Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.
OCR for page 9
Vital Statistics: Summary of a Workshop –2– Uses of Vital Statistics Data FROM THE OUTSET, an intended purpose of the Workshop on Vital Data for National Needs was to provide information on the range of uses of the current vital statistics data and to suggest important uses on the immediate horizon. Given the tight time constraints of a 1-day session, the workshop zeroed in on two major classes of current uses: public health research and the development of population estimates and projections. With regard to health policy and health research, summarized in Section 2–A, workshop presentations focused on two major demographic phenomena of long-standing interest: disparities or inequities in health across different racial and ethnic subgroups and gender differences in mortality. This session of the workshop also contrasted these academic perspectives on the uses of vital statistics data with the use of the data for program and planning purposes by the Maternal and Child Health Bureau (MCHB) in the U.S. Department of Health and Human Services. In Section 2–B, we summarize workshop presentations and discussion on the development of population projections and estimates by the Census Bureau and the Social Security Administration; in the latter case, the decades-long projections of population composition based on vital statistics play a key role in the major policy debates on the long-run viability of Social Security entitlements. In terms of future directions, Section 2–C summarizes the workshop’s session that focused on the emerging field of biosurveillance—monitoring of disease and mortality with fine spatial and temporal precision in order to rapidly detect major disease outbreaks or, perhaps, terrorist attacks using biological agents.
OCR for page 10
Vital Statistics: Summary of a Workshop 2–A USES IN HEALTH POLICY AND HEALTH RESEARCH 2–A.1 Social Inequalities in Health Nancy Krieger (Harvard School of Public Health) spoke on the use of vital statistics and related data to monitor health inequities in the United States—studies of trends in health and health care as they are related to socioeconomic position, ethnicity, and gender. Her remarks summarized findings from her Public Health Disparities Geocoding Project. Detailed information on the project and related publications are available online at http://www.hsph.harvard.edu/thegeocodingproject (April 2009). The project’s objective is to augment data in public health surveillance systems, including the birth and death certificate data, with additional socioeconomic covariate information; the resulting constructs are termed area-based socioeconomic measures (ABSMs). The methodology links geocoded vital statistics and U.S. census data at the block group, census tract, and ZIP code tabulation area levels of geography. Ultimately, the intended goal is to develop a valid, robust, easy-to-construct, and easy-to-interpret ABSM that can be readily used by any U.S. state health department or health researcher for public health monitoring and for studying any health outcome from birth to death for any age, gender, or racial or ethnic group. The project started in 1998, making use of data from the Massachusetts Department of Public Health and the Rhode Island Department of Health; the data were for a set of years centered around the 1990 census, and the socioeconomic data in the ABSMs made use of information from that census. To test robustly whether choice of ABSM and geographic level matters, Krieger said that she focused on a wide variety of health outcomes, including mortality (all cause and cause specific), birth (specifically, low birth weight) and also cancer incidence (all sites and site specific), childhood lead poisoning, sexually transmitted infections, tuberculosis, and nonfatal weapons-related injuries. Each outcome was analyzed in relation to 19 different AB-SMs, capturing diverse aspects of socioeconomic position. Eleven of the measures were single-variable measures (e.g., percent working class, percent crowded household) and eight were composites (e.g., deprivation indices developed in previous research). Analyses were performed for the total population and also stratified by race, ethnicity, and gender. Krieger summarized four key findings from the geocoding project. First, measures of economic deprivation were most sensitive to the expected socioeconomic gradients in health. Second, census-tract-level analyses yielded the most consistent results, with maximal geocoding, compared to the block group and ZIP code data. Third, these findings held for separate analyses conducted for white, black, and Hispanic men and women; they also held for those outcomes that could be meaningfully analyzed among the
OCR for page 11
Vital Statistics: Summary of a Workshop smaller Asian, Pacific Islander, and American Indian populations. Fourth, the single-variable measure of percentage of persons below poverty performed as well as more complex composite measures of economic deprivation, including the Townsend index.1 The research suggested that socioeconomic inequalities in health are best monitored with a census-tract poverty measure; Krieger said that one advantage of this approach is that the measure can be applied to all persons, regardless of age, gender, current individual-level educational status, or current employment status. Krieger presented socioeconomic gradients for several health outcome measures to illustrate that the technique provides a way for routine documentation and monitoring of trends using existing vital statistics and public health surveillance data. Specifically, her graphic displays divided census tracts into categories based on percentage of the population below the poverty level (e.g., less than 5 percent, 20 percent or greater). The figures suggested clear poverty gradients in terms of low birth weight, the risk of which was two times higher among births occurring in the most versus least impoverished tracts, that is, 7.5 percent versus 3.6 percent; children with elevated lead levels, with a seven-fold excess among those living in the most versus least impoverished census tracts (33 versus 5 percent); syphilis, with excess risk for the most impoverished tracts being 17 times higher than for the least impoverished tracts; cervical cancer, the incidence of which was twice as high for the most impoverished areas (18 versus 9 per 100,000 population); nonfatal gunshot injury, with an 11-fold increase (22 versus 2 per 100,000 population); and heart disease mortality, with a 1.4-fold excess risk found, resulting in an excess of nearly 100 deaths per 100,000 population. Moving to analysis of racial, ethnic, and gender health disparities, Krieger presented 1989–1991 data on premature mortality (death before age 65). As context, the data indicated that fully half of the black and Hispanic populations lived in census tracts with 20 percent or more of the population below the poverty level whereas, by contrast, almost 50 percent of white men and women live in census tracts with less than 5 percent below poverty. Against this demographic backdrop, the researchers found evidence 1 The Townsend index (Townsend, 1987; Townsend et al., 1988) is a composite index score based on four area-based census measures: percentage of households with no car, percentage of households not owner-occupied, percentage of persons unemployed, and percentage of households overcrowded.
