Benjamin S. Bradshaw

Variation in Life Expectancy during the Twentieth Century in the United States

The National Center for Health Statistics (NCHS) reports life expectancy at birth (LE) for each year in the United States. Censal year estimates of LE use complete life tables. From 1900 through 1947, LEs for intercensal years were interpolated from decennial life tables and annual crude death rates. Since 1948, estimates have been computed from annual life tables. A substantial drop in variation in LE occurred in the 1940s. To evaluate these methods and examine variation without artifacts of different methods, we estimated a consistent series of both annual abridged life tables and LEs from official NCHS age-specific death rates and also LEs using the interpolation method for 1900–1998. Interpolated LEs are several times as variable as life table estimates, about 2 times as variable before 1940 and about 6.5 times as variable after 1950. Estimates of LE from annual life tables are better measures than those based on the mixed methods detailed in NCHS reports. Estimates from life tables show that the impact of the 1918 influenza pandemic on LE was much smaller than indicated by official statistics. We conclude that NCHS should report official estimates of intercensal LE for 1900–1948 computed from life tables in place of the existing LEs that were computed by interpolation.

A cohort study of tuberculosis and influenza mortality in the twentieth century

Abstract We reviewed period and cohort mortality for tuberculosis and influenza and pneumonia over the twentieth century and data on the roles of influenza and tuberculosis as underlying and contributory causes of death. As would be consistent with long‐term trends, each cohort had lower tuberculosis mortality but there was no decisive downturn in age specific tuberculosis mortality for any male cohort until after 1945. Tuberculosis mortality among females fell steadily from cohort to cohort as well as within each cohort In every cohort born from around 1890 to around 1930, tuberculosis mortality was higher among women than among men at ages under 30, suggesting that prevalence in women was also higher, but death rates of females crossed under those of males at about age 30. Tuberculosis death rates rose more for males than females around 1918; however, any unusual increase that could be attributable to the 1918 influenza pandemic must have been brief. Contrary to expectations in the medical community, tuberculosis mortality did not rise following the 1918 influenza pandemic Some portion of the rise in death rates around 1918 may have been associated with the influenza, but a comparison of the increase in male tuberculosis mortality during and after World War II, when there was no influenza pandemic, with male mortality in a similar period during and after World War I suggests that any excess in tuberculosis mortality among males in both periods may have been due to wartime mobilization rather than influenza.

Motor vehicle driver death and high state maximum speed limits: 1991-1993

OBJECTIVE To measure the association between motor vehicle crash (MVC) driver death and high state maximum speed limits. METHODS This study used a case-control design and assessed driver deaths from three major types of MVCs: non-collision; collision with motor vehicles in transit; and collision with stationary objects. The study period was 1991-1993. For each type of crash, case subject populations of fatally injured drivers were obtained from the U.S. Department of Transportation Fatality Analysis Reporting System. Four control subject populations, each associated with a different cause of death, were obtained from a U.S. national death certificate database (the causes of death were unintentional poisoning, non-Hodgkin lymphoma, drowning, and diabetes mellitus). Subjects were considered exposed if the state in which they crashed (for cases) or died (for controls) had a maximum speed limit greater than 55 mph. Each of the three case subject populations was compared against each of the four control subject populations. Odds ratios (ORs) were adjusted for age and gender. RESULTS For non-collision driver death, ORs ranged from 3.06 to 6.56, depending on the year and control group; all the ORs were significant. For collision with motor vehicles in transit driver death, ORs ranged from 1.12 to 2.22; all the ORs were significant. For collision with stationary objects driver death, ORs ranged from 0.87 to 1.83. CONCLUSIONS There was a moderately strong and significant association between non-collision driver death and high state maximum speed limits. For collision with motor vehicles in transit driver death, the association was somewhat milder but still consistent. For collision with stationary objects driver death, the presence of an association was unclear. During 1991-1993, the effects of high state maximum speed limits may have been different for different types of MVCs.

Decline of tuberculosis mortality in an urban Mexican‐origin population, 1935–1984

Through a series of life table analyses, this paper describes the natural history of tuberculosis mortality in a Mexican-origin community over five decades (1935-84) during which the disease underwent a transition from a major underlying cause of death to a disease conditioned mentioned more often on death certificates as contributing to death than causing death. The decline in death rates from 1940 to 1950 was especially remarkable. Successive birth cohorts of Mexican Americans, separated by as little as five years of age, experienced distinctly lower risk of death from tuberculosis as they entered young adulthood. There was a rapid convergence in age-specific patterns of tuberculosis death rates in Mexican Americans toward those of non-Hispanic whites, so that by 1960 tuberculosis was primarily a cause of death in old age rather than young adulthood. The impact of changing environment, both through improvements of conditions within neighborhoods and through residential mobility, on birth cohorts at risk of tuberculosis needs to be examined in further research.

Postneonatal diarrhea mortality of Mexican American and Anglo American infants: Trends and context

In the 1930s and 1940s, the Mexican origin population of San Antonio, Texas, USA, was desperately impoverished and inhabited some of the worst slums in the USA. Mortality of Mexican origin infants in the 1930s and 1940s was dominated by diarrhea, a cause of death identified with fecal contamination of the environment. The postneonatal diarrhea mortality rate (risk) was 48 per 1,000 Mexican origin infants, but only 7 per 1,000 Anglo infants. By 1970 this cause of death had virtually disappeared in both populations, but in the 35 years of 1935-1969, Mexican origin infants accounted for over 90 percent of all diarrhea deaths in the city. Limited evidence suggests that miserable living conditions without proper water supplies and sanitation in the densely settled Mexican American neighborhoods gave rise to environmental contamination which resulted in high diarrhea morbidity and mortality. The hypothesis is suggested that reduction of mortality from diarrhea was a consequence of specific community interventions.

