Abstract Opioid addiction in pregnancy is a growing concern that has recently received a great deal of attention. When comparing recommended opioid agonist therapies, many currently published studies guiding practice may have been affected by unmeasured confounding by indication. Populations of women who receive methadone are generally different from those treated with buprenorphine. Women treated with methadone frequently have more severe and uncontrolled addiction than buprenorphine-treated patients; however, these factors are typically unmeasured or unavailable in large observational data sets. Consequently, findings of superior perinatal outcomes with buprenorphine may in truth be a result of an overall healthier profile of women taking this medication. In this issue of the Journal, Brogly et al. (Am J Epidemiol. 2018;187(6):1153–1161) describe an approach utilizing detailed data from an external cohort (n = 113) to account for confounding by indication in a larger Medicaid population (n = 1,020) in order to more accurately compare opioid agonist therapies in pregnancy. They found that the decreases in risk of preterm birth and length of infant hospitalization associated with buprenorphine as compared with methadone were attenuated after accounting for the additional confounding. Brogly et al. should be commended for providing a novel method with which to address this bias in future studies. buprenorphine, methadone, opioid dependence, pregnancy, unmeasured confounding As evidenced by the 3-fold increase in opioid-dependent deliveries between 2006 and 2011 (1), the opioid epidemic in the United States is affecting all demographics, including pregnant women. The American Congress of Obstetricians and Gynecologists currently recommends 2 opioid maintenance therapies for opioid-dependent pregnant women: buprenorphine (Subutex; Reckitt Benckiser Pharmaceuticals Inc., Richmond, Virginia) and methadone (2). Recent research supports treatment with buprenorphine over methadone for optimizing maternal and child health outcomes, including decreased neonatal withdrawal, increased birth weight, and prolonged gestation (3). However, biases are difficult to eliminate in epidemiologic studies comparing opioid agonist therapies, largely because of differences in federal prescribing regulations. Methadone is a US Food and Drug Administration Class II full opioid agonist with strict prescribing requirements. Federal regulations require that this medication be taken daily under direct observation at a clinic, eliminating the chance of diversion to illicit street sales. It is therefore frequently given to women with more severe addiction and worse expected perinatal outcomes. Conversely, buprenorphine, a Class III partial opioid agonist, is frequently prescribed to women who are less likely to relapse due to less severe addiction, as the women are permitted to keep a limited supply of the drug to take at home. Women treated with buprenorphine also tend to be perceived as more trustworthy and less likely divert the drug. These predictors of treatment, including severity of addiction, are often unmeasured. However, if they are associated with the perinatal outcome of interest, the benefits observed for buprenorphine treatment as compared with methadone may be spurious. Properly measuring and adjusting for these potential confounders is critical, but in the absence of such data, the impact of unmeasured confounding must be estimated before buprenorphine use is expanded and recommended as first-line therapy. In their accompanying paper, Brogly et al. (1) describe how they sought to compare buprenorphine with methadone therapy in relation to preterm birth (<37 weeks’ gestation), gestational-age-standardized low birth weight, and length of infant hospitalization while using a bias analysis to account for unmeasured confounding. Previous research has found benefits associated with buprenorphine for all 3 of these outcomes in comparison with methadone (2). To address these objectives, Brogly et al. used a sample of 1,020 Massachusetts women receiving opioid agonist therapy (OAT) during pregnancy through the Medicaid program (2006–2011). The authors used prescription and therapy databases to define the presence of any exposure to OAT during the period from 294 days before delivery to 30 days after the date of delivery. Following an intent-to-treat approach, women were classified in accordance with the first type of therapy documented. Brogly et al. then obtained detailed addiction severity data from an external cohort of women being treated at a high-risk obstetric and addiction recovery clinic in Boston (n = 113; 2015–2016). The authors first estimated associations between OAT and birth outcomes using generalized linear models and linear regression adjusted for measured confounders. In these conventional analyses, they found that buprenorphine was associated with decreased risk of preterm delivery (risk ratio (RR) = 0.45, 95% confidence interval (CI): 0.34, 0.61), decreased risk of gestational-age-standardized low birth weight (RR = 0.75, 95% CI: 0.51, 1.11), and decreased duration of infant hospital stay (mean difference = −7.35 days, 95% CI: −9.16, −5.55), which is consistent with the extant literature. Because these results may be subject to bias from unmeasured confounding by indication for OAT type, Brogly et al. performed additional collection of detailed severity data that were not available in the Medicaid population. In a smaller cohort of 113 women receiving care