A substantial body of recent research has focused on the prevalence of substance abuse among pregnant women, its effects on maternal and infant health, ways to identify and effectively treat pregnant women, and the cost of care for pregnant women and their infants. However, this research has been hampered by a lack of data, particularly of the state and local levels. The potential benefits of such research for improving treatment and other services to this high-risk group make if important to continue to develop accurate and timely data sources. This study builds on previous work that investigated the quality of birth certificate and Medicaid data as well as previous efforts to merge these two data sources by augmenting the merged birth certificates and Medicaid data with data on a group of known substance abusers - those in substance abuse treatment. The purpose of merging these records is to improve the identification of pregnant substance abusers and, more importantly, to provide an enriched data source of variables from all three sources. Using merged data from two example states, we show that linked birth certificates, Medicaid claims, and state substance abuse treatment files provide a useful source of information for studying the birth outcomes and costs for pregnant women in substance abuse treatment. However, certain groups, such as women who do not obtain any health care or substance abuse treatment, will be underrepresented in such a database. Consequently, efforts at more complete surveillance of pregnant substance abusers should be undertaken, perhaps through increased efforts to improve reporting on the birth certificate, which is a universal document.
BACKGROUND AND RESEARCH ISSUES
A substantial body of recent research has focused on the prevalence of substance abuse among pregnant women, its effects on maternal and infant health, and effective ways to identify and treat pregnant women (Howell et al. 1999; Chasnoff 1988; Horgan et al. 1991 ). However, conducting research on this important subject has been hampered by a lack of data, particularly at the state and local levels. There have been some national-level prevalence studies, including the National Maternal and Infant Health Survey (Centers for Disease Control and Prevention 1994) and National Pregnancy and Health Survey (National Institute on Drug Abuse 1996). The state-specific Pregnancy Risk Assessment Monitoring System is used to measure problem drinking during pregnancy in some states (Bruce et al. 1993). However, these sample surveys cannot be used to measure substance abuse among pregnant women at the local level.
Records that are routinely collected for administrative purposes are increasingly used for research on other topics. These databases, once in computer-- accessible form, can provide a low-cost means of avoiding direct primary data collection. If such databases can be linked, they could provide an even more useful data source because of the additional information that can be obtained from multiple sources. The purpose of this paper is to provide an overview of the process of matching and merging administrative records so that researchers appreciate both the difficulties of using such records and the weaknesses and strengths of the merged data.
There are potential problems with using records that are collected for an administrative purpose for research since the data may be incomplete or of otherwise poor quality. Investigating data quality is consequently one of the most important first steps in using such data for research.
In the case of substance abuse research, one of the most important aspects of data quality is the actual reporting of substance use or abuse in administrative records. A concern for maternal confidentiality may lead providers to underreport substance use. This is particularly true in states that have taken punitive action against pregnant/postpartum women who use drugs or alcohol, since reporting women could make them subject to incarceration or losing custody of their newborns. For example, in South Carolina, a conviction for child abuse associated with substance abuse during pregnancy was recently upheld by the state's Supreme Court and is one of over 200 such cases in 30 states. Most, however, have been overturned on appeal (Terry 1996). Because of the stigma associated with substance abuse, legal sanctions, and the general lack of use of sophisticated screening tools to record substance abuse in such data sources, reporting of substance use by individuals and their providers is likely to be incomplete. In particular, pregnant substance abusers are a highly stigmatized group; consequently surveys, administrative records, and surveillance systems all may underreport the incidence of substance abuse in this population (Sackoff et al. 1992).
While the data in administrative record systems may not be complete, individual patient interviews also do not necessarily provide a "gold standard" for information on a woman's substance abuse. For example, a study of discrepancies between women's personal interview data and medical records found higher rates of reported use from interviews for marijuana use but higher rates in medical charts for cocaine and heroin use (O'Campo et al. 1992). The National Pregnancy and Health Survey was conducted in the hospital following delivery and used blinded urine toxicology screens to confirm rates of reported substance use. It found that mothers underreport their use of illicit drugs during the three days of pregnancy prior to delivery: while only 1.1 percent reported using marijuana, cocaine, methamphetamine, heroin or methadone during that period, 4.4 percent tested positive for one or more of those drugs (National Institute on Drug Abuse 1996).
