Abstract
Objectives. We determined the size and correlates of underascertainment of Hispanic ethnicity on California death certificates.
Methods. We used 1999 to 2000 vital registration data. We compared Hispanic ethnicity reported on the death certificate to Hispanic ethnicity derived from birthplace for the foreign-born and an algorithm that used first and last name and percentage of Hispanics in the county of residence for the US-born. We validated death certificate nativity by comparing data with that in linked Social Security Administration records.
Results. Ethnicity and birthplace information was concordant for foreign-born Hispanics, who have mortality rates that are 25% to 30% lower than those of non-Hispanic Whites. Death certificates likely underascertain deaths of US-born Hispanics, particularly at older ages, for persons with more education, and in census tracts with lower percentages of Hispanics. Conservative correction for under-ascertainment eliminates the Hispanic mortality advantage for US-born men.
Conclusions. Hispanic ethnicity is accurately ascertained on the California death certificate for immigrants. Immigrant Hispanics have lower age-adjusted mortality rates than do non-Hispanic Whites. For US-born Hispanics, the mortality advantage compared with non-Hispanic Whites is smaller and may be explained by underreporting of Hispanic ethnicity on the death certificate.
Hispanic mortality rates published by the Centers for Disease Control are lower than those for non-Hispanic Whites.1 This has been called a paradox because Hispanics are a socioeconomically disadvantaged population for which mortality risks might be expected to be elevated.2–4 However, at least some of this advantage is an artifact of the underascertainment of Hispanic identity on the death certificate relative to the census classifications of Hispanic identity that supply the denominators for vital rate calculations.4–10 Differences in ascertainment of ethnicity arise because census counts of ethnic populations depend on self-identification by a household respondent; death certificate ethnicity is most often coded by a funeral director, who may assign Hispanic status according to his or her own perception.11 One study found a net 7% underascertainment on the death certificate compared with self-classification in responses to the Current Population Survey, whose instrumentation is similar to that of the census.7
Because of this limitation of vital registration data, several recent studies of Hispanic mortality patterns in the United States instead used data from 1 of 2 databases that link large and representative community-based samples (multiyear pools of Health Interview Survey and Current Population Survey and census respondents) to the National Death Index (NDI). These databases eliminate questions about inconsistency of reported ethnic identity, because ethnic classification is taken from the survey responses. These studies generally report a Hispanic mortality advantage compared with non-Hispanic Whites, albeit one that is smaller than the advantage reported in vital statistics data. NDI-linkage studies also report lower mortality for immigrant than for US-born Hispanics.12–15 However, there are concerns about the completeness of ascertainment of Hispanic deaths derived from the NDI.4 Deaths of returned immigrants outside of the United States are not included in the NDI, and Hispanics report Social Security numbers—used for matching to the NDI—at a lower rate than do non-Hispanic Whites.
We investigated the impact of under-ascertainment of Hispanic mortality on the death certificate on vital rates by comparing death certificate classification of Hispanic ethnicity from a geocoded California death master file to a second source of information about ethnicity. For this second source, we adapted a method for ascertaining Hispanic ethnicity developed by Elo et al.9 Foreign-born persons were classified as Hispanic if they were born in a country whose natives, living in the United States, are highly likely to self-identify as Hispanic on the census race item. For persons born in the United States, Hispanic ethnicity was ascertained with a composite score derived from ethnic composition of county of residence and probable “Spanishness” of first and last names determined by Spanish name–coding lists developed from Social Security Administration records. We compared ascertainment of Hispanic ethnicity by this alternative method and as reported on the death certificate, with respect to characteristics of the decedent, including age, gender, US versus foreign birth, and education, and to the percentage of Hispanics in the census tract of residence of the decedent. We then estimated 2 sets of mortality rates for Hispanics by gender, age, and nativity status from the 2 methods of ascertainment. We also enhanced and validated death certificate data on country of birth for the foreign-born by linking death certificates to the Numerical Identification File (NUMIDENT), the Social Security Administration’s master file containing place-of-birth data from applications for Social Security accounts.
