A Case Study of the Impact of Inaccurate Cause-of-Death Reporting on Health Disparity Tracking: New York City Premature Cardiovascular Mortality

Abstract

Objectives. Heart disease death overreporting is problematic in New York City (NYC) and other US jurisdictions. We examined whether overreporting affects the premature (< 65 years) heart disease death rate disparity between non-Hispanic Blacks and non-Hispanic Whites in NYC.

Methods. We identified overreporting hospitals and used counts of premature heart disease deaths at reference hospitals to estimate corrected counts. We then corrected citywide, age-adjusted premature heart disease death rates among Blacks and Whites and a White–Black premature heart disease death disparity.

Results. At overreporting hospitals, 51% of the decedents were White compared with 25% at reference hospitals. Correcting the heart disease death counts at overreporting hospitals decreased the age-adjusted premature heart disease death rate 10.1% (from 41.5 to 37.3 per 100 000) among Whites compared with 4.2% (from 66.2 to 63.4 per 100 000) among Blacks. Correction increased the White–Black disparity 6.1% (from 24.6 to 26.1 per 100 000).

Conclusions. In 2008, NYC’s White–Black premature heart disease death disparity was underestimated because of overreporting by hospitals serving larger proportions of Whites. Efforts to reduce overreporting may increase the observed disparity, potentially obscuring any programmatic or policy-driven advances.

Heart disease remains the number one killer of men and women in New York City (NYC) and the United States.^1,2 In 2003, the age-adjusted coronary heart disease death rate was 1.7 times higher in NYC than nationally; yet, on average, NYC’s heart disease risk profile was better than that of the United States.^3,4 The NYC Department of Health and Mental Hygiene (DOHMH) and the Centers for Disease Control and Prevention (CDC) conducted a cross-sectional validation study to investigate this paradox, comparing the cause of death on the death certificate with a validated cause of death determined by a blinded medical team. In a sample of 444 reviewed cases, coronary heart disease deaths were overreported by 91% overall and increased with decedent’s age: 51% among decedents aged between 35 and 74 years, 94% among decedents aged between 75 and 84 years, and 137% for decedents aged 85 years or older.⁴ Overreporting of coronary heart disease has also been found in other US jurisdictions.⁵

More generally, overreporting of heart disease, comprising rheumatic, hypertensive, and chronic ischemic heart diseases; acute myocardial infarction; cardiomyopathy; and heart failure, varies substantially by hospital in NYC.⁶ This is potentially problematic because patient demographics differ among NYC hospitals because of residential segregation, insurance status, and health services provided.^7,8 As a consequence, the prevalence of heart disease overreporting likely differs by decedents’ demographic characteristics including ethnicity. If this is the case, overreporting of heart disease deaths may distort observed racial/ethnic disparities in heart disease death rates. Premature death (i.e., at ages < 65 years) rates are of particular interest because they contribute disproportionately to the years of life lost from heart disease. Such disparity measures are used extensively by the New York City DOHMH to guide health policy, design public health programs, and measure the impact of local and national public health interventions. In addition, they are a part of the Healthy People 2020 national health objectives and DOHMH’s Take Care New York comprehensive health policy goals.^9,10

The goal of this study was to better understand the impact of heart disease overreporting, which occurs at all ages and among all races, on the measurement of the premature heart disease death disparity between Blacks and Whites in NYC. We assessed the impact of vital data quality issues on health disparity tracking. Our methods will be useful to other jurisdictions faced with similar overreporting issues or other cause-of-death data quality issues in any disease category, and our results may suggest outcomes in other urban settings.

METHODS

We identified overreporting hospitals as those with markedly higher proportions of deaths at any age attributed to heart disease. We identified 3 groups of reference hospitals using 3 different methods: 2 methods utilized 95% confidence intervals for the median proportions of heart disease deaths at any age, and the third method simply aggregated the hospitals not included in the overreporting cluster. We employed these different approaches to ensure that our results were robust to the choice of reference hospitals. Using the distribution of age- and race-specific heart disease deaths at reference hospitals, we estimated corrected counts of premature heart disease deaths citywide—that is, how many heart disease deaths of individuals younger than 65 years would have been reported citywide in the absence of overreporting at the overreporting hospitals. Using these corrected counts, we calculated corrected race-specific, age-adjusted, premature heart disease death rates and examined differences between Black and White populations in NYC (disparities measure). We then compared the uncorrected premature heart disease disparity measure with the corrected premature heart disease disparity measure.

