This Article Abstract Full Text (PDF) All Versions of this Article: 113/17/2071    most recent CIRCULATIONAHA.105.595777v1 Alert me when this article is cited Alert me if a correction is posted Citation Map Services Email this article to a friend Similar articles in this journal Similar articles in PubMed Alert me to new issues of the journal Download to citation manager Request Permissions Citing Articles Citing Articles via Google Scholar Google Scholar Articles by Murray, C. J.L. Articles by Ezzati, M. Search for Related Content PubMed PubMed Citation Articles by Murray, C. J.L. Articles by Ezzati, M. Related Collections Epidemiology

(Circulation. 2006;113:2071-2081.)
© 2006 American Heart Association, Inc.

Epidemiology

Understanding the Coronary Heart Disease Versus Total Cardiovascular Mortality Paradox

A Method to Enhance the Comparability of Cardiovascular Death Statistics in the United States

From the Harvard School of Public Health (C.J.L.M., M.E.), Boston, Mass; Harvard University Initiative for Global Health (C.J.L.M., S.C.K., M.E.), Boston, Mass; and University of California, San Francisco (S.C.K.), San Francisco, Calif.

Correspondence to Dr Majid Ezzati, Harvard School of Public Health, 665 Huntington Ave, Boston, MA 02115. E-mail mezzati{at}hsph.harvard.edu

Received October 16, 2005; revision received February 2, 2006; accepted March 3, 2006.

	Abstract

Top Abstract Introduction Methods Results Discussion References

 
Background— Coronary heart disease (CHD) represents the largest share of cardiovascular disease in the United States, but there are conspicuous discrepancies between CHD and total cardiovascular death rates across the states, possibly due in part to variations in physician assignment of causes of death. Our aim was to identify exogenous individual- and community-level predictors of cause-of-death assignment and variability and to use these predictors to improve the comparability of CHD mortality estimates across states.

Methods and Results— We performed a multinomial logistic regression analysis to estimate the effect of individual- and community-level factors on the likelihood of a death being certified as 1 of 3 ill-defined clusters (general atherosclerosis and unspecified heart disease, heart failure, and cardiac arrest) relative to being certified as CHD. The individual-level variables were the decedent’s race, sex, age, education, and place of death; the community-level variable was the number of cardiologists per capita. We used the model to estimate state-level CHD rates that are standardized with regard to the levels of individual- and community-level determinants of cause-of-death assignment. Decedents who died in hospitals and in counties with more cardiologists per capita were more likely to be assigned to CHD than to the ill-defined categories, as were white males relative to other race-sex combinations. Adjustment for these factors resulted in substantially improved correlation between death rates for CHD and all cardiovascular causes. Increases in CHD death rates across states after adjustment for external predictors of cause-of-death assignment ranged from 2% (North Dakota) to 72% (Washington, DC); New York had a decrease (1%) in CHD death rates after adjustment. Nationally, CHD death rates increased 10% for males and 15% for females. The total number of deaths in 2001 attributed to CHD in patients over 30 years of age rose from 433 625 to 489 836 after adjustment.

Conclusions— Greater presence of medical knowledge at the time of death, reflected by place of death and cardiologists per capita, reduces the use of the ill-defined cardiovascular clusters. Racial and gender effects on CHD assignment may reflect disparities in access to care and quality of care. By adjusting for differentials in these parameters, a comparable and consistent set of CHD mortality estimates can be created. The role of the exogenous predictors in validity and comparability of cause-of-death statistics should be confirmed in carefully designed validation autopsy studies.

Key Words: cardiovascular diseases • coronary disease • statistics • heart diseases • mortality

	Introduction

Top Abstract Introduction Methods Results Discussion References

 
Valid and comparable data on mortality and causes of death are essential for the design and evaluation of health policies and intervention programs. The death certificate serves as a key source of information for both overall and cause-specific mortality, but its utility may be compromised by variability in assignment practices for underlying cause of death between regions and over time. Lack of comparability in cause-of-death assignment is particularly important for cardiovascular diseases, because they are the largest contributors to mortality in the United States. In the year 2001, nearly 38% of all deaths in the United States were due to cardiovascular diseases; of these, 47% were attributed to coronary heart disease (CHD).

