Approach

Previously, the Office of the Assistant Secretary for Planning and Evaluation (ASPE), the U.S. Department of Health and Human Services (HHS) published a 2022 report on findings from a literature review of publications related to linking Medicaid claims with birth certificates and from discussions with representatives from participating states about their linkage processes.² To further the goal of building data capacity for patient-centered outcomes research (PCOR), the objectives of this project and report thereof were to:

• Engage selected state and national technical experts and leaders to develop a standardized methodology to link Medicaid and birth certificate data that could be used by multiple states;
• Develop a standardized procedure to make a proposed linked dataset available to outside researchers; and
• Create a research agenda of maternal health topics, including an example research protocol addressing one research question that could be answered using the proposed linked dataset.

Five states, represented in the prior report, agreed to assist with the implementation of these new objectives. In addition, representatives from four federal agencies (National Institutes of Health, the Centers for Disease Control and Prevention[CDC], the Health Resources & Services Administration, and the Centers for Medicare & Medicaid Services) were also invited to participate and provide input. A series of meetings were convened with state and national technical experts in data linkage methodology, data security, and data access. These experts contributed to developing guiding principles on several aspects of data linkage methodology and data security and access. Input was then sought from state and national leaders from state Medicaid, vital statistics, maternal health, and epidemiology departments during meetings that were convened along with technical experts to discuss and answer questions or concerns prior to finalizing the agenda and approaches described in this report.

Results

The group developed the following guiding principles that could be used to create a multistate linked database for research using birth certificate and Medicaid claims data:

• A standardized methodology to link these datasets that can be used by multiple states
  The standardized methodology for linking mothers’ Medicaid data with infant birth certificate data includes the following:

  ∘

  The same linkage software that allows for probabilistic matches be used by all states. Link Plus, a free, customizable software package developed by the CDC, is expected to facilitate the broadest involvement of multiple states.

  ∘

  The mother’s Social Security Number be used in the first deterministic linkage step.

  ∘

  Mother’s first and last names and 5-digit zip code, and mother’s and infant’s dates of birth be used in a subsequent probabilistic linkage step.

  ∘

  Other variables (e.g., mother’s middle name, father’s last name, and mother’s race) only be used in a manual review of matches.

• A standardized procedure to make a central database with core data elements available to outside researchers
  The standardized procedures include the following:

  ∘

  States submit deidentified, linked claim-level data, which include a core set of data elements, to a central database, beginning with 2016 data and including at least one year of claims data before and after each delivery (where available).

  ∘

  Re-identification risk be assessed for both restricted and public use datasets prior to release for research.

  ∘

  A standardized application be created for researchers to obtain data from multiple states.

  ∘

  Institutional Review Board approval and data use agreement be required for accessing and using restricted data (individual and claim levels); and public use (aggregated) linked data be available online.

In addition, a research agenda was discussed, and the following topics were selected:

• Examining changes in maternal and infant health outcomes and Medicaid-covered service utilization following postpartum Medicaid extension (this topic was selected as the top priority by the group and developed into a sample research protocol).
• Examining changes in telehealth use during pregnancy and postpartum periods during and after the coronavirus disease 2019 (COVID-19) public health emergency for different risk groups.
• Examining how changes in access to reproductive health services impacted maternal health and birth outcomes.

Conclusion

Over the course of multiple meetings and written feedback from a diverse group of state and national experts and leaders, a set of guiding principles for linking Medicaid and birth certificate data and making these linked data available was developed. These guiding principles can be implemented to create a multistate linked database for research on maternal health, including PCOR. Through this standardized and centralized approach, researchers would be able to access these linked datasets through restricted use (claim-level) and public use (aggregated) options to answer important maternal health research questions. The study protocol described in this report (Appendix B) is an example of a study that could be done using the proposed restricted use data. In the future, these proposed data linkage activities could be expanded to include additional states or additional sources of data.