Three types of administrative data files contain records for many pregnant substance abusers and potentially can provide data on the incidence of substance abuse and use of services. These are birth certificates (registration forms completed on each birth in the United States); Medicaid enrollment and claims data (the records for all enrollees and services covered by Medicaid, the health insurance program for low income women and infants); and records of admissions to drug treatment facilities that are collected by state treatment agencies.
This paper investigates whether, in the absence of a complete registry for pregnant substance abusers, existing administrative data sources can be merged and used for observation and research. The purpose of merging these records is to improve the identification of pregnant substance abusers and, more importantly, to provide an enriched data source using variables from all three sources. Two aspects of the merged data are discussed: (1) data quality and (2) the types of variables that are available to study substance abuse and its treatment. Three research questions are addressed:
How well do these three data sources identify pregnant substance abusers?
What research variables are available on these files?
Can the data be used to study treatment patterns?
We first review existing literature on the quality of administrative records and then present results from an analysis of these data sources using data from two example states.
Each infant born in the United States has a birth certificate that is completed at the time of birth, usually by hospital personnel. The U.S. standard certificate (Freedman et al. 1988), adopted with some modifications by all states, has a variety of information on maternal demographics (age, race, education), birth outcomes (gestational age, birthweight), and maternal and infant risk factors. The U.S. Standard Certificate includes a field that indicates alcohol use or abuse (whether a woman drinks and how many drinks) and a check-off box that indicates fetal alcohol syndrome (FAS) in the newborn.
The U.S. standard certificate does not have a specific field for reporting drug use. However, some states' certificates also include a specific place to indicate maternal drug use or infant prenatal exposure to drugs. This may be identified by a physician or nurse from observation or may be detected through a toxicology screen at delivery. Anecdotal information indicates that the amount of reporting varies by hospital policies, especially each hospital's policy regarding toxicology screening of women and newborns.
Most states have adopted the alcohol use field from the Standard Certificate (fifty of fifty-four states and territories) and the FAS check-off box (fifty-two of fifty-four). The states that add some indicator for the mother or newborn of prenatal drug use are Connecticut, Indiana, Nevada, New Jersey, New York, Rhode Island, Tennessee, and Washington. New York, for example, has a checkoff box indicating drug use and, after 1993, the type of drug the mother used. The birth certificates in states without specific fields for coding drug use have open-- ended fields for "other risk factors" where drug use may be written. However, those data are usually not separately coded on the automated files.
Birth certificates linked to other data sources have provided the basis for numerous health services research studies on prenatal care and birth outcomes (Piper et al. 1994; Howell and Brown 1989; Krieger et al. 1992). However, they have not been used extensively for research related to perinatal substance abuse.
While states have the responsibility for collecting and storing birth certificates, they also provide, on a cooperative basis, automated files of all certificates to the U. S. National Center for Health Statistics (NCHS) for national-- level research studies. The NCHS monitors the quality of data on birth certificates and issues periodic reports (National Center for Health Statistics 1985, 1993). As measured by the level of agreement between the birth certificate and an in-person maternal interview, these studies have shown that the quality of basic demographic variables is relatively high but that other variables show less agreement. For example, there is only about a 50-percent agreement in reports of the number of prenatal care visits. The exact gestational age is also poorly reported (Piper et al. 1993), although agreement with hospital clinical records is within two weeks for about 80 percent of records. Another study showed that check-off boxes indicating birth defects are not very reliable (Watkins et al.1996) nor are reports of pregnancy complications (Dobie et al. 1998). Other studies of the quality of birth certificates have shown that item nonresponse is higher for high risk births (Kleinman 1991).
Few studies of the quality of the birth certificates have addressed the reliability and consistency of the reporting of substance abuse. The studies that are available have generally addressed reporting of tobacco and alcohol use, rather than drug use, since these items are on the U.S. standard certificate. A study that compared the agreement between birth certificates and medical records in North Carolina of smoking and alcohol use during pregnancy (Buescher et al. 1993) found relatively high rates of agreement for smoking (84.4 percent) but much lower rates of agreement for alcohol use (56.2 percent). A similar study in Tennessee (Piper et al. 1993) found that birth certificates identified 73.5 percent of women whose medical records said they smoked, but they identified only 30.7 percent of those whose records said they used alcohol. While the medical record may also underreport substance abuse, it is considered a more reliable data source than the birth certificate because it should contain the results of toxicology tests. In Colorado, researchers (Miller et al. 1990 compared birth certificates to a variety of other sources of information on FAS and found that birth certificates identified only 11 percent of cases that were identified from all sources in the state's FAS registry.
