Neonatal Abstinence Syndrome and Associated Neonatal and Maternal Mortality and Morbidity

Information on all singleton hospital live births in Canada (excluding Quebec) between April 1, 2005 and March 31, 2016 (fiscal years 2005–2006 to 2015–2016), was obtained from the Canadian Institute for Health Information (CIHI). CIHI collects data on all hospitalizations in Canada except for those in Quebec, which use a different data collection system. Each hospitalization record includes up to 25 diagnostic codes (International Classification of Diseases, 10th Revision, Canada [ICD-10-CA]) and 20 procedure codes (Canadian Classification of Interventions [CCI]). Related childbirth hospitalizations were identified through a standard CIHI algorithm and linked with newborn hospitalizations. The database included ∼98% of births in Canada (excluding Quebec).¹⁷ Accuracy and completeness of the information on diagnoses was maintained by trained personnel using standardized definitions and forms to collect the data,^18,19 and the information on key data elements in the database has been previously validated.^18,20,21 Ongoing algorithms and checks are in place to ensure data quality.^17,18 

The maternal and infant linked file also included information on maternal age and postal code of maternal residence at the time of birth (maternal records), birth weight, infant sex, and gestational age (neonatal records). Maternal neighborhood at the time of birth was identified by the 6-digit maternal postal code, and the Statistics Canada Postal Code Conversion files were used to link neighborhood indices, including socioeconomic quintile and rural residential area.²² Rural residence was defined as communities with <10 000 inhabitants,²³ and low and high socioeconomic status (SES) were defined as maternal residences in the lowest and highest socioeconomic quintile neighborhoods, respectively, in relation to the median income in Canada. Congenital anomalies and maternal comorbidity were identified by ICD-10-CA codes (Supplemental Table 6).

NAS was identified by ICD-10-CA code P96.1: “Neonatal withdrawal symptoms from maternal use of drugs of addiction.” This code is different from the code P96.2 (“Withdrawal symptoms from therapeutic use of drugs in newborn”), which is used for neonatal withdrawal syndrome because of the therapeutic drugs used to treat neonatal conditions.

The primary adverse health outcomes were as follows: (1) maternal death, (2) a composite outcome of maternal mortality or severe morbidity, (3) neonatal death, and (4) a composite outcome of neonatal mortality or severe morbidity. The secondary outcomes included individual components of severe maternal and neonatal morbidity.

Maternal death was defined as death during delivery hospitalization. The primary health outcome of maternal mortality or severe morbidity included conditions that were potentially life-threatening, conditions leading to serious adverse sequelae, complications requiring prolonged hospitalization, and maternal death. These severe morbidities were identified by using a list of such conditions previously developed by the Canadian Perinatal Surveillance System²⁴ and included eclampsia, acute cardiac morbidity (eg, acute myocardial infarction), cardiomyopathy, shock, obstetric embolism, central nervous system morbidity (eg, cerebral venous thromboembolism), disseminated intravascular coagulation, adult respiratory distress syndrome (RDS), acute abdomen (eg, acute appendicitis), hepatic failure, renal failure, maternal sepsis, antepartum hemorrhage requiring blood transfusion (eg, placenta previa), severe postpartum hemorrhage requiring transfusion, status asthmaticus, status epilepticus, acute psychosis, uterine scar dehiscence or rupture, complications of anesthesia and obstetric intervention, and severe morbidity requiring potentially life-saving intervention (eg, blood transfusion; Supplemental Table 7).

Neonatal death was defined as death before hospital discharge. Severe neonatal morbidity included conditions that are potentially life-threatening or have potentially serious neurodevelopmental sequelae, including bronchopulmonary dysplasia, RDS, intracranial nontraumatic hemorrhage (eg, intraventricular hemorrhage [IVH]), periventricular leukomalacia, retinopathy of prematurity, neonatal sepsis, convulsions, necrotizing enterocolitis, and intestinal perforation (Supplemental Table 8).

Temporal trends in the incidence of NAS were assessed by the Cochran-Armitage test for trend. On the basis of yearly incidence, logistic regression was used to estimate increases in NAS between 2004–2005 and 2008–2009 (reference), 2009–2010 and 2011–2012 (period 1), and 2013–2014 and 2015–2016 (period 2). Interaction terms between each period and SES, rural and urban residence, and maternal age were tested in regression analyses to examine if temporal trends in NAS differed by these characteristics. Odds ratios (ORs) and 95% confidence intervals (CIs) were first adjusted for SES, rural versus urban residence, and age and then also for chronic hypertension, prepregnancy diabetes, male fetus, and parity (missing values were modeled as “missing”). Three-way interactions between age and SES, age and rural residence, and SES and rural residence were tested for additional effects on the increase in NAS in each calendar period. Our aim with the adjusted analyses was to assess temporal trends while controlling for the effect of temporal changes in other demographic characteristics potentially associated with NAS.

