Racial Misclassification in Mortality Records Among American Indians/Alaska Natives in Oklahoma from 1991-2015

Abstract

Objective:

The primary purpose of this study was to compare age-adjusted mortality rates (AAMRs) before and after linkage with IHS records, adjusting for racial misclassification. We focused on differences in racial misclassification by gender, age, and geographic differences, sub-state planning districts (SSPDs), and cause of death. Our secondary purpose was to evaluate time trends in misclassification from 1991-2015.

Design:

Retrospective, Descriptive Study

Setting:

Oklahoma

Participants:

Persons contained in the Oklahoma State Health Department Vital Records

Main Outcome Measures:

To evaluate the age-adjusted mortality ratio (AAMR) pre- and post-IHS record linkage (misclassification rate ratio; MRR), and to evaluate the overall trend of racial misclassification on mortality records measured through annual percent change (APC) and average annual percent change (AAPC).

Results:

We identified two stable trends of racial misclassification upon death for American Indians/Alaska Natives (AI/ANs) from 1991 to 2001 (APC: −0.2%; 95% CI: −1.4%, 1.0%) and from 2001 to 2005 (APC: −6.9%; 95% CI: −13.7%, 0.4%). However, the trend identified from 2005 to 2015 decreased significantly (APC: −1.4%; 95% CI: −2.5%, −0.2%). For the last five years available (2011–2015), the racial misclassification adjustment resulted in higher mortality rates for AI/ANs reflecting an increase from 1 008 per 100 000 to 1 305 per 100 000 with the linkage process (Table 2). There were an estimated 3 939 AI/ANs in Oklahoma who were misclassified as another race upon death in those five years, resulting in an underestimation of actual AI/AN deaths by nearly 29%.

Conclusions:

An important result of this study is that misclassification is improving; however, this effort needs to be maintained and further improved. Continued linkage efforts and public access to linked data are essential throughout the US to better understand the burden of disease in the AI/AN population.

Introduction

Racial misclassification among American Indians and Alaska Natives (AI/ANs) has been well documented in the literature for over 40 years^1–4 and has resulted in a systematic underestimation of the burden of mortality. This misclassification influences states and Tribes/Nations both financially and administratively. It is essential that accurate data be available for national, state and local agencies to determine where disparities exist in order to identify points of intervention, plan and evaluate programs, procure and allocate resources, and measure outcomes.

Misclassification, as well as other factors (i.e., undersampling in surveys), of AI/ANs has been responsible for underestimating the presence and impact of HIV/AIDS, kidney disease, diabetes, cancer, neonatal, and all-cause mortality in many locations throughout the US.^2,5–32 Misclassification varies by region, and the Southern Plains area (Oklahoma, Kansas, and Texas) may experience the greatest impact, with around 37% of AI/AN decedents misclassified as a different race (primarily as white) on death certificates.³³ In Oklahoma, racial misclassification is associated with routinely underestimating morbidity and mortality rates of AI/ANs including areas such as infectious disease, sexually transmitted infections (STI), injury, and cardiovascular disease.^34–36

Despite the long-term recognition of racial misclassification, the problem has been slow to resolve. Multiple methods have been implemented to improve the problem, including regular administrative linkages with Indian Health Service (IHS) and birth, death, and cancer records; linkage with tribal membership or citizenship rolls; increased training of medical examiners/coroners, funeral home directors, and health facilities administrators; and using self-reported race categories.^{2,18,20,23,24,37–40} Studies concerning associations of racial misclassification among AI/ANs beyond the overall impact on morbidity and mortality, however, are virtually non-existent. In addition, the National Death Index linkage with IHS records has not occurred on a national scale since 2009, resulting in nearly ten years of mortality data that were not linked with IHS records (although an updated linkage is planned). Because of this data gap and the concern for inaccurate mortality data for AI/ANs in Oklahoma, staff from the University of Oklahoma Health Sciences Center, Oklahoma State Department of Health, the Oklahoma City Area IHS Office, the Oklahoma Area Tribal Epidemiology Center and CDC staff in New Mexico collaborated to update the linkage with IHS records for Oklahoma mortality files from 2010–2015, which were the most recent data available at the time of the linkage.

