To clarify this issue, we have made the following changes to the Introduction and Discussion:

Transcription

1 Manuscript ID BMJ "Risk of Neonatal Drug Withdrawal after Intrauterine Co-Exposure to Opioids and Psychotropic Medications: A Cohort Study" Please find below our replies to the comments made by the reviewers and the editors, along with an explanation of how we have dealt with them in the paper. Comments BMJ s manuscript committee meeting 1. Please discuss the international context. From a European point of view, the rates of pregnant women prescribed opioids and co-analgesics look extremely high. Although the use of prescription opioids in Europe remains lower than in North America, a strong increase in use has been observed in recent decades in almost all European countries as well as Australia. This increase in use, combined with the fact that the biological relations studied should be generalizable beyond the specific population studied, make the findings and clinical message relevant in an international context. To clarify this issue, we have made the following changes to the Introduction and Discussion: Introduction: A strong increase in the prescription of opioids has also been observed in most European countries (e.g., Germany, Italy, Spain, Scandinavia, United Kingdom, the Netherlands) and Australia, although the number of prescriptions and defined daily doses remains much lower than in the U.S. and Canada. 5 Data from a population-based registry in Norway revealed that 6% of pregnant women filled at least one opioid prescription between 2004 and Discussion: Since there is no reason to believe that the underlying biological relations would be different in this study population compared to other populations of pregnant women, we should be able to generalize from our observations to the broader experience of pregnancies inside the U.S. as well as outside the U.S. where the use of prescription opioids is also on the rise Looking at the tables, this is really about a very small proportion of women taking both types of drugs. Some of the psychotropics are class D (evidence of teratogenicity), and we were surprised to see gabapentin considered a psychotropic, which is also used for pain syndromes and can lead to confounding issues. One in 10 women who take prescription opioids during the last 45 days of pregnancy are also exposed to other psychotropic medications, as can be seen in etable 8. This proportion would likely be higher if we considered a broader exposure window, but we preferred to define the exposure with high specificity since our objective was to assess the causal association (as opposed to documenting prevalence of use). Importantly, our findings also apply to the growing number of women that use opioids illicitly during pregnancy. This combined with the increasing trend of psychotropic polypharmacy (Mojtabai R, Arch Gen Psych 2010; 67(1):26-36) and the fact that pregnancy is such a common condition make this an important public health question. Given the increase in NAS cases over time, in October 1

2 2015, the U.S. government promoted the Protecting Our Infants Act (S.799) to review activities related to prenatal opioid use, with the aims of improving the existing NAS surveillance programs, addressing research gaps, and implementing effective measures for education, prevention, and treatment. Despite the fact that some psychotropics were formerly categorized as class D (e.g., paroxetine), they are used relatively frequently during pregnancy. Although gabapentin is an anticonvulsant approved for epilepsy, it is also being used off label as a mood stabilizer. As such, it is also classified as a psychotropic medication. We included it in the current analyses because of its frequent use as a mood stabilizer and because of prior reports suggesting a potential risk of neonatal withdrawal. It is correct that it is also often used as an analgesic. We would like to note, however, that we account for pain indications, and although these are imbalanced in the unadjusted Table 1, we achieve balance after propensity score stratification. Moreover, the hdps analysis which accounts for proxies of potential unmeasured confounders did not affect the results, suggesting residual confounding is unlikely to explain the observed association. Please see also our response to comment # 4. To address this comment, we have added the following clarification in the text: We distinguished between 5 different exposure groups: antidepressants, atypical antipsychotics (hereafter antipsychotics), gabapentin (included because of its mood stabilizing properties), benzodiazepines, and non-benzodiazepine hypnotics (hereafter Z-drugs). 3. What does pro re nata use mean? Pro re nata is a Latin phrase meaning as needed. We have changed the wording. 4. We agree with Leeman that gabapentin use may well be a marker for pain severity and that additional use of nonprescribed opioids might be playing a role. We did account for pain indications - specifically neuropathic pain, musculoskeletal pain, back or neck pain, fibromyalgia, migraine or headache, and other pain and these pain indications are well balanced in the adjusted analyses (Table 1). We agree, however, that the use of gabapentin could be a marker for pain severity and that additional use of nonprescribed opioids could act as an unmeasured confounding variable. To explore this issue, we conducted a sensitivity analysis, using the rule-out approach (Schneeweiss, Pharmacoepidemiology & Drug Safety 2006; 15(5): ), to assess the strength of the confounding by nonprescribed opioids that would be necessary to fully explain the observed association. Figure 1 shows the strength of the association between the unmeasured confounder (non-prescribed opioids) and the outcome (RR CD ) and between the unmeasured confounder and the exposure (OR EC ) that would be necessary to fully explain the observed adjusted relative risk of 1.61 for three different prevalences of the unmeasured confounder: 0.06, 0.10, and This figure illustrates that the strength of the unmeasured confounding would have to be unrealistically strong to fully explain the observed association; especially considering the broad range of variables we already adjusted for. For example, if 10% of women used non-prescribed opioids (P C =0.10) and non-prescribed opioids increased the risk of NAS four-fold (RR CD =4), gabapentin exposed women would have to be 5 times more likely than unexposed women to use non-prescribed opioids (OR EC =5) for residual confounding to fully explain the adjusted association of

