC) independently reviewed studies for eligibility by screening titles, abstracts, and subsequently full texts. Any disagreement was settled by discussion with a third reviewer (LH). One reviewer conducted data extraction, and assessment of bias (MK). A second reviewer (LH) independently extracted data for a sample of the trials (3 RCTs, and 2 cohort studies), to verify data entry standards. No significant differences were seen in data entry. The data were extracted into an Excel spreadsheet and analyzed in R (version 4.0.3), using the package "meta." Assessment of risk of bias was performed using the Newcastle Ottawa Scale (Wells et al., 2014) for cohort studies and revised Cochrane Risk-of-Bias (RoB 2) tool for RCTs (Sterne et al., 2019). The Newcastle Ottawa Scale consists of three domains to assess the quality and risk of bias. These are: selection, comparability, and outcome assessment. Each cohort study was given a rating of stars for each domain with a maximum star rating of 4 for selection, 2 for comparability and 3 for outcome, with a greater number of stars reflecting a lower risk of bias. For RCTs, the RoB 2 reporting template was used to score each outcome as: "low risk," "some concern," or "high risk." The overall risk of bias is "low" if all domains are low risk, "some concerns" if some concerns are raised but these are not high risk, and "high risk" if any domain has high risk or there are multiple domains with some concerns. Data on risk of bias and overall quality assessment are presented in Supplementary  Tables 5 and 6.
Results for RCT and cohort studies were analyzed separately (Reeves et al., 2022). We used a random effects model (DerSimonion & Laird model) due to heterogeneity between studies. For binary outcomes, we calculated risk ratios. For continuous outcomes we used weighted mean differences. Uncertainty of the estimates (relative risk [RR] and weighted mean difference [WMD]) was expressed by calculating the 95% confidence interval [CI] The I 2 statistic was used to assess study heterogeneity. The I 2 represents the percentage of variance across studies attributable to heterogeneity rather than change and is presented alongside each forest plot for both RCTs and cohort studies. To investigate publication bias we produced a funnel plot if the number of pooled studies was greater than 10 (Supplementary Figure 1). We used adjusted estimates where they were available for cohort studies. As adjusted estimates were only available for 8 of 16 cohort studies, for a total of four outcomes (Table 1), we present the primary results as unadjusted analyses. Estimates for the pooled, adjusted analyses (where available) are included in Supplementary Table 7.
no adjusted results published.
(Continued) Regression model used to examine methadone exposed offspring and length of stay not analyzed as no comparison to buprenorphine published.
(Continued) additional papers were added following screening of citations in previously published papers in the field (Colombini et al., 2008;Gawronski et al., 2014;Konijnenberg & Melinder, 2015;Nørgaard et al., 2015). 408 studies were screened by title and abstract, 129 of which were selected as potentially includible and evaluated as full text articles. 20 papers met criteria for inclusion in the meta-analysis (Table 1). Two papers were included in the results for development outcomes only (Kaltenbach et al., 2012;Konijnenberg & Melinder, 2015), as birth and maternal outcomes were reported from these populations in other papers in this meta-analysis. The CONSORT flow diagram is shown in Figure 1. The 20 studies in this meta-analysis included 7251 patients (methadone n = 4146, buprenorphine n = 3105), in 4 RCTs and 16 cohort studies. The location of the studies was Europe (8), North America (10) and Oceania (2). Of the 16 cohort studies, 8 studies provided adjusted results for a total of 4 outcomes (small for gestational age, prematurity, duration of hospital admission, and NAS treatment). Characteristics of each study are given in Table 1 and results of pooled estimates for both adjusted (where available) and unadjusted analyses for cohort studies are presented in Supplementary Table 7.
The risk of bias was high for all the randomized trials (4/4). The median (IQR) Newcastle Ottawa Scale was 7.5 (6-9) for cohort studies (Supplementary Tables 5 and 6). A funnel plot was produced for outcomes with more than 10 studies, and there was no apparent asymmetry in these plots (Supplementary Figure 1).

Neonatal growth
Birthweight was reported in 14 studies (12 cohort, 2 RCTs). The weighted mean difference in offspring birth weight was 184 g (95% CI: 121-247 g) in cohort studies and 343 g (95% CI: 40-645 g) in RCTs favoring buprenorphine (Figure 2). One paper (Pritham et al., 2012), reported a standard deviation of 2695 g (4 times greater than other studies). When this study was excluded from the results, the weighted mean difference was 186 g (95% CI: 122-250 g) in cohort studies. Length at birth was measured in 9 studies (7 cohort, 2 RCTs), and was 0.65 cm (95% CI: 0.31 cm-0.98 cm) greater in the cohort studies and 2.28 cm (95% CI: 1.06-3.49 cm) greater in RCTs with buprenorphine compared to methadone ( Figure  3). Head circumference was measured in 9 studies (7 cohort, 2 RCTs). Buprenorphine was associated with a 0.42 cm (95% CI: 0.20-0.64 cm) increase in head circumference in the cohort studies, and no change in RCTs (weighted mean difference of 0.80 cm (95% CI: −0.03 to 1.63 cm). None of the cohort studies reporting birth weight or length at birth provided adjusted estimates. Small for gestational age (SGA) was investigated in 5 studies (all cohort studies). The risk ratio for SGA was 0.76 (95% CI: 0.66-0.88), in favor of buprenorphine. When this analysis was restricted to the three studies where the outcome was adjusted for confounding, or when both adjusted and unadjusted estimates were pooled, the risk ratio was no longer significant (Supplementary Table 7).