The primary purpose of this study was to compare age-adjusted mortality rates (AAMRs) before and after linkage with IHS records, adjusting for racial misclassification. We focused on differences in racial misclassification by gender, age, and geographic differences; sub-state planning districts (SSPDs); and cause of death. Our secondary purpose was to evaluate time trends in misclassification from 1991–2015.

Methods

Mortality

We obtained mortality rates per 100 000 population in the state of Oklahoma for AI/ANs and for each of the SSPDs using the Oklahoma State Department of Health’s web-based query system, Oklahoma Statistics on Health Available for Everyone (OK2SHARE).⁴¹ For our time-trend analysis we used AAMRs for AI/ANs from 1991–2015. For all other analyses we used crude, age-specific, and AAMRs for the last five years available (2011–2015). Mortality rates were age-adjusted by the direct method employing the 2000 US standard population using the bridged population estimates. This project was reviewed and approved by the University of Oklahoma Health Sciences Center Institutional Review Board. This manuscript was submitted to the Oklahoma City Area Indian Health Service Institutional Review Board and was approved for publication.

Linkage Methods

The IHS linkage process has been described elsewhere.³³ Briefly, AI/AN individuals must provide proof of membership in a federally recognized tribe to receive healthcare from IHS, Tribal, and Urban Indian health (I/T/U) facilities. A public health database (in this case the Oklahoma vital records mortality data) was linked to the master patient file for IHS. There is a strict protocol and no health information was exchanged with the exception of basic demographic variables that are needed to conduct the record linkage (name, gender, date of birth, address). Oklahoma has participated in this IHS linkage process, primarily for cancer surveillance data, for over 20 years^7,34 and participated in the national linkage project described by Jim et al (2014).³³ For this study, the State of Oklahoma followed the same linkage process to update their death records from 2010 to 2015. OK2SHARE provides both the unadjusted (before linkage) and adjusted (after linkage) data. The unadjusted (original race code) on death certificates were reported by funeral directors and physicians or their staff. These individuals frequently do not ask the family about the racial identification of the deceased individual. The IHS-linked race code classifies individuals as AI/AN if 1) the race reported on the death certificate was AI/AN or 2) the individual record linked to IHS regardless of the original race reported on the death certificate.

Statistical Analyses

We calculated mortality rates prior to adjustment for racial misclassification using the original race code from OK2SHARE and mortality rates adjusted for racial misclassification using the IHS-linked race codes. We then calculated a misclassification rate ratio (MRR) and corresponding 95% confidence interval (CI) of the AAMR adjusted for misclassification compared to the unadjusted AAMR for racial misclassification for all racial groups (White, African American, AI/AN, and Asian/Pacific Islander) and for AI/ANs by gender, age group, and SSPD. We also obtained the AAMR by cause of death for the top 10 causes of death among non-Hispanic AI/ANs in the US from 2015.⁴² We calculated the percent difference in the AAMR pre- and post- IHS linkage.

Time trends analysis

We used the Joinpoint Regression Program (version 4.6.0.0; National Cancer Institute, Bethesda, Maryland) to analyze yearly time trends of the percentage of AI/ANs that were misclassified from 1991 through 2015. Trends in misclassification over time were characterized using annual percent change (APC). We allowed a maximum of four joinpoints in the model. If the slope of the trend was significantly different from zero, an APC estimate was reported as increased or decreased; however, if there was no significant difference, the trend was reported as stable. To account for transition in trends, we also estimated the average annual percent change (AAPC).

Geography

For this study we used the SSPDs in Oklahoma, which are voluntary associations of local governments formed under Oklahoma law that deal with problems and planning needs that cross the boundaries of individual local governments, such as counties, cities, and towns. SSPDs complete regional data collection and analysis, mapping, and coordination of environmental, economic, social program plans, Rural Fire Defense, Capital Improvements Plan, Rural Economic Action Plan Grant programs and Hazard Mitigation Planning. There are eleven SSPDs in Oklahoma with each SSPD encompassing between three counties in the Indian Nations Council of Governments (INCOG) to ten counties in the Southern Oklahoma Development Association (SODA) (Figure 1).

