Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
â4â Methodological Issues and the 2003 Revision of Standard Instruments T HE CONTENTS OF THE RECORDS of vital events in the United States have not been stagnant over time. In its role of coordinating collec- tion at the national level, the National Center for Health Statistics (NCHS) periodically revises the recommended U.S. standard certiï¬cates of birth, death, marriage, and divorce, and related records. However, in the Vital Statistics Cooperative Program, use of these standard certiï¬cates is not legally binding on the states and registration areas, although consistency in the data collection instrument certainly affects the quality of resulting data. In 1998, NCHS and the vital statistics stakeholders began the most re- cent revision of the standard instruments, resulting in the 2003 release of new instruments. The 2003 revision marked the 12th revision of the birth certiï¬cate and the 11th revision of the death certiï¬cate, and both marked the ï¬rst revision of the standard certiï¬cates since 1989. A particularly im- portant feature of the new certiï¬cates was revision of items for collecting information on race and Hispanic origin, in compliance with new U.S. Of- ï¬ce of Management and Budget (OMB) regulations to permit the reporting of multiple-race categories. However, for a variety of reasons, the vital reg- istration areas have been slow to adopt the new form of the certiï¬cates, leading to signiï¬cant methodological challenges in recent years. The patch- work of adopting and nonadopting registration areas has forced attention to how data on race can and should be tabulated in vital statistics and used in such important applications as population estimates; the uneven imple- mentation of other new demographic and health data items on the standard 49
50 VITAL STATISTICS certiï¬cates has also limited capacity to use vital statistics data for research purposes. This chapter describes the 2003 revised instrument in some additional detail, consistent with the material presented at the workshop. Three work- shop presenters from the NCHS staff and one Census Bureau speaker com- mented on the methodological changes inherent in the 2003 revision and the types of analyses made possible by new variables added in the revi- sion. Speciï¬cally, the presentations focused on the complications involved in working with the data in the current situation in which implementation of the revised certiï¬cate is uneven and, hence, states report information in varying formats. 4âA THE 2003 REVISIONS NCHS convened an expert panel in 1998, consisting principally of state vital registration ofï¬cials, as well as representatives from relevant user orga- nizations, to begin the process of evaluating the content of the existing (1989 revision) birth and death certiï¬cates and recommend changes. The panel (Division of Vital Statistics, 2000) developed its ï¬nal recommendations in 1999 and directed that NCHS test redesigned instruments; the resulting documents became the 2003 standard certiï¬cates, and they are reproduced in Appendix D. The major changes to the standard certiï¬cates are described in brief in Box 4-1. As Jennifer Madans (NCHS) observed in her workshop presenta- tion, the 2003 revision continued a long-term push to make the vital records a platform for collecting a variety of public health data items in order to meet real public health needs. The revised 2003 birth certiï¬cate now includes some 60 data items, providing extensive information on pregnancy, labor and delivery, infant health, and maternal health factors; the 2003 round speciï¬cally added queries on risk factors (smoking) and method of delivery. Arguably the most signiï¬cant change made in both instruments was de- scribed in more detail by workshop presenter James Weed (former deputy director, Division of Vital Statistics, NCHS; retired): modiï¬cation of the questions on race and Hispanic origin to reï¬ect new standards promulgated by OMB in 1997. As Weed summarized, the standards (U.S. Ofï¬ce of Man- agement and Budget, 1997): ⢠established a minimum set of race categories that were made manda- tory for statistical data collections: American Indian or Alaska Native; Asian; black or African American; Native Hawaiian or Other Paciï¬c Islander; and white; ⢠deï¬ned a minimum set of categories for collection of Hispanic or eth- nic origin: âHispanic or Latinoâ or ânot Hispanic or Latinoâ; and