Seven studies reported stillbirths (596 methadone-exposed and 759 buprenorphine-exposed offspring) with the relative risk of stillbirth lower in cohort studies (RR 0.38, 95% CI: 0.12-1.20), however the confidence intervals included unity. There was one stillbirth (methadone group) in the three RCTs (86 buprenorphine exposed, 89 methadone exposed), and a relative risk could not be calculated. No offspring deaths were reported in either cohort or RCTs.

Childhood development
Three studies reported development outcomes, however due to heterogeneity in outcome measures these results could not be pooled. Bier et al. (2015) investigated development at 4 months with the Bayley Mental Developmental Index (MDI) and the Alberta Infant Motor Scale (AIMS). There were no significant differences in Bayley MDI scores between methadone (high and low dose groups) and buprenorphine. AIMS scores were different between groups with buprenorphine-exposed offspring having higher scores, compared to methadone (low and high dose groups). The proportion of infants with suspected or abnormal neurological examination was not significant between low dose methadone (n = 19 [23%]), high dose methadone (n = 17 [21%]) and buprenorphine-exposed groups (n = 7 [13%]). Whitham et al. (2010) measured visual evoked protentional at 4-months old, infants exposed to methadone (n = 22) had prolonged latency compared to controls and buprenorphine-exposed offspring (n = 30). Kaltenbach et al. (2018) followed up participants (n = 96, methadone n = 52, buprenorphine n = 44) of the 2010 Jones et al. trial. It was observed that these offspring had normal development at 36 months.
The rate of cesarean section was measured in 11 studies (8 cohort studies and 3 RCTs). Buprenorphine treatment was associated with a reduction in the rate of cesarean sections in cohort studies (unadjusted analyses), relative risk 0.90 (95% CI: 0.84-0.98). There was no difference in RCTs (RR 0.84 (95% CI: 0.52;1.36). No maternal deaths were reported in the studies.

Discussion
This meta-analysis shows that offspring who are exposed to buprenorphine, compared to methadone have greater birthweight, longer length at birth, and lower risk of prematurity in both RCTs and cohort studies. In RCTs, there was a greater risk of maternal adverse events with methadone, but higher drop-out rates with buprenorphine. Analysis of the cohort studies demonstrated greater head circumference, longer gestation, lower requirements for NAS treatment, shorter neonatal hospital stay, and reduced risk for cesarean section, however, these differences were not observed in the RCTs. There was no difference in risk of congenital malformations, small for gestational age, stillbirths or additional maternal opioid use in cohort studies, with insufficient data to analyze these outcomes in RCTs. Longer term childhood outcomes had insufficient data to make robust conclusions. Similarly adjusted estimates accounting for potential confounders were only available in half of the cohort studies and for only four outcomes. Of these only duration of hospital admission remained statistically significant in adjusted analyses. Collectively this data would suggest that buprenorphine may be beneficial compared to methadone, however, larger more robust studies are required.
This meta-analysis updates existing literature to include all available evidence from both RCTs and observational cohort studies comparing methadone and buprenorphine for opioid using mothers. Our findings confirm the results of previous meta-analyses regarding beneficial effects of buprenorphine on growth. Three smaller meta-analyses (n = 271 to 2146) have been published on this topic (Brogly et al., 2014;Minozzi et al., 2020;Zedler et al., 2016). Two meta-analyses (Brogly et al., 2014;Zedler et al., 2016) included RCTs and observational studies, while a third meta-analysis (Minozzi et al., 2020) was conducted including only RCTs (3 studies (Fischer et al., 2006;Jones et al., 2005Jones et al., , 2010 comparing methadone to buprenorphine). The meta-analysis of RCTs and observational studies by Brogly et al. (2014) reported an association with lower NAS treatment risk and treatment duration as well as shorter hospital stay in neonates exposed to buprenorphine compared with methadone, in a sample of 1370 patients. Buprenorphine was associated with greater mean gestational age, higher birthweight, longer length at birth, and larger head circumference at birth, and reduced illicit maternal opioid use near delivery. Adjustment for bias, including confounding by indication, attenuated these findings but there was still clinically and statically significant improvement in the buprenorphine group. In 2016 a meta-analysis including 3 RCTs and 15 cohort studies (2146 patients) reported similar results with buprenorphine exposed offspring having lower risk of preterm birth, greater birth weight, and larger head circumference compared with methadone exposed offspring but did not adjust for confounding . A 2020 Cochrane review by Minozzi et al. included four randomized controlled trials (271 patients), three of which directly compared buprenorphine to methadone, this analysis found no significant differences for maternal or neonatal outcomes between treatments. Evidence was considered of moderate or low quality due to small sample sizes and high drop-out rates as well as a lack of reporting of smoking status and inconsistencies in results. Long-term development outcomes were not included in these meta-analyses.