Clinical Perspective p 2081

Although CHD constitutes the largest share of cardiovascular disease nationally, there are conspicuous discrepancies between CHD and total cardiovascular death rates at the state level (Figure 1). In 2001, the state of New York had the highest death rate from CHD among all the states, but the 17th highest rate of all-cause cardiovascular mortality; Washington, DC ranked 5th highest in overall cardiovascular mortality but 44th for CHD; Alabama had the 7th highest overall cardiovascular mortality but ranked 28th for CHD. Is the poor correlation between CHD and total cardiovascular death rates due to actual epidemiological differences across states in the age-specific mortality from non-CHD cardiovascular causes, or is part of this paradox due to regional differences in cause-of-death assignment practices?¹

View larger version (12K): [in this window] [in a new window]   Figure 1. Unadjusted CHD vs all cardiovascular disease death rates for the states in the United States and the District of Columbia (2001). Deaths among those aged 45 years are included. Rates in each state are age-standardized to the 2000 US population. Unadjusted refers to the observed cause-of-death statistics as recorded in death certificates. r=0.71.

Cardiovascular deaths in the United States are distributed across 8 major clusters of causes: CHD, stroke, hypertensive heart disease, inflammatory heart disease, general atherosclerosis and unspecified heart disease, heart failure, cardiac arrest, and other cardiovascular causes (Table 1; Figure 2). Epidemiological and autopsy studies suggest that a considerable fraction of deaths from 3 of the non-CHD clusters, general atherosclerosis and unspecified heart disease, heart failure, and cardiac arrest, may be due to underlying CHD.^2–6 Variability in the use of these 3 "ill-defined" cause clusters in lieu of CHD may confound the comparability of detailed cause-of-death statistics for cardiovascular disease across the United States.

View larger version (47K): [in this window] [in a new window]   Figure 2. Distribution of non-CHD cardiovascular mortality (2001). CHD constituted 47% of all cardiovascular deaths.

The variability in assigning CHD-related deaths to ill-defined cardiovascular causes could arise from a number of factors: ambiguous wording of the disease classification system; the clinical information available at the time of death, especially for out-of-hospital deaths; the training and skills of the person completing the death certificate; and regional practices and preferences in assigning particular causes of death.^7–9 A number of studies, in the United States and internationally, have examined the validity of death certificates for cardiovascular causes and have concluded that cardiovascular death certificates and clinical diagnoses may disagree with physician reviews and autopsy findings, and that cause-of-death assignment practices vary from region to region.^7–19

An increased number of autopsies and physician follow-ups of death certificates would undoubtedly improve the validity and comparability of mortality statistics, but these measures are limited by financial and practical constraints. In fact, the percentage of deaths with autopsy has trended steadily down over the last 20 years in the United States, and the variable indicating whether an autopsy was performed on a given decedent was dropped from national death records in 1995 due to budget constraints.²⁰ In the absence of autopsy-based information, one option for enhancing the comparability of cardiovascular mortality across the states is to use statistical models that rely on the individual- and community-level factors that influence cause-of-death assignment.

The Global Burden of Disease (GBD) study investigated cross-country variation in the level of ill-defined cardiovascular causes of death (which roughly overlap with general atherosclerosis and unspecified heart disease, heart failure, and cardiac arrest) and their relationship to intercountry variation in CHD mortality.²¹ This analysis found that the proportion of nonstroke cardiovascular deaths coded to ill-defined cardiovascular causes was lowest in New Zealand, Finland, and the United Kingdom (7% to 13%) and highest in Japan, Spain, and Belgium (40% to 60%); the proportion of nonstroke cardiovascular deaths coded to CHD mirrored these values and ranged from 25% to 45% for Japan, Spain, and Belgium to 75% to 80% for New Zealand, Finland, and the United Kingdom. Cross-national comparisons were used to redistribute some ill-defined cardiovascular deaths to CHD,^21,22 but these models did not examine the determinants of cause-of-death assignment. In the present report, we used individual death records in the United States to analyze the effects of individual- and community-level factors on cause-of-death assignment across states in the United States. We then used the model to estimate CHD rates (by state, race, sex, and age) that are standardized with regard to the levels of individual- and community-level determinants of cause-of-death assignment.