OCR for page 12
Vital Statistics: Summary of a Workshop of marked socioeconomic disparities in premature mortality, with the estimated relative risks ranging from 1.6 to 2.8. Within each economic stratum, an excess of premature mortality remained apparent among the black population. Looking a decade later (1999–2001 data), the same trends persisted: for white non-Hispanic, black, and Hispanic men, higher levels of census-tract poverty were associated with an elevated risk of dying prematurely, with black and Hispanic populations most likely to live in the most impoverished census tracts. Krieger noted similar trends in heart disease mortality data from the period 2000–2005 for Massachusetts. Without disaggregation by poverty level, age-standardized heart disease rates among men and women show a basic distinction, with blacks at higher risk than whites. However, stratifying by census-tract poverty level shows more complex gradients: the poorest census tracts have consistently higher risk levels than the least poor, with particularly pronounced gaps for white and black men living in the poorer census tracts. Similar findings follow from an analysis using 2004–2005 Massachusetts birth outcome data involving low birth weight and smoking during pregnancy. The analysis suggests that racial and ethnic disparities again exist within each socioeconomic stratum, with blacks doing worse for low birth weight and whites doing worse in terms of smoking. There are also marked socioeconomic gradients within each racial or ethnic group. Analysis of these data is ongoing, with the final report slated to include data on prenatal care, breast feeding, caesarian sections, preterm deliveries, and infant mortality. Krieger said that sharing data, methods, and publications on the project website is an important part of the project’s goal to enhance the data reported by U.S. state health departments. Project researchers have conducted training sessions of personnel at health departments, and the techniques have been used in special reports issued by several states, including Washington and Maryland. The intent for the project is to expand the state health departments’ use of geographic analysis in analyzing vital statistics. Krieger noted recent work done in collaboration with the Boston Public Health Commission and the Massachusetts Department of Public Health to extend the work to city-defined neighborhoods and to portray socioeconomic and health data on a consistent set of maps. The system developed by the researchers concentrates on premature mortality as the outcome measure; the analysis system is built on modeling premature mortality as a function of fixed and random effects, allowing for statistical smoothing in the estimation of small-area rates, estimation of variance at each of the specified levels, and adjustment for multiple covariates. A particularly interesting finding from this work was based on mapping the population-based proportion of premature deaths that would not have occurred if residents in every census tract enjoyed the same age-specific mortality rates as residents of the
OCR for page 13
Vital Statistics: Summary of a Workshop least impoverished tracts. Krieger said that this proportion exceeded 20 percent for 8 of Boston’s 60 neighborhoods and 68 percent of the city’s census tracts. In two of Boston’s poorest neighborhoods—Roxbury and North Dorchester—the high excess fractions suggest that, in more than half their census tracts, some 25–30 of every 100 deaths among people under age 75 would not have happened if people in those neighborhoods had, at each and every age, the same lower risk of dying as people in the richer areas. Recently, the project considered U.S. national trends and inequities in premature mortality from 1960 through 2002. County-level mortality data from the National Center for Health Statistics (NCHS) were linked to county-level population and median family income data from the Census Bureau. These data were used to calculate and compare premature mortality and infant death rates by county income quintile for the entire study period. The study found that, even as premature mortality declined in all county income quintiles, the gap between the lowest and highest income quintiles persisted over the entire period and it was relatively greatest for premature mortality in 2000. The greatest progress in reducing these income gaps occurred between 1965 and 1980, especially for populations of color; thereafter, the health inequities widened. The same pattern held for infant deaths. The researchers also used an approach similar to that in the Boston neighborhood study, considering excess premature deaths that would not have occurred if the rates in the least impoverished areas were the same as those for the most impoverished areas. Under these assumptions, Krieger said that the research showed that, had everyone experienced the same yearly age-specific mortality rates as whites in the highest-income-county quintile between 1960 and 2002, 14 percent of white and 30 percent of nonwhite premature deaths would have been averted. Going forward, a challenge will be working with a new data source. Unlike the 1990 and 2000 censuses, the 2010 decennial census will not include a long-form sample that obtains additional social and demographic information (including questions used to calculate census-tract-level poverty estimates). Instead, that information is now covered in the Census Bureau’s American Community Survey (ACS). The ACS provides the same data items as the old long form but, because it is collected on a continuous basis (spreading the sample out over several years), the data are in a new format: rolling averages based on 1, 3, or—for small are as such as tracts—5 years of data. Krieger indicated that project researchers are beginning work to explore how best to develop the tract-level characteristics based on ACS data. Krieger concluded that vital statistics are critical for understanding current and changing U.S. patterns of health and health inequities and the story they tell is compelling. Krieger noted that some of these themes were expressed in a 2008 PBS documentary, Unnatural Causes: Is Inequality Making Us Sick? The basic data of vital events are core to these public education ef-
OCR for page 14