Cause-Specific Mortality Rates in Chronic Disease Populations

Random sample surveys have been used for public health surveillance for about 50 years and have grown in size over that time. Such surveys are sufficiently large to estimate the sizes of relatively small subpopulations, such as pa- tients with chronic diseases, helped by pooling multiple years of the same survey. Survey estimates of population sizes have been used as denominators to estimate death rates from specific causes within specific subpopulations. Since these are ratio estimates, with both numerator and denominator being estimates, additional examination is required to evaluate their reliability. We illustrate these methods for rates using deaths due to diabetes and the estimated population of diabet- ics to compute death rates due to diabetes among diabetics by age and sex. Substituting the much smaller population of people at risk for the total population yields useful information about the burden of disease. Modern public health surveillance systems that use random sample surveys of the population have been in use for 50 years and are now large enough to estimate, often with fairly high accuracy, the sizes of small subpopulations of people with a chronic disease. These estimated populations have been used as denominators to compute cause-specific death rates within such subpopulations (1). In this paper, we illustrate the proper statistical use of such ratio estimates, accounting for random variation in the denominator as well as the numerator, with death rates due to diabetes in the subpopulation of persons with diabetes. Substituting the much smaller diabetic population for the total population to compute death rates better informs our understanding of the burden of disease. We compute estimated death rates using the Behavioral Risk Factor Surveillance System (BRFSS), a large, nation- wide sample, to estimate the population of diabetics and counts of deaths due to diabetes in the United States. The estimated death rates due to diabetes within the diabetic population are ratios estimated as the number of deaths due to diabetes over the survey estimate of the size of the diabetic population. METHODS The BRFSS is a random digit dialed (RDD) telephone survey of the U.S. adult population that is conducted by state and territorial health departments under the sponsorship of the Centers for Disease Control (CDC). The BRFSS includes questions on a range of health topics related to chronic diseases and preventive behaviors. A summary description of this surveillance system has been given by Holtzman (2). States

SMITH AND BRADSHAW RESPOND

Between 1980 and 1990, the Hispanic population aged 65 years and older in Texas increased by 75% with the change from surname classification to origin classification. With the same change in classification, deaths in the Hispanic population aged 65 years and older increased 43%. In that decade, the “Hispanic paradox” was born. Using US national data in the only way we are aware it could be used, we found a similar discrepancy between populations and deaths in both 1990 and 2000 from origin classifications in the census and death certificates. The discrepancy persisted when California was excluded from the analysis and when the relatively small Cuban- and Puerto Rican–origin populations were excluded. If we are to understand Hispanic mortality in the United States, the data must be national. If the data are national, the analysis of those data cannot rest simply on the assertion that our assumptions are dubious. We are dealing with the questions of who thinks of themselves as Hispanic when they fill in census forms and who funeral providers think is Hispanic when they prepare death certificates. Texas provides direct evidence that the 2 sources are so inconsistent as to be unusable together in Texas. To find that the problem is not national would require far stronger assumptions than we required. To develop workable Hispanic mortality estimates, any estimates from origin questions must be checked for consistency against estimates from surname classifications. That is not a difficult step. It can still be done for the 2000 US population and deaths and should be an essential part of the analysis of the census and vital statistics for 2010. We need not comment here on Hispanic mortality estimates from sample surveys. Until the national mortality data are better handled, we will not have the benchmark we require to appraise them fairly. That is another cost of the muddle beneath the paradox.

Reconciling heart disease mortality and ICD codes.

Abstract This study uses 61 years of death certificates for Bexar County, Texas, uniformly coded under ICD 9, to describe the transition in heart disease mortality from 1935–1995. We find that Ufe expectancy for persons dying with heart diseases increased throughout this period, with clear differences in rates of increase for males and females, associated with acute ischemic heart disease. Our data point to an epidemic of AIHD in the 1950s and 1960s, which is now abating. Findings are less clear for chronic ischemic heart disease, while other major heart diseases cannot be traced with any confidence owing to changes over time in the emphasis accorded particular causes. Our findings suggest caution with respect to the socioeconomic analysis of heart disease mortality data, particularly where the instability of the coding conventions has been most acute.

State Estimates of Cause and Condition-Specific Mortality Rates for Diabetes Mellitus

Official vital statistics report the death rates due to diabetes in the general population. These are only an indirect indicator of mortality among diabetics since these rates measure deaths due to diabetes in the whole population. A higher rate can mean greater prevalence of diabetes or greater mortality among diabetics or a combination of both. Condition and cause specific death rates directly identify deaths caused by diabetes with the diabetic sub-population in which they occur, giving more specific, focused information since persons diagnosed with diabetes are those most at risk of dying from the disease.

Estimates of Survival and Mortality from Successive Cross-Sectional Surveys

It is hard to estimate death rates among subpopulations that are not defined on death certificates. This paper presents a method for estimating death rates for subpopulations, in this case persons with diabetes, using successive cross-sectional surveys. The method was originally developed to estimate death rates from successive national censuses. Survival ratios use the estimated population in one period as the denominator and the estimated number of survivors at a later time as the numerator. Survival ratios estimated from independent surveys have independent numerators and denominators and their variances are a modification of the usual formulas. We illustrate the method using data from the U.S. Behavioral Risk Factor Surveillance System (1996–1998 and 2001–2003) for persons with diabetes. We estimate annual death rates and their standard errors during the 5 year period between surveys. Useful estimates of death rates for chronic conditions or other small subpopulations can be made from sample surveys of the general population when both status and age of onset are obtained.