METHODOLOGICAL ISSUES 51 Box 4-1 Major Changes to the U.S. Standard Certiï¬cates for Vital Events, 2003 Revision U.S. Standard Certiï¬cate of Live Birth ⢠New items â Fertility therapy â Use of WIC (Special Supplemental Nutrition Program for Women, Infants, and Children) funds to obtain food during pregnancy â Infections during pregnancy â Maternal morbidity â Breast feeding â Principal source of payment for the delivery ⢠Modiï¬ed items â Motherâs and fatherâs race, for compliance with U.S. Ofï¬ce of Management and Budget (OMB) standards â Motherâs and fatherâs education, to record highest degree attained by both â Level of smoking before and during pregnancy â Method of delivery question, to include trial of labor prior to cesarean delivery and other categories â Pre-pregnancy weight of mother and weight and height of mother at delivery â Congenital anomalies ⢠Related documentation â âWorksheetâ to be ï¬lled by mother, giving self-response to such items as personal characteristics and program participation â âWorksheetâ to be ï¬lled by facility (and accompanying guide), based on medical records, covering items such as birth weight and method of delivery â Detailed speciï¬cations and instructions for every element in the electronic birth certiï¬cate system â The U.S. Standard Report of Fetal Death was modiï¬ed for conformity with revised items on the new birth certiï¬cate; the record also queries for other signiï¬cant causes of death in addition to a single initiating cause and adds items on autopsy or histological placental examination U.S. Standard Certiï¬cate of Death ⢠New items â Maternal mortality (pregnancy status at time of death) â Decedentâs role in the event of death due to transportation injury (e.g., passenger, driver) â Tobacco use and contribution to death ⢠Modiï¬ed items â Decedentâs race, for compliance with OMB standards â Decedentâs education, to record highest degree attained â Decedentâs marital status, to distinguish âmarriedâ category from âmarried, but separatedâ â Place of death, to include hospice facility ⢠Related documentation â Revision of funeral directorâs handbook for completing death certiï¬cate â Revision of physiciansâ and medical examiner/coronersâ handbooks to focus on accurate collection of cause-of-death information â Detailed speciï¬cations and instructions for every element in the electronic birth certiï¬cate system â Certiï¬cate includes separate instructions speciï¬cally for funeral director and for person completing medical certiï¬cation portion SOURCES: Workshop presentations summarized in this chapter; Division of Vital Statistics (2002a,b).
52 VITAL STATISTICS ⢠indicated a preference for asking the two questions (race and Hispanic origin) as separate items, with the Hispanic origin question preceding the race question. The ï¬ve base racial categories differed from previous standards by splitting the previous Asian and Paciï¬c Islander category into two and by intending to be inclusive (i.e., there was no âOtherâ option). However, the signiï¬cant change in the 1997 OMB standards was to permit multiple selections rather than just one, allowing respondents to self-identify with more than one race category. Weed commented that the directive also requested that agencies show as much multiple-race detail in its tabulations as possible, subject to data quality and conï¬dentiality standards, and that agencies not ask respon- dents indicating more than one race to pick one as a âmainâ or âprimaryâ identiï¬cation. Many vital records jurisdictions have been slow to fully implement the 2003 revised certiï¬cates. A survey of 52 of the 57 jurisdictions by Friedman (2007:Table 1) suggests that the 1989 revision of the standard birth and death certiï¬cates won rapid acceptance by the local authorities.1 Speciï¬cally, Friedman (2007:Table 1) found that 50 jurisdictions implemented the new standard birth certiï¬cate in 1989 with the other two areas complying within 2 years; 48 jurisdictions supplied data using the new death certiï¬cate format in 1989, three followed in 1990, and only one took longer to implement (in 1997). By comparison, only two and ï¬ve states used the 2003 revised birth and death certiï¬cate formats in 2003, respectively. As of the time of the Friedman (2007) surveyâmore than 3 years laterâ27 and 26 jurisdictions had not yet implemented the new birth and death certiï¬cates, respectively. Several workshop speakers presented maps depicting the current level of implementation of the standard certiï¬cates; this information is summarized in Table 4-1. Speaker Robert Anderson (NCHS) added that six registration areas are planning on implementing the revised death certiï¬cate in 2009 and another seven in 2010. 4âB RACE AND ETHNICITY Weed noted that, at the time of the 1997 establishment of OMBâs revised standard for race and Hispanic origin questions, agencies were required to implement the new standards by January 1, 2003 (with some allowance for a âbridgingâ period between single-race and multiple-race reportingâ discussed further, below). In particular, the Census Bureau implemented the new categories and âmark all that applyâ approach to the questionnaires used in the 2000 census. However, Weed commented that the vital statistics 1 The Friedman (2007) analysis was mentioned and commended by Harry Rosenberg in his workshop presentation.
