Need for Improved Collection and Harmonization of Rural Maternal Healthcare Data

Introduction

Historically, women have been under-represented in clinical and translational research. Because pregnant women were considered a vulnerable population until 2019, they represent an even smaller proportion of the women in clinical and translational research.1 Considering that research is often performed at large academic medical centers, women who live in rural areas are typically not well-represented.2

The results of not including women have been catastrophic. Thalidomide, which was used in pregnant women to battle nausea in the 1950s, caused significant birth defects and fetal death. The thalidomide tragedy was a result of extracting data from men and applying it to pregnant females.3 By not including rural women in clinical and translational studies, we are ignoring their unique experiences and exposures that may affect the outcomes of their pregnancies.

There is a growing gap in maternal morbidity and mortality between women in urban and rural areas.4,5 Numerous reasons can account for these differences, including social determinants of health such as household income, level of educational attainment, and insurance payor type (uninsured, private, or public insurance). There are also structural issues that can affect maternal health access to care due to the closure of rural hospitals and obstetric units, shortage of healthcare providers for pregnant women, lack of coordination of specialty healthcare, lack of internet access for telehealth care, shortage of emergency medical treatment services, and limited access to social and human services. Rural women may also encounter different environmental exposures than urban women; differences in the use of private water systems, pesticides, fertilizer, and animal exposures can result in differences in water and air quality between rural and urban pregnant women.6

The objective of this article is to review the strengths and weaknesses of available data sources for rural pregnancies, the challenges in collating clinical data, and to recommend variables of interest, and describe the approaches that we are taking to facilitate data collection in a midwestern, predominantly rural state.

CURRENT DATA SOURCES

By not including rural women in pregnancy health data sets, we are ignoring the unique challenges that they face. The problem is compounded by a general lack of homogenized, granular data related to pregnancy. Large pregnancy data sets are often derived from only a few sources. These sources of information include the National Vital Statistics System (NVSS), the Pregnancy Risk Assessment Monitoring System (PRAMS), insurance claims data, Medicaid core data sets, and the National Health and Nutrition Examination Survey. There are additional data sets available related to specific life events or specific research studies.7–10 As discussed below, large data sets each have their strengths and limitations. Often there are common limitations, which include the inability to identify early pregnancy losses and terminations, inability to link mother-child records, difficulty identifying gestational ages for critical events in pregnancy, lag time between data collection and availability, and limited available variables.11

The NVSS records include data on births and deaths. The NVSS compiles the birth certificate data at the federal level. It is based on a standard collection form.12 The uniformity of collection on every birth and its standard questions make it a very powerful tool because of the large amount of data with standard variables. However, the birth certificate data are limited to a specific set of questions with standardized answers that can make it difficult to tease apart issues related to maternal health. For example, the question whether gestational hypertension occurred during the pregnancy does not discriminate between pregnancy-induced hypertension, preeclampsia without severe features, preeclampsia with severe features, early-onset preeclampsia, and late-onset preeclampsia. Understanding the incidence of these diseases based on other factors, such as maternal rurality, is important as preeclampsia is the leading cause of maternal morbidity and mortality. Postpartum maternal health issues are also not addressed on the birth certificate. Importantly, studies have also found underreporting of pregnancy complications on the birth certificate and that differences in reporting vary based on maternal characteristics, such as socioeconomic status, race/ethnicity, and maternal English-language proficiency.13,14 Historical flaws in reporting maternal deaths led to the suspension of reporting maternal deaths in 2007. Even with the addition of a pregnancy checkbox on the standard death certificate, accurate reporting of maternal mortality was still challenging. A multi-state analysis of the accuracy of the pregnancy checkbox on the death certificate identified a sensitivity of 62% and a positive predictive value of 68%.15 Indeed problems with the quality of these data resulted in the development of a new method to code for maternal deaths, termed the “2018 method,” to account for errors in reporting.16 Thus, while birth and death certificates can be highly useful, there are limitations in their use because of ambiguous variables and inaccuracies.

