Figure 1. Comorbidities by age and
ethnicity. Percentages of American Indian (AI) and non-Hispanic white
(NHW) patients with an Elixhauser comorbidity index9 of 3 or more by age (P < .01 for patients aged <50, 50-65, and 66-80 years; P = .10 for those aged >80 years).
Figure 2. The continuum of cancer care.
American Indians are subject to disparities at multiple points along
the continuum of cancer care.
Customize your JAMA Network experience by selecting one or more topics from the list below.
Markin A, Habermann EB, Zhu Y, et al. Cancer Surgery Among American Indians. JAMA Surg. 2013;148(3):277–284. doi:10.1001/jamasurg.2013.1423
Author Affiliations: Division of Surgical
Oncology, Department of Surgery, Surgical Outcomes Research Center
(Mr Markin, Drs Habermann, Abraham, Vickers, and Al-Refaie, and Ms
Zhu) and Center for Health Equity (Mr Markin and Dr Ahluwalia), University
of Minnesota, and Department of Surgery, Minneapolis Veterans Affair
Medical Center (Dr Al-Refaie), Minneapolis. Dr Habermann is now with
the Center for the Science of Health Care Delivery, Division of Health
Care Research and Policy, Mayo Clinic, Rochester, Minnesota. Dr Al-Refaie
is now with the Department of Surgery, Lombardi Comprehensive Cancer
Center, Georgetown University Hospital, Washington, DC.
Importance American Indians (AIs) have the poorest cancer survival rates
of any US ethnic group. Late diagnosis, poor access to specialty care,
and delays in therapy likely contribute to excess mortality. Surgery
plays a central role in therapy for solid organ cancer.
Objective To determine whether operative outcomes also contribute to poor
long-term survival among AI patients with cancer.
Design Population-based retrospective cohort study comparing patient-
and hospital-level factors and short-term operative outcomes for AI
and non-Hispanic white patients. Survey-weighted multivariate analyses
assessed the effect of AI ethnicity on hospital location, in-hospital
mortality, and prolonged length of stay.
Setting A 20% stratified sample of all US community hospitals.
Patients Patients undergoing oncologic resection for 1 of 20 malignant
neoplasms in the Nationwide Inpatient Sample from January 1, 1998,
through December 31, 2009.
Main Outcome Measure In-hospital mortality, length of stay, and hospital location
(rural vs urban).
Results Of 740 878 patients who met our inclusion criteria, 3048
were AIs. The AI patients were younger, more likely to undergo cancer
surgery at rural hospitals, and more likely to be admitted for nonelective
procedures and had more comorbidities than non-Hispanic white patients
of similar ages (all, P < .05).
The AI patients had comparable inpatient mortality and length of stay.
Conclusions and Relevance This investigation is the largest study of surgical outcomes
among AIs to date and the first to focus on cancer surgery. This relatively
young cohort does not experience poor outcomes after oncologic resection.
Future research should uncover other factors in the continuum of cancer
care that may contribute to the poor long-term survival of AI patients
with cancer, including delivery of perioperative therapies.
American Indians (AIs) have the worst cancer survival rates
of any US ethnic group,1 and available
evidence suggests that mortality due to cancer among AIs is increasing.2 Relative to non-Hispanic white (NHW) patients,
AI patients present with more advanced stages of cancer,3 are less likely to receive appropriate therapy,4 and experience significant delays between
diagnosis and surgical intervention.5 Indeed, the most recent data from the American Cancer Society indicate
that AI and Alaskan Native women are the only demographic groups not
to benefit from a remarkable decline in cancer death rates during
the past decade.6
Oncologic resection represents a critical component of therapy
for most solid malignant neoplasms. However, a paucity of information
is available regarding the AI cancer surgery experience, although
significant research has been dedicated to disparities in surgical
oncology for other ethnic and racial minorities.7 Several obstacles limit research of this
type among AIs, including small numbers of available subjects, regional
and tribal heterogeneity,2 and racial
miscategorization.8 As such, we do
not know how or whether unique patterns of oncologic resections contribute
to poor long-term cancer survival among AI patients.