Table 4-1 Adoption of 2003 Revised Certiï¬cates and Multiple-Race Reporting for Births and Deaths, by State, 2005 Using 2003 Multiple-Race Using 2003 Multiple-Race Revised Reporting Revised Reporting Certiï¬cate? Allowed? Certiï¬cate? Allowed? State Births Deaths Births Deaths State Births Deaths Births Deaths Alabama Montana Alaska Nebraska Arizona Nevada METHODOLOGICAL ISSUES Arkansas New Hampshire California New Jersey Colorado New Mexico Connecticut New York Delaware North Carolina District of Columbia North Dakota Florida Ohio Georgia Oklahoma Hawaii Oregon Idaho Pennsylvania Illinois Rhode Island Indiana South Carolina Iowa South Dakota Kansas Tennessee Kentucky Texas Louisiana Utah Maine Vermont Maryland Virginia Massachusetts Washington Michigan West Virginia Minnesota Wisconsin Mississippi Wyoming Missouri Total 12 17 17 21 NOTES: , Yes; , No; , Partial (for only part of data year 2005 or only for selected facilities). âNew Yorkâ is New York state less New York City, which reports separately. Reporting status in Puerto Rico and the territories not shown in this table. SOURCES: Data on births from Martin et al. (2007:89, 93); data on deaths, from Kung et al. (2008:105, 107). 53
54 VITAL STATISTICS program was granted a variance or exception to this deadline, because of the difï¬culty in achieving simultaneous compliance by all the vital records re- porting agencies. When changes were being made to the standard birth and death certiï¬cates in 1999â2000 (what would eventually become the 2003 revisions), Weed said that NCHS decided that its best approach would be to try to emulate the phrasing and construction of the census questions to the greatest extent possible. This would not only permit (eventual) compa- rability between the two resulting data sources, but would also allow the use of similar coding and editing processes (e.g., to resolve inconsistent or re- dundant write-in responses) by the two agencies (Division of Vital Statistics, 2004). The certiï¬cate revisions followed the same category breakdowns as the 2000 census questionnaire for the Hispanic origin question but differed slightly from the census instrument by more detailed splitting of the Asian and Paciï¬c Islander categories (the census permitted only one space to write in either an âOther Asianâ or âOther Paciï¬c Islanderâ afï¬liation). At the time of the workshop in 2008, the most recent issues of ï¬nal vital statistics covered data year 2005 (Kung et al., 2008; Martin et al., 2007). For that year, Weed said that multiple-race reporting for deaths had been implemented in 22 of the 57 vital record jurisdictions and in 17 jurisdictions for births.2 For those jurisdictions, multiple-race reporting for decedents (information from next of kin) was indicated on 0.4 percent of the records, more frequently for younger decedents (2.4 percent of those under 25 years of age) than for older decedents (0.3 percent of those over 64). Higher levels of multiple-race reporting are indicated in the birth records for the available statesâ1.5 percent, ranging geographically from 0.4 percent (Texas) to 36.6 percent (Hawaii). Weed noted that NCHS had received multiple-race data for deaths from 30 jurisdictions in 2007 and anticipates compliance by 40 jurisdictions in 2008; for births, 30 jurisdictions reported multiple-race data in 2007, ex- pected to grow to 47 in 2008. Weed observed that, until all states consis- tently report multiple-race data in the same manner, the vital records suffer from two fundamental compatibility problems. First, they are not compat- ible between jurisdictions, which may differ in their reporting schemes. In particular, one approach that has been used in the past (and may still be in use in some localities) for coding multiple-race entries from a birth or death 2 According to Weedâs presentation, multiple-race reporting in 2005 was done for both births and deaths in California, Florida, Hawaii, Idaho, Kansas, Minnesota, Nebraska, New Hampshire, New York state, South Carolina, Utah, and Washington. Multiple races for deaths only were reported in Connecticut, Maine, Michigan, Montana, New Jersey, New York City, Oklahoma, South Dakota, Wisconsin, and Wyoming (with the District of Columbia reporting for part of the year). Reporting for births only occurred in Kentucky, Ohio, Pennsylvania, Tennessee, and Texas (with Michigan and Vermont reporting for part of the year or only for selected facilities).