	Methods

Top Abstract Introduction Methods Results Discussion References

 
Statistical Model
We used multinomial logistic regression to estimate the relative risk ratios (RRRs) of nonstroke cardiovascular deaths being assigned to CHD or 1 of the ill-defined clusters of cardiovascular causes of death (general atherosclerosis and unspecified heart disease, heart failure, and cardiac arrest) for which CHD is more likely to be the actual underlying cause. The multinomial logistic regression estimates the RRRs for observing a dependent variable with >2 outcome categories as a function of independent covariates; the statistical basis for this and applications to epidemiological data are described elsewhere.^23,24 RRRs were estimated for assignment to each of the ill-defined clusters relative to CHD. Therefore, an RRR of 0.48 for in-hospital deaths for general atherosclerosis and unspecified heart disease implies that after controlling for all other variables, in-hospital deaths had roughly half the likelihood of out-of-hospital deaths for being assigned to general atherosclerosis and unspecified heart disease relative to CHD.

The independent variables in the regression were the exogenous individual- and community-level variables listed in Table 2. Individual-level predictors were the decedent’s race, sex, age, education, and place of death (whether death occurred in a hospital or healthcare facility); the community-level variable was the number of cardiologists per capita in the decedent’s county of residence. Race, sex, and education were used as indicators of socioeconomic determinants of access, utilization, and quality of care.²⁵ Place of death was used because deaths in hospitals are likely to be accompanied by more complete medical history before or at the time of death. The number of cardiologists per capita was used as an indicator of specialized medical knowledge used in assigning causes of death. We also examined the effect of change from the 9th to the 10th revision of the International Classification of Diseases (ICD).

Because the 7 million cardiovascular deaths included in the present analysis are assigned to 1 of 4 mutually exclusive cardiovascular cause clusters, the risk ratios can be converted to probabilities of cross-coding, all as a function of the exogenous individual- and community-level variables. We used the coefficients of the multinomial logistic regression to predict the probabilities that every death in the United States would be assigned to each of these 4 cause clusters if the individual and community characteristics were those of a fixed reference category, as defined in Table 2. Although any fixed category could be used for the purpose of comparability, the reference category in Table 2 is one that is expected to represent the highest access to specialized healthcare and hence the greatest likelihood of being diagnosed and assigned accurately, above and beyond comparability. We controlled for age to avoid confounding the relationships that might also be associated with age; however, we did not use a reference age category in the adjusted (predicted) probabilities, because some of the cardiovascular causes listed in Table 1 are expected to have true age patterns, which should be preserved. Predicted probabilities were then aggregated by race, sex, and state to produce comparable estimates of CHD and ill-defined deaths for each state. All analyses were conducted with Stata version 8 software (StataCorp LP, College Station, Tex).

Data Sources
Data for causes of death and for individual-level variables were from the National Center for Health Statistics’ National Vital Statistics System, which maintains records for every death in the United States, including underlying cause of death and standard sociodemographic characteristics. All deaths between 1990 and 2001 that were attributed to CHD or to 1 of the 3 ill-defined cardiovascular clusters were included in the analysis to maximize predictive power (7 million deaths). County identifiers for years after 2001 were not available for analysis because of changes in NCHS policy. The analysis was limited to deaths that occurred at ages 30 years, because very few cardiovascular deaths occur at younger ages.

Noncardiovascular Ill-Defined Causes
In addition to ill-defined cardiovascular causes, a death may also be assigned to other (general) ill-defined causes (ICD-9 codes 780 to 799; ICD-10 codes R00 to R99). Following the approach used in the GBD study,²¹ we reallocated deaths assigned to these causes for people above 30 years of age among noncommunicable diseases, proportional to their observed frequency.

The authors had full access to the data and take full responsibility for its integrity. All authors have read and agree to the manuscript as written.

	Results

Top Abstract Introduction Methods Results Discussion References

 
Regression Analysis
Table 3 shows the RRRs of deaths being assigned to 1 of the 3 ill-defined cardiovascular cause of death clusters relative to CHD for each of the individual- and community-level characteristics, after adjustment for other factors. In-hospital deaths had substantially lower probabilities than out-of-hospital deaths of being attributed to general atherosclerosis and unspecified heart disease (RRR=0.48) and to heart failure (RRR=0.82), consistent with the hypothesis that the availability of clinical information at the time of death in hospitals reduces the use of ill-defined cause groups. Assignment to cardiac arrest was slightly more likely for in-hospital deaths relative to out-of-hospital deaths (RRR=1.07), possibly because of deaths in critical care units with monitoring capabilities being coded to various dysrhythmias.