Study Population

We selected overreporting and reference hospitals from the 66 NYC hospitals that had reported more than 25 deaths at any age in 2008. None of these 66 hospitals specialize in cardiac care. We defined race/ethnicity categories as non-Hispanic Black (Black) and non-Hispanic White (White).

We defined a death as a “heart disease death” if the death was assigned an International Classification of Disease, 10th Revision code (ICD-10 code) for diseases of the heart (I00–I09, I11, I13, I20–I51) for the underlying cause of death based on the information on the death certificate. The ICD-10 codes are determined by applying the standardized World Health Organization ICD-10 revision algorithm to the words written in the death certificate cause-of-death section. The codes are assigned automatically by the National Center for Health Statistics’ Mortality Medical Data System software or manually by a trained nosologist if the software cannot automatically assign a code.¹¹

Overreporting Institutions and Reference Institutions

We used cluster analysis to determine that 13 of the 66 hospitals reported higher than expected proportions of all-age heart disease deaths. Because no universally accepted cluster method exists, we employed several approaches. These included simple visual inspection, nonparametric confidence intervals for the median, cluster analysis using the SAS (version 9.2; SAS Institute Inc, Cary, NC) procedure specifying the average, centroid, and median methods with 3 and 7 cluster specifications, and an experimental method focusing on the percentage difference between ranked and contiguous hospital proportions.¹² Each approach identified the same 13 hospitals, which make up the “overreporting” hospital group.

We took 3 approaches to identify hospitals presumed to not overreport heart disease deaths. A priori, we excluded 3 NYC hospitals specializing in cancer care from the study population of the 66 hospitals because they would have a different distribution of heart disease deaths for reasons unrelated to overreporting, leaving 63 hospitals from which to select reference hospitals.

Reference 1 hospitals.

Reference 1 hospitals comprised the 16 hospitals within the nonparametric 95% confidence interval (24.8, 33.7) for the median proportion of all-age heart disease deaths at the 63 noncancer hospitals described previously, including the 13 overreporting hospitals. We used the median as the measure of central tendency because the distribution of heart disease death proportions at these 63 hospitals was left-skewed. The median and average proportion of all age heart disease deaths at reference 1 hospitals were 29.4% and 28.7%, respectively.

Reference 2 hospitals.

The reference 2 hospitals consisted of the 15 hospitals within the nonparametric 95% confidence interval (22.9, 29.5) for the median proportion of all-age heart disease deaths at the 50 of the 63 noncancer hospitals that did not overreport heart disease. In contrast to reference 1, we a priori excluded the 13 hospitals from the overreporting group before determining the 95% confidence interval. Again, we used the median because the distribution of all-age heart disease death proportions at these 50 hospitals was left-skewed. The median and average proportion of all-age heart disease deaths at reference 2 hospitals were 24.8% and 25.4%, respectively.

Reference 3 hospitals.

In contrast to reference 1 and 2 hospitals, we did not restrict the selection of reference 3 hospitals to those within a 95% confidence interval around a median. Instead, reference 3 includes the 50 of the 63 hospitals that do not specialize in cancer care and are among the 13 hospitals in the overreporting group. The median and average proportion of all-age heart disease deaths at reference 3 hospitals were 24.8% and 26.7%, respectively.

Analyses

We first determined the age- and race-specific premature heart disease death rates and then multiplied the age- and race-specific rates by the age-adjustment weights for each age category (ages < 25, 25–44, and 45–64 years) derived from the standard population (2000 projected US population) and summed across the 3 age categories.¹³

Because the age-specific standard weights are constant and the probability of death in each age group is independent of the probability of death in the others, we calculated the variance of the age-adjusted heart disease death rates (AAR) by¹⁴

where w_i are the age-specific standard weights and R_i are the age- and race-specific heart disease death rates.

Assuming all persons in the same age group i have the same risk of death and the number of deaths within a population follow a Poisson distribution, we then calculated the variance of the age-specific heart disease death rate by¹⁴

where D_i is the number of heart disease deaths in age group i.