from a high-volume external prenatal recovery clinic, they administered the Addiction Severity Index at initiation of OAT or the first prenatal visit for women who conceived while already receiving prescribed treatment. For women initiating treatment, OAT type was determined with consideration of multiple patient-specific factors; the clinic’s protocol recommended that women suffering from more severe dependence receive methadone. As expected, Brogly et al. found that women with less severe dependence and lower overall risk profiles were more often treated with buprenorphine. They then generated propensity-to-treatment-type scores in the external cohort, inclusive only of those variables that were also associated with the infant outcome. Variables retained in the final propensity model included unstable maternal housing in the previous 3 years, gravidity greater than 1, history of regular cocaine injection, and an Addiction Severity Index score (continuous) indicative of dependence. Next, using an approach developed by Lin et al. (4), the authors used this external information to correct for the unmeasured confounding by indication in the larger Medicaid population. After the unmeasured confounding was accounted for, the conventional findings were attenuated but suggested a protective effect of buprenorphine on preterm birth and length of infant hospital stay (RR = 0.53 (95% CI: 0.39, 0.71) and mean difference = −3.66 days (95% CI: −5.46, −1.87), respectively). These findings are supported by a study we conducted with a similar objective of comparing opioid agonist therapies while properly accounting for unmeasured confounding by maternal addiction severity, but in regards to neonatal abstinence syndrome (5). In our study of 716 pregnant women receiving OAT, we implemented a probabilistic bias analysis informed from an internal cohort and found comparable results. The association of methadone with increased risk of neonatal abstinence syndrome, though slightly attenuated from a risk ratio of 1.3 (95% CI: 1.2, 1.5) to a risk ratio of 1.2 (95% CI: 1.0, 1.4), remained and was not fully explained by our measurement of maternal addiction severity (5). Though the findings are in agreement, both analyses were limited by small sample sizes and restricted data collection points. Because of their limited sample size, Brogly et al. had to assume that the impact of confounding was the same regardless of treatment type. Additionally, because factors beyond those measured in their study may be influential confounders, the authors cautioned that the protective association could have been the result of residual unmeasured or mismeasured confounding (1). It is equally possible that the association is true and that this analysis produced an unbiased estimate of the association between therapy and these poor birth outcomes. Regardless, the use of external information to estimate the effects of confounding in similar populations is an innovative approach that can be utilized by other researchers in the future. However, it is not without limitations or opportunities for improvement. Though it is admittedly a difficult decision in all pharmacoepidemiologic research—and is potentially a limitation of the administrative claims data themselves—the authors categorized exposure in a binary manner, considering any prescription received from 294 days prior to delivery to 30 days postdelivery as indicative of treatment type. This approach ignores the complexities of cumulative drug effects, sensitive time windows of exposure, dose response, and potential misclassification. For instance, a woman who takes methadone consistently for years prior to pregnancy and throughout pregnancy will probably have different perinatal outcomes than one who initiates OAT in her 40th week of pregnancy. Equally, a woman who chooses to detoxify early in pregnancy, thereby limiting the fetus’ opioid exposure to the early developmental period, may have different outcomes than a woman whose fetus is exposed in all 3 trimesters. Use of a binary exposure disregarding initiation, frequency, and dose does not allow for investigation of these nuances. These are factors that are essential for all researchers in this field to consider when determining ideal exposure classification. A potential for exposure misclassification also remained in this study. The binary classification of exposure with an intent-to-treat analysis did not account for the misclassification of exposure, an issue acknowledged by the authors (1) and present in many opioid therapy studies. Though it likely had little impact in the Medicaid cohort, as only 26 of those women changed therapies, 20.7% of women in the external cohort converted to a different type of therapy and were exposed to both buprenorphine and methadone during pregnancy. Therefore, the propensity score generated in the external cohort was subject to potentially substantial misclassification. An additional concern with the use of propensity scores in this study is that the factors contributing to the score did not necessarily predate treatment initiation. If their behavior was consistent with that observed in previous studies, many women in the external cohort would have conceived while already prescribed OAT, violating the temporal nature of this approach. For example, if a woman had an Addiction Severity Index score indicative of severe addiction at the time of study enrollment but conceived while on buprenorphine, is this truly a factor predicting treatment choice