In our meta-analysis, we estimate a weighted mean increase in birthweight of 184 grams in cohort studies and 343 grams in RCTs in buprenorphine-exposed, compared to methadone-exposed offspring. This compares favorably to the 174 grams reduction in birthweight seen in offspring of women who smoke during pregnancy to those who do not smoke cigarettes (Delpisheh et al., 2006). The magnitude of improvement should be interpreted with caution due to possibilities of bias, (Brogly et al., 2015) especially as our meta-analysis was unable to control for differences in smoking rates between groups.
There are several strengths to this study. We included both cohort and RCTs, used comprehensive search terms, and reported a wide range of maternal and offspring outcomes including that of longer-term childhood development. We report pooled results, separately for both RCTs and observational studies, reflecting what we believe is the totality of research directly comparing methadone and buprenorphine in pregnancy. The main limitation of this study is that we did not adjust for confounding by indication in the included observational studies, and this may predispose to bias. It is believed that higher-risk opioid-using mothers may preferentially be treated with methadone rather than other agents (Brogly et al., 2016). Higher risk patients are expected to have neonates with worse outcomes due to differences in opioid substitution use as well as other drug use, increased maternal stress and smoking rates.
When correction for confounding by indication was accounted for in a previous meta-analysis, indices of growth differences were no longer significant, but the level of NAS treatment remained reduced (Brogly et al., 2014). A further meta-analysis by Zedler et al. (2016) did not correct for confounding arguing that any corrections were based on largely "subjective"  and potentially severe assumptions for key parameter values. Similarly, to Zedler and colleagues, we did not correct for bias in this study as it is not clear to what direction or extent bias can exist and we were concerned that the introduction of a correction factor biased on prior beliefs might introduce further bias. It is also possible that any existing bias, due to differences in prescription practices, training, and familiarization may change over time. Whilst we agree that ideally correction for confounding would be performed, this methodology requires further development before being widely implemented. We have reported analyses using adjusted estimates where these are available, though we accept that these are reported in only half of the included cohort studies, and for only a minority of outcomes.
A further limitation of this study was the significant risk of bias due to high drop-out rates and lack of a priori published protocols in randomized trials. These limitations are significant but expected when investigating the topics of opioid replacement in pregnancy due to the population studied and side-effects of treatment programs.
This meta-analysis has highlighted that further research is required into longer-term childhood development, specifically looking at any differences between drug groups and formulations. Few studies have investigated developmental outcomes, and meta-analysis has not been performed. Opioid replacement (buprenorphine or methadone) compared to no opioid replacement is negatively associated with a range of offspring developmental outcomes but differences between specific opioids have not been fully investigated (Andersen et al., 2020;Murawski et al., 2015;Nygaard et al., 2015Nygaard et al., , 2017Oei, 2018;Oei et al., 2017;Ross et al., 2015). The measurement of developmental effects is complicated by a multitude of factors including type and timing of testing, and preexisting differences between groups, as well as difficulties in recruitment of participants. Development concerns are increased in the first year of life (Whitham et al., 2010), only to recede in later years (Kaltenbach et al., 2018;Whitham et al., 2015). Different drug formulations such as buprenorphine-naloxone in combination may have additional advantages such as decreasing the risk of diversion and misuse. A recent meta-analysis of buprenorphine combined with naloxone compared with other opioid replacement regimes (methadone, buprenorphine, or long-acting opioids) showed no difference between groups but did not include any longer term follow up (Link et al., 2020). Further randomized controlled trials including larger populations, and with less loss to follow-up is aspirational but may not be feasible. Larger cohort studies using routinely collected healthcare data allow for larger sample sizes compared with RCTs but are limited by their observational nature and potential confounding. Further efforts to control for confounding may be achieved by the collection of detailed data on demographics, social factors, other determinants of health, and other drug use, including smoking and alcohol.
This meta-analysis shows that buprenorphine is associated with improvements in growth when compared to methadone. The priority for opioid replacement care programs remains the delivery of non-judgmental support, addressing of individual needs and maintenance of stability of treatment.