Associations between NAS and mortality and severe morbidity were expressed by using ORs and rate differences (RDs) and 95% CIs. Logistic regression (adjusted analyses) was used to assess whether the elevated risks in mother-infant dyads with NAS, if any, were due to differences in demographic and other prepregnancy characteristics associated with NAS. As such, these analyses were not intended to describe causal associations but rather to describe differences that were above and beyond differences expected as a result of other characteristics.

Covariates in the regression models included maternal age (15–24 and ≥35 years vs 25–34 years), prepregnancy hypertension, prepregnancy diabetes mellitus, year of childbirth, infant sex, hypertension in pregnancy, gestational diabetes, maternal residence (rural versus urban area), and residential SES (lowest and highest SES quintile versus quintiles 2–4). Neonatal outcomes were also adjusted for congenital anomalies and gestational age. Complete-case regression analyses were performed because missing values did not exceed 3% (Supplemental Table 9).

First, symptoms of opioid withdrawal typically appear shortly after birth, with the majority being recognized within the first 48 to 72 hours postdelivery.¹⁰ However, most neonatal deaths also occur shortly after birth, and hence some infants with in utero exposure to opioids may have died before being diagnosed with NAS. We therefore performed sensitivity analyses of neonatal mortality after excluding infants who died in the first 2 days after birth.

Second, information on previous live births (that approximates parity) was not systematically collected in some provinces in Canada; therefore, we examined the effect of parity in sensitivity analyses (excluding 21% of the study population). Logistic regression was used to adjust for parity (primiparous, parity ≥4 vs parity 2–3) in addition to previously listed covariates.

All analyses were conducted by using SAS 9.4 (SAS Institute Inc, Cary, NC). The study was approved by the University of British Columbia and British Columbia Women’s Hospital and Health Center Research Ethics Boards (CW17-0090/H17-00240).

There were 3 040 249 women who delivered a live singleton infant in Canada between April 1, 2005 and March 31, 2016. The study comprised 2 898 883 live singleton birth mother-infant dyads after the exclusion of births before 20 weeks, after 44 weeks, or of unknown gestation; births to mothers aged <15 years; and records that could not be matched between mother and infant (in total, 141 348 records were excluded; 4.65%).

The study included 10 027 mother-infant dyads affected by NAS, and the overall rate of NAS was 3.46 per 1000 live births. Mothers of infants with NAS were more likely to be younger, multiparous, residing in rural areas, have low residential SES, and have prepregnancy diabetes mellitus and anemia (Table 1). They were more likely to have positive HIV test results and less likely to have gestational diabetes. Infants with NAS were more likely to be born preterm (<37 weeks; 20.42% vs 6.33%; P < .001), have lower birth weight, and have higher rates of congenital anomalies as compared with infants without NAS. After adjustment for mutual confounding, significant associations were found between NAS and young maternal age, low SES, prepregnancy diabetes, infant male sex, rural residence, and grand multiparity (≥4 children; Table 2).

The rate of NAS increased from 2.01 per 1000 live births in 2005–2006 to 5.12 per 1000 live births in 2015–2016 (rate ratio 2.55; 95% CI: 2.30 to 2.83); the increase occurred in all demographic groups (all P values for trend <.001; Fig 1).

Adjusted for other covariates, the largest temporal increase in NAS was among mothers with low residential SES (Table 3). This fourfold increase between 2004–2005 to 2008–2009 and 2012–2013 to 2015–2016 was significantly larger than increases among mothers with average or high SES. A significantly larger increase was also seen among rural and younger mothers. Additional effects on the increase in NAS for any combination of age, SES, and rural residence (3-way interactions) were not statistically significant.

The rate of maternal death was 1.99 vs 0.31 per 10 000 women in the NAS group versus the comparison group (adjusted odds ratio [aOR] = 6.53; 95% CI: 1.59 to 26.74; adjusted rate difference [ARD]: 1.71 per 10 000; 95% CI: 0.18 to 7.98 per 10 000; Table 3), and most severe maternal morbidities were also significantly elevated in this group. The rate of composite mortality or severe morbidity was 310.2 vs 134.7 per 10 000 mothers of infants with and without NAS (aOR = 2.21; 95% CI: 1.97 to 2.49; ARD = 163.05 per 10 000; 95% CI: 130.71 to 200.78).

