The burden of disease and injury in the United States 1996

Estimates of the burden of disease and injury were done by gender and seven age groups (0–4; 5–14; 15–24; 25–44; 45–64; 65–74 and 75+) for the total US population, as well as for each of the four official race groups specified by the Office of Management and Budget (OMB): White; Black; American Indian or Alaskan Native; and Asian or Pacific Islander. Whites were the largest population group (82.8 per cent, or 219.7 million). Blacks represented 12.6 per cent of the population (33.5 million); American Indians – 0.9 per cent (2.3 million), and Asians – 3.7 per cent (9.7 million). Estimates by ethnicity were not included in this report because reliable estimates were only available for a subset of the Hispanic population.

The inclusion of race in the analysis posed particular challenges for minority populations because of race misclassification. Two independent data sets were combined to calculate death rates: the number of deaths in the numerator comes from the detailed mortality file, and population numbers in the denominator are from the census. There was no discrepancy in reporting of race in both data sets for Whites and Blacks, but race misclassification was found to be problematic for Asians and American Indians. Self-reporting of race in the census tended to be higher, particularly for American Indians, than was the attribution of race (by a third party) on death certificates – which will yield an underestimate of death rates [16]. However, we did not correct for race misclassification in American Indians and Asians because evaluations showed that discrepancies in race reporting varied from year to year and thus provided supportive evidence that there may be no systematic bias. Therefore race differentials expressed as rates may have been slightly overestimated. Death and DALY rates were age-standardized to the general population.

There were also important gaps in the available epidemiological data for Asians and American Indians. In order to fill information gaps, we assumed that ratios of YLL to YLD by cause, gender and age were similar to that of the total population. Such assumptions introduce a certain level of uncertainty in the estimates, and call for caution in the interpretation of rankings for causes that have small differences in the number of DALYs. For simplicity, American Indians or Alaskan Natives are referred to as "American Indians," and Asians or Pacific Islanders as "Asians" in the text, tables and figures below.

Even though essentially all deaths in the United States are registered and medically certified, a detailed assessment of mortality data was conducted as part of the USBODI. This was done to further explore and refine the utility of the adjustment procedures for misclassification that were used in the GBD, and to provide a contrast to the overall results using the DALY.

The International Statistical Classification of Disease and Related Health Problems, Version 9, (ICD-9) [17] code listed as the underlying cause for each death recorded in the United States in 1996 was attributed to corresponding disease categories listed in the GBD. The GBD classification scheme was developed as a tool to better inform the health policy debate (Table 1). The list of causes selected for the USBODI was amended based on the distribution of causes of deaths in the United States. All ICD-9 reported codes accounting for more than 0.1 per cent of total deaths were examined. This process identified modifications that were needed from the GBD cause list. Several causes of little relevance to the United States were dropped, i.e. malaria and other tropical diseases. Other causes were added, i.e. Sudden Infant Death Syndrome (SIDS), and septicemia. For those codes accounting for more than 0.1 per cent of deaths that were not included in the GBD list the choice was made in consultation with CDC based on two major criteria. If the code represented a true cause of death with significance for health policy, it was added to the cause list. If the code more likely represented a "garbage category," then after consultation with experts in that disease and a review of published autopsy studies on this subject, a redistribution algorithm was proposed and applied. For example, careful consideration was given to the nearly 10.9 per cent of cancer deaths assigned to "unknown primary." There were another 4% of cancers that did not have a code that corresponded to the GBD classification system. The race and sex specific age distribution of cancers attributed to an ill-defined primary source were compared to all other cancer deaths. These distributions were generally similar. Therefore, cancers attributed to an ill-defined primary source were redistributed proportionally to all defined primary sources based on age, race and sex specific distributions. The detailed list of causes selected for the USBODI is provided [see Additional file 3].

The most problematic aspect of cause of death coding pertains to coding of ischaemic heart disease (IHD) (ICD-9 codes 410–414), which is one of the leading causes of premature mortality. The wide cross-national variations that exist in IHD reported mortality rates were explored in the context of the GBD and were convincingly attributed to variations across countries in coding practices. Physicians may use several ICD-9 codes that are actually due to IHD when they assign the cause of death. These include heart failure (428), ventricular dysrhythmias (427.1, 427.4, 427.5), general atherosclerosis (440.9), and ill-defined descriptions and complications of heart disease (429.0, 429.1, 429.2 and 429.9). IHD deaths may be assigned to these ill-defined cardiovascular codes, or "garbage codes" because of insufficient clinical information at the time of death, local medical diagnostic practices or simply by error. The statistical approach developed to correct for likely undercoding resulting from different coding practices in the GBD included a two-step procedure comprising an ordinary least squares (OLS) regression equation predicting the proportion of cardiovascular death for each age group assigned to ill-defined codes as a function of the proportion of deaths assigned to IHD, and the correction of proportions for each country within set constraints, based on the assumption that the cluster of countries where ill-defined coding was low defined the standard coding practices.