Vital Statistics: Summary of a Workshop forts, because they alone can reveal whether population health and health inequities are getting better or worse. 2–A.2 Trends in Mortality Richard Rogers (University of Colorado) began his remarks by commenting that there was a period, in the 1970s and 1980s, when it was generally thought that the important questions related to the study of mortality had already been asked and that the set of factors influencing mortality were well understood. Thirty years of subsequent research demonstrates that the study of mortality remains one of critical importance to understanding health in the United States. As illustrated by Krieger’s presentation, widespread disparities in health and longevity are one important reason for further study of mortality trends. Rogers said that mortality studies are also important because mortality affects a variety of different, broader factors, including social relationships and social institutions; it can have a profound influence on individuals, on families, on communities. It is important to social policies and population forecasting; in thinking of health care financing in the long run, mortality studies are of central importance for administration of Social Security and Medicare. International comparisons are a major emerging motivating factor for studies of mortality. Specifically, Rogers noted a study by Banks et al. (2006) that found a fairly large disparity between the American and English populations. Rogers summarized the study as having two major findings: first, that prevalence rates for disease were generally higher for Americans than for the English and, second, that the socioeconomic health status gradient is a real construct and is evident in both countries. Generally, Rogers said that the fact that the United States is not at the top of the world in terms of life expectancy—there are at least 22 other countries with longer life expectancies—is a basic motivational factor for further study of basic questions: Why are Americans sick and why does the U.S. life expectancy lag behind that of other countries? Because of time constraints, Rogers centered his remarks on sex differences in life expectancy. The data used in his research include mortality data from the vital statistics, particularly a linked mortality file combining records from the National Death Index with survey data from NCHS’s National Health and Nutrition Examination Survey (NHANES). The research also uses data from NCHS’s National Health Interview Survey. Analysis of estimated life expectancy at birth from 1900 to the present shows generally increasing life expectancies for both males and females, though expectancies for males are consistently lower than those for females. Shifts in the data show the effects of infection for several periods, especially the influenza epidemic in 1918. After greater control for infectious diseases,
OCR for page 15
Vital Statistics: Summary of a Workshop mortality becomes less volatile from the 1940s onward. However, the data also show a slow convergence of the male and female trend lines as the gender difference in life expectancy narrows. After peaking at a 7.8-year difference in 1975 (Arias, 2007), the difference between men and women in estimated life expectancy has steadily declined: by 2005, the gap was 5.2 years. Rogers noted that many studies have looked at the differences around the 1978 peak, but fewer studies have examined the motivating factors for the subsequent decline in the gap. He briefly suggested a range of possible factors that contribute to sex differences in mortality: biological factors, health behaviors (smoking, drinking, unsafe driving, exercise), environmental risks, social relations (marriage, family composition), and socioeconomic status (education, employment, income, poverty). Rogers suggested that some as-yet-underresearched possibilities include composite measures that may be difficult to pick up in national data sets. One is addressing the concepts and assumptions of “masculinity” and “femininity”—for instance, the extent to which “masculine” traits of a high pain threshold, reluctance to seek medical help (absent a life-threatening condition), and failure to get regular health checkups affect health outcomes. The differential life expectancy by sex still shows up when mortality rates are disaggregated by age. The biggest age gap between males and females manifests itself in late teens and early adolescence, what Rogers said has been described as the “accident peak” or “testosterone spike.” Cigarette smoking patterns are one variable that seems to be a central contributor to sex differences in mortality, but those patterns have changed over time. Historically, males have tended to smoke in higher proportions than females—about 53 percent of adult men smoked in 1955, compared with 25 percent of adult women. However, over time, rates of smoking have decreased for both sex groups although females have drawn closer to males (an estimated 24 percent of adult men reported smoking in 2004, compared with 18 percent of adult women). Rogers cited previous research in concluding that smoking contributes to some of the sex differences in mortality and life expectancy. Retherford (1972) attributed 47 percent of the sex gap in life expectancy in 1972 to cigarette smoking; Rogers’ own work with colleagues (Hummer et al., 1998; Rogers et al., 2000) suggests that smoking contributed to about 25 percent of the gap as measured in 1990–1995. These estimates are consistent with an overall decline in smoking and a convergence between males and females in their smoking patterns. Rogers summarized work with hazard ratios derived from NHANES data for 1988–2000. Though the original intent of the work was to try to explain away of the sex difference in mortality, the results actually suggest more explanations for a widening of the gap than a narrowing. Relative to males, females in this period had less education, had lower incomes, and were less
OCR for page 16
Vital Statistics: Summary of a Workshop likely to be employed—that is, they were disadvantaged on a number of socioeconomic status measures. Once these factors are controlled, the hazard ratio expands and the gap in mortality widens. Controlling for marital status also widens the gap; this finding can be explained by males’ tendency to marry younger women but die at earlier ages, meaning that females end up living longer in a widowed status. Rogers also noted that religious attendance has some influence on the sex differential (reducing the gap), because females are more religious and attend services more frequently. Physical activity tends to widen the gap, as does disability (as measured by a question on difficulty in walking). Examining causes of death—looking at sex differences in mortality associated with specific causes rather than overall—provides additional insight about the sex gap. Rogers noted that the gap is particularly wide for deaths due to circulatory disease and cardiovascular disease, while cerebrovascular diseases have less role in explaining the differences between males and females. The significant sex differences in terms of deaths due to cancer are mostly a result of cigarette smoking; the major difference (higher rates of lung cancer mortality among men) disappears when smoking is considered. Respiratory diseases do not have a significant difference between the genders, but deaths due to external causes (accidents, homicides, and suicides) do; because of small sample sizes, these effects are hard to examine in detail. Rogers concluded that part of his results are based on specific periods, specific durations, and specific follow-up time periods. Period effects are important—researchers get different results in explaining