PRAMS questionnaires are a project of the Centers for Disease Control and Prevention and health departments. Forty-six states, the District of Columbia, New York City, Northern Mariana Islands, and Puerto Rico, participate in PRAMS. PRAMS samples women who had a live birth in the previous 2 to 6 months. Women are contacted by mail, and if there is no response, then by telephone. The procedures and survey instruments are standardized to facilitate comparison. The PRAMS questionnaire addresses topics such as barriers to prenatal care, obstetric history, alcohol and cigarette use, physical abuse, contraception, economic status, health insurance coverage, and maternal stress. PRAMS surveys are updated every 3 to 5 years. Thus, it can be difficult to address current issues using PRAMS surveys. The CDC provides English and Spanish versions of the surveys. Data become available ~8 to 12 months after the calendar year ends. The broadly available data is aggregated or truncated for specific variables to protect confidentiality. To acquire more granular data, the state must be contacted to request access.17 Because it is self-reported data, the validity and reliability of the granular data must be considered. For example, validation of PRAMS has found that self-report of specific pregnancy complications and pregnancy history was poor.18 Whereas other variables compared well with birth certificate data. Differences in reliability between variables may be a concern in utilizing PRAMS data.

Pregnancy data can also be retrieved from insurance claims data. For example, the Truven Health MarketScan Commercial Claims and Encounters Database (Truven Health Analytics Inc., Ann Arbor, MI) has inpatient and outpatient healthcare data from over 350 United States payers. The benefit of this large dataset is that family identification numbers link family members on the same plan. Therefore, a mother and child on the same insurance plan could be linked. From this data, it can be difficult to determine the vital status of the child at birth, and estimated due dates and gestational ages at delivery are not directly available. Algorithms need to be applied to the data to identify pregnancies and timing and to assign a birth outcome.11,19–21

Obviously, not all pregnant women will have insurance coverage during pregnancy. Medicaid is the single largest payer of pregnancy-related services in the US. In 2018, over 40% of US births were covered by Medicaid. Even with the implementation of the Affordable Care Act, over 25% of mothers covered by Medicaid for their prenatal care did not have insurance before pregnancy.22 Thus, Medicaid remains an important source of data about pregnancy despite the Affordable Care Act.

Taken together, current data sources can provide significant information about pregnancy. However, they face similar changes in their ability to collect timely, granular data.

CHALLENGES IN COLLECTING GRANULAR DATA

While all data is valuable, having granular patient data will be important for improving surveillance, developing better prediction models, and developing better research questions. Extracting, assembling, and displaying granular, row-level pregnancy-related data have several challenges. These challenges include a large amount of healthcare data, linking the events related to pregnancy, linking mother-child outcomes, and harmonization of data between sources.

Pregnancy is a period of high healthcare utilization, which results in a significant amount of data. Currently, the American Congress on Obstetrics and Gynecology recommends that for uncomplicated pregnancies, prenatal visits occur every 4 weeks until 28 weeks, every 2 weeks until 36 weeks gestation, and then every week until delivery (Fig. 1).23 Following these guidelines, an uncomplicated pregnancy would have 10 to 12 prenatal visits, each with vital measurements such as weight, temperature, and blood pressure measurements and provider notes. In general, this period also routinely contains screening tests for structural anomalies, including ultrasound and/or blood tests, a blood test for gestational diabetes, complete blood cell count, urine culture, Group B streptococcus, and depression. Therefore, there can be information from different sources such as imaging results, lab results, and patient-reported screening results. Additional testing may also occur based on patient risk factors, results of screening tests, and patient choice. Thus, even within a single institution, prenatal data will vary between each pregnancy, and there will be a relatively large volume of data for a single pregnancy episode.

F1FIGURE 1:

Average Recommended Perinatal Visit Schedule. Following guidelines established by the American Academy of Pediatrics and American College of Obstetrics & Gynecology, there are ∼14 to 15 healthcare visits in routine prenatal care. Consequently, a large amount of data is generated about pregnancy. (Figure made with BioRender).