Background

Medicaid pays for nearly half of all live births in the United States, including more than half of all births by non-Hispanic Black and Hispanic mothers.¹ Women and infants covered by Medicaid bear a disproportionate burden of maternal mortality and poor infant health outcomes, making them a key population group for the U.S. Department of Health and Human Service’s (HHS’s) efforts to address such outcomes. Medicaid data, which includes eligibility and enrollment data as well as claims and encounter data, (hereafter “claims”) linked with live infant birth certificates (hereafter “birth certificates”) are vital sources of data for maternal and infant health research. Birth certificates contain key information not captured in claims (e.g., parent education level, infant gestational age at birth) or captured inconsistently (e.g., low birth weight) in claims. Many high-quality studies use these linked datasets to investigate maternal and infant health research topics, as highlighted in a 2022 report.² However, most studies reviewed and included in this previous report were single- state studies.

To conduct research using multistate linked Medicaid and birth certificate data, data must be linked using consistent methodologies, and multistate data must be accessible to researchers. If Medicaid and birth certificate data are not linked using consistent methodologies across states, linked data from these states may not be comparable. For example, if one state uses a deterministic (i.e., exact matching) methodology based on the mother’s name and Social Security Number (SSN) and another state uses a probabilistic approach (i.e., assigning a probability of a true match based on the degree of agreement across many variables), the matched populations of the two states may differentially exclude unmatched individuals from the study population. In the first state, mothers with commonly misspelled, non-Anglo names or without a SSN may be disproportionately excluded from the linked dataset, while the second state may have a linked dataset that is more representative of the population. These differences in population representativeness between the two states may impact the results of analyses using data from the two states.

Purpose

The overall goal of this project is to improve data capacity for patient-centered outcomes research (PCOR) and other research, as well as program outcomes measures related to maternal health, by making available a methodology for developing high-quality, multistate linked Medicaid-birth certificate data. In this project, a series of meetings with technical experts and leaders at the state and national level was held to address the following specific goals:

• Develop a standardized methodology to link Medicaid and birth certificate data that can be used by the participating states and potentially other states;
• Develop a standardized procedure to make a central database with core data elements available to outside researchers; and
• Create a research agenda on maternal health, including the development of an example research protocol as a use case.

Improving maternal health is a national priority and has been identified as a top priority by HHS,³ as well as a focus area for the Office of the Secretary Patient-Centered Outcomes Research Trust Fund (OS-PCORTF) portfolio.⁴ The OS-PCORTF portfolio is focused specifically on expanding data capacity to address maternal health through research, and it is aligned with the three goals of this project.

Organization of Report

This report is organized by chapter, starting with a description of the methods used to meet the three goals stated above (Chapter 2). Next, current state linkage efforts and a standardized linkage methodology are described (Chapter 3). Steps to make multistate linked data available to researchers (Chapter 4) and presentation of a set of maternal health research priorities that can be addressed using linked data (Chapter 5) follow. The report concludes with a summary and considerations for the future implementation of the developed approach (Chapter 6).

Participant Selection

Seven states that met the following criteria were invited to participate:

• Had at least 50,000 births in 2021;⁵
• Regularly linked Medicaid and birth certificate data, as defined in the previous report;²
• Had adequate quality Medicaid claims and encounter data (as defined by no “high concern” or “unusable” ratings in the Transformed Medicaid Statistical Information System (T-MSIS) Data Quality Atlas for 2020–2021 inpatient and other therapy fee-for-service claims and managed care encounter data, where applicable);⁶ and
• Had previous collaboration within their state and with outside agencies, as defined in the previous report.²

Initial contacts from these seven states were identified using internet searches and references from the previous report. Representatives from five of the seven states agreed to participate: California, Colorado, Kentucky, North Carolina, and Ohio. Each of these five states was asked to recommend one or more individuals who could fulfill the following roles:

• An expert in the methodology used to link the state’s Medicaid claims with birth certificate data (ideally the person responsible for conducting the linkages);
• An expert in the secure storage of linked Medicaid-birth certificate data and access procedures for internal use and external researchers (e.g., the person in charge of the data use agreement [DUA] process for researchers who want to use the linked data); and
• A leader (e.g., in Medicaid, vital statistics, or maternal health epidemiology) who can provide feedback on the approach developed by the group, as well as institutional buy-in and support to the final agreed-on guidance.