The number of cardiologists was inversely associated with assignment to all ill-defined cause clusters, with an 11% reduction in the probability of being assigned to heart failure and a 7% reduction for cardiac arrest for each additional cardiologist per 100 000 residents. The introduction of the 10th revision of the ICD (ICD-10) in 1999 reduced the probability of death being coded to cardiac arrest (RRR=0.76), but it increased the probability of death being assigned to general atherosclerosis and unspecified heart disease (RRR=1.05) and heart failure (RRR=1.31). This finding about the shift to the new revision requires further investigation.

Compared with white men, white females, blacks, and Native Americans had substantially higher probability of being assigned to all 3 cause groups, and Asians had a higher probability of their deaths being assigned to general atherosclerosis and unspecified heart disease or cardiac arrest. Assignment to ill-defined causes was particularly likely for blacks. Deaths among Asians had a lower probability of being assigned to heart failure, which may reflect assignment practice or differences in the natural history of heart disease in these populations.

General atherosclerosis and unspecified heart disease were more commonly used diagnoses in younger age groups. This finding may have been due to the higher utilization of cardiovascular diagnostic studies among individuals over 65 years of age through Medicare, leading to more specific cardiovascular diagnoses in the older age groups. The use of heart failure as a cause of death increased with age, probably reflecting an age-gradient for this sequela of CHD, hypertension, and other heart diseases. Cardiac arrest, however, was relatively more commonly used in the younger age groups. The observed divergent age patterns of the 3 ill-defined cause clusters require further investigation.

Adjusted Cause-of-Death Statistics (National)
Nationally, CHD, as a proportion of nonstroke cardiovascular deaths, rose from 57% to 64% after we standardized the individual- and community-level determinants to the reference values shown in Table 2. Of the ill-defined categories, "general atherosclerosis and unspecified heart disease" and "heart failure" had larger relative reductions (30% and 38%, respectively, nationally) than cardiac arrest (18%), which made up the smallest share of nonstroke cardiovascular deaths before adjustment. Adjustment resulted in a 10% increase in age-standardized CHD death rates (from 5.5 to 6.1 per 1000) for males and a 14% increase (3.4 to 3.9 per 1000) for females. The total number of CHD deaths above age 30 increased from 433 625 to 489 836 deaths for 2001. Of this increase, 5341 deaths were due to redistribution of deaths certified to noncardiovascular ill-defined codes and the remaining 50 870 to redistribution among CHD and the 3 ill-defined clusters after standardization of the individual- and community-level determinants. CHD as a proportion of nonstroke cardiovascular deaths increased more for females (6 to 8 percentage points) than for males (5 to 7 percentage points) in every age group, possibly because CHD is more commonly assigned to males (Table 4).

CHD proportions of nonstroke cardiovascular deaths increased most for blacks and for Native American females (10 to 12 percentage points) after adjustment (Table 4); Asians and white males had the smallest increase (5 percentage points each). Most of the increase in CHD proportions came from shrinking the shares of "general atherosclerosis and unspecified heart disease" and "heart failure." The relative contributions of the 3 groups of ill-defined causes to the increase in CHD, however, varied with race, with blacks and Native Americans having the largest relative decline in general atherosclerosis and unspecified heart disease. The reduction in the cardiac arrest cluster varied the least across races.

Percent increases between unadjusted and adjusted CHD death rates were lower for whites and Asians (ranging from 7% to 12% from 1990 to 2001) than for blacks and Native Americans (16% to 24%; Figure 3). During the interval 1990 to 2001, observed CHD mortality for blacks and whites was similar, with some divergence only at the end of the interval (11%). After adjustment, black CHD death rates ranged from 13% to 21% higher than those of whites each year.

View larger version (15K): [in this window] [in a new window]   Figure 3. CHD death rates per 1000 by race for 1990 to 2001. Deaths among those aged >45 years are included. Rates in each state are age-standardized to the 2000 US population. Native Am indicates Native American.