There are also many significant details associated with the actual delivery that can vary widely. For example, the attempted modes of delivery may be different than the final mode of delivery, such as an attempted forceps-assisted delivery may still result in cesarean delivery. There can also be multiple modes of delivery for 1 pregnancy. Delivery of twins could be 1 vaginal delivery and 1 cesarean delivery. Any data structure to collect and organize granular data must be able to accommodate “one to many” relationships that occur in pregnancy.

Infrastructure barriers may also impede collecting and assembling prenatal and delivery information. Even when outpatient clinics are associated with a specific hospital or birth center, the electronic health records systems may be segregated between the outpatient and inpatient center reducing information availability. Snapshots of information may be available, but these can exist as faxed, scanned files from the inpatient facility in the outpatient health records. These data can be useful for hand extraction but present a greater challenge in automating data extraction. Interoperability between health record systems may reduce or eliminate this problem in the future; however, it will continue to hamper retrospective data.

Importantly, the data must exist in a format that can be easily extracted and harmonized. For retrospective data from different sources, variable harmonization must be performed to pool data.24 To harmonize data, for each variable, one must assess what it is measuring and the possible answers. This process requires extensive variable alignment and data cleaning. There can be identical, similar, or completely different responses to the same variables in different data sets. Even when the choices are similar, the scales or times for assessment may be different. Thus, the retrospective data harmonization process from multiple data sets can be laborious and time intensive. Protocols have been described for data harmonization, but the literature is scarce regarding pregnancy.24–27 Thus, a common data model representing the pregnancy-child dyad will provide a rich source of data for analysis.

Even when data is designed to be collected easily, the implementation for collection may be long and limit its timely usefulness. For example, a pregnancy question was added during the 2003 revision of the US death certificate. However, its implementation took 16 years across all of the US.16 As a result, the National Center for Health Statistics did not report a maternal mortality rate in the US between 2007 and 2018.28 Quicker implementation of data collection instruments is necessary to adapt to new questions and to be able to analyze whether new activities are having an impact on outcomes.

In an EHR, data in discrete, structured fields with specific choices facilitates the collection and transformation of retrospective and prospective data. Prospectively, data harmonization can be simplified by uniform data collection instruments. Medically, all prenatal providers using the same note template facilitate data extraction from discrete fields and free text fields. Categorical fields with a set of defined responses are the clearest to align. For example, a set of drop-down choices or radio buttons reduces the need to have a content expert, such as a prenatal care provider, assist in data compilation.

There are automated methods to extract and convert unstructured data, such as that found in narratives of medical notes, but they are complex. Natural language processing (NLP) is widely used to handle free text for processing in a format that can be used for analysis.29 In NLP, algorithms detect sentences, convert the unstructured text into a structured format, and assign meaning and relationships to the text.29,30 There are programs and platforms for sharing NLP algorithms and vocabularies to reduce duplication of effort between institutions. Once data is extracted and organized, transformation of data may be necessary to harmonize variables between sites. For example, a cesarean section delivery may be labeled as c-section, c/s, cesarean section, or include other details in the choices for mode of delivery such as repeat c-section, low transverse c-section, or classic c-section. Alternatively, when data is being compiled by hand from disparate sources, someone with content expertise can read the notes and translate the information into fixed responses on a data collection instrument.

While manual data collection reduces the need for algorithms and vocabularies, it can require significant effort and time. Established standards for manual data extraction require using double data extraction. Discrepancies are then resolved with the assistance of a third person. In addition, it may require travel to individual clinics and hospitals and the ability to access their medical records. Data collectors also then need to be facile with multiple electronic health record systems and the local practices of data entry. For example, 1 clinic may indicate the first date of the last menstrual period most often in a first prenatal visit note and another clinic may have it documented on the first ultrasound report. Varied documentation practices among prenatal providers further complicate data extraction.

Another challenge in documenting maternal health outcomes is identifying the health events that are either related to pregnancy or that occur during the pregnancy. The collection of pregnancy data is facilitated using pregnancy episode identification numbers or by employing an algorithm based on the estimated due date (EDD). Only events, such as labs or admissions that are linked to the pregnancy episode or that occurred in this time frame can then be collated. From the EDD, instances within the pregnancy can be labeled with gestational ages and/or trimesters. Care must be taken to account for changes in EDD made during pregnancy. Content experts should iteratively validate results during the development process of automated data extraction.