sex differences in longevity and mortality in the 2000s than were estimated in the 1970s and 1980s. Still, it is important to think about other covariates and, specifically, what other covariates might be important that are not regularly collected in current national surveys and national data sets. Such covariates could include geographic information; they could include better measures of religiosity or religious attendance; and they could also include such factors as altruism, genetics, biology, stress, and refined quantification of socioeconomic status. In discussion, Rogers noted that the existing interview data from the National Health Interview Survey and NHANES are generally restricted to the noninstitutional population: understanding the degree to which these survey measures are conservative estimates (because they exclude major segments of older persons in nursing facilities and younger persons in correctional facilities) is an important consideration for future research. It was also noted that deaths of U.S. citizens overseas—and, particularly, military deaths—are not included in standard vital statistics (and, hence, not in general assessments of health inequities that use those data). Rogers concluded that health disparities are important and reducing them is a critical national objective for the United States; he said that we need more information to more fully un-
OCR for page 17
Vital Statistics: Summary of a Workshop derstand some of the differences, by sex, by age, by race and ethnicity, and by socioeconomic status. 2–A.3 Uses of Vital Statistics by the Maternal and Child Health Bureau Peter van Dyck (MCHB, Health Resources and Services Administration, U.S. Department of Health and Human Services) described the various ways in which vital statistics are used by MCHB: as the basis for both assessing eligibility for and monitoring performance of targeted public health grants; as input to regular publications and policy standards; and as a way of evaluating an agency’s progress toward general objectives. He also commented on MCHB’s role in issuing grants to help states reengineer their vital statistics and child health information systems. Pursuant to Title V of the Social Security Act of 1935, the MCHB is responsible for providing a variety of grant and coordination services. The bureau’s responsibilities make it the oldest continuing health program related to mothers and children in the nation. Each year, MCHB administers about $1 billion in grants, most of which—about $600 million—is provided as block grant allocations to the states and territories. The block grant funds are allocated using a formula based on a state’s percentage of children living in poverty as a share of the national total; the funds support the operation of state-level maternal and child health offices and programs. Van Dyck said that the states are required to provide matching funds (at least $3 in state funds for every $4 in federal funds), which the states usually generate by billing Medicaid or private insurance for the services they deliver to maternal and child health clients. Some counties also provide funds or staff support. In this way, the $600 million in federal money for maternal and child health grants is leveraged to yield a total effort of $5 billion to 6 billion. To qualify for and obtain the MCHB Title V block grants, state applicants must annually report on a series of 18 specific performance measures; see Box 2-1. Van Dyck noted that vital statistics are essential to this performance and evaluation effort, because several of the performance measures are obtained directly from vital records data (as indicated in italics in the box). State grantees are also directed to provide regular information on a set of six national performance outcome measures, also shown in the box; all of these are directly computed from vital statistics. The Title V block grant program also makes use of a set of “health system capacity indicators” (HSCIs) and “health status indicators” (HSIs) in program evaluation, several of which are keyed directly to vital statistics: HSCI #04: Percentage of women ages 15–44 with a live birth during the reporting year for whom the ratio of observed to expected prenatal
OCR for page 18
Vital Statistics: Summary of a Workshop Box 2-1 Performance and Outcome Measures for the Maternal and Child Health Bureau Block Grant Program Performance Measures Percent of screen positive newborns who received timely follow-up to definitive diagnosis and clinical management for condition(s) mandated by their state-sponsored newborn screening programs Percent of children with special health care needs age 0–18 whose families partner in decision making at all levels and are satisfied with the services they receive Percent of children with special health care needs age 0–18 who receive coordinated, ongoing, comprehensive care within a medical home Percent of children with special health care needs age 0–18 whose families have adequate private and/or public insurance to pay for the services they need Percent of children with special health care needs age 0–18 whose families report the community-based service systems are organized so they can use them easily Percent of youth with special health care needs who received the services necessary to make transitions to all aspects of adult life, including adult health care, work, and independence Percent of 19–35-month olds who have received full schedule of age appropriate immunizations against measles, mumps, rubella, polio, diphtheria, tetanus, pertussis, haemophilus influenza, and hepatitis B Rate of birth (per 1,000) for teenagers ages 15–17 years Percent of third-grade children who have received protective sealants on at least one permanent molar tooth Rate of deaths to children ages 14 years and younger caused by motor vehicle crashes per 100,000 children Percent of mothers who breast-feed their infants at 6 months of age Percent of newborns who have been screened for hearing before hospital discharge Percent of children without health insurance Percent of children, ages 2–5 years, receiving WIC services that have a Body Mass Index (BMI) at or above the 85th percentile Percent of women who smoke in the last 3 months of pregnancy Rate (per 100,000) of suicide deaths among youths 15–19 Percent of very-low-birth-weight infants delivered at facilities for high-risk deliveries and neonates Percent of infants born to pregnant women receiving prenatal care beginning in the first trimester Outcome Measures Infant mortality rate per 1,000 live births Ratio of the black infant mortality rate to the white infant mortality rate Neonatal mortality rate per 1,000 live births Postneonatal mortality rate per 1,000 live births Perinatal mortality rate per 1,000 live births plus fetal deaths Child death rate per 100,000 children ages 1–14 NOTE: Italics indicate that the measure is derived from vital statistics data. SOURCE: Workshop presentation by Van Dyck; http://mchb.hrsa.gov/training/performance_measures.asp (April 2009).