Figure 1:

Sub-state planning districts (SSPDs) in Oklahoma 2018

ACOG Association of Central Oklahoma Governments, ASCOG Association of South Central Oklahoma Governments, COEDD Central Oklahoma Economic Development District, EODD Eastern Oklahoma Development District, GGEDA Grand Gateway Economic Development Association, INCOG Indian Nations Council of Governments, KEDDO Kiamichi Economic Development District of Oklahoma, NODA Northern Oklahoma Development Authority, OEDA Oklahoma Economic Development Authority, SODA Southern Oklahoma Development Association, SWODA South Western Oklahoma Developmental Authority

Results

Time Trends Analysis

There was a decrease in the percentage of AI/ANs who were misclassified from 1991–2015 (Figure 2). The AAPC for percentage of racially misclassified AI/ANs over the entire study period was −1.8% (95% CI: −3.1%, −0.5%) (Table 1). We identified two stable trends from 1991 to 2001 (APC: −0.2%; 95% CI: −1.4%, 1.0%) and from 2001 to 2005 (APC: −6.9%; 95% CI: −13.7%, 0.4%). However, the trend identified from 2005 to 2015 decreased significantly (APC: −1.4%; 95% CI: −2.5%, −0.2%).

Figure 2.

Percent of American Indians/Alaskan Natives misclassified as another race by year with 95% confidence intervals, Oklahoma 1991-2015

Table 1.

Percent of American Indians/Alaska Native (AI/AN) misclassified as another race by year, Oklahoma 1991-2015.^a

Joinpoint*	Years	APC (95% CI)	p-value
Trend 1	1991-2001	−0.2 (−1.4, 1.0)	0.74
Trend 2	2001-2005	−6.9 (−13.7, 0.4)	0.06
Trend 3†	2005-2015	−1.4 (−2.5, −0.2)	0.03
Overall‡	1991-2005	−1.8 (−3.1, −0.5)	0.007

Abbreviations: APC, annual percent change; AAPC, average annual percent change;

^*Two joinpoints were identified for AI/ANs.

^†Annual percent change is significantly different from 0 at alpha of 0.05.

^‡Average annual percent change is significantly different from 0 at alpha of 0.05.

Mortality Rates

For the last five years available (2011–2015), the racial misclassification adjustment resulted in a higher all-cause mortality rate for AI/ANs reflecting an increase from 1 008 per 100 000 to 1 305 per 100 000 with the linkage process (Table 2). There were an estimated 3 939 AI/ANs in Oklahoma who were misclassified as another race upon death in those five years, resulting in an underestimation of actual AI/AN deaths by nearly 29%. Prior to IHS record linkage, African Americans had the highest all-cause AAMR; however, after adjusting for misclassification through IHS-linkage, the all-cause AAMR for AI/ANs was the highest of any racial group. When accounting for misclassification, AI/ANs had a significantly higher all-cause mortality rate compared to whites (868 per 100 000) with an all-cause mortality rate ratio of 1.5 (95% CI: 1.4, 1.6). AI/ANs experienced an all-cause mortality rate ratio of 1.3 (95% CI: 1.2, 1.4) when compared to African Americans (all-cause AAMR 1 006 per 100 000) and an all-cause mortality rate ratio of 2.6 (95% CI: 2.6, 2.9) when compared to Asian/Pacific Islanders (all-cause AAMR 475 per 100 000) (data not shown).

Table 2.

Number of deaths, crude death rates, age-adjusted mortality rates per 100 000 (AAMRs) and misclassification rate ratios for American Indians/Alaska Natives before and after Indian Health Service (IHS) linkage by demographic groups, Oklahoma 2011-2015.