There has been almost no research on the quality of birth certificates for drug abuse reporting. A single study in Oregon (Slutsker et al. 1993) compared reports from birth attendants to birth certificates and found that the certificates identified only 41 percent of women that the attendants knew to be drug users. Consequently, while information is scarce, the data that are available suggest that birth certificates may be of limited use for studying pregnant substance abusers and their infants unless the data are linked to other records.
Medicaid claims. Medicaid claims are another potential source of data on perinatal substance abuse since Medicaid is used to finance many services for pregnant substance abusers. One of the major reasons that researchers choose to use Medicaid records is to study the cost of treatment and other services. In addition, states have the option to further expand eligibility for pregnant women and infants. As of October 1998, thirty-seven states had expanded the eligibility level for pregnant women and infants above the federal mandate. These expansions have increased the federal and state financial involvement in prenatal, delivery, and postnatal care, and in 1996 Medicaid covered over a third of all births in most states (National Governors Association 1999). Although welfare reform has reduced Medicaid enrollment overall, the Medicaid program continues to provide potential coverage for all pregnant women and infants living at or below 133 percent of the Federal poverty level, regardless of whether they are on welfare.
Despite these expanded eligibility levels for pregnant women, Medicaid does not offer extensive coverage of substance abuse treatment during pregnancy. Historically, substance abuse services have been considered to be under the purview of state substance abuse agencies. As abuse of illicit drugs continued to increase and treatment options expanded, concurrent with an expansion in Medicaid eligibility for pregnant women, states began to cover some substance abuse treatment services for pregnant and parenting women under existing Medicaid-mandated and optional services. However, most residential treatment is still excluded from either mandatory or optional services unless it is provided in a hospital. In particular, services in institutions for mental disease (IMDs) -- residential facilities with sixteen or more beds that provide behavioral health care including substance abuse treatment services - are excluded for people between twenty-two and sixty-four years of age.
In fee-for-service reimbursement under Medicaid, most health services are documented individually in state claims files and have a code for the diagnosis (lCD-9-CM) and for the service provided. To the extent that the diagnosis of substance abuse is made and coded on the claim (whether or not the claim is for substance abuse treatment), the claim can be used to identify a substance abuser. However, ambulatory claims often contain only a single ("primary") diagnosis; if substance abuse is only an underlying condition and not the primary reason for the visit, it may not appear on the claim. Other codes such as provider types and categories of services can also be used to identify substance abuse providers and services. When substance abuse treatment is covered by Medicaid and the appropriate codes are used, the claim will show which treatments women receive during their pregnancy and the post-partum period. Claims data are used widely for health services research, and recently some studies have used claims to study the cost of substance abuse services (Garnick et al. 1996; Howell et al. 1997).
Interest in using Medicaid data for research has grown for several reasons. First, the automated data that are a byproduct of claims payment provide a profile of health service use for a large group of people while they are enrolled in Medicaid. Small, high-risk populations - many of whom would not be included in large enough numbers in population-based studies - can be identified and studied.
However, there are major problems for researchers who attempt to use Medicaid data for research. The large size of the files and issues surrounding the accuracy of the data present great challenges. Several articles and reports have highlighted these issues (Ray and Griffin 1989; Weiner et al. 1990; Romano 1993; McDonald and Hui 1991; Alpha Center 1995; Garnick, Hendricks, and Comstock 1994; Cherlow et al. 1991 ). One of the more important issues for studies of pregnant substance abusers that is pointed out by these authors is problems with the completeness and accuracy of diagnostic codes. For example, Cherlow et al. (1991) examined Medicaid data from twenty states and found that six states had serious problems with the completeness of inpatient hospital primary diagnosis codes and nine states had problems with completeness of primary diagnosis codes on ambulatory claims. Consequently, researchers recommend using other types of codes, such as procedure codes for the services received, in addition to diagnosis codes. In particular, since providers are not paid on the basis of diagnosis codes, they have a lower incentive for providing complete and accurate diagnosis codes than for procedure codes. None of the studies of the accuracy and completeness of claims data have addressed the issue of identifying pregnant substance abusers in Medicaid claims data.