METHODOLOGICAL ISSUES 55 certiï¬cate is to record only the ï¬rst one mentioned. Second, the state-level records are not compatible with the data collected in the decennial census (or the new American Community Survey) or with the Census Bureauâs in- tercensal population estimates. Relatively little is yet known about the characteristics of persons who report multiple race categories in comparison with those who report a single race, the internal cognitive weightings that may go into such determinations (e.g., consistent reporting of American Indian or Native Hawaiian ancestry), or whether the multiple categories accurately capture an individualâs sense of racial and ethnic identity. 4âB.1 Bridging Single-Race and Multiple-Race Data at NCHS Weed mentioned some insights that have been derived from analysis of data from the National Health Interview Survey (NHIS) and work on âbridgingâ generally: using statistical modeling of multiple-race responses to attempt to identify the one single response that an individual would have reported under the old single-race standard. The NHIS (ï¬elded by NCHS) began permitting multiple responses to race questions in 1982; respondents could choose up to two categories through 1996 and as many as ï¬ve categories beginning in 1997. Those respondents checking more than one category were also prompted to name their primary or main race afï¬liationâthe one they say would best repre- sent them. With the promulgation of the new OMB standards, NCHS won a special variance for the NHIS to continue to ask the follow-up question on a preferred single-race category. The resulting data have played a key role in developing methodology for bridging multiple-race responses to single- race categories: using the percent distribution of the âpreferredâ/single-race categories for each multiple-race combination to proportionally allocate per- sons to a single race. Weed presented three speciï¬c examples from 4 years (1997â2000) of the NCHS data: ⢠For persons reporting the two-race combination of black and white, 45.4 percent identiï¬ed black as the single-race category that best rep- resents them. Persons indicating white as a primary race category and those indicating no preferred single race were about evenly divided, at 26.9 and 27.7 percent, respectively. ⢠For persons reporting the two-race combination of American In- dian/Alaskan Native and white, the distribution was more lopsided: 74 percent identiï¬ed white as their primary race afï¬liation, 21.2 per- cent American Indian or Alaskan Native, and 4.8 percent no primary race.