We used the death statistical master file issued by the California Department of Health Services for the years 1999 and 2000.16 Supplemental information about place of birth was drawn from NUMIDENT and matched to the death statistical master file by Social Security number, name, and other identifying information.17 Census tract of residence at the time of death, derived from a geocoding of residence address in the master file, was reported in a supplemental file supplied by the Department of Health Services.18
We derived denominators for vital rates from the National Center for Health Statistics bridged race and Hispanic origin files from the 2000 census for July 1999 and July 2000.19 To derive estimates by immigrant status, we applied the proportions of Hispanic-origin persons who were native or foreign-born within each age and gender group; estimates were calculated with the use of the 2000 census 5% public-use microdata sample for California.20 We excluded Hispanics who were Asian and Pacific Islanders from both numerators and denominators to limit false identification of Filipino and Chamorro decedents as Hispanics by the name-coding method.
We coded a decedent as Hispanic if he or she was reported to be born in 1 of 15 countries (Argentina, Bolivia, Chile, Colombia, Costa Rica, Cuba, Dominican Republic, Ecuador, El Salvador, Guatemala, Honduras, Nicaragua, Mexico, Peru, and Venezuela) or the US territory of Puerto Rico; 75% or more census 2000 respondents who were born in these places said that they were of Hispanic origin. These persons accounted for 98.6% of foreign-or Puerto Rico–born Hispanics in the 2000 census; 98.3% of persons born in these countries reported a Hispanic identity on the census. Substituting the birthplace population of these countries for the census self-identified population yielded an estimated foreign- or Puerto Rico–born Hispanic population 0.3% larger than that actually enumerated in the census by self-identification of Hispanic origin.20
The death master file reported 3 of the 16 places of birth explicitly: Mexico, Cuba, and Puerto Rico. Together these countries accounted for 82% of the foreign- or Puerto Rico–born Hispanics counted in the 2000 census in California. The remaining countries are reported in a “rest of the world” category. To identify countries of origin for these decedents, we merged records from the Social Security NUMIDENT file17. To complete the count of foreign-born decedents for whom birthplace information was not available from NUMIDENT, we ascertained probable Hispanic ethnicity with the county and name algorithm. The county and name-coding method contributed 3% of deaths of foreign-born Hispanics in the study.
For US-born decedents and foreign-born decedents for whom the country of birth was not known, we matched first and last names separately to lists of first names and last names from Social Security number applications. Decedents were given a separate score for first name and for last name that reflected the percentage of persons with that name who identified as Hispanic in their Social Security application. A third score was assigned to represent the percentage of county residents who were Hispanic in 2000 within the age range of the decedent (0–19, 20–39, 40–59, ≥ 60 years). Persons with a total score of 96 or higher summing across these 3 categories were classified as Hispanic. For female decedents, we substituted the surname of the father for decedent’s own surname where the names were different, to minimize attributions based on a husband’s surname. Because of fluctuations by birth cohort in the Spanishness of certain first names, the contribution of the first name component was capped at 40 points to prevent classification by first name only. See Elo et al.9 for more description and validation of this method.
We calculated the number and ratio of deaths ascertained by country of birth or the county and name algorithm and as reported in the death statistical master file from funeral director reports of Hispanic ethnicity on the death certificate. We also compared ascertainment of Hispanic identity by the alternative classification method and as reported by the funeral director by estimating the percentage of Hispanics by birthplace or the county and name algorithm who were identified as such by the funeral director (predictive value positive) and the percentage of Hispanics identified by the funeral director and identified as such by birthplace or the county and name algorithm (sensitivity). All analysis was stratified by nativity status and by age, gender, education, and quartiles of the percentage of Hispanics in the census tract of residence at the time of death.
We calculated 2 sets of age and gender–specific and age-standardized rates, one from Hispanic origin as reported on the death certificate and the other from ascertainment by birthplace or the county/name algorithm. We reported mortality rates separately for Hispanic immigrants and for US-born Hispanics. We calculated mortality rate ratios and 95% confidence intervals to compare these rates with those for non-Hispanic Whites. Confidence intervals for the rate ratios reflected sampling variability for population denominators from the public-use microdata sample that used high- and low-bound estimates of populations in age, gender, and nativity groups.