APPROACHES

We have taken several innovative approaches to improve the collection of pregnancy-related granular, patient-level data in Iowa to drive quality improvement, better surveillance of health outcomes, and expanding research to improve health outcomes. Our methods are centered on standardizing the collection practices and variable language, which have been shown to improve population health.31,32 This approach is also in line with Priority 1 of the recently published “CMS Framework for Health Equity 2022-2032” to “Expand the Collection, Reporting, and Analysis of Standardized Data.” Our initiatives include developing a clinical enterprise data warehouse to store pregnancy-related information for the mother and child at the University of Iowa Hospitals & Clinics. On the basis of our experience in manual extraction of medical record data from multiple clinics throughout Iowa,33 we are now developing standardized forms that can be used in the medical record for the delivery information and for the discharge summary.33

Our first initiative was developing a clinical enterprise data warehouse and viewer for maternal and infant health. In the first phase, it was called the Maternal-Child Knowledgebase. In the second phase, the enterprise data warehouse was called the Intergenerational Health Knowledgebase (IHK). In the first phase of this multi-phase project, we have automated abstracting medical record information from the University of Iowa Hospitals & Clinics (UIHC). This includes retrospective information on all pregnancies receiving care at UIHC since 2009. In addition, it includes all infants born at UIHC or who transferred into UIHC within the first 30 days of birth. We began Phase 1 at 1 institution because UIHC is the only academic medical center in Iowa. Phase 1 was a multi-year project that involved a diverse team, which included Biomedical Informatics, Obstetrics & Gynecology, and later Pediatrics. Information between the birthing parent and child are linked; future iterations will also include the co-parent (Fig. 2). In addition, the IHK provides information about the timing of events and about the family relationships (Fig. 2). Given the complexity of this information, institutional commitment was critically necessary for designating the personnel resources and hardware needed for development. In Phase 2, we deployed and tested the IHK. In testing the IHK, we identified more variables that needed to be incorporated and changes in the queries that would facilitate extracting commonly requested data elements. In future phases, we plan to incorporate data from hospitals outside UIHC that use the same electronic health record system as UIHC. Later, we also plan to include data from hospitals using other electronic health record systems.

F2FIGURE 2:

The IHK provides information about the timing of events and the relationships between family members. Data between parents and children are linked. In addition, the timing of the data elements can be analyzed, such as what medications were given after a specific procedure.

Our implementation plan addresses a common barrier in granular data collection. Although academic health centers may have the capacity for the collation of data, small rural centers may not have this capacity. As a result, the data gap around rural maternal health would only widen. However, if an academic medical center develops the data model, creates drafts of forms for data collection, and builds the infrastructure for data collection and storage, then it could simplify the process for rural health centers. The inclusion of implementation scientists on our team also facilitates identifying impracticalities or “road bumps” for these centers in contributing data and identifying mechanisms truly working collaboratively.

There are clear legal, institutional review board, and privacy issues to overcome to combine the data. The key to this process will be transforming data to be useful and also to protect patients’ privacy simultaneously. One potential future option is to utilize the original data set to generate a simulated one that can be shared widely because it would not contain any identifying information; however, it would retain the distributions and correlations from the original data. The ultimate goal is to increase the amount of accurate granular data from rural pregnancies to improve the ability to perform timely surveillance work, quality improvement and assurance projects, and research. By having this maternal health data resource, new data sets would not need to be built for each new question, nor would different groups be duplicating work by extracting the same information. Furthermore, rural health inequities could be identified by comparing results with patients living in more urban areas. The granular data would be useful for identifying the factors that place patients most at risk.

To increase the usability of the IHK, the biomedical informatics group designed and built a secure web-based viewer with predesigned reports. This viewer has 2-factor authentication and can only be used on the UIHC network by authorized personnel. The viewer allows people with little to no structured query language knowledge to access data. The first version has been extensively beta tested, and revisions have been made based on adding new variables, the ability to filter results, and broaden data acquisition.