State substance abuse treatment data. The third type of data set that can potentially be used to identify pregnant substance abusers, at least those who present for treatment, is the state substance abuse treatment administrative data that are collected when clients are admitted to state-funded treatment facilities. These data differ substantially across states, but a common recommended data set that was developed by the Substance Abuse and Mental Health Services Administration (SAMHSA) has helped to generate some degree of uniformity across states (Substance Abuse and Mental Health Services Administration 1999). The data set is known as the Treatment Episode Data Set ("TEDS").
This file contains, for those pregnant/post-partum women and others who receive substance abuse treatment, a great deal more information than from the other sources on their substance use and prior treatment history. It also contains more contextual information such as whether the mother lives with her children and whether she is employed. However, SAMHSA estimates that only about 60 percent of all substance abuse treatment facilities provide data to the state reporting systems (Substance Abuse and Mental Health Services Administration 1999). These data have rarely been used for research, and there is little known about the quality of the data.
Medicaid and state substance abuse files will underreport the number of pregnant substance abusers because women will only appear in these administrative records when they receive substance abuse treatment or some other kind of service. For example, those substance abusers who do not have prenatal care or any other Medicaid-funded service during the prenatal and post-partum period may not appear in the Medicaid files at all, except at the time of delivery. Unless signs of substance abuse are evident at delivery (e.g., fetal alcohol syndrome, a positive toxicology screen or some other clinical signs in the mother or infant), the substance abuse will not be reported on the delivery claim and probably not on the birth certificate as well.
Data linkage. As mentioned, one of the ways that administrative databases can be enhanced is through linkage of multiple types of data. In this way, variables that are in one file can be added to another.
Merging existing data sources is generally a lower-cost method to enhance a particular data source than collecting primary data. In the case of research on pregnant substance abusers, for example, the birth certificate might provide data on the infant's birth weight, while the Medicaid claims for mother and infant would provide information on their use and cost of services. Even though substance abuse is likely to be underreported in both data sources, adding identifiers from both will allow for more complete identification. Augmenting these two data sources with state substance abuse treatment data further enhances identification, since those pregnant women admitted for treatment form an important subset of all pregnant substance abusers.
Since there is a high interest in pregnancy outcomes among policy makers and researchers, there have been numerous record linkage studies that have used merged birth certificate and Medicaid files. These studies have developed various ways to perform the linkage from simple matches on name or unique number to complex probabilistic matching algorithms (Bell, Keesy, and Richards 1994). Studies have shown that there should be a very careful matching process using multiple matching variables and incorporating some hand matching. Without such efforts, merged files will be incomplete (Piper et al.1990; Buescher 1999). Since identifiers are needed for constructing linked files, there are serious confidentiality concerns with such research. Institutional Review Boards should be consulted regarding the protocol for obtaining and merging records, and negotiations with state agencies are likely to be lengthy and costly.
The current study provides an opportunity to build on previous work investigating the quality of birth certificate and Medicaid data, as well as efforts to merge these two data sources, in order to create a research file for studying the service use and cost of pregnant substance abusers. In this study, birth certificates and Medicaid data were augmented with data on a group of known substance abusers, those in substance abuse treatment. Creating and evaluating the quality of the linked database was part of a larger project that evaluated the impact of five demonstration programs to provide innovative prenatal care and substance abuse treatment in five states: Maryland, Massachusetts, New York, South Carolina, and Washington (Howell, et al. 1997). For purposes of this paper, we describe the methods used to identify pregnant substance abusers in all five states and then focus on two states for a more in-depth discussion. The two states were chosen because they offer a contrast in their types of administrative records and treatment systems.
For confidentiality reasons, the state staff merged the three data files prior to submitting them to the national evaluator. While procedures for merging differed somewhat by state, they all involved interactive matching using multiple matching variables such as social security number (generally not available in all sources), mother's date of birth, and mother's name (or an abbreviated version of the name), at a minimum. Usually some manual matching, which visually examined names, for example, was needed.