The rate of neonatal death was 1.20 per 1000 infants with NAS as compared with 1.89 per 1000 infants without NAS (aOR = 0.28; 95% CI: 0.15 to 0.53; ARD = −1.36 per 1000; 95% CI: −1.61 to −0.89 per 1000; Table 4). The rate of mortality or severe morbidity was 63.63 per 1000 infants with NAS and 17.29 per 1000 infants without NAS (aOR = 2.27; 95% CI: 2.06 to 2.50; ARD = 21.96 per 1000; 95% CI: 18.33 to 25.94 per 1000). NAS was associated with a sixfold risk of periventricular leukomalacia (aOR = 6.0; 95% CI: 3.1 to 11.6; Table 5).

First, mortality rates restricted to infants who survived 2 days after birth were higher among infants with NAS than those without NAS (0.70 vs 0.38 per 1000; OR = 1.84; P = .112). Although not statistically significant, this association is reversed from the inverse association that was observed in main analyses of all live-born infants (OR = 0.64), suggesting that infants that were exposed to opioids in utero may have died in the first 2 days (before diagnosis of NAS) at a higher rate than nonexposed infants.

Second, the results were not largely different when parity was added to the regression models for adverse maternal and infant outcomes (Supplemental Tables 10 and 11). The associations between NAS and acute psychosis and hysterectomy were no longer statistically significant.

To our knowledge, this population-based study of NAS included the largest number of mother-infant dyads to date and provided description and quantification of severe adverse maternal and neonatal outcomes associated with NAS. The rate of NAS increased in Canada (excluding Quebec) from 2.01 in 2005–2006 to 5.12 per 1000 singleton live births in 2015–2016, and the increase was larger among infants of mothers with low SES, rural residence, and young age. NAS was associated with a significantly elevated risk of severe neonatal morbidity, maternal death, and severe maternal morbidity.

Similar increases in NAS were reported in the United States from 1.20 in 2000 to 3.39 per 1000 hospital births in 2012,⁶ in England from 1.0 in 1997 to 2.7 per 1000 live births in 2011,⁴ and in Western Australia from 0.01 in 1990 to 2.7 per 1000 live births in 2008.²⁵ In the United States, incidence of NAS among infants of Medicaid-insured mothers was 8.0 per 1000 hospital births in 2014.²⁶ In Ontario, Canada, the increase was from 0.3 in 1992 to 4.3 per 1000 live births in 2011,²⁷ and a more recent study in Canada (excluding Quebec) revealed an increase from 1.8 to 5.4 per 1000 live births between 2003 and 2014.³ The latter study was focused solely on unadjusted temporal and provincial comparisons and costs associated with NAS treatment; our study provides new details on rates of mortality and severe maternal and neonatal morbidity.

Although the demographic profile of mother-infant dyads in our study is consistent with the literature,^1,28,–30 our findings of elevated prevalence of prepregnancy diabetes and grand multiparity among women with infants affected by NAS is new. These associations may be due to unmeasured confounding, for example, by ethnicity.^25,30 It is also possible, however, that chronically elevated glucose levels and associated obesity^31,32 may contribute to painful chronic conditions that can lead to overuse of prescribed opioids during pregnancy and NAS. However, these hypotheses require additional research. The lower rate of gestational diabetes among mothers of infants with NAS was potentially due to fewer prenatal visits and opportunities for screening in this population.³⁰ The strong association between NAS and periventricular leukomalacia is a new finding that warrants additional research to exclude the possibility of a false-positive result due to multiple comparisons.

Infants with NAS had better prognosis with respect to neonatal mortality as compared with infants without NAS. One possible explanation for this paradox is that infants who would have otherwise shown withdrawal symptoms died soon after birth before the symptoms of NAS manifested. Our sensitivity analyses suggest that this “survival bias” is likely responsible for lower mortality among infants with NAS. Another possible explanation is that NAS symptoms may not be well recognized and therefore not diagnosed to a larger extent in preterm infants who are severely ill and die before discharge. This differential misclassification of NAS (ie, infants are less likely “classified” as having NAS among those with severe morbidity) can lead to ascertainment bias and contribute to inverse association between NAS and neonatal mortality.