Current variables are focused on maternal and neonatal health. These variables are centered on several areas, including demographics, diagnoses, obstetric history, laboratory results, procedures, inpatient and outpatient medications, admission details, delivery details, smoking and alcohol use, and depression screening results. Recommended variables are described below.

In our previous work, to develop a perinatal network including non-academic clinics, we developed a data extraction tool. Study data were collected and managed using Research Electronic Data Capture (REDCap) electronic data capture tools hosted at the University of Iowa.34,35 REDCap is a secure, web-based software platform designed to support data capture for research studies, providing (1) an intuitive interface for validated data capture; (2) audit trails for tracking data manipulation and export procedures; (3) automated export procedures for seamless data downloads to common statistical packages; and (4) procedures for data integration and interoperability with external sources. We built collection instruments based on the obstetric patient record forms from the American College of Obstetricians and Gynecologists. Trained data extractors retrieved data from local electronic record systems of patients who provided informed consent to participate in this network. Data were entered into the REDCap forms at least 6 weeks after participants delivered. When the COVID-19 pandemic developed, we adapted our collection instruments to capture new information related to vaccination, screening, infection, and treatment for COVID-19. The benefit of this approach is that we were able to include patients who were receiving care in community practice clinics. The clinics do not have dedicated research personnel and do not normally participate in research studies. Moreover, the catchment areas of the participating clinics include rural Iowa. However, this approach was time and personnel intensive. We were able to collect information about fewer variables than we did in the IHK because of the time needed to identify and input data. Once the variable set had been decided, it was difficult to change the questions and answers because it would require re-extracting data. COVID-19 was an exception because we added the questions at the beginning. However, answers to questions had to be added as care standards changed and new preventative and therapeutic options became available.

Because of the difficulty in identifying all data elements in the various health record systems and even with the variability between providers, the benefits of standard forms became evident. Thus, our second initiative centered around designing common forms that could be used by any prenatal care provider. The use of common forms helps to overcome several barriers in collecting granular data. First, it ensures that all groups are collecting the same data. Secondly, the variables are harmonized at the point of collection to eliminate or at least significantly reduce the need for clean data to synthesize results from different hospitals and clinics. Third, it reduces the burden of completing other required paperwork, such as the birth certificate form, by ensuring that the fields for the required forms are included at the point of data collection in clinical care. Fourth, it eliminates outdated choices and questions, which results in the reduced collection of erroneous data.

For these common forms, we focused our efforts on 2 significant events in pregnancy: delivery and discharge. To design these forms, we asked several hospitals from different systems in Iowa to provide a copy of their template. We then aligned similar questions and listed all possible answers. In addition, questions from the birth certificate were also included in the alignment. For questions and answers that were not uniformly included, we sent a survey to stakeholders regarding their preference for whether they should be included in the final version. Surveys were sent, built, and distributed by email using REDCap to stakeholders at facilities participating in our state Alliance for Innovation on Maternal Health on Maternal Health program. For the delivery summary survey, we had a final response rate of 54% (35 responses from 65 invitations). On the basis of the survey results, a draft of the delivery summary was generated. We then asked 3 to 5 different stakeholders and content experts to beta test several different rounds of drafts and iteratively made changes based on their feedback. Our next steps are to distribute our delivery summary and assist hospitals with implementation. We are using a similar process to construct and implement a common discharge summary.

These data can inform multilevel implementation research. Harmonized data will allow state-level agencies, health systems, and researchers to characterize intersectional disparities in access to and delivery of quality care and understand clearly how interventions and policies reduce or increase disparities in health outcomes across the state. Harmonized data on pregnancy conditions, care, and outcomes at the state level can be used to identify rural areas and guide the implementation of evidence-based practices that usually are not developed or trialed in rural communities. Engaging multilevel stakeholders over the course of the development of standardized data collection tools through surveys and iterative rounds of piloting makes implementation of the tools more likely to be adopted and sustained over time across systems.