Ascertainment of Hispanic ethnicity through country of birth or the algorithm that used percentage of Hispanics in the county of residence and Spanishness of name identified 58 670 deaths of Hispanics. This figure was 4.6% higher than the 56 059 deaths of Hispanics identified by the ethnicity reported on the death certificate. The discrepancy was larger for US-born decedents (a 7% increase) than for foreign-born decedents (a 2% increase; Table 1). The net difference between the 2 sources of ethnicity information also varied sharply with age, particularly for US-born Hispanics. The counts of Hispanic deaths found by the 2 methods agreed for US-born decedents under the age of 35, whereas the algorithm yielded an estimate that was 12% higher for decedents aged 75 to 84 years and 23% higher for decedents aged 85 or older. The increase in Hispanic ascertainment derived from the county and name algorithm was larger for decedents with higher education and for counties with decreasing tract percentage of Hispanics.
Algorithm-based classification is typically compared with a standard that identifies predicted value positive (specificity), or the percentage of persons classified with an attribute for which the classification is correct according to the standard, and sensitivity, or the percentage of persons with the attribute according to the standard who are ascertained by the algorithm. In the context of our study, the preferred standard—census classification as Hispanic—was not available. To compare classification by the 2 sources we arbitrarily assigned death certificate classification as the standard, without implying that the death certificate classification was correct.
For immigrants, the correspondence between death certificate coding of Hispanic ethnicity and country-of-birth ascertainment is high. Among persons identified as born in 1 of 15 Spanish-identifying countries, 95.7% were reported as Hispanic on the death certificate; among immigrants identified as Hispanic on death certificates, 97.2% were reported to be born in 1 of these countries (Table 2). For the US-born, 84.5% of those identified as Hispanic by the county and name algorithm were identified as Hispanic on the death certificate; 91.0% of those identified as Hispanic on death certificates were classified as Hispanic by the algorithm. For both the US-born and immigrants, correspondence decreased for decedents aged older than 65 years, for persons with more education, and for those residing in a census tract with a low percentage of Hispanics.
Death certificate–based estimates of mortality for immigrant Hispanics showed a substantial advantage for this group compared with non-Hispanic Whites (standardized rate ratio = 0.71 for both men and women). The alternative rates, based primarily on the birthplace component of the alternative algorithm, were very close to the rates derived from death certificates, with very small upward adjustments at older ages (Table 3).
Mortality rate ratios for US-born Hispanics derived from death certificate reports of Hispanic ethnicity showed an advantage compared with non-Hispanic Whites for men (standardized rate ratio = 0.90) and for women (standardized rate ratio = 0.79). The apparent advantage for both men and women was attributable to much lower mortality rates for Hispanics after 75 years of age. By contrast, the alternative rate ratios derived from the county and name algorithm eliminated the age-standardized differences from non-Hispanic Whites for US-born men and reduced them by approximately half for US-born women.
The results for foreign-born Hispanics imply that Hispanic ethnicity is correctly measured on the California death certificate for immigrant Hispanics if birthplace is correctly measured. To investigate whether this is the case, we compared the death certificate classification of nativity status to nativity status as reported in Social Security’s NUMIDENT file. We found that 97.3% of matched records agreed on birthplace status (foreign- or US-born) for persons who were Hispanic by either death certificate or algorithm coding and that US birth was underreported on the death certificate in a net 1.3% of cases. Substituting NUMIDENT birthplace for death certificate birthplace, we estimated revised standardized rates that were within ±2.5% of the rates derived from the birthplace or county and name algorithm (reported in Table 3) for each nativity and gender subgroup. These adjustments slightly lowered the estimated mortality rates for immigrants and slightly increased them for US-born Hispanics.