OCR for page 19
Vital Statistics: Summary of a Workshop visits is greater than or equal to 80 percent on the Kotelchuck Index (which is related to the mother’s age at time of prenatal care entrance and the birth weight of the baby if the baby is born early) HSCI #05: Comparison of infant deaths between Medicaid and non-Medicaid recipients, using information associated with prenatal care, low birth weight, and infant mortality. (Van Dyck added that the MCHB’s website, which posts these indicators for all grant recipients, is the only ongoing data site that provides the rate of infant deaths for Medicaid clients compared with the infant deaths for non-Medicaid clients.) HSCI #09A and B: Self-scores by the states on their data capacity for implementing four types of data linkages: annual linkage of infant birth and infant death certificates annual linkage of birth certificates and Medicaid eligibility or paid claims files annual linkage of birth certificates and WIC eligibility files annual linkage of birth certificates and newborn screening files HSI #01A: Percent of live births weighing less than 2,500 grams HSI #01B: Percent of live singleton births weighing less than 2,500 grams HSI #02A: Percent of live births weighing less than 1,500 grams HSI #02B: Percent of live singleton births weighing less than 1,500 grams HSI #03A: Death rate per 100,000 due to unintentional injuries among children ages 14 years and younger HSI #03B: Death rate per 100,000 for unintentional injuries among children ages 14 years and younger due to motor vehicle crashes HSI #03C: Death rate per 100,000 for unintentional injuries for youth ages 15 through 24 years due to motor vehicle crashes MCHB also administers the $100+ million, county-based Healthy Start program, which is intended to reduce infant mortality rates in vulnerable or poor communities. The program is intended to facilitate service delivery in selected areas, including easing access to prenatal health care and promoting positive prenatal health behaviors. As with the block grants, Healthy Start administrators depend on vital statistics—in this case, detailed disaggregation of infant mortality rates—to target program activities and evaluate progress. Looking at county-level plots of infant mortality rates is a particularly important diagnostic tool for Healthy Start, allowing MCHB to pinpoint areas in the nation that might be eligible to apply for grants (grant
OCR for page 24
Vital Statistics: Summary of a Workshop for women in their 20s but increasing rates for women in their 30s—part of a longer-term trend toward higher average maternal age at birth. Goss said that OCA understood that much of the 1965–1990 decline in the fertility rate was likely attributable to this transition to birth at higher ages and, consequently, not one that would continue to drop forever. Hence, Goss said that OCA has never assumed birth rates lower than 1.9 for the total fertility rate. Though some European countries do project a continued decline in fertility rates, OCA generally assumes a steady, average fertility rate of 2.0 for the U.S. population into the future. Clearly, Goss said, the cost implications of shifts in birth rates for Social Security are substantial. The range of OCA’s current projections at the end of 75 years—a total fertility rate estimated at 2.0, within an interval of 1.7 to 2.3—maps to a estimated cumulative effect of about 15.5–19.8 percent of payroll. That is, Social Security would require somewhere between 15.5–19.8 percent of total payroll earnings in order to pay all of its scheduled benefits. Goss demonstrated that changing fertility assumptions even slightly can have major effects on the estimates (and on the uncertainty relative to those estimates) of Social Security’s funding needs. Goss noted that OCA acquires its birth data from the NCHS-compiled vital statistics. In terms of data quality, Goss said that OCA is always concerned about the potential for underreporting, given the potential for distortion of the basic fertility rate that underlies so much of Social Security’s fiscal projections. Immigration and Emigration Goss said that OCA resolves migration into four basic components and draws its data from a variety of sources. Legal immigration: OCA uses data from the U.S. Department of Homeland Security (DHS) on legally admitted immigrants by age and sex. OCA typically bases its assumptions on averages of these data over the past 10 fiscal years. Though most of the categories of legal immigrants are numerically limited or capped by law, one category that is not numerically limited is new immigrants who are immediate relatives of citizens. From its discussion with DHS staff, OCA has concluded that this category has been growing. Reconciling this information with some shifts in other categories (e.g., an increased tendency for persons acquiring legal permanent resident status to be people adjusting their immigration status rather than new entrants), OCA raised its standard assumption of 800,000 gross legal immigrants per year to 1,000,000. Legal emigration: OCA uses historical estimates of legal emigration produced by the Census Bureau, which have ranged from 20 to 30 percent of the level of legal immigration. OCA’s current assumption
OCR for page 25
Vital Statistics: Summary of a Workshop for this category is 25 percent of the level of immigration. However, OCA does make some adjustments to this working rule. In particular, people can leave the borders of the United States but retain their insured status for Social Security benefits; hence, for purposes of projecting Social Security needs, OCA needs to recognize this group in its calculations. Accordingly, OCA lowers its assumed number of emigrants at older ages—effectively treating them as non-emigrants for estimation purposes. Other immigration (undocumented and temporary): Historically, OCA relied on estimates of net immigration of U.S. residents. However, starting in 2008, OCA began working with separate estimation of both inflows and outflows in undocumented and temporary residents, with separate age structures. OCA’s new calculations are based on analysis of 2000 census data, combined with estimates generated by DHS in 2006; the age distribution at entry (and exit) is based on unpublished Census Bureau tabulations for the net “other immigrant” count for 1975–1980. On the basis of this work, OCA’s current annual assumption is about 1.5 million new other immigrants per year. Other emigration: On the basis of OCA’s analysis, the office assumes that about 0.5 million of the 1.5 million other immigrants each year become legal permanent residents within 5 years. The other 1 million either stay (in undocumented or temporary status) or emigrate; OCA currently assumes that about 700,000 of that 1 million eventually exit the country. In discussion at the workshop, Goss noted that, in making its projections of undocumented immigrants, OCA has to make assumptions about the extent to which the undocumented immigrants work for wages and, if they do, whether they pay taxes. OCA’s current projections are that about half of new undocumented immigrants do pay into the system (Social Security and other taxes) but that the fraction will decline over time. In part, Goss said, this is due to the increased documentation requirements to obtain a Social Security card. OCA currently projects that only a relatively small fraction (10–20 percent) of undocumented individuals will go through the process of acquiring legal residence and actually receive benefits. Goss commented that the implications of immigration for Social Security projection are relatively modest, with only about a 1 percentage point swing in the Social Security cost rate over the 75-year projection period being attributable to immigration.