VARIABLES OF INTEREST

In addition to the technical challenges in collecting and harmonizing data, users with content expertise need to agree on the variables of interest and the level of granularity needed. The key to reducing medical disparities for rural pregnant women is having a high degree of granular data from numerous women of different backgrounds and from different areas. Because rural hospitals and birth centers each often have a relatively low volume of deliveries per year, it will be important to collect these variables from as many rural locations as possible to represent the state of rural maternal health.

Our maternal variables of interest are broadly divided into several categories. These categories include demographics, diagnoses, medications, procedures, results, and delivery information. Demographic information includes medical record number, last name, first name, middle name or initial, date of birth, race, ethnicity, gender, current address (street address, city, county, state, zip code + 4), phone number, email address, total number of newborns and pregnancy episodes documented in our center’s EHR, each pregnancy episode ID, and whether that episode is linked to a delivery that occurred at UIHC. Demographics for each person also include the rural-urban commuting area codes for their address. These codes are frequently used to define rural versus urban locations. The list is available online and is matched to each patient based on their zip code.36

For maternal diagnoses, we have a report with true/false columns for several disorders that are frequently of interest in pregnancy-related studies. These include gestational diabetes, preeclampsia, eclampsia, and Rh sensitization. Our system indicates whether the diagnosis is found on the patient’s diagnosis list and/or problem list. Any “true” results also include the diagnosis code that caused the result. In addition, we have generated a second report that lists each diagnosis by ICD-10 code and by name. Because the number of diagnoses is quite large (currently over 6 million), this report is cumbersome and is limited by web browser capability.

In our medication lists, we separate prescribed medications for outpatient use and those that were administered for inpatient. Medication names and generic name are both listed. The outpatient includes whether medications were prescribed before the pregnancy or added to the medication list during the pregnancy episode. The dosage for all medications and start/end dates are noted.

Procedures in pregnancy are also tracked. They are listed by their current procedural terminology codes. The date of the procedure is also included. Information about the delivery procedure is listed separately, with information that is largely extracted from our delivery summary template. One report also identifies whether some commonly used medications were administered; all medications can also be searched for by name. It also contains numeric information for gravidity, parity information, term deliveries, preterm deliveries, pregnancies not resulting in a live birth, and living children. Importantly, granular details about the delivery are collected. These details include whether the patient was transferred from another facility, if and how the delivery was induced or augmented, whether it was a trial of labor after a previous c-section, if a forceps and/or vacuum-assisted delivery was attempted, the type of anesthesia used, gestational age at delivery, indications for induction, indications for c-section delivery, type of uterine incision, type of laceration occurred, date and time of rupture of membranes, the method for rupture of membrane, fluid color at rupture of membranes, how long the membranes were ruptured before delivery occurred, date and time of delivery, number of prenatal outpatient visits, amount of blood loss, length of the hospital stay for delivery, the number of hospital admissions during pregnancy, the number of hospital readmissions within 6 weeks of delivery, and descriptions of the primary insurance and payer type.

Results include both laboratory-generated and patient-provided information. In addition, the results of vital exams are included. The results of any inpatient and outpatient laboratory tests are available. Laboratory tests are identified by their name and the logical observation identifiers, names, and codes identifier value; units are included for all results. The use of the logical observation identifiers, names, and codes identifier ensures that any lab results being compared are from the same test. The patient-provided information includes screening results from the Patient Health Questionnaire-9 for depression and the Edinburgh Postnatal Depression Scale for postpartum depression. It also includes patient-reported results for alcohol and tobacco use.

Outcomes for the delivery are also recorded. These outcomes include infant at 1 minute, 5 minutes, and 10 minutes (if applicable) Apgar scores, birth weight, birth length, and sex. Similar to the birthing parent, laboratory values, procedures, and diagnoses are recorded for the infants.

Thus, the architecture of the IHK is novel in that it is a transformative data set that aggregates patient data from several sources. In addition, data elements are harmonized between sources (Fig. 3). We transform health record data using concepts relevant to obstetrics to output standardized data that is granular, timely, and adaptable for use in surveillance, quality improvement and assurance, and research. In future phases, we plan to include data from outside of UIHC. Data will be scrubbed of identifiers or transformed to be able to be used by a larger community to improve health outcomes.