To examine disparities, we also calculated absolute differences in citywide age-adjusted premature heart disease death rates between Blacks and Whites as this is New York City’s primary racial/ethnic disparity indicator for heart disease.¹⁰ Furthermore, premature heart disease deaths result in more years of life lost than nonpremature (≥ 65 years) heart disease deaths and thus are of great interest for public health prevention purposes. The absolute difference is represented in the following formula:

where is the age-adjusted premature death rate attributed to heart disease among Whites and is the age-adjusted premature death rate attributed to heart disease among Blacks, i is the index for age groups 1 through N, is the standardized age group weight, and is the rate of heart disease deaths in each Black age group and similarly for Whites.

Corrected heart disease death counts.

We calculated the corrected premature heart disease death counts separately with reference 1, reference 2, and reference 3 hospitals. To calculate the age- and race-specific heart disease death counts for each reference group, we first aggregated all premature deaths within each reference group and then stratified by age at death category (< 25, 25–44, or 45–64 years) and race category (Black or White). Within each age–race stratum, we calculated the proportion of heart disease deaths. We then multiplied these proportions by the counts of deaths from all causes within each age–race stratum at the overreporting hospitals to obtain 3 estimates of counts of premature deaths attributed to heart disease at the overreporting hospitals, 1 for each reference group.

We defined “surplus” heart disease deaths at overreporting hospitals as the number of reported heart disease deaths in excess of the corrected count. We calculated the corrected count of citywide NYC heart disease deaths within each age–race stratum for each reference standard by subtracting the estimated surplus heart disease deaths from the reported number of heart disease deaths citywide.

Age-adjusted corrected and reported heart disease death rates.

To obtain corrected and reported age-specific heart disease death rates for Blacks and Whites, we divided the corrected and reported citywide heart disease counts in each age–race stratum by the corresponding age- and race-specific population in NYC in 2008.¹⁵ We multiplied the age-specific rates by the population weights to obtain the age-adjusted rate for the 2 racial categories.

Corrected and reported disparity measures.

We subtracted the age-adjusted premature heart disease death rate among Whites from that among Blacks to obtain the premature heart disease death disparity indicator by using reported and corrected counts. We assessed the effects of heart disease overreporting by comparing the disparities measure based on the reported causes of death with the measures corrected by using the 3 different reference hospitals.

RESULTS

Table 1 describes the distribution of deaths by age–race strata within different hospital groups. Of the 16 331 citywide premature deaths, 11 687 occurred at the 66 hospitals with more than 25 deaths. The overreporting category includes 13 hospitals. Reference groups 1, 2, and 3 include 16, 15, and 50 hospitals, respectively. Overall, the race distribution of decedents at hospitals with greater than 25 deaths was comparable to that of decedents citywide: 36% versus 35% Black and 32% versus 34% White, respectively. The age distribution at hospitals with greater than 25 deaths and citywide was also comparable: 6.0% versus 7.3% of decedents were younger than 25 years, 11.5% versus 10.8% were aged between 25 and 44 years, and 51.3% versus 49.6% were aged between 45 and 64 years.

TABLE 1—

Premature (Age < 65 Years) Heart Disease and All-Cause Deaths by Racial Group and Age Category Within Institution Groupings: New York City, 2008

