Medication for Opioid Use Disorder During Pregnancy — Maternal and Infant Network to Understand Outcomes Associated with Use of Medication for Opioid Use Disorder During Pregnancy (MAT-LINK), 2014–2021

Methods

CDC, in collaboration with the Public Health Informatics Institute (PHII) and with funding from the Assistant Secretary for Planning and Evaluation (ASPE)’s Office of the Secretary Patient-Centered Outcomes Research Trust Fund (OS-PCORTF), established the Maternal and Infant Network to Understand Outcomes Associated with Medication for Opioid Use Disorder During Pregnancy (MAT-LINK) in 2019 (Figure 1). MAT-LINK is a surveillance network of clinical sites that collect data on persons with OUD during pregnancy. In this report, the term “maternal” is used to identify the person who is pregnant or postpartum. Pregnancy is not equated with the decision to parent, nor do all parents who give birth identify as mothers.

CDC is responsible for day-to-day project management and surveillance network design. CDC also leads data collection and cleaning, develops analytic protocols and guidance documents, and monitors project performance and spending. PHII, the implementation partner, subcontracts with the clinical sites and manages the surveillance infrastructure according to relevant federal regulations under CDC’s direction. PHII also ensures the integrity and safeguarding of data and provides technical assistance to the clinical sites.

The MAT-LINK partners group comprises federal, clinical, and public health partners. Federal partners provide subject matter expertise and individual input into the implementation of MAT-LINK. The clinical and public health partners consult with the CDC project team to share perspectives from their relevant constituencies. A CDC steering committee composed of leadership and subject matter experts from multiple CDC centers provides guidance and oversight of MAT-LINK.

Participating Clinical Sites

MAT-LINK was established in 2019 and initially comprised four clinical sites: Boston Medical Center, Kaiser Permanente Northwest, The Ohio State University, and the University of Utah. These clinical sites were selected because they had an advanced and robust data infrastructure, clinical care protocols that included multiple MOUD regimens, and the capacity to capture postpartum and childhood outcome data up to age 2 years. They also were required to demonstrate their ability to integrate or link maternal and child data (i.e., dyads) as well as their authority to access and share these data with CDC. In 2021, with more funding for expansion from ASPE’s OS-PCORTF, three additional clinical sites were selected: the University of New Mexico, the University of Rochester, and the University of South Florida. This expansion of MAT-LINK also included the collection of childhood data through age 6 years to enable assessment of neurodevelopmental outcomes and outcomes for school-age children.

Case Ascertainment

The inclusion criteria for MAT-LINK were all known pregnancy outcomes from January 1, 2014, through August 31, 2021, and an International Classification of Diseases, Ninth or Tenth Revision, Clinical Modification (ICD-CM) code for OUD diagnosis during that pregnancy (Table 1). Clinical site staff members worked with their principal investigators and clinical care teams to identify all eligible pregnancies. Each clinical site then assigned a unique identifier to each dyad of linked maternal and child electronic health record (EHR) data before submitting data. If a duplicate identifier was generated, clinical site staff members informed CDC immediately, deleted any previously submitted duplicative data from the data repositories, and excluded the duplicated dyad identifier from any future data submissions. CDC does not receive personally identifiable data (e.g., names or addresses) but does receive potentially identifiable and sensitive data (e.g., birth dates and hospital admission dates). However, the unique dyad identifiers do not include personal, institutional, or geographic information.

Data Sources, Collection, and Processing

Data for MAT-LINK are collected by clinical sites from existing health services and medical records including EHRs, pharmaceutical management systems, laboratory records, public health reports, and state surveillance data. External sources for additional data might include MOUD-related visits from an outside clinic or administrative data sources. Health data at clinical sites are collected in accordance with state and local policies and procedures. Six clinical sites use Epic software and one clinical site uses Cerner software as their EHR system. Clinical sites have variations in how data are abstracted (i.e., manually retrieved from various chart sources), extracted (i.e., queried from relational data tables within an EHR), and reported (e.g., the use of custom data warehouses). Each clinical site has a unique approach to using and transforming its EHR data, including which fields are populated and how data are organized. Site-specific variations include which data are stored in structured format and which data are in text form (e.g., clinical notes or reports). Data were obtained by abstraction and extraction (Figure 2). Clinical sites used the data dictionary CDC provided to determine which variables would need to be abstracted or extracted at their individual clinical sites. A preset algorithm for data extraction was not used because of differences between sites in how data are entered and stored. In addition, a comparison group of pregnant persons not taking MOUD was sampled.

Information Technology Infrastructure

The information technology (IT) infrastructure for MAT-LINK was synchronously developed by CDC and PHII. The initial four clinical sites piloted the infrastructure using a collaborative approach for abstracted and extracted data. The main components included Microsoft SharePoint for project collaboration and support; Research Electronic Data Capture (REDCap) for entry of abstracted data; PHII-developed XML schemas for various data concepts; a MAT-LINK broker application to perform the extraction, transfer, and loading of abstracted and extracted data into a secure SQL server database; and a data extractor for generating and validating the schema-compliant XML files (Supplementary Figure, https://stacks.cdc.gov/view/cdc/126230) (7,8).