There were several different approaches that states used to identify pregnant substance abusers and merge their data. For example, one state began with the birth certificate to create its file and matched it to the Medicaid enrollment file, claims, and the state's substance abuse treatment file. Three states began with Medicaid delivery records and matched to the other sources. After the phase-in of managed care under Medicaid (beginning with 1993 data), one state reversed its process, choosing to begin with birth certificates rather than Medicaid delivery claims and then matching to any claim in the Medicaid files. This avoided the situation where delivery claims might be excluded from claim files but substance abuse treatment claims might be available due to a "carve out" of those services. The fifth state began with a group of women known to be pregnant substance abusers (identified in detox facilities) and matched them to the other data sources.
There was substantial variation in the ability of states to identify deliveries and match to all available data sources. For example, one state had substantial difficulty matching Medicaid claims to birth certificates, and another state had great difficulty finding all infant records.
The two states presented in this paper include one, "State A," with minimal managed care at the time of the study, that began its merger with Medicaid delivery claims. The second state, "State B," had some managed care and began with the birth certificate.
Once the merged files were available, we developed a complex, state-specific algorithm for identifying substance abusers from Medicaid claims using diagnosis codes, service codes, and codes for provider and residential treatment level. The algorithm used diagnosis codes that indicate alcohol or drug use, abuse, or dependence in either the mother or infant and service codes for detoxification, residential and outpatient chemical dependency treatment programs. The algorithms included only conditions directly associated with alcohol or drug use (such as withdrawal or methadone) and not other services that might identify substance abusers (such as AIDS or general psychiatric care). Since the standard procedure coding systems developed for reporting physicians' services such as CPT-4 contain almost no codes for substance abuse treatment, state-specific service codes were very important to include in the algorithm. Consequently, modifying the algorithm for each additional state required substantial effort.
HOW WELL DO THE THREE DATA SOURCES IDENTIFY SUBSTANCE ABUSERS?
We investigated the extent to which women were likely to be identified as substance abusers in each data source. Table 1 shows the total number of pregnant substance abusers identified in each ofthe two example states according to the data source that was used to identify the woman as a substance abuser. As shown, for example, in State A, 41.4 percent of the 451 pregnant substance abusers in the study were identified in the claims files only and not in any other source. Fourteen percent were identified from birth certificates only, and only 2.2 percent were identified only from substance abuse treatment data. Another 24.6 percent were identified in both claims and birth certificates and a smaller number in both claims and state substance abuse files. In contrast, in State B, fewer records were identified by birth certificates and more records were identified from the state substance abuse system. The .z2 test for differences in the two distributions confirms that the states are significantly different in how pregnant substance abusers are identified from the three sources.
It is not possible to know precisely why the two states are different without a more detailed record abstraction effort. However, two differences in the states' data systems may explain some of the differences. First, State A has more detailed questions regarding drug use on the birth certificate, and urine toxicology screens at delivery are more common in that state. This may explain the apparently more complete recording of drug use on birth certificates there. On the other hand, State B has a more established reporting system from substance abuse facilities, apparently improving reporting from that system and facilitating links to Medicaid files.
Table 1 shows (in the last line) the percentage of total delivering women identified as pregnant substance abusers in each state. The percentage was 8.3 percent in State A and 5.9 percent in State B. The lower reported prevalence is plausible for State B where the demonstration area was primarily rural.
In order to investigate how accurately birth certificates and Medicaid claims identify pregnant substance abusers, we examined linked records for pregnant women who were identified in state substance abuse treatment files. These women are certainly substance abusers because they have been admitted for treatment. Among these substance abusers, we looked in both birth certificate and claims files for an indication of substance use (Table 2).
As shown, almost all of these pregnant substance abusers in treatment were identified as substance abusers in Medicaid files (over 85 percent in both states). There was more variability between the states in the degree to which birth certificates identified substance abusers than in the degree to which claims did so. This suggests that - at least for women with serious substance abuse problems that lead to identification by providers - a linked Medicaid/birth certificate file identifies approximately 90 percent of those women, at least in the two states examined. However, at least 10 percent of these women were not identified in the linked Medicaid/birth certificate file, even in these two states that require drug use reporting on the birth certificates.
WHAT RESEARCH VARIABLES ARE AVAILABLE ON THE FILES?