	Before IHS Linkage			After IHS Linkage
Category	Deaths	Crude Rate	AAMR*	Deaths	Crude Rate	AAMR*	Misclassification Rate Ratio† (95% CI)
Race
White	163 083	1 079.5	889.2	159 369	1 054.9	868.3	1.0 (0.9, 1.0)
African American	12 175	723.4	1 023.3	11 959	710.6	1 006.0	1.0 (0.9, 1.1)
Asian/Pacific Islander	1 217	267.8	477.4	1 208	265.9	475.0	1.0 (0.9, 1.1)
American Indian/Alaska Native	13 821	690.4	1 008.2	17 760	887.2	1 304.8	1.3 (1.2, 1.4)
American Indian/Alaska Native Only
Gender
Male	7 503	757.8	1 198.1	9 399	949.3	1 509.7	1.3 (1.2, 1.3)
Female	6 318	624.4	851.3	8 361	826.3	1 132.2	1.3 (1.3, 1.4)
Age Group (years)‡
Under 1	298	895.6	--	315	946.7	--	1.1 (0.9, 1.2)
1-4	61	41.0	--	73	49.1	--	1.2 (0.8, 1.6)
5-14	80	21.1	--	105	27.7	--	1.3 (0.8, 1.9)
15-24	343	98.8	--	457	131.6	--	1.3 (1.1, 1.6)
25-34	590	207.6	--	758	266.7	--	1.3 (1.1, 1.5)
35-44	821	343.4	--	1 046	437.5	--	1.3 (1.1, 1.4)
45-54	1 737	768.9	--	2 194	971.3	--	1.3 (1.2, 1.4)
55-64	2 567	1 401.1	--	3 242	1 769.5	--	1.3 (1.2, 1.3)
65-74	2 756	2 671.6	--	3 544	3 435.5	--	1.3 (1.2, 1.4)
75-84	2 544	5 813.8	--	3 319	7 584.9	--	1.3 (1.3, 1.4)
≥85	2 022	14 815.4	--	2 703	19 805.1	--	1.3 (1.3, 1.4)
Sub-State Planning District‡
ACOG	1 638	533.9	895.3	2 312	753.5	1 269.0	1.4 (1.3, 1.5)
ASCOG	1 068	753.9	1 018.1	1 275	900.0	1 221.5	1.2 (1.1, 1.3)
COEDD	1 315	816.6	1 184.5	1 542	957.6	1 387.2	1.2 (1.1, 1.3)
EODD	2 726	721.8	1 038.2	3 342	884.9	1 282.2	1.2 (1.2, 1.3)
GGEDA	2 053	758.6	1 056.8	2 609	964.1	1 362.2	1.3 (1.2, 1.4)
INCOG	2 085	649.7	1 014.9	2 899	903.3	1 417.3	1.4 (1.3, 1.5)
KEDDO	1 063	675.0	852.2	1 377	874.3	1 107.2	1.3 (1.2, 1.4)
NODA	375	676.7	1 067.4	479	864.4	1 396.1	1.3 (1.2, 1.4)
OEDA	60	539.3	1 068.4	72	647.1	1 316.7	1.2 (0.9, 1.5)
SODA	1 243	734.2	1 030.4	1 593	940.9	1 326.3	1.3 (1.2, 1.4)
SWODA	187	627.3	952.4	249	835.3	1 303.9	1.4 (1.2, 1.6)

AAMR – Age-Adjusted Mortality Rate

^*Age-adjusted mortality rate only calculated for overall and by gender. Age-specific rates were calculated by age group and are, therefore, not age-adjusted.

^†The misclassification rate ratios by age group are not age-adjusted, but age-specific rate ratios

^‡Unknowns: n=2 for age groups before IHS linkage, n=4 after IHS linkage; n=8 for sub-state planning districts before IHS linkage and n=11 after IHS linkage

We observed little difference in racial misclassification by gender for AI/ANs, with a male MRR of 1.3 (95% CI: 1.2, 1.3) compared to 1.3 (95% CI: 1.3, 1.4) among females (Table 2). When stratifying by age for AI/ANs, we observed elevated MRR for all age groups above 1 year of age, with statistically significant increases among AI/ANs over age 15 years of age at death.

Of the 11 SSPDs, all districts had an increase in mortality after IHS linkage (Table 2). While the increase in the Oklahoma Economic Development Authority (OEDA; rural northwestern Oklahoma) was not statistically significant, the number of deaths was smallest in this SSPD and the population was also smaller than other SSPDs. We observed the highest MRR in the two districts that encompass the two major metropolitan areas of Oklahoma: The Association of Central Oklahoma Governments (ACOG; Oklahoma City) (MRR 1.42; 95% CI: 1.32, 1.51) and INCOG (Tulsa) (MRR 1.40; 95 %CI: 1.31, 1.49). In addition to experiencing one of the highest rates of misclassification, AI/ANs in the INCOG SSPD had the highest all-cause AAMR in the state after adjustment for racial misclassification. Before the IHS linkage however, the INCOG area AI/ANs all-cause AAMR was in the bottom 50%.