56 VITAL STATISTICS ⢠For persons indicating the three-race combination of white, black, and American Indian/Alaskan Native, the modal response was to indicate no primary race afï¬liation (57 percent); 27.6 percent indicated black as the primary preference and 8.5 percent and 6.9 percent chose white and American Indian, respectively. For multiple-race combinations with enough reports in the 4 years of NHIS data to support further analysis, NCHS developed its algorithm for bridging or allocating a single-race response using multinomial logistic re- gression. The regression models included county-level covariates (urban- ization level, log percentage of single main-race reporting in the county, percentage of multiple-race reporting in the county); NCHS has shared its bridging algorithm with the Census Bureau to use in analyzing and develop- ing its population estimates. In Weedâs assessment, the algorithm provides reasonably consistent numerators and denominators for such estimation pur- poses until all birth and death counts are available in the new multiple-race format. NCHS staff also contributed to an analysis of the performance of the bridging algorithm on the multiple-race entries to the 2000 census; see Ingram et al. (2003). Going forward, Weed suggested that vital statistics data would be an im- portant proving ground for better understanding the characteristics of per- sons who report multiple-race afï¬liations. In large part, this is because of sheer numbers: along with the decennial census and the American Commu- nity Survey, data on births and deaths are one of the data sources in which multiple-race combinations are reported in sufï¬cient quantities to support detailed analysis. Weed noted that NCHS colleagues had already begun some studies in this regard, including the analysis by Hamilton and Ven- tura (2007) of births to mothers who report a single race or multiple-race afï¬liations in California and selected other states. 4âB.2 Bridging Single-Race and Multiple-Race Data at the Census Bureau In her workshop remarks, Victoria Velkoff (U.S. Census Bureau) com- mented on the challenges of working with the new multiple-race categories in the Census Bureauâs population estimates program and reconciling them with the different race measurement in the vital statistics program. Velkoff acknowledged that the need to bridge from the âoldâ race categories in vital statistics records to new categories complicates the estimation process. Velkoff indicated that most of the birth data that the Census Bureau re- ceives directly from the states do not use the current race categories. Hence, the Census Bureau uses models to develop the birth component of the popu- lation estimates using the race information for parents included on the birth certiï¬cate and the distribution of family composition for multiracial families from the 2000 census. For death data, Velkoff noted the fundamental chal-
METHODOLOGICAL ISSUES 57 lenge that the decedent race information on death certiï¬cates is, by default, a proxy report rather than a self-report. Although the race data are sup- posed to be provided by a family member or next of kin, it sometimes falls to a funeral director or medical examiner to ï¬ll in that category, resulting in inconsistencies. These inconsistencies (including, perhaps, incomplete re- porting of American Indian heritage or multiracial combinations) carry over to affect the intercensal population estimates. For its adjustment to the base population in constructing estimates, Velkoff indicated that the Census Bureau is also still grappling with another feature of race reporting in the 2000 census. The 2000 census question- naire permitted respondents to choose multiple options from the ï¬ve race categories deï¬ned in the 1997 OMB standards but also included a sixth op- tion, âsome other race.â For its population estimates, Velkoff said that the Census Bureau distributes the âsome other raceâ responses into the standard categoriesâa second level of bridgingâso that the estimates do not include tabulations for âsome other race.â3 The problem of working with varying race categories has been more acute in the area of population projections because, historically, the projec- tions have used a coarser categorization. Fred Hollmann (Census Bureau) said that, prior to 1986, the Census Bureau performed projections for only three categoriesâwhite, black, and other racesâwith âotherâ implied as a residual. In 1986, the Bureau performed a special projection of the Hispanic population through 2080, using vital statistics and birth certiï¬cate data for Hispanic births in 22 states. At that time, the Census Bureau deemed the death certiï¬cate reporting of Hispanic origin to be insufï¬cient in detail, and so it relied instead on life tables developed for the state of California (aug- mented by Medicare data). By 1993, the population projection program began the practice of crossing Hispanic origin with four race categories. As of the 2004 release of interim projections, the Census Bureau is using the same 31 race and multiple-race categories, crossed with Hispanic or not Hispanic, as the population estimates program. Its products use what has been called a âmin-maxâ approach, deï¬ning only two values for each of the ï¬ve race categories: those who report the race as the only race and those who report the race either alone or in combination with (any number of other) categories. 