Table 3 shows some basic descriptive information on identified pregnant substance abusers and other women who delivered and indicates the source of each descriptive variable. The table also compares identified pregnant substance abusers to other pregnant women, illustrating that merged birth certificate/Medicaid files provide important comparison information. For example, the data can be used to show that pregnant substance abusers were older on average than other pregnant women in State A (but not in State B), had smaller infants, and cost considerably more to the Medicaid program. The difference in total costs (including infant hospital costs) ranged from 60 percent greater in State A to 45 percent greater in State B.
CAN MEDICAID AND SUBSTANCE ABUSE TREATMENT DATA BE USED TO STUDY TREATMENT PATTERNS
In order to study treatment patterns, it was first necessary to develop and implement a uniform treatment categorization across states and files since different programs and states label services differently. Table 4 shows the treatment categories that we developed for the following services: (1) detoxification, (2) short-term residential, (3) long-term residential, (4) methadone, (5) least intensive outpatient, (6) moderately intensive outpatient, (7) most intensive outpatient, and (8) other.
The method for defining these categories varied somewhat by state and type of file. For the Medicaid files, we used procedure codes and other indicators to define detoxification, residential care, and ambulatory care. We did not have indicators of intensity in claims-based state procedure codes, so we used length of stay for residential care and number of visits for ambulatory care to distinguish between various levels of intensity. The "other" category includes claims from other ambulatory care providers that had substance abuse diagnoses. These are important to include because care for substance abusers may be provided by hospital clinics, psychologists, psychiatrists, social workers, and other providers, and these would otherwise be excluded. Another possible reason for the lack of overlap in the two systems could be that the crosswalk for types of treatment (Table 4) is inaccurate, and the two coding systems were not properly reconciled into common codes.
Treatment could be identified by duplicate treatment records from the Medicaid and state substance abuse treatment data sources since a record could appear in both sources for the same treatment episode. Table 5 shows that there was not a great deal of overlap in the monthly treatment records identified from the two sources, revealing that it is important to use both Medicaid and state treatment records to develop a full profile of treatment. In State A, only 5.4 percent of monthly treatment records were identified in both sources; the comparable percentage was also low in State B (7.4 percent). In State A, more records were identified from Medicaid files (74.2 percent); in State B more were identified from state substance abuse files (50.8 percent). Recall that earlier we speculated that the more established substance abuse reporting system in State B led to greater identification of pregnant substance abusers in that source. It is possible that treatment records from the state treatment system are also more complete in State B and are underreported in State A.
In order to illustrate the types of analyses that are possible with the merged data, Table 6 shows the types of treatment received by pregnant women in the two states in 1992. The types of treatments differed between states. For example, 21.5 percent of women received detoxification in State A, but only 4.3 percent did so in State B. If the detoxification and short-term residential categories are grouped, however, the state rates are similar, suggesting that either the short-term residential level is used in place of detoxification in State B or that the definitions of detoxification services are inconsistent between the states. Both states had low rates of use of long-term residential treatment, 2.4 percent of women in State A and 9.3 percent of women in State B. This level of care was not covered by Medicaid before the demonstration, but these services could have been provided in the state substance abuse treatment system; apparently they were received rarely during pregnancy. State A had a much higher proportion of women receiving methadone treatment during pregnancy (23.0 percent) than did State B (2.9 percent).
Because of their high cost and poor birth outcomes, pregnant substance abusers are a group that is of interest to researchers. However, a lack of data has hampered such research. This paper has shown that it is possible to link three data sources available in all states - birth certificates, Medicaid files, and state substance abuse treatment records - to create a cross-state analysis file for studying the birth outcomes, use of services, and cost for pregnant substance abusers. The database can be used to identify many, but not all, pregnant substance abusers. In particular, those who do not obtain any services from the Medicaid program or from the state substance abuse treatment system will be underidentified since birth certificates do not provide reliable coding of substance abuse at the current time. Consequently, efforts at more complete identification of pregnant substance abusers should be undertaken, perhaps through increased efforts to improve reporting on the birth certificate, which is a universal document.
The paper further shows that combining the three data sources provides a rich analytical file that is not available from any single administrative data source. Such a merged file also provides data on a comparison population, Medicaid pregnant women who are not identified as substance abusers or who are not in treatment.
