Identification of Prenatal Opioid Exposure Within Health Administrative Databases

In this population-based cross-sectional study, we used linked health and demographic administrative data obtained from ICES. ICES is an independent, nonprofit research institute whose legal status under privacy law allows it to collect and analyze health and demographic data, without consent, for health system evaluation. Data sets were linked by using unique encoded identifiers. Study data sets are detailed in Supplemental Table 6 and described elsewhere.^23–26  Ethics approval was granted from the University of Toronto and Hospital for Sick Children research ethics boards. We followed Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) ²⁷  and Reporting of Studies Conducted Using Observational Routinely-Collected Data (RECORD)²⁸  reporting guidelines.

We included all hospitalizations for a livebirth or stillbirth among women aged 12 to 50 years, in Ontario, Canada, in 2014–2017. Ontario has an estimated population of 14 million²⁹  and ∼140 000 births annually, 97% of which occur in a hospital.³⁰  Women were included who had continuous eligibility for Ontario’s public health insurance during pregnancy and <100 days of ineligibility in the 2 years preconception.³¹  Ineligibility could arise from relocation or recent immigration to Ontario. Included were stillbirths starting at 20 weeks' gestation³²  and livebirths between 23 and 42 weeks' gestation.³³  To restrict our cohort to clinically meaningful opioid use, we excluded women with 1 to 10 cumulative days of opioid analgesic use during pregnancy and those with opioids prescribed only for cough. The 10-day threshold corresponded to the 75th percentile of the distribution of opioid analgesic use in pregnancy among the study cohort and was made by expert opinion through consensus among a pediatrician (A.G.), high-risk pregnancy internist (J.G.R.), and expert in pharmaceutical use (T.G.). Infants with iatrogenic NAS were identified by using International Classification of Diseases, 10th Revision (ICD-10) P96.2 (withdrawal symptoms from therapeutic use of drugs in newborn), in addition to previously published methods,³⁴  and excluded.

A comprehensive measure of POE was developed by using 3 ascertainment methods including prenatal prescription opioid data, maternal opioid-related hospital records, and newborn hospital records with NAS (Supplemental Table 6). First, prenatal prescription opioid use during pregnancy was identified and categorized as opioid analgesics for pain and opioid agonist therapy (OAT) for opioid use disorder (OUD). Second, we identified maternal opioid-related hospital records (hospitalizations and emergency department [ED] visits) during pregnancy using ICD-10 diagnostic codes for mental and behavioral disorders due to opioid use, opioid poisoning, or opioid-related adverse drug reactions. Last, as in previous work, POE was identified among neonates in the newborn birth episode or infant hospital readmission during the first 14 days of life by using ICD-10 diagnostic codes for NAS P96.1 (withdrawal from maternal drugs of addiction) or P04.4 (infant affected by maternal drugs of addiction).^16,35  The use and validity of ICD-10 codes to identify NAS was previously described.¹⁶ 

To compare and contextualize the measure of POE developed herein, we replicated existing measures of maternal opioid use and NAS, as described in studies with administrative data (Supplemental Table 5). Existing measures included types of prenatal prescription opioid use, OUD ascertained from maternal hospital records, and NAS from newborn hospital records.

Additional variables characterized women by ascertainment method, many used and validated elsewhere (Supplemental Table 6). Socio-demographic variables were maternal age at first³⁶  and current birth, area-level low income (quintile 1),³⁷  rural residence (<10 000 inhabitants), immigrant status, and previous livebirths. Indicators of maternal social risk were receiving medical care while in the criminal justice system, health care related to violence or homelessness, receipt of social assistance for prescriptions, and newborn apprehension at birth hospitalization.³⁸  Maternal substance use included hospital care related to tobacco, alcohol, opioid, or nonopioid and/or multidrug use. Mental health and addiction-related hospital care included any ED visit or hospitalization.³⁹  Preconception morbidity included chronic pain conditions (low back, joint, migraine or headache, rheumatoid arthritis,⁴⁰  fibromyalgia, joint pain, chronic pancreatitis, peripheral neuropathy, sickle cell disease and renal calculus),^11,41  and common chronic conditions from validated registries (diabetes,^42,43  hypertension,^43,44  asthma⁴⁵ ). Johns Hopkins ACG System Aggregated Diagnosis Groups (v10) scores of 10+ were used to describe significant medical comorbidities. Measures of appropriate access to care included ultrasound by 20 weeks³⁸  and first trimester prenatal care by a physician. We describe common delivery and newborn characteristics to allow comparisons with other jurisdictions.