To better understand the origins of excess long-term cancer
mortality among AIs, this study used a large national database to
examine short-term operative outcomes. We compared patient characteristics,
hospital characteristics, and hospital course for AI and NHW patients
undergoing cancer surgery. We hypothesized that AI ethnicity affects
patterns of surgical oncologic procedures and operative outcomes.
As the largest all-payer source of hospital stay data in the
United States, the Nationwide Inpatient Sample (NIS) is uniquely suited
to answer these questions. The NIS contains nationally representative
data from a 20% stratified sample of nonfederal community hospitals
and includes 5 million to 8 million inpatient discharges per year.
We used the NIS to identify admissions for cancer surgery from January
1, 1998, through December 31, 2009. Alaska does not submit data to
the NIS, so the present study does not include information on Alaskan
natives, who frequently are grouped with AIs for research purposes.
In addition, although the NIS is well suited to studying perioperative
outcomes, it does not include cancer-specific variables or long-term
outcomes. All NIS data are de-identified; this study was deemed exempt
by the institutional review board of the University of Minnesota.
Data were extracted for all admissions that contained matching
procedure and diagnosis codes from the International
Classification of Diseases, Ninth Revision, for 1 of 22 solid
organ tumor resections. Patient-level information included age, sex,
insurance status, Elixhauser comorbidity index,9 zip code–level annual income, inpatient
mortality, disposition at discharge, length of stay (LOS), and cost
of stay. Hospital-level data were extracted for size, teaching status,
urban vs rural location, and region in the United States. We excluded
from analysis patients identified as African American, Hispanic, Asian/Pacific
Islander, or other and those missing race/ethnicity information. Fewer
than 10 AI patients were diagnosed with bone or adrenal cancers, so
these 2 sites were not included in the subsequent analyses.
Bivariate analyses were performed to compare patient- and hospital-level
variables and inpatient operative outcomes by ethnicity (AI vs NHW).
Association was assessed with the χ2 and 2-tailed t tests for categorical and continuous variables,
respectively, with P ≤ .05 considered
statistically significant. The 2 groups differed in the distribution
of age, so stratified analyses compared comorbidities among AI and
NHW patients for each age group. We stratified mortality by rural
location to explore the effect of hospital location on operative outcomes
for both ethnic groups.
Survey-weighted logistic regression models were constructed
to assess the effect of AI ethnicity on hospital location and postoperative
mortality using factors that emerged as significant in the bivariate
analyses and clinically relevant factors. We created a prolonged LOS
variable and developed regression models to assess whether AI ethnicity
predicted prolonged LOS. Prolonged LOS was defined as a hospital stay
greater than or equal to the 75th percentile of the cohort.
The information included in the NIS database changed during
the study period. Of particular relevance to the present investigation
was the addition of the Elixhauser comorbidity index9 in 2002 and adjustment of the quartiles
used for the income factor in 2003. To account for these variations,
multiple regression models were developed for each outcome. The models
reported herein correspond to the years 2003 through 2009 and most
accurately fit the data as assessed using the C statistic.
Finally, we performed additional tests for interactions among
ethnicity, income, and insurance status. Owing to the colinearity
of income and insurance status, the regression was repeated after
separately removing each factor. Additional sensitivity analyses included
repeating all models after removing patients older than 80 years.
Data were analyzed using commercially available software (SAS, version
9.1; SAS Institute, Inc).