Variable	Citywide, No. (%)	Institutions > 25 Deaths, No. (%)	Overreporting Institutions, No. (%)	Reference 1 Hospitals,a No. (%)	Reference 2 Hospitals,b No. (%)	Reference 3 Hospitals,c No. (%)
Premature all-cause deathsde
Non-Hispanic Black, age, y	5688 (35)	4201 (36)	280 (23)	1196 (46)	1285 (45)	3658 (39)
< 25	607 (11)	542 (13)	16 (6)	172 (14)	236 (18)	520 (14)
25–44	1099 (19)	776 (18)	51 (18)	210 (18)	239 (19)	684 (19)
45–64	3982 (70)	2883 (69)	213 (76)	814 (68)	810 (63)	2454 (67)
Non-Hispanic White, age, y	5593 (34)	3720 (32)	610 (51)	668 (26)	594 (21)	2553 (27)
< 25	376 (7)	322 (9)	29 (5)	56 (8)	101 (17)	276 (11)
25–44	817 (15)	489 (13)	82 (13)	86 (13)	91 (15)	330 (13)
45–64	4400 (79)	2909 (78)	499 (82)	526 (79)	402 (68)	1947 (76)
Other race/ethnicity, all ages	5050 (31)	3766 (32)	316 (27)	710 (28)	964 (34)	3103 (33)
Total deaths	16 331	11 687	1206	2574	2843	9314
Premature heart disease deathsdf
Non-Hispanic Black, age, y	1236 (22)	715 (17)	106 (38)	220 (18)	218 (17)	604 (17)
< 25	24 (3)	23 (4)	0 (0)	9 (5)	10 (4)	23 (4)
25–44	151 (14)	76 (10)	10 (20)	23 (11)	19 (8)	64 (9)
45–64	1061 (27)	616 (21)	96 (45)	188 (23)	189 (23)	517 (21)
Non-Hispanic White, age, y	1255 (22)	717 (19)	271 (44)	152 (23)	121 (20)	441 (17)
< 25	13 (4)	12 (4)	3 (10)	2 (4)	5 (5)	9 (3)
25–44	104 (13)	65 (13)	22 (27)	16 (19)	13 (14)	43 (13)
45–64	1138 (26)	640 (22)	246 (49)	134 (25)	103 (26)	389 (20)
Other race/ethnicity, aged <65 y	915 (18)	574 (15)	123 (39)	124 (18)	278 (29)	448 (14)
Total premature deaths attributed to heart disease	3406 (21)	2006 (17)	500 (41)	496 (19)	496 (17)	1493 (16)

^aInstitutions included in nonparametric 95% confidence intervals for the median constructed by using all 63 institutions.

^bInstitutions included in nonparametric 95% confidence intervals for the median constructed by using 50 nonoverreporting institutions.

^cNonoverreporting institutions aggregated.

^dAll percentages rounded to the nearest whole number.

^ePercentage of total deaths.

^fPercentage of age- and race-specific total deaths that are attributed to specific cause.

The racial distribution of decedents at the overreporting hospitals differed from the distribution citywide and in the 3 reference groups. Whites were overrepresented at the 13 overreporting hospitals, which filed 51% of death certificates among White decedents younger than 65 years. By contrast, overreporting hospitals filed only 23% of death certificates among Black decedents younger than 65 years (Table 1). The racial distribution of decedents varied somewhat across the 3 reference groups, although all reference groups reported fewer premature deaths of Whites (ranging from 21% to 27%) than did the overreporting hospitals and relatively more premature deaths of Blacks (ranging from 39% to 46%).

Reported Heart Disease Deaths

Table 1 shows the proportion of premature deaths attributed to heart disease overall and stratified by race citywide, at hospitals with greater than 25 deaths, at overreporting hospitals, and at the hospitals making up the 3 reference groups. We used these proportions to calculate the race-specific rates, which are relative to the composition of the living population and are the basis of measures of health disparities. In all groups, the proportion of premature heart disease deaths increased with increasing age.

Citywide, 1236 premature heart disease deaths were reported among Blacks and 1255 were reported among Whites. Most of these were reported by the 66 hospitals with greater than 25 deaths (715 Black, 717 White). Citywide, heart disease deaths make up 22% of all premature deaths among both Whites and Blacks. At hospitals reporting greater than 25 deaths, this proportion is slightly smaller—19% and 17%, respectively—but similar to the heart disease death proportions in the 3 reference groups. The proportions of premature deaths attributed to heart disease at overreporting hospitals were 38% among Black decedents and 44% among White decedents.

Corrected Rates

Table 2 shows the corrected citywide premature heart disease death rates using each of the 3 groups of reference hospitals. Overall, correcting for overreporting of heart disease reduced the reported premature age-adjusted heart disease death rate (42.9 per 100 000). Using reference 1 reduced the rate to 39.8 per 100 000. The most meaningful rate changes were confined to the group aged 45 to 64 years with the magnitude of the decrease depending on race and reference group used. When we used reference 1 premature heart disease death proportions, the premature heart disease death rate decreased 15.1 per 100 000 among Whites aged 45 to 64 years and 9.6 per 100 000 among Blacks aged 45 to 64 years.