OCR for page 26
Vital Statistics: Summary of a Workshop Disability Though not commonly thought of as a vital event in the usual sense, Goss noted that disability is certainly an important and life-changing factor—with real implications for the cost of Social Security—and so is incorporated into the fiscal projections. In the absence of firm national data on disabilities, the Social Security Administration draws its data and assumptions on disability from its internal data. Specifically, OCA draws on Social Security data on incidence (based on entitlements and awards) and reported medical and work terminations. Deaths For data on deaths, OCA augments NCHS-compiled vital statistics with Medicare data. For deaths of persons under age 65, the vital statistics of death by cause are the exclusive source, with Census Bureau population estimates as the denominator. For persons age 65 and over, Goss said that OCA tends to work with its own statistics, based on Medicare enrollments; although these data are limited to those people who are fully insured in the Social Security system, OCA has concluded that this approach gives it consistency in the numerator and denominator used in death rates and, moreover, helps minimize misstatement of age at time of death (as is a lingering concern with death certificate data). However, the vital statistics mortality data for persons age 65 and older are still an important input through their information on the distribution of death by cause. Goss observed that OCA’s death rates are projections by specific causes of death. To make such projections, Goss said that OCA pays careful attention to historical trends in mortality, but its final assumptions may reflect slightly differing expectations. Though mortality has historically declined rather rapidly at young ages and not very much at older ages, OCA tends to assume more rapid acceleration of mortality for the oldest ages (85 and above) than some figures would suggest. Goss indicated that the cost sensitivity of its fiscal projections to assumptions on mortality is very significant—about a 3 percent swing over the 75-year projection period between its low- and high-end projections. However, fertility measures remain the most sensitive part of OCA’s overall projections. Marriage and Divorce Marriage and divorce are critical to consider because of their effects on both benefits and employments. Though NCHS no longer compiles marriage and divorce data in the national vital statistics, OCA continues to base it assumptions on age distributions from the last available national
OCR for page 27
Vital Statistics: Summary of a Workshop numbers—1995 for marriages and 1988 for divorce. OCA believes that it has a reasonably good handle on the total number of marriages and divorces, and its projections for both marriage and divorce are effectively flat, constant trends. Still, Goss said that OCA would clearly benefit from more recent and detailed data of the form that used to be compiled in the national vital statistics program. 2–B.2 Population Estimates and Projections at the Census Bureau Victoria Velkoff and Fred Hollmann (both of the U.S. Census Bureau) addressed the workshop on the use of vital statistics data in the Census Bureau’s intercensal population estimates and its projections of U.S. population trends. By law (13 USC §181), the Census Bureau is required to produce basic estimates of population and demographic characteristics between decennial censuses: During the intervals between each census of population … the Secretary, to the extent feasible, should annually produce and publish for each State, county, and local unit of general purpose government which has a population of fifty thousand or more, current data on total population and population characteristics…. Velkoff commented that the population estimates are used to allocate over $300 billion in federal funds each year, and they are also used by some states in their funding formulas. The Census Bureau also uses the intercensal population estimates as controls or weights in major household surveys such as the Current Population Survey, ACS, the Survey of Income and Program Participation, and the American Housing Survey. The Bureau of Economic Analysis uses the population estimates in its estimates of per capita income, and they play significantly into calculations by other federal agencies. At the workshop, Kenneth Prewitt (Columbia University) pointed out one particular federal use that illustrates the circularity and feedback loops in the broader statistical system: because vital statistics on births and deaths are a critical component of the population estimates, vital statistics drive both the numerator (incident counts) and denominator (population) of NCHS’s calculated birth and death rates. As the Census Bureau’s system has evolved, national- and state-level population estimates are released by the end of the reference (estimate) year. Estimates disaggregated by demographic groups and for smaller geographic areas are rolled out over the course of the year; Velkoff noted in particular that the Bureau’s nation and state demographic estimates for specific demographic categories for 2007 were slated for release the day after the April 30, 2008, workshop.
OCR for page 28
Vital Statistics: Summary of a Workshop The Census Bureau’s population estimates program generates on an annual basis: national-level estimates, total and disaggregated by age, sex, race, and Hispanic origin categories; state-level estimates, total and disaggregated by age, sex, race, and Hispanic origin categories; estimates for the 3,141 counties (or county-level equivalents), total and disaggregated by age, sex, race, and Hispanic origin; estimates for about 39,000 incorporated places (cities and towns) and minor civil divisions (county subdivisions), total population only; and estimates for Puerto Rico and its county-level municipios, by age and sex. Consistent with the U.S. Office of Management and Budget (OMB) guidelines, the full disaggregation by race and Hispanic origin involves 62 categories: the 31 combinations of the five race categories crossed with two Hispanic origin categories (Hispanic or not Hispanic). Velkoff described the basic cohort component method used to generate Census Bureau estimates: updating the most recent decennial census count by adding births, subtracting deaths, and adding an estimate of net international migration. The Census Bureau generally relies on matches of Internal Revenue Service records from year to year to estimate domestic migration (supplemented by Social Security and Medicare data) and the Bureau’s own ACS for estimating international migration rates. The NCHS vital statistics data are the basis for the estimates of births and deaths used in the cohort method. The Census Bureau also periodically releases long-term population projections to describe the demographic character of the future U.S. population. Hollmann said that these projections are used by states and localities for specific planning objectives, such as assessing demand for roads, schools, and other infrastructure improvements. Among federal users, the Bureau of Labor Statistics uses the population projections as the basis for its own projections of the future labor force, and the National Center for Education Statistics uses them to plan education estimates. The population projections program does not approach the level of geographic detail of the population estimates program; it produces only national- and state-level projections (though Hollmann noted that projections for metropolitan areas are sometimes discussed as a future improvement). Like the full suite of population estimates, the release of population projections is also staggered over time (albeit a longer time range than the annual estimates): At the time of the April 2008 workshop, Hollmann indicated that the most recent (interim) national projections were released in
OCR for page 29