3 The Census Bureau tested alternatives to the race and Hispanic origins in its 2003 and 2005 tests that omitted âsome other raceâ as a response option. Howeverâpresumably con- cerned that respondents who strongly self-identify with their Hispanic origin as a race (rather than a separate ethnicity category) choose âsome other raceâ as the most ï¬tting optionâ congressional appropriators have forbidden such changes. First enacted in the Consolidated Appropriations Act of 2005 (P.L. 108-447) and repeated in some subsequent reports, appro- priators directed that ânone of the funds provided in this or any other Act for any ï¬scal year may be used for the collection of Census data on race identiï¬cation that does not include âsome other raceâ as a [category].â
58 VITAL STATISTICS Internally, though, the Bureauâs models of fertility and mortality rates are still principally based on three main categories: white, non-Hispanic black, and other. Hollmann suggested that the Census Bureau plans to revisit its projection strategy after all the states adopt the 2003 revised standards for birth and death certiï¬cates and report race in a consistent fashion. Hollmann said that NCHSâs switch in 1989 from tabulating vital statis- tics on birthâfrom the (imputed) race of the child to the race of the motherâwas a signiï¬cant difference for the population projections pro- gram. Initially, the Census Bureau tried a âworkaroundâ of adapting race of mother to infer race of child using some information on racial characteris- tics of families from a fertility supplement to the Current Population Survey. However, that approach proved inconvenient and burdensome; accordingly, the Bureau elected to base its projections primarily on the race of the mother in determining the âage 0â population of newborns in each subsequent year. With the promulgation of the 1997 standards, the Bureau changed again to an imputation strategy, deriving a childâs race from the race of both par- ents. However, Hollmann reported that the imputation process continues to be somewhat problematical (though arguably more so for the population estimates than for the projections) and an area of continued work. 4âC FETAL DEATHS AND INFANT HEALTH RISK FACTORS In the U.S. vital records system, reports of fetal deaths are completed sep- arately from certiï¬cates of birth and death. The fetal death report contains additional questions on the cause and condition of the death, demographic and health information on the mother, record of previous prenatal care, and risk factors involved in the pregnancy. The Report of Fetal Death is period- ically revised in the same manner as the Standard Certiï¬cates of Birth and Death; as shown in Box 4-1, the standard record was revised in 2003 to ex- pand collection of cause-of-death data and include additional questions also added to the standard birth certiï¬cate. Stephanie Ventura (chief, Reproductive Statistics Branch, NCHS) noted that the process of adoption of the 2003 revision of the U.S. Report of Fetal Death has been slower than that of the standard birth certiï¬cate (as shown in Table 4-1). By 2006, only 19 states had adopted the 2003 revision of the birth certiï¬cate; Ventura said that those states account for about half of all births in the country. NCHS anticipated that adoption of the new form in about eight additional states and New York City by 2008 would push that coverage to about two-thirds of all births. The states that are using the 2003 format and questions are an incom- plete set, but NCHS has generally found them (particularly a 12-state subset as of 2005) to be representative of the national population in terms of racial
METHODOLOGICAL ISSUES 59 120 Singletons Multiples1 106.8 90 86.2 Per 1,000 births 72.6 60 56.8 57.8 38.7 41.7 30.5 30 17.3 12.8 0 All ages Under 20 20â29 30â39 40 and over 1 Years Twins, triplets, and other higher order multiple births. SOURCE: CDC/NCHS National Vital Statistics System. Figure 4-1 Rates of gestational diabetes by age of mother and plurality, 12-state reporting area, 2005 SOURCE: Menacker and Martin (2008:Fig. 1). and ethnic composition. Ventura summarized ï¬ndings from recent NCHS reports that use the new additional information included on both the birth certiï¬cate and record of fetal death. In addition to restriction to the 12-state subset, Ventura discussed what NCHS terms âreleasableâ dataânew check- box items in existing data items that are already part of the national data set and that the states have authorized NCHS to release. ⢠DiabetesâPrior to