Adjusted Cause-of-Death Statistics (State Patterns)
Washington, DC, Alaska, and Nevada had the greatest increase in CHD death rates after adjustment, with adjusted rates 51% to 72% higher than the unadjusted (observed) rates (Figure 4). The smallest increases in CHD death rates occurred in North Dakota, Wisconsin, and Missouri (2% to 4% of the original values). New York was the only state with a reduction in CHD death rate after adjustment (–1%). After adjustment, the proportion of nonstroke cardiovascular deaths coded to CHD across states ranged from 59% to 69% compared with 33% to 68% in unadjusted values. The proportion of ill-defined causes was between 14% and 16% compared with 15% to 41% before adjustment. The remaining variation between CHD and ill-defined cardiovascular causes was due to regional differences in other nonstroke cardiovascular deaths: inflammatory heart disease, hypertension, valvular disorders, and cardiac dysrhythmias. After standardization for the effects of exogenous determinants of cardiovascular cause-of-death assignment, the relationship between all cardiovascular mortality and CHD mortality (Figure 5) across states was strengthened substantially.

View larger version (11K): [in this window] [in a new window]   Figure 4. Adjusted vs unadjusted CHD death rates for the states in the United States and the District of Columbia (2001). Deaths among those aged 45 years are included. Rates in each state are age-standardized to the 2000 US population.

View larger version (11K): [in this window] [in a new window]   Figure 5. Adjusted CHD vs all cardiovascular disease death rates for the states in the United States and the District of Columbia (2001). Deaths among those aged 45 years are included. Rates in each state are age-standardized to the 2000 US population. r=0.95.

Mississippi had the highest adjusted CHD death rate and the highest death rates due to all cardiovascular diseases; Oklahoma ranked second for both (Figures 5 and 6). Generally, there was a more consistent relationship between CHD and total cardiovascular death rates after adjustment. Alabama, which had ranked 28th for highest CHD mortality before adjustment, ranked 5th and 7th for CHD and total cardiovascular death rate after adjustment, respectively. The highest adjusted CHD death rates for males occurred in the Mississippi River Delta and Appalachia regions, whereas the lowest CHD death rates were in the southwestern and northwestern states. Adjusted rates for females showed a cross-state pattern similar to that for males but with less variation.

View larger version (32K): [in this window] [in a new window]   Figure 6. CHD death rates per 1000 (2001) for the the states in the United States and the District of Columbia with and without adjustment for exogenous predictors of cardiovascular cause-of-death assignment. Deaths among those aged 45 years are included. Rates in each state are age-standardized to the 2000 US population. A, Males; B, females.

	Discussion

Top Abstract Introduction Methods Results Discussion References

 
Definitive identification of CHD deaths that might have been coded as being due to ill-defined cardiovascular causes requires detailed prospective autopsy studies. In this analysis, we retrospectively investigated some of the determinants of variation in cardiovascular death assignment with the aim of enhancing the comparability of CHD death rates across race, sex, and state. Adjusting the assignment of cardiovascular deaths in this analysis led to significant changes in the CHD pattern across states and race-sex groups and partially addressed the "CHD versus total cardiovascular mortality" paradox. The findings are consistent with the clinical literature that suggests that a substantial fraction of deaths assigned to general atherosclerosis and unspecified heart disease, heart failure, and cardiac arrest have CHD as the underlying pathology.^2–6

The current literature on healthcare disparities almost invariably suggests that whites, males, and people of higher socioeconomic status have better access to healthcare and higher quality of healthcare.²⁵ Even after we controlled for place of death and number of cardiologists per capita, which are proxies for access to care and medical center infrastructure at the community level, race and sex had significant associations with assignment to ill-defined cardiovascular causes, with blacks and Native Americans being the most likely to be assigned to ill-defined causes. The finding may reflect differences in quality of care experienced by minority patients and differential practices in cause-of-death assignment by race. At the same time, some of the observed racial and gender variation of recorded causes of death may represent actual epidemiological differences in the natural history of CHD, including differential prevalence of risk factors, which cannot be captured in our model. Determination of the exact contributions of cause-of-death assignment versus biological and risk factor determinants would require further epidemiological investigations that include autopsies for cause-of-death assignment. We also examined the interactions between individual-level sociodemographic variables and place of death (results not shown). Some of the interactions were statistically significant, but the final results of the analysis, including the relationships by race and state, were not sensitive to these additional analyses.

Whether a death occurred in the hospital and the number of cardiologists per capita were valuable proxies for clinical information and diagnostic skills at the community level. However, better information on the attributes of the hospitals in which individuals died (eg, training and diagnostic facilities) would allow a more direct assessment of these factors on assignment practices. Such information would become available if death records were linked to hospital characteristics.