SharePoint is a web-based platform for sharing files and documents that allows the clinical sites to collaborate on data collection and receive support from the project team. CDC developed custom forms and workflows in SharePoint so users could submit questions confidentially with automated business processes to facilitate communication and collaboration.

Data entry and extraction are standardized by using REDCap instruments and the data extractor, respectively. A MAT-LINK–specific REDCap project, which contains approximately 1,000 variables, was developed for entry of data abstracted through medical chart review. A duplicate copy of the REDCap project is implemented at all MAT-LINK clinical sites. The data extractor is a customized tool with 28 XML schemas that support a standardized format for extracted data from data warehouses, EHR systems, and other sources. It includes data sets (e.g., ICD-CM codes and laboratory data) that can be automated for extraction from health care data systems. To reduce the need for previous XML knowledge at a clinical site, the data extractor package includes multiple support documents, such as a data dictionary, standard CSV templates for compiling the data, and validation scripts to generate XML files.

Data Quality and Sharing

After data abstraction and extraction, the clinical sites conduct both automated and manual checks to review the data for quality. These steps include internal quality control checks that perform dual entry verification for 10% of abstracted dyads and compare the data with what is expected from clinical experiences. The clinical sites also run their extracted data through a schema-specific validator in the data extractor to check for errors before submission. Any errors in the extracted EHR data are spot checked and rectified using chart reviews. Finalized data are shared with CDC via Secure Data Exchange, a secure file transfer platform. Data received are automatically subjected to a series of validation processes by the MAT-LINK broker before being deposited into a raw data server as a relational database. Next, CDC conducts a series of automated data quality checks. If no issues are detected, the data are imported into the production database. If the data set does not pass the data quality checks, the data team contacts the clinical site for clarification or correction of the data. Finally, CDC analysts conduct a manual review of the data to look for anomalies.

The IT architecture was designed to expand as the project needs develop. Additional topic areas and dyads can be added to the MAT-LINK REDCap project and data extractor. XML schemas in the data extractor that can be used by non-MOUD projects include demographics, procedures and diagnoses, medication schemas, and child follow-up schemas.

After data are collected and organized, each clinical site receives a final version of its own data (i.e., without data from other clinical sites). MAT-LINK partners and CDC are collaborating to analyze and report findings from the pooled data combined across clinical sites. Site-specific data will not be reported unless permission is provided by the clinical site.

A partial data set that is consistent with all institutional policies and the Assurance of Confidentiality will be available for analysis by external investigators after submission of a proposal to and approval from the National Center for Health Statistics’ Research Data Center. Guidance for external researchers interested in requesting access to a MAT-LINK data set will be available at https://www.cdc.gov/ncbddd/aboutus/mat-link.html.

Data Protection

Data provided to CDC within the MAT-LINK network are protected under an Assurance of Confidentiality (9). This activity was reviewed by CDC and was conducted consistent with applicable federal law and CDC policy.* Each clinical site received approval or exemption from the respective institutional review board before beginning data collection.

Exposure Definitions and Comparison Group

For this analysis, MOUD included buprenorphine with or without naloxone, methadone, and naltrexone. Buprenorphine-based medications specifically approved for management of chronic pain were not included as MOUD; however, they were included with other medications as a potential co-exposure. All persons with a diagnosis of OUD during pregnancy who received MOUD in the pregnancy during the surveillance period were included in the MOUD group for analysis, including persons in remission or with a history of OUD before the pregnancy. Each person could have more than one pregnancy included in MAT-LINK and these pregnancies could be included in the same or different MOUD groups. Remission is designated by ICD-CM codes and includes persons with a previous diagnosis of OUD currently taking MOUD.

To provide a comparison group representative of pregnant persons with OUD who did not receive MOUD, clinical sites were asked to include extracted data on all pregnant persons with an active diagnosis of OUD during pregnancy who did not receive MOUD (the non-MOUD group). Clinical sites with >60 pregnancies among persons not taking MOUD were asked to provide a simple random sample of complete data (abstracted and extracted) on 60 dyads over the surveillance period.

Description of Variables

MAT-LINK includes longitudinal data for the person who is pregnant from entry into prenatal care through 1 year postpartum and procedure and diagnosis codes (e.g., ICD-CM and Current Procedural Terminology [CPT] codes) up to 6 years postpartum. Longitudinal data for the child are collected from EHRs from birth through age 6 years. Pregnant person and child data are linked by the clinical sites. Determining the variables to collect was an iterative process that included literature reviews and input from subject matter experts. The ability to answer key analytic questions, ease of availability from existing data sources, and reasonable standardization across clinical sites were considered. The initial clinical sites provided a minimum of 25 records for all variables as pilot data; these were examined for data structure, quality, and compliance with data requirements. The additional clinical sites provided pilot data on key variables that were identified as difficult to ascertain from piloting efforts at the initial clinical sites. After pilot data were reviewed, all clinical sites provided final data.