the 2003 revision, only a single check box was available to indicate diabetes as a risk factor; the new certiï¬cate in- cludes a distinction between preexisting and gestational diabetes. Fig- ure 4-1 illustrates reported gestational diabetes by the age of the mother and plurality of birth (i.e., whether the birth is a singleton or a multiple delivery). The data suggest that the risk of gestational diabetes is elevated for older mothers, regardless of plurality. ⢠Racial disparitiesâSome important racial disparities can be found through analysis of birth certiï¬cate data. Figure 4-2 shows the receipt of surfactant replacement therapy by gestational age, a procedure to directly provide surfactant before the lungs are capable of producing it naturally, to make lung expansion and breathing easier. As is to be expected, the data show that the procedure is most frequently used for extremely preterm births for which it is most necessary to prevent respiratory distress syndrome. However, the data also suggest that white births are much more likely to receive this therapy than black or Hispanic births, regardless of gestational age. Similarly, Figure 4-3 illustrates receipt of steroids by the mother, prior to delivery, for fe- tal lung maturation. Again, the procedure is most frequent for very
60 VITAL STATISTICS Non-Hispanic white Non-Hispanic black Hispanic 180 150 Per 1,000 births 120 90 60 30 0 All gestations Less than 34 Less than 28 28â31 32â33 Weeks of gestation SOURCE: CDC/NCHS National Vital Statistics System. Figure 4-2 Rate of surfactant therapy by gestational age and race and Hispanic origin of mother, 12-state reporting area, 2005 SOURCE: Menacker and Martin (2008:Fig. 6). All gestations Non-Hispanic white Non-Hispanic black Hispanic 180 Per 1,000 births 120 60 0 All gestations Less than 34 34â36 Less than 37 37 or more Weeks of gestation SOURCE: CDC/NCHS National Vital Statistics System. Figure 4-3 Steroids for fetal lung maturation received by the mother prior to delivery, by gestational age and race and Hispanic origin, 12-state reporting area, 2005 SOURCE: Menacker and Martin (2008:Fig. 4). preterm births, but white births are much more likely to receive this therapy regardless of gestational age. ⢠Admission to neonatal intensive care unitsâAnalysis of distribution of admissions (for these purposes, restricted to singleton births) to neonatal intensive care units (NICUs) by gestational age shows an un- expected trend. The natural expectation is that preterm births would dominate those admissions, requiring the additional specialized care. However, almost half (47.5 percent) of those admissions were term or full term (37 or more weeks) at birth. Ventura noted that the factors leading to NICU admission are important to understand in planning for health care services, and NICU admission was not included as a speciï¬c factor on the standard birth certiï¬cate until the 2003 revision.
METHODOLOGICAL ISSUES 61 In addition to its work with the new data items, Ventura noted that NCHS is pursuing research and improvement efforts in its traditional birth and infant health data. Particular effort is being put toward improving the data quality of the fetal death ï¬les. In addition, the generation of a linked birth and infant death data set (described in Box 1-1) has been a particular asset for research. Ventura said that the 2006 data suggest that about one in eight births in the United States is preterm, a percentage that has been grow- ing and that heightens the importance of studying preterm-related causes of infant deaths and the factors that contribute to successful deliveries. Still, Ventura said that NCHS is beginning to focus predominantly on the revised data, even though transition to the new certiï¬cates is still in progress. Part of NCHSâs strategy in this regard is continued improvements in data availability. Birth and perinatal data are available for online tabulation and mapping at NCHSâs VitalStats site (http://www.cdc.gov/nchs/VitalStats.htm), and more data sets and documentation are directly downloadable from the NCHS site as well (births from 1968 through 2005; linked birth-infant death ï¬les from 1983 through 2004). However, Ventura emphasized that the major goal for the revision is not simply adding new data items, but also improving the data quality of the overall vital statistics series. Although resources for evaluation have been limited for a number of years, Ventura argued that the addition of the new data makes a program of data validation essential; it is critically important to assess the consistency of reporting in new and old data items and their comparability. She said that NCHS would like to be able to more regularly initiate special projects with individual states or groups of states to focus on data collection and data quality issues. 