F3FIGURE 3:

Example of IHK data architecture and harmonization. A, Multiple sources of data are integrated and transformed. Elements are shared using pre-defined reports. B, Discrete variables, such as gender, may have disparate possible answers in distinct health record systems. Data transformation harmonizes results across systems. Data harmonization before output reduces the amount of preparation necessary before analysis.

USING DATA TO IMPROVE RURAL MATERNAL HEALTH

Many organizations have ongoing efforts to eliminate rural maternal health disparities. Data needs to be available to drive and evaluate these efforts. Having a comprehensive, granular data set will improve the ability to identify needs, target interventions, and evaluate whether the outcomes improved for rural maternal health.

References 1. Biggio JR Jr. Research in pregnant subjects: Increasingly important, but challenging. Ochsner J. 2020;20:39–43. 2. Seidler EM, Keshaviah A, Brown C, et al. Geographic distribution of clinical trials may lead to inequities in access. Clinl Invest. 2014;4:373–380. 3. Kim JH, Scialli AR. Thalidomide: the tragedy of birth defects and the effective treatment of disease. Toxicol Sci. 2011;122:1–6. 4. Hansen A, Moloney M. Pregnancy-related mortality and severe maternal morbidity in rural appalachia: Established risks and the need to know more. J Rural Health. 2020;36:3–8. 5. Rossen LM, Ahrens KA, Womack LS, et al. Rural-urban differences in maternal mortality trends in the US, 1999-2017: Accounting for the impact of the pregnancy status checkbox. Am J Epidemiol. 2022;191:1030–1039. 6. Institute of Medicine (US) Roundtable on Environmental Health Sciences, Research, and Medicine. Rebuilding the Unity of Health and the Environment in Rural America: Workshop Summary. Washington (DC): National Academies Press (US); 2006. 7. Margulis AV, Setoguchi S, Mittleman MA, et al. Algorithms to estimate the beginning of pregnancy in administrative databases. Pharmacoepidemiol Drug Saf. 2013;22:16–24. 8. Devine S, West S, Andrews E, et al. The identification of pregnancies within the general practice research database. Pharmacoepidemiol Drug Saf. 2010;19:45–50. 9. Society for Maternal-Fetal Medicine. Data Sets for MFMs. 2021. Available at: https://www.smfm.org/research/datasets. Accessed May 1, 2022. 10. Myers JE, Myatt L, Roberts JM, et al. Global Pregnancy C. COLLECT, a collaborative database for pregnancy and placental research studies worldwide. BJOG. 2019;126:8–10. 11. MacDonald SC, Cohen JM, Panchaud A, et al. Identifying pregnancies in insurance claims data: Methods and application to retinoid teratogenic surveillance. Pharmacoepidemiol Drug Saf. 2019;28:1211–1221. 12. Centers for Disease Control and Prevention. National Vital Statistics System. 2022. Available at: https://www.cdc.gov/nchs/nvss/births.htm. Accessed 2022. 13. Haghighat N, Hu M, Laurent O, et al. Comparison of birth certificates and hospital-based birth data on pregnancy complications in Los Angeles and Orange County, California. BMC Pregnancy Childbirth. 2016;16:93. 14. Reichman NE, Schwartz-Soicher O. Accuracy of birth certificate data by risk factors and outcomes: analysis of data from New Jersey. Am J Obstet Gynecol. 2007;197:32 e31–32 e38. 15. Catalano A, Davis NL, Petersen EE, et al. Pregnant? Validity of the pregnancy checkbox on death certificates in four states, and characteristics associated with pregnancy checkbox errors. Am J Obstet Gynecol. 2020;222:269 e261–269 e268. 16. Hoyert DL, Minino AM. Maternal mortality in the United States: Changes in coding, publication, and data release, 2018. Natl Vital Stat Rep. 2020;69:1–18. 17. Centers for Disease Control and Prevention. PRAMS. 2022. Available at: https://www.cdc.gov/prams/index.htm. Accessed May 25, 2022. 