TABLE 2—

Reported and Corrected Citywide Heart Disease Death Rates by Race Group and Age Category for Three Reference Standards: New York City, 2008

	Reference 1 Hospitalsa			Reference 2 Hospitalsb			Reference 3 Hospitalsc
Variable	Reported Rated (SD)	Corrected Rated (SD)	Changee	Reported Rated (SD)	Corrected Rated (SD)	Changee	Reported Rated (SD)	Corrected Rated (SD)	Changee
Non-Hispanic Black, age, y
< 25	3.4 (0.7)	3.4 (0.7)	0.0	3.4 (0.7)	3.4 (0.7)	0.0	3.4 (0.7)	3.4 (0.7)	0.0
25–44	27.2 (2.2)	26.4 (2.2)	−0.8	27.2 (2.2)	26.2 (2.2)	−1.0	27.2 (2.2)	26.3 (2.2)	−0.9
45–64	218.3 (6.7)	208.7 (6.6)	−9.6	218.3 (6.7)	208.8 (6.6)	−9.5	218.3 (6.7)	207.8 (6.5)	−10.5
Age-adjusted	66.2 (1.9)	63.4 (1.8)	−2.8	66.2 (1.9)	63.4 (1.8)	−2.8	66.2 (1.9)	63.2 (1.8)	−3.0
Non-Hispanic White, age, y
< 25	1.7 (0.5)	1.5 (0.5)	−0.2	1.7 (0.5)	1.7 (0.5)	0.0	1.7 (0.5)	1.4 (0.4)	−0.3
25–44	11.7 (1.1)	10.9 (1.1)	−0.8	11.7 (1.1)	10.5 (1.2)	−1.2	11.7 (1.1)	10.4 (1.1)	−1.3
45–64	145.0 (4.3)	129.9 (4.1)	−15.1	145.0 (4.3)	129.9 (4.5)	−15.1	145.0 (4.3)	126.4 (4.0)	−18.6
Age-adjusted	41.5 (1.2)	37.3 (1.1)	−4.2	41.5 (1.2)	37.3 (1.2)	−4.2	41.5 (1.2)	36.3 (1.1)	−5.2
Total population, age, y
< 25	2.0 (0.3)	1.9 (0.3)	−0.1	2.0 (0.3)	1.9 (0.3)	−0.1	2.0 (0.3)	1.9 (0.3)	−0.1
25–44	14.3 (0.7)	13.4 (0.7)	−0.9	14.3 (0.7)	13.5 (0.7)	−0.8	14.3 (0.7)	13.3 (0.7)	−1.0
45–64	146.5 (2.7)	135.5 (2.6)	−11.0	146.5 (2.7)	135.3 (2.6)	−11.2	146.5 (2.7)	133.5 (2.6)	−13.0
Age-adjusted	42.9 (0.7)	39.8 (0.7)	−3.1	42.9 (0.7)	39.8 (0.7)	−3.1	42.9 (0.7)	39.2 (0.7)	−3.7

Note. We calculated the corrected age-specific heart disease death rates after we applied the age- and race-specific heart disease proportions from each of the 3 reference groups to the overall deaths within the overreporting institutions.

^aInstitutions included in nonparametric 95% confidence intervals for the median constructed by using all 63 institutions.

^bInstitutions included in nonparametric 95% confidence intervals for the median constructed by using 50 nonoverreporting institutions.

^cNonoverreporting institutions aggregated.

^dAll rates per 100 000 and rounded to the nearest tenth.

^eAbsolute difference between the reported and corrected rates.

Correcting the count of premature heart disease deaths at overreporting hospitals led to a greater decrease in the citywide premature heart disease death rate among Whites than among Blacks, regardless of the reference group used. The decrease among Whites ranged between 10.1% and 12.5% depending on the reference group used. For example, the rate decreased 10.1%, from 41.5 to 37.3 per 100 000, with reference 1. The decrease among Blacks ranged between 4.2% and 4.5%; decreasing 4.2%, from 66.2 to 63.4 per 100 000 with reference 1.