For foreign-born Hispanics, the algorithm classification as Hispanic depended primarily on birth in 1 of 15 countries for which a large majority of immigrants to California self-identify as Hispanic. The Hispanic population identified by birthplace was 0.3% larger than the Hispanic foreign-born population determined by self-identification in the California 2000 census 5% microdata file, with 97.4% agreement on classification of individual cases. To test for sensitivity of results to the small differences in denominators, we reestimated the rates for the foreign-born with a denominator derived from the population born in the 15 countries used to identify Hispanic deaths. The resulting adjustments corresponded to a 1.5% decrease in the estimated standardized mortality rate for immigrant Hispanic males and a 1.0% increase in the estimated standardized mortality rate for immigrant Hispanic females.
In 1980, the US Census started to collect data on self-identification of Hispanic identity by the household respondent. Starting with the 1990 census, published Centers for Disease Control and Prevention and state mortality estimates for Hispanics were created by linking vital event counts from deaths of persons identified as Hispanic in death registration data to population denominators calculated from responses to the census Hispanic-origin question. These data showed mortality rates for Hispanics that were generally lower than those calculated for non-Hispanic Whites. Many epidemiologists and demographers have been skeptical of these rates, suggesting that death certificates underascertain Hispanic identity relative to the census.1,4,7–10,15
We investigated the distribution of this underascertainment for Hispanics in California, a state that is home to almost one third of the Hispanic population of the United States. We stratified the analysis by foreign birth versus birth in the United States. Both the substantive and data findings were different for the foreign-born and the US-born, so we have discussed the findings separately for these two groups.
For the foreign-born, Hispanic ethnicity was almost always reported on the death certificate for persons who were also reported to be born in a Spanish-identifying country. Country of birth information on the death certificate was confirmed by an external source—the Social Security NUMIDENT file. There was a small bias toward over-reporting foreign birth relative to US birth compared with the NUMIDENT source, implying a slight overascertainment of mortality of foreign-born Hispanics in vital registration data. This finding was consistent with studies that compared census and vital registration country of birth.5,21 The strong concordance between country of birth and ethnic classification in death registration data is similar to the concordance found in census responses.
The implication of these findings is that Hispanic ethnicity is not underascertained on the death certificate for foreign-born Hispanics. The importance of this finding is increased because the data also suggest that the majority of the Hispanic mortality advantage compared to that of non-Hispanic Whites is attributable to the foreign-born population. Whether we based mortality rate estimates on birth country or reported ethnicity, we found that age-adjusted mortality rates for both men and women were approximately 25% to 30% lower than those for non-Hispanic Whites. This advantage was generally consistent across adult age groups for those aged between 25 and 65 years, declining somewhat at older ages.
The estimate from the National Longitudinal Mortality Study of a 7% net under-ascertainment of Hispanic ethnicity on the death certificate compared with survey self-classification7 has gained acceptance as the best available information about the existence and magnitude of underreporting of Hispanic ethnicity on the US death certificate.9,15 Whether this figure is precisely the one that should be applied to recent California death registration is not certain. However, if ethnic classification of foreign-born Hispanics is trustworthy, as we found, then the need for correction for underascertainment is primarily attributable to errors in the classification of Hispanics born in the United States.
We estimated a 7% underascertainment of Hispanic ethnicity on the death certificate for US-born decedents with a classification algorithm based on county and name. We also found that death certificate underascertainment of Hispanic ethnicity was concentrated at the oldest ages, among more educated persons, and in places where the Hispanic population is well integrated with non-Hispanics. Findings derived from this coding method suggest patterns for further investigation rather than present compelling proof of a particular level of underascertainment.
The pattern of underascertainment that we found is consistent with theory-informed expectations about the mechanisms that produce the underascertainment of Hispanic ethnicity by funeral directors. Birth in the United States, higher education, and spatial integration are all markers of assimilation that may make recognition of Hispanic ethnicity more difficult. These markers are likely correlated with the same characteristics among the family members and friends with whom funeral directors interact. We also speculated that very old age among decedents is correlated with diminished flow of information about ethnicity from a knowledgeable informant to the funeral director as funerals become less elaborate and the contact of the funeral director may be with a less informed source—for example, a child or nursing home administrator rather than a spouse.