Vital Statistics: Summary of a Workshop 2004, the most recent projections of total state populations dated to 2005, and national estimates disaggregated fully by race and Hispanic origin categories was slated for release in summer 2008. Until recently, the Census Bureau’s population projections were particularly dependent on input from vital statistics because they relied on an essentially census-independent population base. The projections program relied on the so-called demographic analysis population based on historical vital statistics and estimates of international migration. However, Hollmann said that the Census Bureau abandoned that practice in 2000 as a result of comfort with the smaller level of aggregate undercount in the census. Hence, the Census Bureau is using a census base for its rates and projections rather than a purely demographic basis. Still, the Census Bureau’s population estimates and projections both hinge on the NCHS-compiled vital statistics to provide birth and death components in various formulas. Velkoff and Hollmann both commented on the experience of using the vital statistics data and analyses: Velkoff noted that the time lag in the availability of final vital statistics raises some concerns for the Census Bureau’s work. Given the Bureau’s internal timeline of producing some national estimates by the end of the target year, the Census Bureau typically finds itself in the position of “projecting” vital events for about a year and a half. Although Velkoff indicated that the Census Bureau has adjusted to this situation and that these internal projections typically turn out to be of acceptable quality, there are some instances in which they do not—the significant population shifts due to Hurricane Katrina in 2005 being a prominent example. For the purposes of computing the estimates, Velkoff said that the Census Bureau makes the assumption that local reporting of births and deaths is 100 percent complete; however, Census Bureau researchers who develop the Bureau’s demographic analysis estimates have conducted studies that relax the assumption of constant, complete reporting. The Census Bureau has also recently become concerned with the quality of data on reported age at death, particularly at the oldest ages, and is conducting work to evaluate its internal model for mortality at the oldest ages. Citing Census Bureau research on the components of change within the national population projections (Mulder, 2002), Hollmann commented that the Bureau has found that, historically, the largest source of error in its projections comes from projecting fertility. He noted that this is not to say that the Bureau has made the largest-magnitude errors in projecting fertility rate, but rather that variability in the birth component tends to have the largest effect on the final estimates. Over the long term, Census Bureau projections both overprojected fertility
OCR for page 30
Vital Statistics: Summary of a Workshop (e.g., in the late 1960s and 1970s on the back side of the baby boom) and conservatively underprojected fertility (in later years). Errors have tended to be less acute in projecting death and international migration (even though the Census Bureau generally projects the latter as a constant, and thus is almost certainly consistently low). Both Velkoff and Hollmann commented on the challenge of using vital statistics data in their current methodologies, given current variation among registration areas in the format of race and Hispanic origin data; this specific methodological discussion is summarized in Section 4–B. Discussion at the workshop centered on the trends projected for the Hispanic population and, in particular, on research on differential trends in mortality and general health among Hispanics by generation. Nancy Krieger (Harvard School of Public Health) asked whether nativity in the United States, versus foreign-born, is factored into the projections. Hollmann indicated that neither nativity nor specific type of Hispanic origin (e.g., Mexican or Cuban) is directly factored into the models but that it is picked up to the extent that the projections are based on a historical series driven in part by such differential trends. Session moderator Samuel Preston (University of Pennsylvania) commented that lower mortality rates among Hispanics are evident in Social Security Administration data, which may be less immune to data reporting effects in the vital statistics data (i.e., without birth and death certificates playing such a major role in both the numerator and denominator of calculated rates). Since the early 1970s, the Census Bureau has worked with a network of state agency contacts to assist in the production of the annual estimates; later in the decade, this Federal-State Cooperative Program for Population Estimates (FSCPE) was joined by a companion Federal-State Cooperative Program for Population Projections (FSCPP). Under both programs, the states designate an agency as their representative. In some cases, Velkoff noted, the state FSCPE contacts directly provide vital statistics to the Census Bureau (particularly inasmuch as some states’ designated FSCPE or FSCPP agency is also its participant in the Vital Statistics Cooperative Program). 2–B.3 Discussion In discussion of this session of the workshop, moderator Samuel Preston (University of Pennsylvania) commented that one of the values of vital statistics, as they are produced for analysis of fertility and mortality, is that they can be arrayed by birth cohort. Cohort analysis suggests interesting patterns in both fertility and mortality that are not possible to observe by just considering period behavior. For births and fertility, no individual cohort was as extreme in magnitude as the peak in the (period-based) total fertility rate would suggest, and the cohort-level trends have less variability than in
OCR for page 31
Vital Statistics: Summary of a Workshop period behavior. For deaths and mortality, cohort mortality patterns raise interesting findings looking as far back as the 1930s and the first studies of cohort mortality patterns. In particular, when sex differentials in mortality are arrayed on a cohort rather than a period basis, the sex differential peaks for the cohort born around 1905—exactly the cohort for which sex differences in smoking behaviors also reached a peak. Goss concurred that this suggests that vital statistics—and cohort analyses of them—provide great opportunities for investigation of patterns in mortality and fertility. Goss said that significant work had been done internationally on this, particularly in the United Kingdom. Goss said that OCA had also compared its data with counterparts in Canada; Canada has not seen the same broad improvement in its national rate of mortality as in the United States, but further analysis of trends could be useful. 2–C GROWING AND EMERGING USES: VITAL STATISTICS AND BIOSURVEILLANCE A workshop session moderated by Kenneth Prewitt (Columbia University) considered important applications of vital statistics beyond the health care planning domain. Michael Stoto (Georgetown University) spoke of his recent work in health surveillance for national security, also known as syndromic surveillance or biosurveillance. Although originally focused on the detection of terrorist attacks using biological agents, Stoto argued that biosurveillance has come to be interpreted more broadly, as a means for situational awareness for public health emergencies. In either event, Stoto noted that the data systems he was discussing have a much more exacting standard for timeliness than the current vital statistics collections—timeliness measured in weeks and days, and sometimes hours, rather than years. Indeed, the basic point of near-real-time acquisition and use of prediagnostic health data is that waiting until people are diagnosed with diseases or, in the case of vital statistics, die from them would be too late to inform possible interventions. Still, he said, there are important linkages between biosurveillance and the current vital statistics. The central statistical challenges in biosurveillance are, first, obtaining and integrating accurate data from a variety of sources in a timely way and, second, determining whether something is “unusual.” The latter task is complicated by high variability in the background and a possible unstable process generating the data; it involves making critical trade-offs among sensitivity (i.e., false negatives), specificity (i.e., false positives), and timeliness. Current work in biosurveillance has sought to build on existing data systems in the health care world—such as emergency department reports, sales of over-the-counter medication, and absenteeism from work and school.