We calculated positive percent agreement (PPA), analogous to sensitivity in the absence of a gold standard, and 95% confidence intervals (CIs), to measure agreement between conventionally used definitions of POE and the comprehensive method for identifying POE developed herein, using all 3 linked data sources combined. At study outset, it was understood the latter would serve as an imperfect reference standard.

Baseline characteristics by POE were described by using means or frequencies and compared by using standardized differences, where a difference >0.10 is considered clinically meaningful.⁴⁶ 

We then replicated and contrasted conventional definitions of different types of maternal opioid use and NAS with our comprehensive measure of POE and estimated the additional cases of POE identified by our approach (Supplemental Table 5). To be consistent with other definitions of prescription opioid use, we removed our previous exclusion of prenatal exposure to 1 to 10 days of opioid analgesics.^11,47,48  Analyses were performed by using SAS software, version 9.4 (SAS Institute Inc, Cary, NC).

There were 471 798 Ontario hospital livebirths or stillbirths between 2014 and 2017. We excluded 17 004 births with prenatal exposure of 1 to 10 days of opioid analgesics only, or opioids for cough only (3.6%) (Supplemental Fig 2). We identified 115 newborns with potential iatrogenic NAS. Using linked hospital records and prescription opioid data, we found evidence of POE in 67 births identified with iatrogenic NAS. Forty-eight newborns with no history of maternal opioid use who met the criteria for iatrogenic NAS were excluded. The final cohort comprised 454 746 births: 9624 with, and 445 122 without, POE.

Compared with mothers without POE, those with exposure were more likely to be disadvantaged, with notable disparities in demographic and social factors, hospital care for mental illness and medical morbidities, and neonatal health outcomes (Table 1).

Using all 3 linked data sources (prenatal prescription opioid data, maternal opioid-related hospital records, and newborn hospital records with NAS), we identified 9624 births with POE (21.2 per 1000 births). Compared with ascertainment of POE by using all 3 data sources, PPA was highest with prenatal prescription opioid data alone (79.0%, 95% CI: 78.2–79.8), followed by newborn hospital records with NAS alone (44.7%, 95% CI: 43.7–45.7) and maternal opioid-related hospital records alone (19.0%, 95% CI: 18.2–19.8) (Fig 1).

Ascertainment of POE varied by data linkage of mothers and newborns and across health records. For example, maternal opioid-related hospital records during pregnancy were used to identify 1825 births with POE (4.0 per 1000). Newborn hospital records with NAS increased the estimate of POE to 5024 births (11.0 per 1000). Further addition of prenatal opioid prescriptions notably increased case ascertainment to 9624 births (21.2 per 1000).

Table 2 further illustrates how POE estimates vary by method and provides additional numbers of cases identified by using our comprehensive method. The largest difference was with maternal opioid-related hospital records alone. Compared with opioid-related hospital care captured at the birth hospitalization (1584 births [3.5 per 1000]), our comprehensive method identified an additional 8040 cases (Table 2). The smallest gain was in comparison with prenatal prescription opioid data alone (7605 births [16.7 per 1000]) wherein the comprehensive method identified an additional 2019 births.

Table 3 describes measurement implications of more restrictive measures of maternal opioid use and NAS, in previous studies, compared with a more comprehensive measure of POE. After replicating existing definitions with data from Ontario, estimates of maternal opioid use ranged from 3.4 to 47.1 per 1000 births. We identified an additional 1811 to 8079 cases using our method. Studies designed to capture maternal OUD from maternal hospital records identified the fewest cases. Variations in OUD definitions by ICD code (OUD and poisoning) and timing (pregnancy and delivery) had a minor impact on ascertainment (Table 3).