Researchers are cautiously encouraged to use linked administrative records for research on substance abuse treatment, cost, and outcomes for those women enrolled in Medicaid who are receiving treatment. Since identifiers are needed for constructing linked files, there are serious confidentiality concerns with such research. The potential benefits of research for improving treatment and other services to this high-risk group make it important to develop methods of using existing data while working to improve accuracy and timeliness of data sources for future research and policy development. The methods described here provide one such approach to data development.
1995 Working with large insurance databases: Avoiding and overcoming the pitfalls. Data Resource Guide.
Bell, R., J. Keesey, and T. Richards
1994 The urge to merge: Linking vital statistics records and medicaid claims. Medical Care 32(10):1004-1018.
Buescher, P., K. Taylor, M. Davis, and J. Bowling
1993 The quality of the new birth certificate data: A validation study in north carolina. American Journal of Public Health 83(8):1163-1165.
1999 Method of linking medicaid records to birth certificates may affect infant outcome statistics. American Journal of Public Health 89(4):564-566.
Bruce, F.C., M. Adams, H. Shulman, M. Martin, and the PRAMS Working Group 1993 Alcohol use before and during pregnancy. American Journal of Preventive Medicine 9(5):267-273.
Centers for Disease Control and Prevention
1994 From Data to Action: CDC's Public Health Surveillance for Women, Infants, and Children. Washington, DC: U.S. Department ofHealth and Human Services.
Chasnoff, I. J.
1988 Drug use in pregnancy: Parameters of risk. The Pediatric Clinics of North America 35 (6):1403-11.
Cherlow, A., M. Ellwood, E. Howell, K. Miller, and S. Dodds
1991 Data Quality in the Medicaid Statistical Information System (MSIS). Washington, DC: Mathematica Policy Research, December.
Dobie, S., L. Baldwin, R. Rosenblatt, M. Fordyce, C. Holly, A. Andrilla, and L. Hart
1998 How well do birth certificates describe the pregnancies they report?
The Washington state experience with low-risk pregnancies. Maternal and Child Health Journal 2(3):145-154.
Freedman, M.A., G.A. Gay, J.E. Brockert, et al.
1988 The 1989 revisions of the U.S. standard certificates of live birth and death and the U.S. standard report of fetal death. American Journal of Public Health, 78(2): 168-172.
Garnick, D., A. Hendricks, and C. Comstock
1994 Measuring quality of care: Fundamental Information from administrative datasets. International Journal for Quality in Health Care 6(2):163-177.
Garnick, D. W., C. M. Horgan, A. M. Hendricks, and C. Comstock
1996 Using health insurance claims data to analyze substance abuse charges and utilization. Medical Care Research and Review 53(3):350-68. Horgan, C., M. Rosenbach, E. Ostby, and B. Butrica
1991 Targeting special populations with drug abuse problems: Pregnant women. In Drug Abuse Services Research Series, No. 1. 123-42. Washington, DC: U.S. Department of Health and Human Services. Howell, E. M., and G. Brown
1989 Prenatal, delivery and infant care under medicaid in three states. Health Care Financing Review 10(4):1-15.
Howell, E., N. Heiser, C. Thornton, D. Ewell, and M. Mason
1997 Final Report: Evaluation of the Demonstrations to Improve Access to Care for Pregnant Substance Abusers. Washington, DC: Mathematica Policy Research.
Howell, E., N. Heiser, and M. Harrington
1999 A review of recent findings on substance abuse treatment for pregnant women. Journal of Substance Abuse Treatment 16(3):195-219. Kleinman, J.
1991 Methodological issues in the analysis of vital statistics. In Reproductive and Perinatal Epidemiology, ed. M. Kiely, CRC Press, Inc. Kotelchuck, M.
1994 An evaluation of the kessner adequacy of prenatal care index and a proposed adequacy of prenatal care utilization index. American Journal of Public Health 84(9):1414-1420.
Krieger, J. W., F. A. Connell, and J. P. LoGerfo.
1992 Medicaid prenatal care: A comparison of use and outcomes in fee-for-- service and managed care. American Journal of Public Health 82(2):185-90.