3 Figure 1 Sensitivity analysis of residual confounding (rule-out approach) for an estimated relative risk (RR) of 1.61 and different levels of confounder prevalence. Each line splits the area into two. The upper right area represents all combinations of an association between exposure and confounder (OREC) and an association between confounder and outcome (RRCO) that would create confounding by an unmeasured factor strong enough to move the point estimate of RR to the null value (RR = 1) or beyond. The area to the lower left represents all parameter combinations that would not be able to move the estimated RR toward the null OREC PC=0.10 PC=0.06 PC= RR CD To address this comment, we have added the following text to the Discussion: Since gabapentin is frequently used off-label as an analgesic, use of gabapentin may be associated with more severe pain and illicit opioid use, which could lead to residual confounding. However, sensitivity analyses indicate that the strength of the association between illicit opioid use, and both gabapentin exposure and neonatal drug withdrawal would have to be unrealistically high for confounding to fully explain the observed association, especially since we adjusted for pain indications and indicators of substance abuse and dependence (etable 2). The above analysis was added to the Appendix. 5. Please consider if you could be more cautious in your conclusions. We have adjusted the conclusions in the Abstract as well as the main text, using track changes as recommended. 6. Paediatricians may be more likely to label symptoms as withdrawals if they know the mother has been taking more and combinations of medications. Please consider if this is worth a comment in the discussion. Exposure to these medications has not traditionally been thought of as risk factors for neonatal withdrawal, making this explanation perhaps less likely. Nevertheless, to address this concern, we have added the following comment to the Discussion: Although we observed a doubling of the risk in infants who had been exposed to 2 or more psychotropic medications in additions to prescription opioids, we cannot exclude the possibility that pediatricians are more likely to diagnose withdrawal in infants with known exposure to polypharmacy. 3

4 7. The committee agreed with the reviewers concerns. Please see below for a point-by-point response to the reviewer comments. 8. Results are given from an observational cohort study and yet inferences made are causal. The language needs to be changed. We carefully checked the manuscript and removed any language that could be interpreted as causal, using track changes. 9. The PPV estimates of outcome definition should be given with confidence intervals. For the prescription opioid subgroup, 7 were sampled, and all had the correct diagnosis. Whilst this was 100%, the 95% confidence interval ranges from 65% (to 100%). How often was there discrepancy between the two reviewers? We have added the requested information: Two reviewers (D.B. and J.C.), blinded to the exposure status, adjudicated these cases; discrepancies were resolved by consensus (4 out of 50 cases with prescription opioids, 0 out of 7 cases without prescription opioids). We estimated a positive predictive value (PPV) of 91% (95% CI, 82% - 97%) overall, and 100% (95% CI, 65% - 100%) in the subset of infants with intrauterine exposure to prescribed opioids. 10. Two levels of propensity scoring are used to identify subgroups of individuals who are similar between the two exposure groups. The adjustment variables should be clarified. We have clarified the variables included in both levels of adjustment in the text and the tables: Statistical Analysis: In the first adjusted analysis, we accounted for the specific opioid compound (i.e., generic name and form [immediate release or extended release] as listed in etable 1) and dose expressed in morphine equivalents, which is expected to be the most important confounder. In the second adjusted analysis, we also accounted for additional potential confounders, as described above (etable 3). In etable 3, we now list the variables by level of adjustment: Exposures Antidepressants Antipsychotics Benzodiazepines Gabapentin Z-drugs Potential confounding variables First level of adjustment Specific Opioids Morphine equivalents in mg, Codeine, Hydrocodone, Oxycodone (IR/SA and ER/LA), Other opioids [Butorphanol (IR/SA), fentanyl (IR/SA and ER/LA), 4