The death certificate itself contains other sources of error that may undermine the applicability of this type of model. The examples of potential errors in the death certificate include educational attainment and race, which are provided by next of kin.^26,27 These errors may have affected the findings of our regression analysis. We did not include deaths assigned to the other 3 nonstroke cardiovascular clusters (hypertensive heart disease, inflammatory heart disease, and other) because these clusters were considered less likely to be cross-coded with CHD. The percentage of nonstroke cardiovascular deaths assigned to these other categories ranged from 15% in Oklahoma to 29% in Washington, DC. Some deaths in these categories may be related to CHD and should be investigated in detailed investigations that rely on autopsies.

Our statistical model to enhance the comparability of cardiovascular cause-of-death estimates might be applied periodically to produce estimates of trends by state or race. As with any statistical model, the results may depend in part on the years from which the data are taken. When year of death was included in the model, a given death was consistently more likely to be assigned to heart failure relative to CHD later in the interval (RRR=0.72 in 1990 compared with RRR=0.95 in 2000, with 2001 as the reference year). There was no association with year for general atherosclerosis or cardiac arrest.

Above and beyond comparability, improving the overall accuracy of death certificates is important for public health planning. Improving accuracy would require autopsy validation studies, especially those that examine cause-of-death assignment practices and their determinants across states and races rather than simply for 1 community. However, the number of autopsies, the "gold standard" for accuracy, has declined in the United States over the last 20 years. This unfortunate trend makes the use of statistical models essential in accounting for possible biases. A related strategy for improving cardiovascular death assignment would be more extensive training and better standardization of assignment practices for those who complete death certificates.

The multinomial logistic model used in the present report can be used for a number of other purposes, including the variation of cardiovascular cause-of-death assignment across countries. Such application requires unit record data that are currently difficult to obtain for many countries. The analytical strategy may also be useful for enhancing the comparability of other clusters of causes, such as diabetes mellitus, renal failure, and related heart disease.

	Acknowledgments

 
This research was supported by a cooperative agreement from the Centers for Disease Control and Prevention (CDC) through the Association of Schools of Public Health (ASPH) grant No. U36/CCU300430-23 and by the National Institute on Aging (grant No. PO1-AG17625). The contents of this article are solely the responsibility of the authors and do not necessarily represent the official views of the CDC or the ASPH.

Disclosures

None.

	References

Top Abstract Introduction Methods Results Discussion References

 

1. Centers for Disease Control and Prevention. Trends in ischemic heart disease deaths: United States, 1990–1994. MMWR Morb Mortal Wkly Rep. 1997; 46: 146–150.[Medline] [Order article via Infotrieve]
   
2. He J, Ogden LG, Bazzano LA, Vupputuri S, Loria C, Whelton PK. Risk factors for congestive heart failure in US men and women: NHANES I epidemiologic follow-up study. Arch Intern Med. 2001; 161: 996–1002.[Abstract/Free Full Text]
   
3. Rich MW. Epidemiology, pathophysiology, and etiology of congestive heart failure in older adults. J Am Geriatr Soc. 1997; 45: 968–974.[Medline] [Order article via Infotrieve]
   
4. Saxon LA. Sudden cardiac death: epidemiology and temporal trends. Rev Cardiovasc Med. 2005; 6 (suppl 2): S12–S20.[Medline] [Order article via Infotrieve]
   
5. Gheorghiade M, Bonow RO. Chronic heart failure in the United States: a manifestation of coronary artery disease. Circulation. 1998; 97: 282–289.[Free Full Text]
   
6. Kasper DL, ed. Harrison’s Principles of Internal Medicine. 16th ed. New York, NY: McGraw-Hill; 2005.
   
7. Messite J, Stellman SD. Accuracy of death certificate completion: the need for formalized physician training. JAMA. 1996; 275: 794–796.[Abstract]
   
8. Sorlie PD, Gold EB. The effect of physician terminology preference on coronary heart disease mortality: an artifact uncovered by the 9th revision ICD. Am J Public Health. 1987; 77: 148–152.[Abstract/Free Full Text]
   
9. Zumwalt RE, Ritter MR. Incorrect death certification: an invitation to obfuscation. Postgrad Med. 1987; 81: 245–247, 250, 253–254.
   