Representatives from the following four federal agencies were also invited to participate because of their roles in supporting maternal health research, vital statistics, and Medicaid programs: the National Institutes of Health (NIH), the Centers for Disease Control and Prevention (CDC), the Health Resources & Services Administration (HRSA), and the Centers for Medicare & Medicaid Services (CMS).

The state and national experts in linkage methods and in data security and access participated in smaller technical expert groups, which met to discuss the current linkage methodologies and processes in the five participating states and to develop the guiding principles for one consolidated approach. The technical experts also participated in larger group meetings with state and national leaders where these proposed guiding principles were reviewed prior to being finalized.

Appendix A lists the participants, their affiliations, and their roles in this project.

Process for Developing the Linkage Methodology

After an initial meeting introducing this project’s goals and all participants, a series of meetings with state and national technical experts in data linkage methodology was convened to discuss state-specific processes and develop standardized linkage and data-sharing processes.

Specifically, representatives from the five participating states submitted written details on the methods used to link Medicaid and birth certificate data in their states and their processes for securely making data available to researchers. Based on state input regarding the importance of different methodological decisions and the feasibility of implementation in their states, guiding principles were drafted on data linkage methodology. Similarly, technical experts in data security and access discussed current processes for securing linked data, making linked data available to researchers, and what data would be included in a central database consisting of linked, deidentified datasets with core data elements from five states. They discussed how these data could be shared through restricted and public use models. The larger group reviewed the linkage methodology, provided written feedback, and discussed it further during subsequent meetings.

To develop a research agenda for the use of the linked data, a list of research priorities was compiled based on suggestions from participants and based on the literature review conducted for the 2022 report.² These potential research priorities were discussed at a subsequent large group meeting with both technical experts and leaders, where the group also determined the top research priority and identified other important research topics. Subsequently, the smaller technical expert group discussed options for a research protocol and provided input on the final draft presented in this final report as Appendix B.

The state and national technical experts and leaders provided their input on the final report. The smaller technical expert group met a final time to make minor clarifications to the linkage methodology and security and access procedures. The larger group convened one final time to discuss participants’ feedback on a draft of this report and to clarify any questions or concerns. Altogether, participants met nine times to provide input at multiple stages of the project.

Additional detail on the linkage methodology development process and a schedule of meetings for this project is provided in Appendix C.

Populations and Datasets Linked

Variables Used in Linkages

Software, Scoring/Weighting Methods, and Validation Methods

Linkage Timeline and Updating Linkages

Implementation and Next Steps

In building on the findings of this project, states that agree to implement the principles detailed in this chapter could do so to create a database of linked Medicaid and birth certificate data for research. The implementation process could be iterative, with states sharing additional details on data cleaning and standardization, testing different score cutoffs to generate comparable match quality (in terms of confidence level in matches) across the states’ datasets, and developing procedures for the manual validation of matches. In the future, participating states could contribute deidentified data to a centralized, multistate linked database (described in Chapter 4) that could be used by researchers and policy analysts to generate stronger, more actionable cross-state insights into the impact of policies on maternal and child health. The methodology could also be used in the future by additional states, and the database could be expanded to include data from more states. States also indicated that an expert group on data linkage would be valuable to assist states in implementing the guiding principles in this report.

Models of Data Access for External Researchers

Participating state and national experts reviewed different potential models for securing and accessing linked Medicaid-birth certificate data. These models reflect existing ways that other types of multistate health data are centralized and shared with researchers. Participants noted that both restricted use access models for claim-level data (i.e., requiring a formal application and review process) and public use access models for aggregated data (i.e., data available online with no application required) could be valuable to researchers. Table 4.1 presents the potential variables to be included in restricted and public use data (described in detail later in the chapter).