5 Second level of adjustment Demographics Obstetric Characteristics Pain Psychiatric Diagnoses Other Diagnoses Indicators of Substance Abuse or Dependence Other psychotropic medications Other medications General Markers of Disease Burden hydromorphone (IR/SA), meperidine (IR/SA), morphine (IR/SA and ER/LA), oxymorphone (IR/SA and ER/LA), pentazocine (IR/SA), propoxyphene (IT/SA), tapentadol (IR/SA), tramadol (ER/LA)] Age, White race, Black race, Hispanic race, Other or unknown race Multifetal gestation, Preterm birth, Preeclampsia Neuropathic pain, Musculoskeletal pain, Back or neck pain, Fibromyalgia, Migraine or headache, Other pain Bipolar disorder, Schizophrenia, Depression, Anxiety, Sleep disorder, Psychosis, Personality disorder, Adjustment disorder, ADHD, Dementia, Delirium, Other psychiatric disorders Diabetes, Epilepsy / convulsions, Nausea / vomiting, Obesity / overweight, Restless legs syndrome Alcohol abuse or dependence, Drug abuse or dependence, Tobacco use, Methadone, Methadone extended release, Methadone maintenance therapy, Buprenorphine, Naloxone, Naltrexone Anxiolytics, Barbiturates, Other hypnotics, Stimulants, Primidone, Phenobarbital, Pregabalin Antidiabetic, Insulin Number of generics, Number of diagnoses, Number of outpatient visits, Hospitalization, Number of hospitalizations, Number of days in hospital, Number of emergency department visits, Maternal comorbidity index 11. A confidence interval should be given for the increase in mean hospital length of stay (last sentence exploratory analyses). The confidence interval has been added: 1.13 (95% CI -0.03, 2.28) 12. There is a typo in figure 2: ci for benzodiazepines PS adjusted, lower limit is 1.35 not 1.65 Thank you for pointing this out. We have corrected the typo. Reviewer: 1 General comment: 1. This paper aims to test the association of in utero exposure of psychotropic medications and opioids with neonatal abstinence syndrome incidence and severity using a large cohort of maternal/infant dyads enrolled in the Medicaid program. Overall, the paper is very well written and compelling. The data are somewhat dated ( ); however, I do not suspect this would impact the findings. Overall, the paper may benefit from a broader discussion of where this paper fits in the overall literature. The discussion could be expanded to include more clinical context as well. Please refer to our response to comment #13 for details on how we addressed the request for including a broader discussion of where the paper fits in the overall literature and the request for more clinical context. 5