The corrected disparity measure, i.e., the difference in age-adjusted premature heart disease death rates among Whites and Blacks, was wider than what was actually reported, and this was true for each reference group used (Table 3). Correction using reference 1 and reference 2 increased the disparity measure from 24.6 to 26.1 deaths per 100 000. Correction using reference 3 resulted in the largest increase, from 24.6 to 27.0 deaths per 100 000 (Table 3).

TABLE 3—

Reported and Corrected Citywide Age-Adjusted Premature Heart Disease Death Rates by Race Group and Disparity Measure for Three Reference Standards: New York City, 2008

	Reference 1 Hospitalsa			Reference 2 Hospitalsb			Reference 3 Hospitalsc
Variable	Reported Rated (SD)	Corrected Rated (SD)	Changee	Reported Rated (SD)	Corrected Rated (SD)	Changee	Reported Rated (SD)	Corrected Rated (SD)	Changee
Non-Hispanic Black	66.2 (1.9)	63.4 (1.8)	−4.2	66.2 (1.9)	63.4 (1.8)	−4.2	66.2 (1.9)	63.2 (1.8)	−4.5
Non-Hispanic White	41.5 (1.2)	37.3 (1.1)	−10.1	41.5 (1.2)	37.3 (1.2)	−10.1	41.5 (1.2)	36.3 (1.1)	−12.5
Heart disease disparity measure	24.6	26.1	+6.1	24.6	26.1	+6.1	24.6	27.0	+9.8

Note. We calculated the corrected age-adjusted premature heart disease death rates after we applied the age- and race-specific heart disease proportions from each of the 3 reference groups to the overall deaths within the overreporting institutions.

^aInstitutions included in nonparametric 95% confidence intervals for the median constructed by using all 63 institutions.

^bInstitutions included in nonparametric 95% confidence intervals for the median constructed by using 50 nonoverreporting institutions.

^cNonoverreporting institutions aggregated.

^dAll rates per 100 000 and rounded to the nearest tenth.

^ePercentage change between the reported and corrected rates.

DISCUSSION

To our knowledge, this is the first analysis of the impact of inaccurate cause-of-death reporting on a health disparity measure. Specifically, we examined the impact of heart disease death overreporting on the reported White–Black premature heart disease death disparity in NYC. We found that the overreporting of heart disease deaths in NYC results in an underestimate of the true White–Black disparity because decedents younger than 65 years at overreporting hospitals were more than twice as likely to be White than Black. As a result, estimating a corrected count of premature heart disease deaths at these facilities had the net effect of decreasing the citywide rate of premature heart disease deaths among Whites, thus increasing the disparity.

The error in the disparities measure is not caused by a citywide differential error in reporting heart disease deaths by race, as evidenced by the similar proportions of premature heart disease deaths among Blacks and Whites within the overreporting hospitals and reference hospitals. The race-specific proportions at reference institutions (Table 1) were similar to national data showing that 19% of premature deaths were attributed to heart disease in 2008. By contrast, overreporting hospitals had had higher rates than nationally and tended to overrepresent White decedents. One suggested explanation for this is a desire among overreporting hospital staff, under influence of local funeral directors, to document a cause of death that will not elicit scrutiny during registration as such scrutiny potentially delays funeral proceedings. Such delays are particularly undesirable when cultural or religious practices require that funerals take place within a constrained time period as required in Islamic and Jewish custom.⁶ In fact, 9.6% of decedents at intervention hospitals were documented as having an ancestry from an Islamic majority country, or having Hebrew, Jewish, or Israeli ancestry, in contrast to 3.5% among reference 1 hospitals.^6,15,16

Heart disease is the leading cause of death in NYC and nationally, and has been an area of intense local public health programmatic focus in recent years, including smoking reduction efforts, restaurant food trans-fat bans, and population-level efforts to increase physical activity.¹⁰ Overreporting of heart disease as a cause of death is a known problem in NYC and other US jurisdictions. These findings have important programmatic implications, as successful efforts to reduce this racial disparity could be obscured by efforts to reduce heart disease overreporting, which are occurring in NYC.⁶ More broadly, these findings indicate that errors in vital events registration can obscure or even exacerbate health disparities seen between population subgroups. Other jurisdictions should consider exploring cause-of-death data quality’s impact on key mortality disparity measures, especially when interventions to improve data quality are planned. Our novel methods could be used in these other settings to assess subpopulation disparity indicators.