Table 4.1

Potential Variables and Level of Detail Included in Restricted and Public Use Data.

Characteristics of Restricted and Public Use Data

After discussing the overall models of how data could be stored and accessed, group participants discussed specific characteristics of the restricted and public use datasets.

Participants noted the following considerations:

• Importance of the right data time frame—such as having data on the preconception period available (where possible) and including at least one year of data after delivery (where possible)—would allow longitudinal research on maternal health or study of the impact of such policies as states’ recent postpartum Medicaid expansion.
• Inclusion of an encrypted Medicaid record number for the mother in the restricted use dataset to allow researchers to follow mothers longitudinally across multiple annual files and pregnancies.
• Submission of claims data begin in 2016, because this was the first full year that all states had transitioned to T-MSIS, which included some new standardization of Medicaid data. This was also the first year a new standardized birth certificate was used by all states and the first full year in which the ICD-10 diagnosis codes were used.
• Careful consideration of the variables and level of detail that should be included in restricted and public use data. Participants noted that their initial viewpoints were based on their experiences making data available for researchers but that a full review of reidentification risk for both restricted and public use files would still be necessary and that certain values of variables might need to be suppressed (e.g., public use files might include only the ten most common ICD-10 diagnoses and group the rest as “other”). They also noted that each state might want to do its own initial assessment of reidentification risk the first time data were submitted, but that if the central data repository (e.g., ResDAC) was able to meet each state’s standards, it could perform this action moving forward. For public use files, at a minimum, frequencies of variable combinations <11 would be suppressed in query results. Different groupings of values would need to be explored and assessed to ensure that individuals could not be identified through combinations of aggregated data reports.
• Allowance of some variation in the level of detail provided for a specific variable. For example, in the restricted use data file, it could be possible to use more-granular 5-digit zip codes in urban areas but only provide county geographic identifiers in more-rural areas, where the risk of reidentification is higher. Participants also noted that merging the Social Vulnerability Index (SVI) score,¹³ based on census tract data and including the SVI score instead of the identified census tract could be a way to prevent patient reidentification while including details about important neighborhood characteristics.

Application Process and Data Use Agreement for Restricted Data

Currently, all participating states have processes in place to make their data available to external researchers in some form, although not all states provide individual-level linked Medicaid-birth certificate data.

States that make these linked data available to researchers typically require Institutional Review Board (IRB) approval and a DUA for restricted use data, even when direct identifiers, such as names, have been removed. States also require a description of the research or a business agreement. Some states also have training requirements for researchers, such as the Collaborative Institutional Training Initiative human subjects training.

In addition, states have requirements regarding the environment in which the data need to be stored, the signatures (i.e., required from the data custodian and an authorized representative of the requestor), the data elements shared, the length of data access, the dissemination process, and the review of research output by the state. States pointed to their requirements related to the deletion of data at the end of the research project but with consideration for the research peer-review and publication timeline.

Although there are many similarities across the five participating states, there are also differences in their processes. To consolidate these processes, participants discussed whether it would be possible to have a single IRB review process for accessing multistate data. Two states raised concerns about the possible centralization of the IRB process and suggested that state leaders would not be willing to cede their authority to approve individual research proposals to a central body. However, participants noted that if the central data repository met states’ security and access standards, a centralized IRB process could be considered. Other participants emphasized the value of only requiring a single IRB approval and DUA to increase the likelihood that researchers use multistate linked data. They also noted that using a central data repository with which states were familiar, such as ResDAC, could increase confidence in the process. States emphasized that any shared data would need to be backed up with language on penalties for data release and rules on what researchers must do before submitting derivative work to journals to assure state leaders that data will be used properly. One additional common agreement among participating states was related to the standardization of and signoff on the process for data destruction by external researchers once the projects were complete.