6 Specific Comments: 2. Abstract: Clear. Accurately states findings. Thank you; no changes made 3. What is known on this topic: Page 5, Line 9: Neonatal withdrawal from psychotropic medications is somewhat unclear and inconsistent. The literature most strongly supports a withdrawal syndrome from benzodiazepines. Some exposures are less clear. For example, neonatal clinical signs after SSRI exposure, however, likely represent a toxidrome rather than a withdrawal syndrome. We agree that for SSRI exposure, the symptoms might be better explained by toxicity than withdrawal in some cases. We felt this was too much detail to include in the What is known on this topic section, especially since the statement refers to psychotropic medications more broadly, not just SSRIs ( but other psychotropic medications can also cause signs of withdrawal) and it is a correct statement that SSRIs can cause signs of withdrawal. However, to address this comment, we have added the following to the limitations section of the manuscript: It should be acknowledged that in some cases, the constellation of neonatal symptoms following in utero exposure to SSRIs may be better explained by potential toxicity associated with increased serotonin concentrations rather than withdrawal (attributable to a relative hypo- serotonergic state) Introduction: Overall, the introduction is clear and well written. The introduction could be improved by more clearly highlighting the differences between this paper and previous work, including by this research team, and the present paper. For example, the team s previous work highlighted an increased risk of NAS with opioid exposure + exposure to third trimester psychotropic medications. This paper appears to have greater specificity to drug class, but this clarification would be useful. This does appear in one sentence in the introduction, but may be best positioned in the introduction. As recommended, we have moved the clarification of how this paper fits within the content of he existing evidence to the Introduction: Whereas intrauterine exposure to psychotropic medications has been used as a risk stratification factor 14, the risks of neonatal drug withdrawal in opioid exposed pregnancies with or without coexposure to individual and combinations of psychotropic medications have not previously been examined. Little is therefore known about how this risk varies by the type of psychotropic medication or about the risks associated with psychotropic polypharmacy. 5. Methods: Overall, clear. Sensitivity analyses well done and thought out. More detail on final model and variables in propensity score would be informative. We have clarified the specific variables included in both PS models. Please refer to our detailed response to comment #10 from the Committee Meeting. We have also provided more detail on the final model (SAS Proc Genmod with a weight statement and log link function). 6

7 6. Page 8, line 38: Were cases of iatrogenic drug withdrawal identified? We defined the outcome based on ICD-9 diagnostic code (Drug withdrawal syndrome in newborn) which should be reserved for withdrawal following in utero exposure. Iatrogenic drug withdrawal should be coded using Our internal validation study showed a high PPV for code In response to this question, we have assessed how many of the cases identified also have a code for (see Table). Given the small proportion and given that fact that the % is lower among the psychotropic medication exposed than among those exposed to opioids alone (with the exception of antidepressants), we do not believe that the fact that some of the cases might be iatrogenic withdrawal wrongly coded as withdrawal following in utero exposure would have affected the findings. Opioids + Psychotropic Medications Opioids alone Withdrawal codes Withdrawal codes and and N % N % Antidepressants , Antipsychotics , Benzodiazepines , Gabapentin , Z-drugs , Page 10, Line 18: Was opioid type considered (e.g. IR, SR)? Some opioids may have the same generic name, but have different preparations. Yes, we did account for opioid type. We have now clarified this in the footnote in Table 1, etable 1 and etable 3. We also added a clarification to that extent in the text itself: In the first adjusted analysis, we accounted for the specific opioid compound (i.e., generic name and form [immediate release or extended release] as listed in etable 1) and dose expressed in morphine equivalents, which is expected to be the most important confounder. 8. Page 10, lines 6-46: It would be informative to know final variables in final models and/or used in propensity scores. We have clarified the specific variables included in both PS models. Please refer to our detailed response to comment #10 from the Committee Meeting. 9. Page 10, lines 6-46: Were interaction terms tested? For example opioid type x MME? The goal of using stratification on the PS is to achieve balance in terms of the potential confounding variables. As can be seen from the descriptive tables, we achieved good balance on all variables as judged by a an absolute value of the standardized mean difference <0.1 (with exception of a few variables for antidepressants). Opioid type and MME were well balanced for all. We therefore did not 7