These findings point to the usefulness of distinguishing between foreign-born and US-born Hispanic populations in investigating Hispanic mortality patterns. Both substantive and data-quality explanations of Hispanic mortality patterns are different for these 2 groups. The leading substantive explanations of the Hispanic mortality advantage compared with non-Hispanic Whites include health-selective in-migration and out-migration and cultural-protective social structures and behavioral patterns.4,13,22–24 Each of these mechanisms should operate most strongly in the foreign-born.
For NDI-linkage studies, concerns about the completeness of NDI coverage and the validity of the matching algorithm pertain primarily to the foreign-born, who are at greatest risk for out-migration and for whom identifiers such as US Social Security numbers and English-style naming conventions are most problematic. By contrast, we found that for vital registration data, Hispanic identification was most trustworthy for the foreign-born. Our findings of a significant mortality advantage for this portion of the Hispanic population are consistent with reports from NDI-linkage12–14 and other passive follow-up studies9 and provide corroboration that those reports are not artifacts of data-quality issues.
Finally, the pattern that we found in uncorrected vital registration data of sharply lower mortality at the oldest ages for US-born Hispanics compared with both foreign-born Hispanics and non-Hispanic Whites is hard to explain by any substantive theory relying on either cultural differences or migration selection but is readily explained by the existence of an age-specific underascertainment of Hispanic ethnicity suggested by our algorithm-based findings.
The chief limitation of our study is the absence of a conclusive source of information about the ethnic identification of US-born decedents. This portion of our findings is theoretically plausible and suggestive rather than a definitive result. It should be noted that although we used the county and name algorithm to ascertain 3% of foreign-born Hispanic deaths in records for which birthplace information was not available from the linkage to NUMIDENT, too few records were subject to this treatment to materially affect the mortality estimates for the foreign-born.
A second limitation is that we did not correct for other errors in creating rate and rate ratio estimates. Possible errors include ethnic differentials in census undercounting and age misstatement. The accuracy and coverage evaluation survey subsequent to the 2000 census suggested a small and not statistically significant undercounting of Hispanics and a slight overcounting of non-Hispanic Whites. In view of the small size and uncertainty about the distribution of the undercounting for the 2000 census, it was not clear what corrections to apply or whether they would improve the estimates.25,26 Previous evidence about differential age overstatement suggests that this is primarily likely to affect mortality estimates at very old ages.27,28 Because we aggregated decedents after age 85, the influence on our results should have been minimal.
A third limitation is that results are specific to the state of California. This study did not investigate directly whether the quality of reporting of Hispanic ethnicity by funeral directors is better or worse than in other states. Because the quality of reporting of Hispanic ethnicity appears to be in part a function of the density of Hispanic settlement, the quality of reporting in California may be typical of other southwestern states such as Texas and Arizona with large Mexican American populations. Because California is home to just less than one third of the Hispanic population in the United States, mortality patterns for this state contribute a substantial share to national Hispanic mortality patterns.
Foreign-born Hispanics in California have substantially lower mortality rates than do non-Hispanic Whites in the same state. This advantage is not an artifact of underascertainment of Hispanic ethnicity by funeral directors. There is less evidence of a mortality advantage for US-born Hispanics in California and suggestive evidence of substantial underascertainment of Hispanic ethnicity by funeral directors for older and more assimilated US-born decedents.
Investigation of the causes of the Hispanic mortality advantage in California should focus on the evaluation of hypothesized explanations of this advantage that pertain to the foreign-born: health-selective in-migration, return migration, and social and cultural advantages in health-related behavior. For US-born Hispanics, underascertainment of Hispanic ethnicity on the death certificate may be creating a false appearance of a Hispanic mortality advantage.
Source. Denominators for rates are from National Center for Health Statistics19 bridged race estimates for 1999 and 2000 for the non-Asian Hispanic population. Foreign vs native distribution data by age and gender is from the Census 2000 5% public use microdata file for California.20 aInfant mortality estimates are suppressed for immigrants because neonatal mortality is not measurable for this population group. bAge-standardized rates are directly standardized with the National Center for Health Statistics 2000 population standard.