McDonald, C., and S. Hui
1991 The analysis of humongous databases: Problems and promises. Statistics in Medicine 10:511-518.
Miller, L., T. Shaikh, C. Stanton, A. Montgomery, R. Rickard, S. Keefer, and R. Hoffman
1995 Surveillance for fetal alcohol syndrome in Colorado. Public Health Reports 110:690-697.
National Center for Health Statistics
1985 Comparability ofReporting Between the Birth Certificate and the 1980
National Natality Survey. Washington, DC: U.S. Department of Health and Human Services.
National Center for Health Statistics
1993 Comparability of the Birth Certificate and 1988 Maternal and Infant Health Survey. Washington, DC: U.S. Department of Health and Human Services.
National Governor's Association.
1999 http://www.nga.org/Pubs/IssueBriefs/1999/ 990210MCHUpdate.asp. National Institute on Drug Abuse
1996 National Pregnancy and Health Survey: Drug Use among Women Delivering Livebirths: 1992. NIH Publication Number 96-3819. Rockville, MD: U.S. Department of Health and Human Service.
O'Campo, P., M. A. DeBoer, R. R. Faden, A. Gielsen, N. Cass, and R. Chaisson 1992 Discrepancies between women's personal interview data and medical
record documentation of illicit drug use, sexually transmitted diseases, and hiv infection. Medical Care 30(10):965-71.
Piper, J., W. Ray, M. Griffin, R. Fought, J. Daughtery, and E. Mitchell, Jr.
1990 Methodological issues in evaluating expanded medicaid coverage for pregnant women. American Journal of Epidemiology 132(3):561-571. Piper, J., E. Mitchell, M. Snowden, Jr., C. Hall, M. Adams, and P. Taylor
1993 Validation of 1989 Tennessee birth certificates using maternal and newborn hospital records. American Journal of Epidemiology 137(7):758-768.
Piper, J. M., E. F. Mitchell, and W. A. Ray
1994 Presumptive eligibility for pregnant medicaid enrollees: its effects on prenatal care and perinatal outcome. American Journal of Public Health 84( 10):1626-30.
Ray, W., M. Griffin
1989 Use of Medicaid data for pharmacoepidemiology. American Journal of Epidemiology 129(4):837-849.
1993 Can Administrative data be used to compare the quality of health care? Medical Care Review 50(4):451-477.
Sackoff, J., J. Kline, A. Kinney, and A. Grunebaum
1992 Cocaine Use in Obstetric Patients Underreported. American Journal of Public Health 82(7):1043.
Slutsker, L., R. Smith, G. Higginson, and D. Fleming
1993 Recognizing illicit drug use by pregnant women: Reports from Oregon birth attendants." American Journal of Public Health 83( 1 ):61-64. Substance Abuse and Mental Health Services Administration
1999 National Admissions to Substance Abuse Treatment Services: The Treatment Episode Data Set (TEDS) 1992-1997. Washington, DC: U.S. Department of Health and Human Services.
1996 In Wisconsin, a rarity of a fetal-harm case: Attempted-murder charges for alcoholic. New York Times, 17 August.
Watkins, M., L. Edmonds, A. McClearn, L. Mullins, J. Mulinare, and M. Khoury 1996 The surveillance of birth defects: The usefulness of the revised US standard birth certificate. American Journal of Public Health 86(5):731-734.
Weiner, J., N. Powe, D. Steinwachs, and G. Dent
1990 Applying insurance claims data to assess quality of care: A compilation of potential indicators. Quality Review Bulletin 16(12):423-438.
-EMBRY M. HOWELL, NANCY HEISER, ANN CHERLOW, MEI-LING MASON, DAISY EWELL, SUZANNE ROTWEIN
Embry M. Howell is a vice president at Mathematics Policy Research. She has been the principal investigator for several studies that have used linked Medicaid data, vital statistics, and other secondary data sets. Nancy Heiser is a doctoral student in psychology at the University of Maryland. At the time of this study, she was a research analyst at Mathematics Policy Research. Ann Cherlow is a senior researcher at Mathematics Policy Research, specializing in Medicaid claims and encounter data. Mei-Ling Mason is a senior programmer at Mathematics Policy Research. She has been involved in several research projects that have linked Medicaid data sets with other files. Suzanne Rotwein is a research analyst at the Center for Beneficiary Services at the Health Care Financing Administration, U.S. Department of Health and Human Services.…