8 consider it necessary to include interaction terms for these variables. This has now been clarified in the manuscript: Only main effects were included in the models. 10. Page 10, line 47: Because some areas of the country discharge infants home on medication weans, LOS may under-represent severity of withdrawal. Thank you for pointing this out. We have now acknowledged this possibility in the Discussion: In some parts of the U.S., infants are discharged home while they are still being weaned off opioids or other medications, which would result in an underestimation of the severity of withdrawal. 11. Results: Page 12, line 45: To clarify, by absolute risk, unadjusted risk is presented. So for example, those populations presented here are also the populations with higher MMEs? Yes, that is correct. These are unadjusted absolute risks. 12. Page 13, line 15: More detail as to what was included in the model would be informative. We have clarified the specific variables included in both PS models in etable 3. Please refer to our detailed response to comment #10 from the Committee Meeting. 13. Discussion: Could be expanded to discuss findings in context of the existing literature and discuss applicability to clinical practice. Overall, clear and concise. Limitations are well written. Thank you for raising this point. We have added the following section to the Discussion: Comparison with other studies Psychotropic medications, notably benzodiazepines and SSRIs, have been identified as exposures that increase the incidence and severity of neonatal drug withdrawal in studies of pregnant women with addiction. In a study of 450 drug-misusing women prescribed substitute methadone, maternal use of benzodiazepines significantly increased the likelihood of neonatal drug withdrawal requiring treatment (OR=1.73, 95% CI ) in univariate analyses. Maternal use of SSRIs was associated with a small, non-significant increase in risk (OR=1.30, 95% CI ). 25 Similarly, a small study in an urban drug treatment center found that nearly twice as many newborns (83% versus 42%) of women exposed to methadone and other illicit substances including benzodiazepines (N=19) received pharmacotherapy for more severe neonatal drug withdrawal than newborns born to women receiving methadone only (N=18). There was no difference, however, in the number of days of pharmacotherapy required to treat the withdrawal, indicating that while there was a difference in the incidence of withdrawal, there was no difference in severity among affected infants. 26 In contrast, a study of infants with drug withdrawal born at Boston Medical Center between found that maternal use of prescribed methadone for opiate addiction and benzodiazepines (N=56), compared to methadone alone (N=158), increased the length of stay for infants with drug withdrawal by 6 days. Maternal use of methadone and SSRIs (N=51) increased the length of stay by 4.5 days. 27 Based on these findings, intrauterine exposure to psychotropic medications has been used as a risk stratification factor in a study examining the risk of neonatal drug withdrawal with exposure 8

9 to prescription opioids. 14 We built on this work by examining the absolute and relative risk and the severity of withdrawal with co-exposure to specific psychotropics and combinations of psychotropic medications. We have expanded the section on the clinical context as follows: Chronic pain management is a serious global health priority with chronic pain afflicting around 20% of the population. Women are more frequently affected than men. The use of prescription opioids has grown significantly over the last two decades, including among pregnant women. The increased risks of neonatal drug withdrawal in women concomitantly using opioids and antidepressants, benzodiazepines, or gabapentin suggest that clinicians should be cautious in prescribing these medications together in late pregnancy and in prescribing psychotropic medications to women with known or suspected illicit opioid use. Because pain and mental health conditions often co-occur, co-exposure will be unavoidable in many instances. In those cases, our findings may be helpful in risk stratification for exposed infants. Our findings also imply that it will be important for neonatologists and pediatricians to rethink treatment protocols for infants born to women who were prescribed multiple drugs during their pregnancy. 14. Page 16, Line 12: This is an excellent point. Thank you; no change needed 15. Table 3 It is somewhat unclear that these variables exhibit disease severity. Iatrogenic withdrawal may influence these results if not accounted for. For example, lower gestational age infants are more likely to have respiratory complications and have longer LOS. We agree that these are proxy variables at best; which is why we describe and present these analyses as exploratory only. The caveats related to interpretation of these analyses are described quite extensively in the Discussion: Our data source did not contain a quantitative measure of severity such as the Finnegan score. 24 We therefore explored the effect of exposure on the severity of withdrawal using indirect proxies which are less sensitive to changes in severity than a direct quantitative measure, and which may actually reflect other co-morbidities. Likewise, length of stay as documented in healthcare utilization data is an imperfect measure of severity as it can be affected by unrelated medical conditions or social concerns such as the need to arrange for foster care prior to discharge. In some parts of the U.S., infants are discharged home while they are still being weaned off opioids or other medications, which would result in an underestimation of the severity of withdrawal. The relatively short hospital length of stay and 1- day increase due to concomitant exposure to psychotropics should be interpreted in that context. If the reviewer and/or editors would prefer us to remove these exploratory analyses, we would be willing to do so, although we do feel they provide useful information and are cautiously interpreted in the manuscript. 9