10. Lloyd-Jones DM, Martin DO, Larson MG, Levy D. Accuracy of death certificates for coding coronary heart disease as the cause of death. Ann Intern Med. 1998; 129: 1020–1026.[Abstract/Free Full Text]
    
11. Every NR, Parsons L, Hlatky MA, McDonald KM, Thom D, Hallstrom AP, Martin JS, Weaver WD. Use and accuracy of state death certificates for classification of sudden cardiac deaths in high-risk populations. Am Heart J. 1997; 134: 1129–1132.[CrossRef][Medline] [Order article via Infotrieve]
    
12. Coady SA, Sorlie PD, Cooper LS, Folsom AR, Rosamond WD, Conwill DE. Validation of death certificate diagnosis for coronary heart disease: the Atherosclerosis Risk in Communities (ARIC) Study. J Clin Epidemiol. 2001; 54: 40–50.[CrossRef][Medline] [Order article via Infotrieve]
    
13. Naruse Y, Nakagawa H, Yamagami T, Sokejima S, Morikawa Y, Nishijo M, Tabata M, Semma M, Miura K, Kagamimori S. Ischaemic heart disease deaths in a Japanese rural area evaluated by clinical records. J Epidemiol. 1997; 7: 71–76.[Medline] [Order article via Infotrieve]
    
14. Sundman L, Jakobsson S, Nystrom L, Rosen M. A validation of cause of death certification for ischaemic heart disease in two Swedish municipalities. Scand J Prim Health Care. 1988; 6: 205–211.[Medline] [Order article via Infotrieve]
    
15. Cameron HM, McGoogan E. A prospective study of 1152 hospital autopsies, I: inaccuracies in death certification. J Pathol. 1981; 133: 273–283.[CrossRef][Medline] [Order article via Infotrieve]
    
16. Chen L, Walker S, Tong S. The impact of the variation in death certification and coding practices on trends in mortality from ischaemic heart disease. Aust Health Rev. 2002; 25: 189–197.[Medline] [Order article via Infotrieve]
    
17. Engel LW, Strauchen JA, Chiazze L Jr, Heid M. Accuracy of death certification in an autopsied population with specific attention to malignant neoplasms and vascular diseases. Am J Epidemiol. 1980; 111: 99–112.[Abstract/Free Full Text]
    
18. Smith CJ, Scott SM, Wagner BM. The necessary role of the autopsy in cardiovascular epidemiology. Hum Pathol. 1998; 29: 1469–1479.[CrossRef][Medline] [Order article via Infotrieve]
    
19. Iribarren C, Crow RS, Hannan PJ, Jacobs DR Jr, Luepker RV. Validation of death certificate diagnosis of out-of-hospital sudden cardiac death. Am J Cardiol. 1998; 82: 50–53.[CrossRef][Medline] [Order article via Infotrieve]
    
20. Hanzlick R. National autopsy data dropped from the National Center for Health Statistics Database. JAMA. 1998; 280: 886. Letter.[Free Full Text]
    
21. Murray CJL, Lopez AD. Estimating causes of death: new methods and global and regional applications. In: Murray CJL, Lopez AD, eds. The Global Burden of Disease. Cambridge, Mass: Harvard School of Public Health (on behalf of the World Health Organization and the World Bank); 1996.
    
22. Lozano R, Murray CJL, Lopez AD, Satoh T. Miscoding and Misclassification of Ischaemic Heart Disease Mortality. Geneva, Switzerland: World Health Organization; September 2001. Global Programme on Evidence for Health Policy Working Paper No. 12.
    
23. Rothman KJ, Greenland S. Modern Epidemiology. Philadelphia, Pa: Lippincott-Raven; 1998.
    
24. Stata Corporation. Stata Base Reference Manual. College Station, Tex: Stata Press; 2005.
    
25. National Healthcare Disparities Report. Rockville, Md: US Department of Health and Human Services, Agency for Healthcare Research and Quality; July 2003.
    
26. Hahn RA, Eberhardt S. Life expectancy in four U.S. racial/ethnic populations: 1990. Epidemiology. 1995; 6: 350–355.[Medline] [Order article via Infotrieve]
    
27. Rosenberg HM, Maurer JD, Sorlie PD, Johnson NJ, MacDorman MF, Hoyert DL, Spitler JF, Scott C. Quality of death rates by race and Hispanic origin: a summary of current research, 1999. Vital Health Stat 2. Sept 1999: 1–13.