Maternal History

Maternal history includes demographic and pregnancy-related data (e.g., encounter information, medications, laboratory data, ultrasounds, substance exposure laboratory results, infectious disease testing, procedures, and diagnoses). Race and ethnicity were categorized and reported based on standards specified by CDC’s Office of Management and Budget. Race categories are American Indian or Alaska Native (AI/AN), Asian, Native Hawaiian or other Pacific Islander (NH/OPI), Black or African American (Black), White, and other races; ethnicity categories are Hispanic or Latino (Hispanic) and not Hispanic or Latino (non-Hispanic). Asian and NH/OPI racial groups were consolidated to meet patient confidentiality requirements. Race and ethnicity are not mutually exclusive, and persons are accounted for in both categories. All multiracial and other race persons have been grouped under “other” races. Covariates, including psychosocial support, behavioral therapy, and peer support program participation as documented in the EHR, also are collected.

MOUD

Data collected on MOUD can be compared by timepoints during pregnancy. Information about how each clinical site provides MOUD to its patients will be used to understand possible differences between MOUD treatment groups. All clinical sites have data about MOUD integrated into their data collection system, regardless of whether the clinical site is the prescriber of the MOUD.

MOUD initiation, duration, and dosing patterns are included as well as other medications, including known teratogenic medications and medications associated with drug–drug interactions with MOUD. Data about inpatient or residential stays, substance use, and overdose events also are collected.

Delivery Birth Hospitalization

Variables related to pregnancy and delivery outcomes include pain management during and after labor; delivery type; newborn measurements; newborn care; laboratory data about infections and substance exposure; neonatal abstinence syndrome, neonatal opioid withdrawal syndrome, or both; and timing and other data on discharge, readmissions, and emergency department visits. Variables for pregnancy outcomes include live birth, intrauterine fetal death (≥20 weeks’ gestation), spontaneous abortion (<20 weeks’ gestation), termination, ectopic pregnancy, and other unspecified non-live births.

Child Follow-Up

Data are collected to examine short- and long-term outcomes for children, including physical growth and development, diagnoses of acute or chronic conditions, health care use, vaccinations, neurodevelopmental outcomes, medications, and referrals. Sources of this information include pediatric routine follow-up visits, acute care visits, and hospitalizations.

Postpartum

Additional data are collected postpartum. These variables include information about the person’s anxiety, depression, contraception, substance use screening, substance exposure laboratory results, MOUD, and inpatient or residential stays.

Analysis

Selected maternal sociodemographic characteristics among pregnancies from the seven clinical sites by MOUD treatment status are presented. Variables include maternal age, race, ethnicity, insurance status, and urbanicity. Data were collected for persons identifying as more than one race, and MOUD treatment status of multiracial persons were reported under “other” race, thereby maintaining five minimum categories for data on race (10). All persons with reported ethnicity as Hispanic are grouped as Hispanic, regardless of race; persons with a reported race are grouped separately by race category. Groups are not mutually exclusive.

Urbanicity was determined using Rural-Urban Commuting Area Codes (RUCAs) associated with U.S. Postal Service zip codes (11,12). RUCA codes use urbanization measures, daily commuting, and population density to classify U.S. Census Bureau tracts into urban and rural areas (11,12). To identify the differences between core urban areas and their peripheries, the 33 RUCA codes were aggregated into the categories of urban core, other urban, and rural (Supplementary Table, https://stacks.cdc.gov/view/cdc/126230) (11).

These analyses were performed at the pregnancy-episode level because certain characteristics could change between pregnancies. Results might include data for more than one pregnancy per person; however, multiple-gestation pregnancies were only accounted for once in this analysis.

Differences in characteristics by MOUD treatment status were examined by chi-square tests with the Rao-Scott correction for categorical variables and regression using Taylor series linearization for continuous variables to account for clustering by persons with more than one pregnancy and by clinical site. Statistically significant differences between MOUD and non-MOUD groups were indicated if p values were <0.05. SAS (version 9.4; SAS Institute) was used for data analysis. The data cleaning and analysis processes were internally replicated by an independent analyst to correct any errors and minimize bias.

Evaluation of Surveillance System

After the data schemas and infrastructure were built and initial pilot data were received, CDC conducted a formal evaluation using the CDC Guidelines for Evaluating Public Health Surveillance Systems (13,14). The evaluation consisted of two parts: 1) a review of existing documentation on the surveillance system and 2) key informant interviews with collaborators (13). Existing documentation included a published manuscript describing the rationale for the surveillance system and internal documentation including data collection tools, standard operating procedures, IT architecture diagrams, funding proposals, and pilot data (4). Sixteen key informant interviews were conducted, representing obstetricians, pediatricians, research scientists, and informaticians at the initial four clinical sites; data informatics contractors; CDC clinicians, epidemiologists, and health scientists working on MAT-LINK; a scientist from the CDC MAT-LINK Steering Committee; and the primary funder, ASPE’s OS-PCORTF. Interviews were conducted by a program evaluator without previous engagement in MAT-LINK who used semi-structured guides for each interview that were tailored to the relevant components of CDC Guidelines for Evaluating Public Health Surveillance Systems. A thematic analysis of interviews was performed for each evaluation component. The interview guides are available at https://stacks.cdc.gov/view/cdc/126231.