Implementation and Next Steps

To implement sharing of the linked data underlying the restricted and public use datasets, participants emphasized that they would need buy-in and approval from top state leaders at departments of health care services or public health. Participants noted that they would also need legal review and approval on several of the topics discussed, including the processes for external researchers to access data, publish results, and destroy data once projects were completed. Current participating state processes for approving data access to external researchers varied greatly, so substantial effort would be needed to harmonize these processes across all five states. Generally, states suggested that the more layers of security to access a dataset, the likelier state leaders were to approve the process (i.e., a model like that used for ResDAC would be more likely to get buy-in).

To help get buy-in, states proposed that emphasizing the use cases and the concrete insights that could be gained (i.e., the ability to assess the impact of specific policies) would help encourage state leaders to overcome the challenges associated with sharing data. Specific use cases of linked data proposed by the technical experts and state and national leaders are discussed in the next chapter. States also suggested limits on the amount and level of detail of data available in datasets to balance their usefulness for research with risk of reidentification.

Motivation

The third goal of this project was to develop a research agenda including several high-level topics and research questions for the use of the linked Medicaid-birth certificate data, as well as a research protocol for one of these research questions. This effort included identifying pressing maternal and child health research topics that could be answered using the linked data and that would benefit from comparison across several states. This research agenda and protocol could serve as a demonstration of OS-PCORTF efforts to improve maternal health through expanding data capacity and research.

Approach

Prior to the meeting discussing the national research agenda, participants in the large group proposed potential topics and research questions. In addition, representatives from ASPE and RAND proposed other topics and research questions based on a previous literature review conducted of studies using linked data.² Participants then decided on the highest priority research question for inclusion in the national research agenda.

Identified Topics for the Research Agenda

Prior to the above-mentioned meeting, participants noted general interest in topics related to the postpartum Medicaid extension, telehealth use during pregnancy, and substance use during pregnancy. The topic of Medicaid expansion under the Affordable Care Act (ACA) was also identified as a broad topic of interest, using linked data in the previous literature review.² The topic of access to reproductive health services was also brought up by participants during the meeting. Participants discussed these five broad research topics, then proposed and refined one or more research questions within each topic. The large group determined what they viewed as the top research priority of the five presented for development into a research protocol. Most participants selected “How did postpartum Medicaid extension under the American Rescue Plan Act impact maternal postpartum service use and health outcomes?” Table 5.1 below presents the topics, research questions, and key discussion points about the feasibility of answering these questions with the linked Medicaid-birth certificate data and the potential benefit to policymakers of answering these questions.

Table 5.1

Key Considerations for Identified Research Topics on Maternal Health Using Linked Multistate Medicaid-Birth Certificate Data.

Elements of Research Protocol

Once the highest priority research question had been identified, the technical expert group met to discuss specific aspects of a research protocol to address this question, such as definition of the study population, outcome and covariate measurement, and approaches to analysis. Each of these aspects of the research protocol were considered in developing a preliminary set of options and considerations that the participants reviewed (e.g., measures of service utilization to consider for outcomes, variables that could be included as control variables in the analyses, and potential analytic methods) prior to finalizing the sample research protocol (Appendix B). This protocol serves as a motivating use case for linking data across multiple states and could be used in the future for a study conducted with linked data by state and federal government researchers, as well as external researchers, as appropriate.

Considerations

Participants noted that the use of a consistent data linkage methodology and availability of variables will be key to implementing this research protocol. Medicaid data must include enrollment, eligibility, and encounter data. Additional variables include type of Medicaid coverage (i.e., traditional, pregnancy and delivery, or emergency), number of months enrolled pre- and post-delivery, and family income/size. Birth certificate data must include the mother’s age, race/ethnicity, and education; presence of certain maternal complications or infections; abnormal conditions of the newborn; infant APGAR score, and the interpregnancy interval.

Guiding Principles

With regard to the research protocol for the highest priority research topic, the following principles emerged:

• The first question addressed with the linked data set be: “How did postpartum Medicaid extension under the American Rescue Plan Act impact maternal postpartum service use and health outcomes?”
• This study to include the population and key variables of interest, specified in Appendix B.
• The protocol be refined based on feedback from other stakeholders and experts invited to participate in the proposed implementation phase.