or rather a result of the treatment choice (e.g., an insufficiently strong maintenance therapy)? Application of these measures of confounding may be better suited to informing other bias quantification methods, such as a probabilistic bias analysis, due to this temporal violation of propensity scores. Another adaptation future researchers may consider when utilizing this approach is using a subsample of the total population as the cohort when available, and incorporating detailed information from this subpopulation. Though Brogly et al. justifiably chose to use the highest-volume clinic in New England as the source for their external cohort, use of an internal cohort would minimize misclassification of the propensity score by ensuring that the factors driving the propensity to receive treatment were similar across groups. The internal cohort would simultaneously eliminate provider practice differences and demographic disparities, such as the variation in the patient race/ethnicity distribution between the Medicaid cohort and the external cohort in Brogly et al.’s study (1). An issue integral to both this article and the population at large, addressed very soundly in the Discussion section of the paper (1), is that the majority of the population identified as opioid-dependent in the Medicaid database received no medication. The mere existence of this group is a public health crisis. Although OAT is available for this population of Massachusettsans, countless known and unknown barriers persist in restricting access to care for these women. Limitations such as a lack of health-care providers, difficulty complying with regulations, and the inherent stigma of seeking care for drug addiction while pregnant must be addressed. This abysmal situation is worsened by the racial trend noted by Brogly et al. (1), which was also present in the population we studied (5), in which non-Hispanic black women diagnosed with opioid-use dependence are less likely to seek to treatment. While overcoming these barriers and providing appropriate care to all women in need should undoubtedly remain the ultimate goal, optimizing perinatal outcomes for those receiving treatment is nevertheless essential. Brogly et al. have admirably demonstrated one approach that allows researchers to quantify the effect of unmeasured confounding by prescribing preferences for OAT based on severity of opioid dependence. Their work is a valuable addition to the tool set necessary for addressing this public health catastrophe. With an expanded repertoire of methods for more accurately comparing opioid agonist therapies, we can begin to provide optimal medication options to pregnant women seeking to treat their addiction. ACKNOWLEDGMENTS Author affiliation: Department of Clinical Analytics, University of Pittsburgh Medical Center, Pittsburgh, Pennsylvania (Lara S. Lemon). Conflict of interest: none declared. Abbreviations CI confidence interval OAT opioid agonist therapy RR risk ratio REFERENCES 1 Brogly SB, Hernández-Diaz S, Regan E, et al. . Neonatal outcomes in a Medicaid population with opioid dependence. Am J Epidemiol . 2018; 187( 6): 1153– 1161. 2 ACOG Committee on Health Care for Underserved Women; American Society of Addiction Medicine. ACOG Committee Opinion no. 524: opioid abuse, dependence, and addiction in pregnancy. Obstet Gynecol . 2012; 119( 5): 1070– 1076. CrossRef Search ADS PubMed 3 Brogly S, Saia K, Walley A, et al. . Prenatal buprenorphine versus methadone exposure and neonatal outcomes: systematic review and meta-analysis. Am J Epidemiol . 2014; 180( 7): 673– 686. Google Scholar CrossRef Search ADS PubMed 4 Lin DY, Psaty BM, Kronmal RA. Assessing the sensitivity of regression results to unmeasured confounders in observational studies. Biometrics 1998; 54( 3): 948– 963. Google Scholar CrossRef Search ADS PubMed 5 Lemon LS, Caritis SN, Venkataramanan R, et al. . Methadone versus buprenorphine for opioid use dependence and risk of neonatal abstinence syndrome. Epidemiology . 2018; 29( 2): 261– 268. Google Scholar PubMed © The Author(s) 2018. Published by Oxford University Press on behalf of the Johns Hopkins Bloomberg School of Public Health. All rights reserved. For permissions, please e-mail: firstname.lastname@example.org. This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices)
American Journal of Epidemiology – Oxford University Press
Published: Nov 16, 2017
It’s your single place to instantly
discover and read the research
that matters to you.
Enjoy affordable access to
over 18 million articles from more than
15,000 peer-reviewed journals.
All for just $49/month
Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly
Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place.
Get unlimited, online access to over 18 million full-text articles from more than 15,000 scientific journals.
Read from thousands of the leading scholarly journals from SpringerNature, Elsevier, Wiley-Blackwell, Oxford University Press and more.
All the latest content is available, no embargo periods.
“Hi guys, I cannot tell you how much I love this resource. Incredible. I really believe you've hit the nail on the head with this site in regards to solving the research-purchase issue.”Daniel C.
“Whoa! It’s like Spotify but for academic articles.”@Phil_Robichaud
“I must say, @deepdyve is a fabulous solution to the independent researcher's problem of #access to #information.”@deepthiw
“My last article couldn't be possible without the platform @deepdyve that makes journal papers cheaper.”@JoseServera