Immigrant Born in United States Characteristic Birthplace or County and Name Death Certificate Ratio BCN to Certificate Birthplace or County and Name Death Certificate Ratio BCN to Certificate All 27 972 27 496 1.02 30 698 28 563 1.07 Age, y 0–14 141 141 1.00 3645 3696 0.99 15–24 1076 1071 1.00 1477 1469 1.01 25–34 1798 1789 1.01 1190 1181 1.01 35–44 2443 2409 1.01 2033 1919 1.06 45–54 2882 2840 1.01 2867 2689 1.07 55–64 3418 3369 1.01 3470 3244 1.07 65–74 4657 4609 1.01 6253 5880 1.06 75–84 5473 5372 1.02 6498 5822 1.12 ≥ 85 6078 5889 1.03 3264 2661 1.23 Gender Male 15 233 14 944 1.02 17 383 16 420 1.06 Female 12 739 12 552 1.01 13 315 12 143 1.10 Education, y (aged ≥ 25 only) 0–6 13 552 13 408 1.01 3899 3760 1.04 7–11 4593 4541 1.01 7872 7383 1.07 12 4791 4735 1.01 8966 8075 1.11 ≥ 13 2795 2684 1.04 4110 3593 1.14 Unknown 1024 916 1.12 729 587 1.24 Tract % Hispanic < 30 6368 6109 1.04 10 029 8592 1.17 31–53 6218 6126 1.02 7277 6860 1.06 54–75 6932 6885 1.01 6559 6378 1.03 ≥ 76 8454 8376 1.01 6833 6733 1.01 Immigrant Born in the United States Characteristics Predicted Value Positive Sensitivity Predicted Value Positive Sensitivity Total 95.7 97.3 84.5 91.0 Age, y 0–14 97.9 97.9 93.0 91.8 15–24 98.3 99.0 91.5 92.0 25–34 98.1 98.9 87.1 87.7 35–44 96.7 98.1 84.0 89.1 45–54 97.0 98.1 84.8 90.9 55–64 96.8 98.1 85.4 91.5 65–74 96.2 97.1 86.8 92.4 75–84 95.2 97.0 81.0 90.5 ≥ 85 92.7 95.7 72.6 89.2 Gender Male 96.1 97.9 86.6 91.8 Female 95.2 96.6 81.8 89.8 Education, y (for those aged ≥ 25 only) 0–6 97.9 98.9 92.4 95.8 7–11 96.0 97.1 86.3 92.1 12 93.7 94.8 80.1 88.9 ≥ 13 89.5 93.2 75.7 86.6 Unknown 88.3 97.0 69.8 93.4 Tract % Hispanic < 30 89.6 93.4 72.6 84.9 31–53 96.2 97.7 86.0 91.3 54–75 97.6 98.2 91.1 93.8 ≥ 76 98.3 99.1 94.1 95.6 Mortality Rate (Deaths/100 000) Rate Ratio (95% Confidence Interval) vs Non-Hispanic Whites Death Certificate Birthplace or County and Name Death Certificate Birthplace or County and Name Immigrants Men, age, y 0a . . . . . . . . . . . . 1–4 15 12 0.63 (0.28, 1.02) 0.42 (0.21, 0.86) 5–14 18 19 1.07 (0.81, 1.43) 1.09 (0.83, 1.45) 15–24 93 94 1.19 (1.08, 1.30) 1.22 (1.11, 1.34) 25–34 96 97 0.85 (0.79, 0.91) 0.86 (0.80, 0.92) 35–44 150 152 0.66 (0.62, 0.70) 0.68 (0.64, 0.73) 45–54 308 313 0.63 (0.59, 0.67) 0.65 (0.61, 0.69) 55–64 704 714 0.65 (0.61, 0.69) 0.66 (0.62, 0.70) 65–74 1845 1878 0.67 (0.63, 0.71) 0.69 (0.64, 0.73) 75–84 4672 4792 0.71 (0.66, 0.77) 0.73 (0.68, 0.79) ≥ 85 13 300 13 780 0.79 (0.72, 0.87) 0.83 (0.75, 0.91) Age standardizedb 695 712 0.71 (0.68, 0.75) 0.74 (0.69, 0.77) Women, age, y 0a . . . . . . . . . . . . 