Mothers with POE ascertained through maternal and newborn opioid-related hospital records generally displayed higher levels of disadvantage than those identified through prenatal prescription opioid data (Table 4). Both groups had notably higher rates of maternal polysubstance use, mental health care, and markers of social instability. Mothers identified through maternal opioid-related hospital records were more likely to reside in rural and low-income areas, have a higher parity, and have a history of mental health hospital care compared with those ascertained through newborn hospital records with NAS and prenatal prescription opioid data. Prescription opioids were dispensed to 68.8% of mothers identified through maternal opioid-related hospital records and 61.5% of mothers identified through newborn hospital records with NAS, and OAT use was more common in both groups compared with those ascertained through prenatal prescription opioid data.

Prescription opioid exposure was notably higher among newborns diagnosed with ICD-10 P96.1 (withdrawal from maternal drugs of addiction) compared with P04.4 (infant affected by maternal drugs of addiction) (69.0% vs 41.7%, respectively) (Supplemental Table 7). Mothers of both had similar levels of socio-demographic disadvantage. Newborn characteristics between groups were largely similar, except for neonatal intensive care and hospital length of stay (Supplemental Table 7).

With this large population-based study, we add new and important information regarding the ascertainment of births with POE using health administrative data for public health surveillance and research. Use of 3 linked data sources (prenatal prescription opioid data, maternal opioid-related hospital records, and newborn hospital records with NAS) offered the most comprehensive approach to estimating POE (21.2 per 1000 births). When using different methods, estimates of POE ranged from 4.0 per 1000 births from maternal opioid-related hospital records alone, 9.5 per 1000 births from newborn hospital records with NAS alone, and 16.7 per 1000 births from prenatal prescription opioid data alone. Previously reported measures of prevalence of maternal opioid use and NAS ranged from 3.4 to 47.1 per 1000 births. Using a comprehensive approach to identify POE, we ascertained an additional 1811 to 8079 cases. Case ascertainment through maternal and newborn opioid-related hospital records identified high-risk mothers, indicating different ascertainment methods yield different study populations.

We compared differences in POE estimation by replicating existing methods of ascertaining types of maternal opioid use and NAS using our data. Although researchers in previous studies used more restrictive measures of POE, through choice or necessity, these comparisons are useful to contextualize differences in POE measures to understand if various cohorts represent different groups of women and describe differences in case capture. Our method was most similar to a cohort study of mother-infant dyads enrolled in Medicaid in Tennessee.¹¹  In this sample, Patrick et al¹¹  reported 279.9 per 1000 pregnant women filled 1 or more opioid prescriptions. Using Patrick et al’s¹¹  method in Ontario, we estimated 22 103 births with POE (47.1 per 1000 births) compared with our method, which identified 23 914 births with POE (51.0 per 1000 births). Congruency between methods was due to the inclusion of prescription opioid analgesics and OAT, and the majority of cases derived from maternal prescription opioid records. Sole reliance on OUD codes in maternal hospital records resulted in a smaller cohort of higher-risk mothers. Haight et al³  presented the prevalence of maternal OUD at 6.5 per 1000 delivery hospitalizations in the United States. Using this method in Ontario, we estimated 1545 births (3.4 per 1000 births) compared with our estimate of 9624 births with POE (21.2 per 1000 births). We have shown a minority of cases are derived from maternal hospital records. In our cohort, case ascertainment based on maternal OUD in hospital records resulted in an underestimation of OUD and inclusion of a higher-risk study population. We identified mothers with OUD in 18% of births through maternal hospital records alone, compared with 41% using both maternal hospital records and prescription data. Finally, we have shown when NAS (ICD-10 P96.1) is used as a proxy for POE, POE is underestimated by 6385 births. Efforts were made to closely replicate definitions; however, there were not always corresponding conversion codes from International Classification of Diseases, Ninth Revision (ICD-9) to ICD-10, which may have resulted in minor differences in case ascertainment.