When stratifying by the top 10 causes of death, the percentage difference in the AAMR pre- to post- IHS records linkage ranged from a 15.2% increase (Chronic Liver Disease and Cirrhosis) to 40.1% increase (Chronic Lower Respiratory Disease) in the AAMR for AI/ANs (Table 3). For whites, the AAMR decreased from 1.6% (Cerebrovascular Disease) to 4.1% (Unintentional Injuries and Chronic Liver Disease and Cirrhosis). Preceding the IHS record linkage, the white population had a higher AAMR for Chronic Lower Respiratory Disease, Cerebrovascular Disease, and Suicide compared to AI/ANs. However, after the IHS linkage, the data now reflect a more accurate burden of death due to disease and injury for AI/ANs (Table 3).

Table 3.

Number of deaths, crude death rates, age-adjusted mortality rates per 100 000 (AAMRs) and rate ratios of before and after Indian Health Service Linkage by for the top 10 causes of death in American Indians/Alaska Natives, Oklahoma 2011-2015.

	White					American Indian/Alaska Native
	Before IHS Linkage		After IHS Linkage			Before IHS Linkage		After IHS Linkage
Cause of Death	Deaths	AAMR	Deaths	AAMR	Percent Difference in AAMR	Deaths	AAMR	Deaths	AAMR	Percent Difference in AAMR
Diseases of the Heart	42 112	225.0	41 277	221.0	−2.0	3 136	246.0	4 019	316.2	28.6
Cancer	34 753	184.9	33 932	181.0	−2.4	2 692	193.0	3 560	256.1	32.8
Unintentional Injuries	9 951	61.4	9 584	58.9	−4.1	1 218	72.2	1 605	95.6	32.4
Diabetes	4 895	26.3	4 765	25.6	−2.7	789	58.2	927	69.1	18.7
Chronic Liver Disease and Cirrhosis	2 194	12.3	2 111	11.8	−4.1	539	32.3	624	37.4	15.8
Chronic Lower Respiratory Disease	12 444	65.9	12 186	64.6	−2.0	667	51.4	930	72.0	40.1
Cerebrovascular Disease	8 096	43.4	7 967	42.7	−1.6	519	42.0	659	53.9	28.3
Suicide	3 050	19.8	2 934	19.1	−3.5	307	16.7	429	23.3	39.5
Kidney Disease and Infections	2 319	12.4	2 265	12.1	−2.4	277	21.4	336	26.5	23.8
Pneumonia and Influenza	3 150	16.9	3 087	16.6	−1.8	273	22.3	342	27.7	24.2

AAMR: Age-Adjusted Mortality Rate per 100,000 population

Discussion and Conclusions

Mortality data are a primary source of information used in public health to understand the population burden of disease and injury, identify disparities among population subgroups, monitor trends over time, and prioritize programs and resource allocation. With many decisions relying on the assumption of accurate data, misclassification of race on death certificates is a significant problem when describing health disparities in Oklahoma and throughout the US. Although this most recent study shows that racial misclassification on death certificates has improved through time, it continues to be a problem in Oklahoma with 30% more deaths among AI/AN residents being identified through linkage with IHS records. Misclassification remains high in all regions of Oklahoma and in most disease categories. Furthermore, we know that there is still misclassification for which the process of IHS-linkage does not account, as not all AI/ANs seek care at I/T/Us. Other mortality linkage studies^2,43 and commentaries³ have previously described the limitations of using IHS-linked data to correct for racial misclassification, particularly in places where IHS services are less accessible and for AI/ANs who were eligible for but never accessed IHS services.

Oklahoma has invested in continued IHS-linkage to improve mortality data, and, to our knowledge, is the only state that makes this misclassified data available on a public web query site (OK2SHARE). The Centers for Disease Control and Prevention make their IHS-linked cancer incidence data publicly available through Wide-ranging ONline Data for Epidemiologic Research (CDC WONDER); however, the unlinked data are not available to examine misclassification. Publicly available IHS-linked data have been very important for tribal epidemiology centers, tribes, and researchers and even more important for program managers and policy makers who are unable to measure the extent of conditions or the impacts of their programs.