4âD MORTALITY AND CAUSES OF DEATH Robert Anderson (chief, Mortality Statistics Branch, NCHS) commented on the major methodological issues faced by NCHS with regard to the mor- tality data, in particular the handling of cause-of-death data. In contrast with birth data, fewer states have adopted the 2003 revised standard certiï¬cates for death; hence, Anderson noted that NCHS is not quite to the point that it is with the birth data in terms of transitioning to ârevisedâ data. Anderson said that the timeliness of release of vital statistics mortality data has been reduced not only by the slowness of adoption of the revised certiï¬cates, but also by the ongoing challenge of implementing the 10th revision of the International Classiï¬cation of Diseases by local authorities. He said that the adoption of electronic death registration systems has been slower and more difï¬cult than electronic birth registration systems. These challenges to timeliness have manifested themselves in the lags between data collection and publication of ï¬nal national-level statistics. An-
62 VITAL STATISTICS derson presented data on the past 12 years of data releases, showing the times between the end of the data year and the release of estimates, ï¬rst as a preliminary data set and then as a ï¬nal data release. The data showed increased lag timesâparticularly in the time until production of the prelim- inary estimatesâsince 1999. From 1995 to 2005, including a projection for release of 2006 data, the mean time to preliminary release was 13.8 months; the average time to release of ï¬nal data was 21.5 months. Anderson said that electronic death registration systems promise to greatly improve the timeliness of mortality data, after the initial growing pains of implementation. As an example, he noted that New Hampshire was the ï¬rst state to implement such a system and is now able to provide the state epidemiologist with death certiï¬cate data, including cause of death, within 2â3 days from the date of death. A map displayed by Anderson suggested that electronic death registration systems were functioning in 22 states, the District of Columbia, and New York City; 8 states were said to be in de- velopment as of January 2008 and another 8 in planning or requirements stages. The presence of an electronic registration system in a state does not necessarily mean that it has complete coverage, but that the system has at least begun to be used. Cause-of-death certiï¬cation by physicians is an area of concern for the quality and consistency of mortality data, both in general and with speciï¬c regard to the development of electronic death registration systems. Ander- son said that physicians are not always fully aware of the way in which death certiï¬cate data are compiled at the national level and used for public health and resource allocation purposes. For example, âcardiac arrestâ is still listed as the sole cause of death on about 12,000 certiï¬cates each year (though that number has decreased somewhat over time): from the data standpoint, this is undesirable because the terminal event of cardiac arrest provides no infor- mation on the mechanism of death or the factors that contributed to death. Disseminating training materials to the local levelâincluding the develop- ment of online training tutorialsâis an important step in correcting these problems. Anderson also said that states are experiencing some problems in getting physicians to accept the concept of electronic certiï¬cation, which is obviously an important part of a fully ï¬edged electronic death registration system. Regarding the new data items added in the 2003 certiï¬cate revision, An- derson noted that the change in the education variableâfrom years of ed- ucation to highest degree attainedâhas been somewhat problematic. The variable is important as a summary measure of socioeconomic status, essen- tially the only such measure available on the mortality data ï¬les. Because no way to exactly bridge between the two formats is available, NCHS is currently putting both education items on the data ï¬le (depending on the reporting area) so that users can consider both.
METHODOLOGICAL ISSUES 63 The perennial problem of death certiï¬cate data is that it is necessarily proxy-reported information; in particular, funeral directors are typically re- sponsible for the sociodemographic information on the certiï¬cate. Though these data items, such as education, race, and Hispanic origin, are supposed to be obtained from a knowledgeable informant such as a family member, data quality and consistency problems arise because the data items are some- times ï¬lled by observation. This phenomenon is thought to produce under- reporting of Hispanic ethnicity and American Indian and Asian and Paciï¬c Islander races; Anderson also said that it may be responsible for overre- porting of high school completion. Anderson said that NCHS is hoping to conduct a national longitudinal mortality study, linking death certiï¬cate information with the Current Population Survey or other source, to try to better assess the difference between self-report and proxy-report items on death certiï¬cates.