10 Reviewer: 2 The issue of the effect of maternal psychotropic medications on the development of NAS and need for pharmacological treatment of NAS is an important one. Evidence from other studies suggests an effect from SSRIs. 1. My primary concern with this study is the ability to accurately determine the exposure and the subsequent potential for confounding. An example is the possibility that since Gabapentin is used for pain while Z drugs and atypical antipsychotics are not, it is likely that the group using Gabapentin has a higher utilization of opioids for pain. The authors control for this by including the variable of amount of prescribed opioids in their adjusted analysis. This has a major effect and brings the adjusted RR into the 1.0 to 1.5 range where residual confounding is common. The residual confounding here may be the illicit use of additional opioids either heroin or "street" prescription opioids. Given the very small proportion of infants with NAS, even a small confounding effect that would occur if women using heroin were more likely to use gabapentin or benzodiazepines could explain these RR. I would consider this an exploratory study with a more rigorous study best done with a pregnant population using medication assisted treatment with close follow-up including frequent drug screening. We agree with the reviewer that with every non-randomized study, the potential for systematic error exists and careful analyses are needed to minimize the potential for misclassification and confounding. In this specific study, adjustment for the type of prescribed opioid, form and dose, reduced the relative risk for gabapentin to 2.68 ( ). Adjustment for all other potential confounding variables further reduces the relative risk to 1.61 ( ), above the threshold the reviewer refers to (1 to 1.5). Nevertheless, to address concerns about residual confounding by illicit opioid use, we conducted a quantitative bias analysis to assess the strength of the confounding that would be needed to fully explain this observed increased risk if the true underlying association was null. Please refer to our response to comment #4 from the Committee Meeting for details on this analysis. From this analysis, we conclude that the strength of the confounding would have to be unrealistically high, especially in light of the fact that we already account for a broad range of pre-specified confounders (PS-adjusted), including the pain conditions noted by the reviewer, and 200 empirically identified confounders and proxies for unmeasured confounders (hdps-adjusted). We now refer to the potential for residual confounding in the Discussion and have included this sensitivity analysis in the Appendix. Please refer to our response to comment #4 from the Committee Meeting for details on how we incorporated this additional information. 2. Recommend that the limitation due to the lack of assessment of non-prescribed opiates needs to be more clearly presented. We agree this is an important point. We included the following information in the Discussion: Despite our focus on prescription opioids, the possibility remains that the level of exposure was underestimated if some women also consumed opioids illicitly, or was overestimated if some women diverted their opioids and therefore did not consume them. In that regard it is reassuring that we did not observe an increased risk for women exposed to antipsychotics, since the likelihood of abuse is high in this group of more severely ill women. Since gabapentin is frequently used off-label as an analgesic, use of gabapentin may be associated with more severe pain and illicit opioid use, which could lead to residual confounding. However, sensitivity analyses indicate that the 10

11 strength of the association between unmeasured illicit opioid use and both gabapentin exposure and NAS would have to be unrealistically high for confounding to fully explain the observed association, especially in light of the fact that we adjusted for pain indications and indicators of substance abuse and dependence (efigure 2). 3. The large number of authors appears excessive for a database study. A carefully conducted epidemiologic study using secondary data draws on the expertise of many people with different backgrounds and experiences, and requires a multidisciplinary team. All members of the team contributed specific expertise and met authorship criteria, as documented in the author contributions. 4. The finding of a very low proportion of infants with perinatal opiate exposure having NAS is indeed important. Recommend this be included as important secondary outcome. The absolute risks of NAS in infants exposed to prescription opioids are presented in Table 2, as the first step of the analyses. More detailed analyses of risks of NAS with prescription opioids have previously been presented (e.g., Desai R et al. BMJ 2015;350:h2102). The focus in this contribution is on the risk increase due to concomitant exposure to psychotropic medications. We agree this is an important point, however, and indicate in the Discussion that the low risk of withdrawal with exposure to prescription opioids is an important finding in itself: The fact that in many of these women, fetal exposure may be minimal and will not lead to withdrawal is an important finding in itself. 5. Determine the RR of NAS at different exposure levels Although we agree with the reviewer that such analysis would be of interest in principle, we do not believe it is feasible in this case. Dose-response analyses for prescription opioids themselves have previously been published (Desai R et al. BMJ 2015;350:h2102). With respect to the psychotropics, analyses are presented at the drug class level (with the exception of gabapentin). In contrast to opioids, no generally accepted conversion factors exists for these other psychotropic medication classes, making such analysis not feasible. The numbers are too small to create stable estimates for individual generics. 11