The key findings of this study are that (1) the leading causes of premature mortality were largely preventable: among men, these were HIV/AIDS, suicide, drug overdose, homicide, and alcohol use disorder; and among women, these were lung cancer, breast cancer, hypertensive heart disease, colon cancer, and diabetes mellitus; (2) leading causes of premature death differed remarkably between ethnic groups (Tables A-1–A-4); (3) a large health disparity was measured between African Americans and other ethnic groups: African American age-adjusted overall and cause-specific Y LL rates are notably higher, especially for homicide among men (Figures 1, 2, and 3); and (4) except for homicide among Latino men, Latinos and Asians had comparable or lower Y LL rates among the leading causes of premature death compared to whites (Figures 2 and 3). These results illustrate how death registry data can be used to measure, rank, and monitor the leading causes of premature mortality for a local geographic region. Such studies can be used to monitor the local mortality burden of disease and injury over time. For example, our results were compared to our previous San Francisco Y LL study for the period 1990–1995 . While the burden of HIV/AIDS deaths decreased remarkably, the ethnic health disparities remained, with African Americans continuing to suffer the largest burden. This was especially striking for homicides among African American men. The generally better health status of Asians and Latinos has persisted.
Several of these findings mirror those from national studies . For example, the U.S. Burden of Disease and Injury Study  found many of the same preventable causes of premature death among the leading causes, and that the YLL ranking for each ethnic group was unique. Like our study, there were large disparities, measured as DALY s, between African Americans and other ethnic groups, and they reported better health outcomes among Asians than whites. The Eight Americas Study [22, 23] also found large disparities, measured as life expectancy, between Asian Americans and African Americans. A recent examination of the U.S. black-white disparity in life expectancy during the period 1983–2003  found, like our study, that cardiovascular disease (both males and females), homicide (males), and HIV/AIDS (males) were leading contributors to the gap in recent years.
Three measures were used in this study: Y LL s, average Y LL s, and ASY R s. The Y LL is a stand-alone measure of mortality burden not requiring population estimates. It was used to rank the 15 leading causes of death for men and women (Table 7). However, these 15 leading causes were influenced by the larger number of deaths among older residents. To highlight premature, preventable causes of death, we then ranked these top 15 causes by their average Y LL s. Notably, many of the leading causes of death have strong social determinants. Alternatively, the ASY R could have been used to rank the leading causes of death; however, this was not our first choice because it requires population estimates, and the rankings would still be influenced by older deaths. Given our availability of population estimates, ASY R s were used to make comparisons among ethnic groups (Table 6 and Additional file 1). However, only the Y LL s (including average Y LL s) were necessary to rank the leading causes of premature death. Similar analyses were conducted for each ethnic group [Additional file 1].
This study has several strengths. First, we used a simple measure of premature mortality – expected years of life lost – that can be calculated from death registry data that is readily available, population-based, and complete for the whole population. Second, Y LL estimates can be calculated for a comprehensive list of causes of death. Third, Y LL calculations do not require population estimates, allowing leading cause of deaths to be ranked for parts of the population (such as specific ethnicities or geographic areas) for which population estimates are not available. Fourth, subranking by average Y LL s identifies leading causes of premature death, bringing attention to preventable deaths that contribute most to the mortality burden. Fifth, these analyses can be repeated periodically to monitor changes, guide and inform policy makers, and to direct and evalute interventions.
Sixth, except for motor vehicle accidents , we used the Global Burden of Disease ICD-10 cause of death categories, making our methods similar to national and international studies [15, 21]. Seventh, our study included Latinos/Hispanics, an important segment of the population that was not included in a similar national study . Eighth, with the availability of ethnic-specific population estimates, we were able to age-standardize the Y LL s to measure, compare, and monitor the ethnic health disparities in the burden of premature deaths. And ninth, our study findings are directly relevant and can be adapted to the diverse and unique needs of our communities, and to our local government and policymakers.
This study also has several limitations. First, the accuracy of data recorded on death certificates (e.g., underlying cause of death and ethnicity) varies by region and underlying cause . Additionally, analyses using underlying cause of death categories may underestimate the mortality burden for selected contributing causes of death listed on the death certificates (e.g., diabetes mellitus) . Second, the Y LL metric does not measure well conditions that cause significant disease and disability, but are difficult to measure (e.g., mental illness) or do not result in death (e.g., osteoarthritis). Third, on average, there may be a 10-month or longer delay from the time a calendar year ends and the availability of ICD-10-coded death registry data.
Fourth, the ranking of a specific cause of death depends on its individual Y LL magnitude as well as its relative contribution compared to other causes; changes in ranking for a cause over time may be due either to changes in the occurence of that cause, or to changes in the occurences of other causes ranked above or below it. Fifth, the average Y LL could be large for a specific cause of death but only involve a small number of deaths (small burden). To avoid this problem, we only evaluated the average Y LL for the highest ranked causes of death based on Y LL s. Sixth, the Y LL measure is not age-standardized and cannot be used to compare specific causes of death between groups with different age compositions. (With population estimates, Y LL can be age-standardized as described in Methods.) And seventh, because of the uncertainty of population estimates, age-standardized rates must also be interpreted with caution. In spite of these limitations, using Y LL s to rank the leading causes of premature death provides community residents, community-based organizations, policy makers, public health authorities, and researchers with local, representative, objective, and informative data to guide and inform public health priorities, and to direct and evaluate public health interventions.
This study has the following key implications: First, we provide the methodological details for calculating Y LL to measure the burden of premature mortality for any geographic area that has death registry data. We provide both the ICD-10 cause of death classifications used for this study [Additional file 2] and the computational program code for calculating age-interval-specific expected years of life lost that can incorporate discounting (used in this study) and age weighting (not used in this study) [Additional file 3]. This code can be executed in a freely available, open source program for statistical computing and graphics . And second, we demonstrate how these results can be used to rank the leading cause of premature death for major ethnic groups. The rankings can be use to guide, inform, and monitor public health priorities and programs for each group. These analyses can become part of routine public health surveillance for local health jurisdictions, as we have done in San Francisco.