18. Dietz P, Bombard J, Mulready-Ward C, et al. Validation of self-reported maternal and infant health indicators in the pregnancy risk assessment monitoring system. Matern Child Health J. 2014;18:2489–2498. 19. Jovanovic L, Liang Y, Weng W, et al. Trends in the incidence of diabetes, its clinical sequelae, and associated costs in pregnancy. Diabetes Metab Res Rev. 2015;31:707–716. 20. Ailes EC, Simeone RM, Dawson AL, et al. Using insurance claims data to identify and estimate critical periods in pregnancy: An application to antidepressants. Birth Defects Res A Clin Mol Teratol. 2016;106:927–934. 21. Martin AS, Monsour M, Kissin DM, et al. Trends in severe maternal morbidity after assisted reproductive technology in the United States, 2008-2012. Obstet Gynecol. 2016;127:59–66. 22. Johnston EM, McMorrow S, Alvarez Caraveo C, et al. Post-ACA, more than one-third of women with prenatal medicaid remained uninsured before or after pregnancy. Health Aff. 2021;40:571–578. 23. American Academy of Pediatrics, American College of Obstetricians and Gynecologists. Guidelines for perinatal care, Eighth edition. ed. Elk Grove Village, IL Washington, DC: American Academy of Pediatrics ; The American College of Obstetricians and Gynecologists; 2017. 24. Adhikari K, Patten SB, Patel AB, et al. Data harmonization and data pooling from cohort studies: a practical approach for data management. Int J Popul Data Sci. 2021;6:1680. 25. Rolland B, Reid S, Stelling D, et al. Toward rigorous data harmonization in cancer epidemiology research: one approach. Am J Epidemiol. 2015;182:1033–1038. 26. Schaap LA, Peeters GM, Dennison EM, et al. European Project on OSteoArthritis (EPOSA): methodological challenges in harmonization of existing data from five European population-based cohorts on aging. BMC Musculoskelet Disord. 2011;12:272. 27. Fortier I, Raina P, Van den Heuvel ER, et al. Maelstrom Research guidelines for rigorous retrospective data harmonization. Int J Epidemiol. 2017;46:103–105. 28. Creanga AA, Thoma M, MacDorman M. Value and disvalue of the pregnancy checkbox on death certificates in the United States-impact on newly released 2018 maternal mortality data. Am J Obstet Gynecol. 2020;223:393 e391–393 e394. 29. Mehta N, Pandit A. Concurrence of big data analytics and healthcare: A systematic review. Int J Med Inform. 2018;114:57–65. 30. Koleck TA, Dreisbach C, Bourne PE, et al. Natural language processing of symptoms documented in free-text narratives of electronic health records: a systematic review. J Am Med Inform Assoc. 2019;26:364–379. 31. Bhalla R, Yongue BG, Currie BP. Standardizing race, ethnicity, and preferred language data collection in hospital information systems: results and implications for healthcare delivery and policy. J Healthc Qual. 2012;34:44–52. 32. Dorsey R, Graham G, Glied S, et al. Implementing health reform: improved data collection and the monitoring of health disparities. Annu Rev Public Health. 2014;35:123–138. 33. Santillan DA, Brandt DS, Sinkey R, et al. Barriers and solutions to developing and maintaining research networks during a pandemic: An example from the iELEVATE perinatal network. J Clin Transll Sci. 2022;6:e56. 34. Harris PA, Taylor R, Thielke R, et al. Research electronic data capture (REDCap)--a metadata-driven methodology and workflow process for providing translational research informatics support. J Biomed Inform. 2009;42:377–381. 35. Harris PA, Taylor R, Minor BL, et al. The REDCap consortium: building an international community of software platform partners. J Biomed Inform. 2019;95:103208. 36. Agriculture UDo. Rural-Urban Commuting Area Codes. 2020. Available at: https://www.ers.usda.gov/data-products/rural-urban-commuting-area-codes/. Accessed April 20, 2022.

留言 (0)

沒有登入
gif