Findings have additional implications for defining NAS in surveillance and research. Our results suggest almost half of newborns with ICD-10 P04.4 are exposed to opioids, with notable prenatal polysubstance use, common among women with opioid dependence.^51,52  Newborns with P96.1 or P04.4, regardless of withdrawal severity, are important to include in studies examining long-term effects of POE. Results suggest methods to exclude iatrogenic cases of NAS may be excluding births with POE,³⁴  because 58% of potentially iatrogenic cases of NAS had evidence of POE.

This is the first study to explore a range of maternal demographic, social, and medical factors by data source. Using administrative data, we identified important differences in maternal characteristics by data source suggesting selection bias may occur in the design of observational studies. Births with POE ascertained through maternal and newborn opioid-related hospital records had higher-risk profiles versus those with prenatal prescription opioid data. This was evident for demographics, social risk factors, and mental health and addiction care. We did not differentiate between exposure to prescription opioid analgesics and OAT. There is inevitably within group heterogeneity as maternal characteristics differ by type of opioid. Group differences will be explored in subsequent work designed to develop maternal phenotypes by type of opioid(s) used during pregnancy. The decision to use a comprehensive measure of POE, as we developed, over more-specific measures of opioid type, OUD or NAS, will depend on the construct of interest. Findings are meant to aid researchers in their selection of measures and data sources while providing context to assess the potential for bias.

This descriptive study has several limitations. Information reported is limited to women who sought health care for themselves or newborns. Some cases may have been missed and constructs may have been underestimated, notably polysubstance use and social risk factors. Methods used represent progress in capturing the true universe of POE; however, misclassification may have been introduced in our approach. POE may be overestimated through inclusion of cases with NAS from intrauterine exposure to nonopioid substances.⁵³  Cases may have been missed if illicit opioid use during pregnancy went unnoticed by health care practitioners or newborns were not in clinical withdrawal, resulting in underestimation of prevalence. The true universe of women who use opioids during pregnancy is not known. Thus, we could not evaluate our approach against a gold standard. Furthermore, because all conventional measures of POE were included in our overall definition of POE, we could not calculate negative percent agreement. We took an as-prescribed approach assuming women took the full course of medications prescribed. This may result in overestimation of POE; however, women with 1 to 10 days of opioid analgesic use were excluded, which mitigates this impact. Dose and timing were not considered in the exclusion of women with 1 to 10 days of opioid analgesic use; however, most women had 1 to 5 days of opioid analgesic use (60%), which is likely not clinically meaningful POE. Nevertheless, results from this study are widely generalizable. Findings are population-based and reflect women with hospital births who were eligible for Ontario’s universal health care system. Definitions in this study are easily replicable and can be applied to other settings with health administrative data.

This study demonstrates the importance of multiple linked health administrative databases to estimate the prevalence of POE and facilitate research of maternal and child health outcomes. Differences in data availability and linkage impact the number of cases and study population ascertained. Samples derived from different ascertainment methods may differ by important characteristics. Jurisdictions and public health researchers should be aware of potential biases and underestimation related to ascertainment method.

Parts of this material are based on data and information compiled and provided by MOH, Canadian Institute for Health Information, IMS Brogan Inc, and Immigration, Refugees and Citizenship Canada. The analyses, conclusions, opinions, and statements expressed herein are solely those of the authors and do not reflect those of the funding or data sources; no endorsement is intended or should be inferred. We thank IMS Brogan Inc for use of their Drug Information Database. We thank Alison L Park, MSc (ICES) and Javaid Iqbal, MD (ICES) for their technical help with this study and Kinwah Fung, MSc (ICES) for her methodological guidance.

CI

confidence interval

ED

emergency department

ICD

International Classification of Diseases

ICD-10

International Classification of Diseases, 10th Revision

ICD-9

International Classification of Diseases, Ninth Revision

NAS

neonatal abstinence syndrome

OAT

opioid agonist therapy

OUD

opioid use disorder

POE

prenatal opioid exposure

PPA

positive percent agreement