The strengths of this study include large population data consistently collected and periodically checked for accuracy. We were able to discern between NAS resulting from a long-term prenatal exposure and an abstinence syndrome resulting from a pharmacologic treatment of the neonate. This distinction is a significant advantage over previous studies using only 1 International Classification of Diseases, 9th Revision code for “drug withdrawal syndrome in a newborn.”^1,6 A large validation study in the United States revealed hospital administrative data suitable to ascertain NAS, especially when using International Classification of Diseases, 10th Revision (ICD-10) codes (positive predictive value: 98.2%).³³ 

Although the implicit causal exposure is prolonged opioid exposure during pregnancy, this study was not designed to examine causal associations between opioid use in pregnancy and adverse outcomes but rather to describe demographic and clinical factors among infants with NAS and to quantify mortality and severe morbidity in the mother-infant dyads. The study has several limitations. First, we compared adverse health outcomes in mother-infant dyads with and without NAS diagnosis. Such diagnosis depends on clinically recognizable abstinence signs that appear in 60% to 80% of opioid-exposed neonates, depending on total fetal exposure, the amount and purity of the drugs taken by the mother, and the individual kinetics of metabolism and placental drug transfer.^2,34 Second, substances other than opioids (for instance, alcohol or antidepressants) can result in withdrawal symptoms similar to NAS and may be diagnosed as such.⁸ Previous studies have revealed that the proportion of NAS resulting from these exposures is relatively small.³⁵ Third, we did not have information on illicit versus nonillicit opioid use and on opioid-substitution treatment. NAS resulting from methadone or buprenorphine treatment exposure is likely to be associated with lower rates of severe morbidity because the treatment requires periodical visits at a treatment center, which includes medical checks and pregnancy monitoring.³⁶ Fourth, increased rates of opioid use and an escalating number of deaths due to opioid overdose have been widely reported in the media in Canada since 2008, and therefore the diagnostic suspicion for NAS has likely increased over time and influenced temporal trends. Fifth, coding errors and omissions could occur in the hospitalization database, which may have led to an underestimation of the rates of severe morbidity and nondifferential misclassification (ie, bias toward null) in the association between NAS and mortality and severe morbidity. Authors of a previous validation study of CIHI data found that sensitivity for postpartum hemorrhage was 90%, for preterm delivery 91%, for chronic hypertension 83.3%, and for IVH (grade 3 and 4) 90%, whereas specificity values were all >98%.²⁰ This study also revealed that severe conditions were reported more accurately; for instance, the International Classification of Diseases, 9th Revision code for RDS had sensitivity 51% for all RDS but much higher sensitivity for severe RDS (96%).²⁰ Hence, the rates of RDS in our study were likely underestimated for overall RDS but relatively accurate for severe RDS. Seventh, information on medication use, alcohol use, smoking, BMI, and ethnicity (including Aboriginal status) was not available. Higher rates of NAS in rural areas may reflect differences in NAS incidence due to its association with ethnicity or differences in accessibility of health care services.³⁰ Lastly, pregnancies that ended in stillbirth were not included in our study. This potentially leads to lower neonatal mortality among women with prolonged opioid use during pregnancy because their compromised fetuses (eg, those with intrauterine growth restriction) may die while still in utero at a higher rate as compared with other mothers.² 

The incidence of NAS more than doubled in Canada between 2005–2006 and 2015–2016, following similar trends observed in the United States, England, and Australia. Elevated risks of maternal mortality and maternal and neonatal severe morbidity have implications for postpartum care. Treatment of the mother-infant dyad should be the goal because optimal health care provision during this crucial period contributes to harm reduction and prevention of adverse sequelae. Severe morbidity occurring in infants with NAS and their mothers can hinder rooming-in, resulting in a need for additional strategies to encourage mother-infant bonding and breastfeeding. Despite these challenges, pregnancy and the postpartum period present a unique opportunity for compassionate multidisciplinary care to improve the health and well-being of these vulnerable mothers and their infants.

aOR

adjusted odds ratio

ARD

adjusted rate difference

CCI

Canadian Classification of Interventions

CI

confidence interval

CIHI

Canadian Institute for Health Information

ICD-10

International Classification of Diseases, 10th Revision

ICD-10-CA

International Classification of Diseases, 10th Revision, Canada

IVH

intraventricular hemorrhage

NAS

neonatal abstinence syndrome

OR

odds ratio

RD

rate difference

RDS

respiratory distress syndrome

SES

socioeconomic status