Health measures must be accurate for states and healthcare systems to plan for the use and costs of health care services. Accurate data are needed to evaluate the trends related to the quality and delivery of health care services, as well as the capacity of various components of health care. Although the corrected data are available in OK2Share, not all health care providers are familiar with or access this data source when planning service delivery. This also demonstrates the need to educate health care providers on the impacts of racial misclassification in Oklahoma and on data sources, like OK2Share, where they can obtain the adjusted mortality rates.

Mortality records are not the only healthcare record system affected by racial misclassification. Morbidity data (hospital discharge data, Emergency Department data, etc.) can be affected, as well. To date, hospital discharge data has not been linked with IHS records in Oklahoma. Underestimating AI/AN morbidity can cause additional problems for healthcare entities, health program managers and policy makers because they are unable to obtain accurate data to evaluate and measure the effects of public health initiatives for the AI/AN population. Accurate hospital discharge data and Emergency Department data could be used to illustrate where AI/ANs seek acute care outside of the IHS system and allow for IHS to better respond to needs.

There are several limitations in this study. One limitation is the change in reporting quality over time. During the reporting period, there have been changes in the quality of some variables collected by the Oklahoma Vital Records. There have also been changes to categorization within certain variables, one of which is how race is collected. In particular we observed that after 2000, when multiple race categories could be reported, we observed dramatic improvement in racial misclassification. Furthermore, there continues to be misclassification in the AI/AN population in Oklahoma. Not all AI/ANs seek care through IHS or I/T/U facilities during their life and may still have their race reported as another race as opposed to AI/AN. A final limitation was the lack of tribal specific data. Because of the large number of Tribal headquarters and Tribal heterogeneity in the state, Tribal members live throughout the state not just in Tribal jurisdictional areas. The geographic dispersion of the AI/AN population leads to an additional challenge in reviewing any health data for AI/AN populations in Oklahoma.

This analysis highlights the need for several additional studies and recommendations. Estimates of the number of racially misclassified cases among individuals who do not use IHS services is necessary. The use of geographic information systems (GIS) analysis could be implemented to determine accuracy of coding by tribal boundaries. The most important recommendation from this analysis is to continue the enhancement systems that support IHS linkages with surveillance systems throughout the US. Finally, we recommend public access to linked data so that the data remain accessible. The linkage projects are of no value to program planners, evaluators, and researchers if the data are not accessible. As a final point, it is important to determine how to handle the adjusted race variable and determine whether we provide dual estimates (as in OK2Share) or if there is a more appropriate method. In addition, tribal nations need AI/AN-specific information to plan health services and outreach. Strides cannot be made to improve the health status of a population when inaccurate information prevents effective health planning. Public and private entities require this information so that the conditions which most profoundly affect the health of AI/ANs can effectively be targeted in order to provide higher quality of care, improve access, targeted health facilities, cost containment, and allow the pursuit of more resources for prevention and treatment. A final issue that needs to be made clear is that these data reflect statewide population-based information. They do not represent tribal health services or the IHS user population, and thus, do not reflect the quality of care provided by tribal health services, tribal programs, or IHS.

As mentioned above, an important result of this study is that misclassification is improving; however, this effort needs to be maintained and further improved. Continued linkage efforts and public access to linked data are essential throughout the US to better understand the burden of chronic disease in the AI/AN population. Accurately recorded mortality and morbidity records can also allow for better understanding of acute causes of illness (i.e., Influenza), injuries, and newly emerging epidemics like the current opioid crisis. Having more accurate data for AI/ANs will help inform healthcare providers and funders about the true burden of disease and other health outcomes to ultimately improve the health of AI/ANs.

Implications for Policy & Practice.

• American Indian and Alaska Native surveillance data are important for program planning and evaluation of health services, particularly those services designed to reduce health disparities.

• Linkages to Indian Health Service data systems are an established way of improving the well-known issues of misclassification of American Indians and Alaska Natives in the US.

• Rapid and ongoing linkages to Indian Health Service data systems would allow for a more accurate identification and can support effort to improve access to health care resources, particularly for vulnerable populations.

• Access to linked data is important for program planning and evaluation staff throughout the US.