Background

As part of the 2021 American Rescue Plan Act,¹⁴ states had a new option to extend Medicaid coverage to 12 months postpartum. Prior to this extension, pregnant individuals typically lost Medicaid coverage 60 days after delivery if they were not otherwise eligible for Medicaid based on their income or other qualifying reasons. Extending Medicaid coverage to a full year postpartum is significant because mortality and other severe complications stemming from pregnancy and delivery often occur well beyond 60 days postpartum.¹⁵

As of June 22, 2023, 35 states and the District of Columbia have implemented this extension, another nine states are planning to implement this extension, three states are planning more-limited extensions (i.e., less than 12 months), and three states have no plans to extend postpartum Medicaid coverage.¹⁶

All five of the states participating in this project have implemented the 12-month postpartum Medicaid extension, although there were some slight state-specific variations in the timing and implementation. California extended Medicaid to 12-months postpartum for Medicaid-enrolled mothers with a diagnosed mental health condition, effective August 1, 2020, and then expanded such coverage for all Medicaid-enrolled mothers, effective April 1, 2022. Kentucky, Ohio, and North Carolina all extended Medicaid coverage to 12 months postpartum, effective April 1, 2022, while Colorado’s postpartum Medicaid extension had an effective date of July 1, 2022.

It is important to note that these policy changes also occurred during the COVID-19 public health emergency. During this time, the Families First Coronavirus Response Act (FFCRA) required states to maintain enrollment of most Medicaid enrollees,¹⁷ meaning that many individuals who typically would have lost Medicaid coverage 60 days postpartum did not. Congress removed the FFCRA continuous eligibility condition and, as of March 2023, required states to begin redetermining Medicaid eligibility and terminate enrollment for individuals who were no longer eligible.¹⁸ The primary objective of this proposed research study is to determine how extending postpartum Medicaid coverage to one year impacted the access and utilization of maternal health care. Extending postpartum Medicaid coverage may have several benefits to mothers. First, the changes in postpartum Medicaid policies in states will likely result in individuals staying enrolled in Medicaid longer and retaining access to needed medical care in the postpartum period. However, changes in postpartum length of Medicaid coverage immediately following the American Rescue Plan Act will likely be modest due to the FFCRA. Postpartum Medicaid extension may also lead to modest increases in Medicaid-covered services used, such as postpartum visits, mental health visits, substance use treatment, and long-acting contraceptive (e.g., intrauterine device, injectable contraceptives) in the year following delivery. This primary objective includes two aims:

• Aim 1 (Primary): Examine how extending postpartum Medicaid coverage to one year impacted maternal Medicaid enrollment during the COVID-19 public health emergency.
• Aim 2 (Primary): Examine how extending postpartum Medicaid coverage to one year impacted Medicaid-covered service utilization among mothers with a Medicaid-covered delivery during the COVID-19 public health emergency.

The secondary objective is to examine longer-term impacts of postpartum Medicaid extension after the end of the COVID-19 public health emergency. One outcome of interest for this objective is the interpregnancy interval (the number of months between pregnancies) following index pregnancy. Knowing this is important because short interpregnancy intervals (less than 18 months) are associated with an increased risk of infant mortality, low birth weight, and pre-term delivery.¹⁹ Individuals with extended postpartum Medicaid coverage may more easily obtain highly effective forms of birth control and subsequently increase their interpregnancy intervals.

Given the policy context of FFRCA during the initial implementation of postpartum Medicaid extension, it is also of interest whether postpartum Medicaid extension policies helped prevent the loss of Medicaid coverage following the end of the COVID-19 public health emergency. Thus, the secondary objective includes two aims:

• Aim 3 (Secondary): Examine how extending postpartum Medicaid coverage to one year impacted interpregnancy intervals among mothers with a Medicaid-covered delivery during the COVID-19 public health emergency.
• Aim 4 (Secondary): Examine whether the impact of extending postpartum Medicaid coverage during the COVID-10 public health emergency on maternal Medicaid enrollment was sustained after the end of the public health emergency and how the results may differ by state.