1–4 14 14 0.64 (0.31, 1.32) 0.63 (0.31, 1.29) 5–14 15 15 1.22 (0.88, 1.70) 1.29 (0.94, 1.79) 15–24 25 25 0.74 (0.62, 0.88) 0.73 (0.61, 0.87) 25–34 32 31 0.61 (0.54, 0.68) 0.60 (0.53, 0.68) 35–44 74 75 0.58 (0.53, 0.63) 0.60 (0.55, 0.66) 45–54 174 176 0.60 (0.56, 0.64) 0.61 (0.57, 0.66) 55–64 461 468 0.63 (0.59, 0.68) 0.65 (0.61, 0.70) 65–74 1214 1217 0.64 (0.60, 0.68) 0.65 (0.61, 0.69) 75–84 3267 3306 0.70 (0.65, 0.74) 0.71 (0.66, 0.76) ≥ 85 11 925 12 281 0.82 (0.76, 0.88) 0.85 (0.79, 0.92) Age standardized 499 507 0.71 (0.67, 0.74) 0.72 (0.69, 0.75) US-born Men, age, y 0 651 643 1.23 (1.13, 1.35) 1.22 (1.11, 1.33) 1–4 35 35 1.24 (1.04, 1.49) 1.19 (0.99, 1.42) 5–14 16 16 0.96 (0.82, 1.12) 0.95 (0.82, 1.11) 15–24 100 101 1.26 (1.16, 1.38) 1.31 (1.21, 1.43) 25–34 126 128 1.11 (1.02, 1.22) 1.13 (1.04, 1.24) 35–44 273 284 1.21 (1.12, 1.30) 1.27 (1.18, 1.37) 45–54 614 656 1.25 (1.17, 1.34) 1.36 (1.27, 1.45) 55–64 1241 1311 1.14 (1.06, 1.23) 1.22 (1.13, 1.30) 65–74 2742 2887 1.00 (0.94, 1.06) 1.06 (1.00, 1.12) 75–84 5484 6042 0.84 (0.78, 0.90) 0.93 (0.86, 1.00) ≥ 85 10 601 12 732 0.63 (0.54, 0.74) 0.76 (0.66, 0.89) Age-standardized 871 952 0.90 (0.85, 0.95) 0.99 (0.93, 1.05) Women, age, y 0 560 550 1.23 (1.12, 1.36) 1.19 (1.08, 1.31) 1–4 26 26 1.23 (0.99, 1.52) 1.18 (0.96, 1.46) 5–14 12 12 1.02 (0.85, 1.23) 1.04 (0.86, 1.25) 15–24 31 30 0.89 (0.78, 1.03) 0.85 (0.74, 0.98) 25–34 49 49 0.95 (0.82, 1.08) 0.94 (0.82, 1.08) 35–44 121 133 0.95 (0.86, 1.05) 1.07 (0.97, 1.18) 45–54 282 300 0.97 (0.89, 1.06) 1.04 (0.96, 1.13) 55–64 704 768 0.97 (0.89, 1.05) 1.07 (0.99, 1.15) 65–74 1731 1865 0.92 (0.86, 0.98) 1.00 (0.94, 1.06) 75–84 3790 4284 0.81 (0.75, 0.86) 0.92 (0.86, 0.98) ≥ 85 8607 10 687 0.59 (0.53, 0.67) 0.74 (0.67, 0.83) Age-standardized 559 632 0.79 (0.75, 0.84) 0.90 (0.85, 0.95)
This research was supported by the University of Texas Medical Branch Center for Population Health and Health Disparities (P50 CA10563-02).
The linkage of California mortality records to NUMIDENT records was provided by Bertrand Kesten-baum of the Social Security Administration. Diane Lauderdale provided comments on an early draft.
Human Participant Protection This study was approved by the institutional review board of the University of Texas Medical Branch, Galveston.