OCR for page 32
Vital Statistics: Summary of a Workshop These data are usually electronically gathered and highly networked. Using these data, statistical analyses can detect sudden changes that might suggests a disease outbreak or maybe a covert bioterrorist attack. As an example, Stoto described analysis of emergency room data for seven Washington, DC–area hospitals during winter 2003; for those data, detection algorithms suggested certain excesses of gastrointestinal diseases at several hospitals at several points (e.g., early February, early March, mid-April). Though not definitive—Stoto said that there is no way to know whether, for example, the early February increase is a sudden escalation due to random chance or the beginning of something big, that all that is known is that there are differences from what happened before February 1—the work provides clues to follow in ferreting out causes. Biosurveillance research also involves dealing with a number of practical issues. These include privacy concerns about the patients to whom the data refer, as embodied in the Health Insurance Portability and Accountability Act Privacy Rule and related state laws. Other practical challenges include proprietary concerns (who owns the data and with whom can they be shared), concerns and possible prohibitions on secondary use of the data, and operational costs for personnel and information technology. These formidable practical challenges have partially contributed to the recent shift to cast bio-surveillance more broadly as a “situational awareness” technique, Stoto said. Though the purely statistical questions of detecting significant increases in activity continue to draw attention, interest has shifted toward public health activities such as “case finding”: making it easier for physicians and other health care providers to report individual cases that might be of concern and when and where something might be going on, as well as ways to aid outbreak investigations and monitor outbreaks. Given the emphasis on timeliness that is central to biosurveillance systems, Stoto asked what kind of contributions vital statistics—and the principles and practices of vital statistics systems—can bring to bear. Mortality due to pneumonia and influenza is an instructive example to consider for a number of reasons, among them a lengthy history of analysis of such data, experience with the challenge of distinguishing between routine seasonal influenza and wider pandemic outbreaks, and the fact that exposure to many biological agents that could be used in a terrorist attack would initially cause flu-like symptoms. Stoto began by noting recent work by Mills et al. (2004) analyzing data concerning the 1918 Spanish influenza epidemic in the United States. That research found big differences from city to city in terms of the timing and extent of excess influenza deaths. Stoto said that having such data available during an epidemic would certainly have been useful. Close analysis of data (Collins et al., 1930) suggests further insights on public response to the disease outbreak. Stoto noted that a classic example that has been raised from the data is the difference in response by two cities—Philadelphia,
OCR for page 33
Vital Statistics: Summary of a Workshop which did relatively little in response, and St. Louis, which was much more interventionist in terms of closing schools and reducing public activities. Pneumonia and influenza mortality data for 122 cities that have relatively rapid vital statistics reporting are still a key component of public health influenza surveillance. Like the historical data, these data provide useful insight on the timing and extent of flu in one year relative to previous years. Timeliness, however, is a problem: Data downloaded on one Saturday refer to deaths in the week ending the previous Saturday. Yet death from influenza is the end point of a process that typically runs from 1 to 2 weeks, and so even with the most timely data, insight on the time and nature of infection is limited by an effective 3-week lag. The example of influenza monitoring raises the question of how modern information technology—such as electronic vital records collection and electronic death records—might make mortality data more useful for near-real-time monitoring. Vital statistics may never achieve the near-hourly temporal resolution that is needed for outbreak detection, but Stoto suggested that there is still a great deal of value in being able to frame assessments based on what is going on in cause-specific mortality data on a monthly or weekly basis. The question of geographic representation is an important issue: the flu surveillance data from which Stoto drew in his example is based on the cities that, more than a decade ago, were at the forefront in being able to gather electronic death records; more development, and assessment of the coverage represented by such cities and areas, is essential. Stoto closed by noting that vital events reporting had advanced technologically—going beyond the postcards used for reporting in the 19th century to data compilation by fax, phone, and the Internet. However, in its basic character, the collection of information on vital events and disease has not progressed much from the old postcard system. He argued that there is great benefit in the basic structure of the current vital statistics system, which leaves “ownership” of case reports with the state and local authorities, but which provides for federal ownership of the system for gathering vital records and compiling them. Stoto urged that the nation consider a notifiable diseases cooperative program, akin to but broader in scope than the existing vital statistics cooperative program. Ed Hunter’s presentation (Centers for Disease Control and Prevention) focused principally on the demands on birth certificates and the issuance process due to antiterrorism legislation (see Section 3–A), but he concurred with key parts of Stoto’s presentation. Hunter said that it is still unclear whether birth certificate requirements will be the final major impetus for a fully electronic, rapid, standardized vital registration system for births and deaths (given legally mandated matches of birth and death records). If it is, however, Hunter said that the system improvements required by the security provisions will do all the things that Stoto mentioned were necessary for
OCR for page 34
Vital Statistics: Summary of a Workshop situational awareness and surveillance, including rapid availability of vital records data and, ideally, marked decreases in the time lags to issuance of birth and death data. This development would also have a variety of beneficial spillover effects for the general study of health information. Hunter concurred in the usefulness of pandemic influenza as an initial study and development area; he said that the pandemic funding stream is another opportunity to build on the state electronic registration systems and to try to advance their timeliness.