Methods

Limitations

A key limitation of using Medicaid claims data for research is that only care paid for by Medicaid is observed. For example, it is possible that some individuals who were no longer enrolled in Medicaid during the postpartum period obtained private insurance and received postpartum services paid for by their new insurer or received other financial assistance in paying for care from a health system or federally qualified health center. Even when services are paid for by Medicaid, the use of bundled payment codes for postpartum visits can make counting the number and timing of visits difficult.²⁴

When considering potential policy impacts, the most important limitation is the passage of the FFCRA, which may decrease or even eliminate the observed magnitude of the effect of postpartum Medicaid extension in each state. Furthermore, use of health services and health outcomes were strongly impacted by the COVID-19 public health emergency. While the use of a difference-in-differences design accounts for some of these national trends, other impacts of the COVID-19 public health emergency were more localized and would not be accounted for in the study design. The proposed use of a regression discontinuity design in the secondary analysis could address this limitation if no policy effect is observed in the difference-in-differences design proposed in the primary analysis.

Policy Implications

The results of this study can be used to inform policy deliberations within states that have not yet implemented postpartum Medicaid extension. It may also inform policymakers’ understanding of the impact of having Medicaid coverage on mothers’ access to and utilization of postpartum care or other health care services and outcomes. In addition, the results from extending Medicaid coverage in the postpartum period may inform policy deliberations related to addressing churn in the Medicaid program more broadly.

Populations and Datasets Linked

California begins by linking mothers’ Medicaid delivery claims with the Medicaid eligibility file and hospital delivery data either deterministically if SSN is present (using SSN and delivery month) or probabilistically using the LINKS SAS package if SSN is not present (using delivery date, demographic, and diagnosis/procedure variables). The mother’s linked Medicaid-hospital discharge data are then probabilistically linked to infant birth certificate data using a combination of birth event times and place variables and the mother’s name and demographic variables.

Colorado begins with a mother blocking step (i.e., deterministic matching of a mother’s Medicaid delivery claim to birth certificate on any of a broad set of criteria: exact Medicaid number, exact SSN, or partial name plus DOB match). Each mother’s Medicaid delivery claim and each birth certificate is allowed to have multiple matches at this stage. Next, Colorado executes an infant blocking step similar to the mother blocking step in which infant Medicaid claims are linked to birth certificates. Finally, there is a selection step in which infant and mother Medicaid matches are given a score based on the number and types of variables matched, and the best score is retained for each match.

Kentucky was unable to provide detailed information on linkages, but the state noted that it begins with infant birth certificates and matches them to Medicaid claims and enrollment data using IBM Infosphere, a software that allows for probabilistic matching. Kentucky also confirmed the variables used in their linkages.

North Carolina links both infant Medicaid and maternal Medicaid data to infant birth records using probabilistic matching in Link Plus. This involves blocking steps, in which records are subset to match on blocking variables, reducing the total number of pairwise matches to be scored, and a scoring step that is used to select the most likely record pair when a birth certificate has been linked to multiple Medicaid records. Matches that meet certain more-stringent criteria are automatically retained, while matches that match on other criteria are manually reviewed by at least one reviewer.

Ohio begins by restricting the dataset to mothers who have an SSN on both Medicaid claims and birth certificates. These mothers are linked in SAS using SSN, DOB, and Soundex mother’s first and/or last name. The remaining unmatched mothers are then linked probabilistically using Link Plus software with a combination of the variables mentioned above plus address, zip code, phone number, and delivery date. Mother and infant Medicaid data are then linked using a similar process, but also including the Medicaid case ID number. Finally, infants from Medicaid claims data are linked to infant birth certificates. This final step helps ensure that both the mother and the infant (already linked through Medicaid claims) are matched to the same birth certificate. Matches are selected for manual review based on their Link Plus score or if they match to multiple records (i.e., linked mother-infant Medicaid claims link to multiple birth certificates).