Although our study was restricted to NYC data, we would expect similar issues in any jurisdiction where the quality of vital event data varies by reporting units, such as hospitals or local death registration offices. Similar to NYC, other ethnically diverse urban areas may also expect demographics to vary by hospital catchment area or other geographic unit, thus biasing measures of disparity. However, we are unable to predict whether such biases will underestimate or overestimate the true disparity, as the direction of the error will depend on the determinants of error across the geographic areas and demographic characteristics. The methods in this article demonstrate that jurisdictions can and should determine whether errors in vital event reporting cluster by hospital or registrar’s office and, if so, whether such clusters of error could affect measures of disparity.

To some extent, the larger problem of error in vital events reporting results from the lack of resources necessary to maintain and improve the data quality within both state and local vital statistics systems. Inadequate jurisdiction resources coupled with the fact that vital events must be reported to the department of health within a constrained time period may affect the quality of data that are provided and disseminated from state and local vital record offices.¹⁷ Poor quality of vital data at the state level has implications for the quality of the national vital statistics data set.¹⁸ Thus, to ensure high accuracy and reliability of state- and national-level vital events data, funding should be directed to quality assurance and improvement activities within vital statistics and vital records departments, and reporting laws should be revisited. Many low-cost or free resources are also available, including an e-learning module created by New York City DOHMH in cooperation with the National Association for Public Health Statistics and Information Systems, and cause-of-death documentation handbooks provided by the CDC.^19–22

Effective methods to remedy error in vital events reporting are needed. One method found to be effective is physician training in death certificate completion to increase the accuracy of cause-of-death data.^23–25 The NYC Office of Vital Statistics is intervening on this issue, specifically targeting heart disease death reporting. From 2009 to 2010, an intervention was implemented within 8 NYC hospitals responsible for 25% of the city’s heart disease death reporting in 2008.⁶ After the intervention, the proportion of heart disease deaths reported at the intervention hospitals decreased 50% to the level of those at the nonintervention hospitals.⁶ Thus, through the expansion and continuation of this intervention, NYC will expect to see changes in the heart disease disparity measure unrelated to the changes in the actual health of the population based on our analysis. This successful intervention will have an impact on the measurement of heart disease disparity in NYC by increasing the measured heart disease disparity as well as the reported rates for other causes of death that will now rise with more accurate reporting.

Some limitations of our analysis include the lack of a widely accepted method for identifying the overreporting hospitals and “reference” hospitals. To determine how our results would vary with the selection of the reference, we chose 3 different reference groups to use to “correct” for overreporting. The most conservative approach included reference 1, the group that was determined by constructing a 95% confidence interval about a central tendency of heart disease proportions for the entire study population of hospitals. Reference 2, although producing similar results to reference 1, was not as representative of the citywide age and race composition and not all hospitals were used in the determination of this group. Reference 3 appeared to be the least conservative approach, resulting in the greatest changes in the heart disease death rates and the disparity measure. Perhaps this is because of the contribution of hospitals with the lowest heart disease death proportions to reference 3. Heart disease, however, may have been overreported at reference hospitals, making our analysis conservative. Furthermore, we focused on premature heart disease deaths because this disparity is a focus of our agency’s interventions.¹⁰ Heart disease overreporting increases substantially with age.⁴ A similar study examining all heart disease deaths might have found a greater percentage difference between the measured disparity and estimated “true” disparity on that basis. Finally, this study was not designed to determine the reasons for overreporting of heart disease deaths. Other reports provide further information on the root cause of this problem and effective remedies.⁶

These findings have important implications for research and policies based on disparities data; they stress the importance of accurate vital event data and, more specifically, accurate cause-of-death data. Errors in vital data reporting are often acknowledged as a limitation, but the extent to which these errors have an impact on public health research, policy development, and program planning has not been previously investigated to our knowledge. Given the potential impact on population health disparity measurement, such errors may lead to certain populations not receiving the health resources they need. On this basis, jurisdictions should allocate resources to ensure the quality of vital event data, including resources to allow routine outreach and training to physicians about proper cause-of-death documentation for death registration.