An exploration of cardiovascular death coding in the United States showed important differences in coding practices between states. Indeed, the proportion of all cardiovascular deaths (minus stroke) coded to cardiovascular "garbage" codes ranged from 14% in New Mexico to 37% in Alabama and New Jersey (Table 2). Figure 1 illustrates the enormous variation across US states in coding practices with respect to these ill-defined cardiovascular codes. For each state, the fraction of cardiovascular deaths (excluding stroke) that are assigned to ICD-9 codes 410–414 is shown on the y-axis. On the x-axis the fraction of cardiovascular deaths (excluding stroke) that are assigned to the ill-defined cardiovascular codes is measured. The strong negative relation between IHD mortality and that from ill-defined cardiovascular codes supports the suggestion that the quality of IHD death certification varies substantially across states, as it does across countries in the world.

This preliminary analysis confirmed the need to correct for under-registration of IHD in the US. To estimate the fraction of IHD deaths assigned to ill-defined cardiovasular codes, the regression equation applied in the GBD was revised. Age and sex specific lung cancer death rates were added to the model. Lung cancer mortality rates measure the cumulative effects of tobacco exposure as a risk factor for IHD [18].

The regression model for the US included age and sex specific lung cancer death rates, and ill-defined cardiovascular disease (CVD) rates for Blacks and Whites. These regression equations predicted the proportion of ill-defined CVD deaths by age and sex for Whites and Blacks. We applied results of regression equations for Whites to American Indians and Asians, which were not included in the regressions due their small population size.

The finding that the extent of miscoding increased in older age groups is consistent with GBD regression results: R-square increases with increasing age, which provides further evidence that ill-defined codes are indeed being used for IHD which is more common in older ages. Differences in coding practices observed by race as well as gender are not fully explained (Table 3). A recent study concluded that "the greater presence of medical knowledge at the time of death, reflected by place of death and cardiologist per capita, reduces the use of ill-defined cardiovascular clusters. Racial and gender effects on coronary heart disease (CHD) assignment may reflect disparities in access to care and quality of care." [19]

Ten countries with comparable levels of development and a population greater than 10 million: Australia, Canada, France, Germany, Greece, Italy, Japan, Netherlands, Spain and the United Kingdom, were selected for international comparisons. YLL by cause were obtained directly from the World Health Organization (WHO). YLD and DALY estimates were only available at the regional level, with the exception of Australia, where a national burden of disease study applying the GBD methodology had been conducted [21].

International comparisons may address two sets of issues – the difference in the magnitude of YLL (expressed as YLL rates), or differences in the distribution of major causes of YLL. We examined differences in rankings of major causes of YLL and YLL rates between the United States and other comparable countries.

Rankings for the twenty leading causes of mortality burden in the United States were compared to rankings for these conditions in the ten selected countries. We made one change in the list of conditions adopted for the United States to ensure comparability among countries, which was to combine mortality burden due to lymphomas and multiple myelomas. These two conditions are different forms of reticuloendothelial malignancies. Estimates were not available for these conditions separately in several of the selected countries. This change slightly altered rankings in the United States for several conditions, as the two conditions combined had a higher mortality burden than lymphomas alone. Lymphomas and multiple myelomas ranked 14^th for males and 13^th for females, and the mortality burden for lymphomas ranked 19^th and 17^th respectively. We plotted the rankings for each of the twenty leading causes of mortality burden in the United States (horizontal bars) against the range of rankings observed for each of these conditions in the selected countries (vertical bars), for each sex. The lowest and highest rankings observed in the countries other than the US define the bounds of vertical bars for each condition. Rankings, from one to twenty, are inversely related to the magnitude of mortality burden. Thus, IHD, which caused the largest number of YLL in the United States, ranked 1^st. We also compared YLL rankings for the twenty leading causes of YLL for each race and sex against the ranges observed in the ten selected countries.