Variables Used in Linkages

All five states used the mother’s SSN in some way. However, California only uses it in the Medicaid-hospital discharge linkage step, not in the Medicaid-birth certificate linkage because SSN is not in the birth certificate data they receive. SSN was used by states in both deterministic and probabilistic matching steps. States generally used SSN as an exact matching variable (i.e., no partial credit is given for transposed or missing digits), although Ohio considered inexact matches in its manual review process. States noted that SSN is sometimes missing, and probabilistic matching on other variables is needed.

All states use the mother’s first and last names. Some states also use the mother’s middle and pre-marital names and the father’s last name (sometimes, the father’s last name on the birth certificate can be matched to the mother’s last name on Medicaid claims if the mother’s name on the birth certificate is missing or incorrect). The mother’s first and last names were used in both deterministic and probabilistic matching steps. Some states allow for fuzzy matching in probabilistic matches using Soundex or the first three or four characters of a name.

All states use the mother’s and infant’s DOB for matching. The mother’s DOB is in the Medicaid enrollment file, the infant’s DOB can be estimated from the mother’s’ Medicaid delivery claims data, and both the mother’s and infant’s DOB are on the birth certificate. Some states allow for fuzzy matching on DOBs. For example, Colorado allows for two of three date elements to be correct in its blocking steps. Ohio allows a two-week window around the DOB, and North Carolina allows a five-day window around the DOB in probabilistic matching steps. Most states used the mother’s geographic information, such as full residential address, street, city, zip code, or county. These variables were used in probabilistic steps and in manual review steps.

Other variables used by only one or two states include the mother’s’ Medicaid record number, the mother’s’ phone number, the mother’s’ race/ethnicity, and delivery method (i.e., vaginal or C-section). Other variables, such as infant name, infant SSN, infant DOB from infant Medicaid enrollment or claims data and hospital ID number were used by some states, but they are not discussed further here because they can only be used when the mother’s Medicaid data are merged with the infant’s Medicaid data or hospital discharge data from the delivery.

Software, Scoring/Weighting Methods, and Validation Methods

Ohio and North Carolina use Link Plus,²⁵ free software developed and supported by the CDC, to probabilistically match records. The software generates match scores based on the strength of the match. Two cutoffs are selected: a higher cutoff, above which matches are accepted as is, and a lower cutoff, for which matches should be manually reviewed between the lower and the higher cutoff. Kentucky uses IBM InfoSphere MDM to perform probabilistic matching.²⁶ This process assigns match scores for each variable used for matching and aggregates such scores into a total match score. California uses LINKS, a SAS package developed by the University of Manitoba that performs a staged matching process that generates and uses scores for probabilistic matching.²⁷ The highest scores at each stage are used to remove duplicates from ties, and staff manually review to ensure the best matches were kept. Colorado uses an in-house-developed SAS program to perform deterministic linkages only and generates scores based on exactly matched variables. States also use in-house-developed SAS programs before and after using the linkage software or programs for data preparation and cleaning.

All states do manual review to some extent, although the criteria and process for manual review differ by state. Ohio and North Carolina manually review a subset of matches based on its Link Plus match score; North Carolina also has two reviewers review some matches and a third reviewer address disagreements between the two reviewers. California manually reviews records that have multiple records and also verifies frequencies of births and demographic variables in matched data against national vital registry sources. Colorado sometimes manually validates all matches for a special purpose analysis of a smaller population and periodically validates a random sample of matches to assess accuracy.

Linkage Timeline and Updating Linkages

Currently, states vary in the frequency with which linkages are performed. California and North Carolina perform linkages annually, and Colorado and Ohio perform linkages quarterly. As Medicaid and birth certificate data might become available at different times due to enrollment or reporting delays, states typically use a runout period (i.e., only match births that happened after a certain amount of time). Ohio and North Carolina also include previously unmatched records in new rounds of matching.