Less than 1% of our cohort were AIs (n = 3052). Bivariate
analyses (Table 1) indicated
that, compared with NHW patients, AI patients were younger (P < .001) and more likely to reside
in lower-income zip codes (P < .001),
to be covered by Medicaid (P < .001),
and to be admitted to the hospital for nonelective procedures (P < .001). Cross-tabulation of the
comorbidity index by age revealed that, despite comparable comorbidities
for AI and NHW patients for all ages combined, AIs had more comorbidities
in each age group (Figure 1). Relative to NHW patients, the oncologic resections performed among
AI patients included a greater proportion of tumors of the stomach,
testicle, brain, uterus, kidney, thyroid, and gallbladder. Conversely,
AI patients were less likely to undergo lung, bladder, or colon resections
On bivariate analysis (Table 1), AI patients were more likely to receive their cancer surgery at
rural hospitals. In addition, hospital region within the United States
varied significantly by ethnicity, with a significantly greater proportion
of AI patients found in the western United States compared with NHW
patients. Our multivariate analysis (Table 3) confirmed that AI ethnicity was associated with
receipt of cancer surgery at rural hospitals (odds ratio [OR], 1.31
[95% CI, 1.16-1.50). Of note, mean annual income in the zip code of
residence was the strongest predictor of receipt of surgery at rural
hospitals; those in the lowest-income zip codes were more than 20
times as likely to be treated at rural hospitals than those residing
in the highest-income zip codes (OR, 22.72 [95% CI, 21.28-23.81];
data not shown).
An interaction was observed between ethnicity and income (P = .02). Stratification by income revealed
that, among residents of low-income zip codes, AI ethnicity predicted
care at rural hospitals; this association lost significance for high-income
patients (data not shown).
Overall, our unadjusted analysis showed no significant differences
in inpatient mortality between AI and NHW groups, but AI patients
exhibited a nonsignificant trend toward longer hospital stays (mean
LOS, 6.2 vs 6.1 days; median, 4 days for both groups [P = .41]). After adjustment for covariates using
multivariate regression, AI ethnicity did not predict overall poorer
in-patient mortality or prolonged LOS (Table 3). When we examined each type of malignant neoplasm
individually, we found that unadjusted mortality was comparable for
all but 2 tumor sites. The AI patients had poorer in-hospital mortality
than NHW patients after surgery for cancer of the rectum (4.55% vs
1.75%; P = .005 [data not shown])
or esophagus (25.00% vs 7.22%; P = .02
[data not shown]); however, fewer than 10 AI patients died after either
To ensure that these results were not influenced by our choice
of modeling, we conducted repeated sensitivity and interaction analyses.
We found no significant interactions between ethnicity and income
or insurance status for inpatient mortality. Similarly, ethnicity
did not interact significantly with insurance in the prediction of
prolonged LOS. However, an interaction was observed between race and
income in the prediction of prolonged LOS. Stratified analyses revealed
that residence in low-income zip codes was a significant predictor
of prolonged LOS for NHW but not for AI patients. Given the colinearity
of insurance status and income, we repeated our models after separately
removing each. The impact of AI ethnicity did not change (data not
shown). Our choice to use income was largely owing to a better model
fit compared with insurance status (C statistics
for income vs insurance, 0.79 vs 0.68). Finally, because of the potential
confounding by differences in age distribution, we repeated our analyses
after removing patients older than 80 years from the models and found
that our results remained unchanged (data not shown).
The present investigation is, to our knowledge, the largest
study of AI surgical outcomes to date and the first to focus on cancer
surgery. In this large national cohort, AI ethnicity significantly
influenced patterns of oncologic procedures. The AI patients undergoing
cancer surgery were younger, lived in lower-income zip codes, and
had more comorbidities than did similarly aged NHW patients. The AI
patients were more likely to undergo operations at rural hospitals
and more likely to undergo nonelective procedures; however, AI patients
did not demonstrate worse inpatient mortality or longer LOS.
To our knowledge only 6 other studies have examined surgical
outcomes among AIs,10-15 including only 2 unique samples of more than 160 AIs. Our results
are consistent with previous reports that AI surgery patients are
on average younger,11-14 a finding potentially attributable to lower overall life expectancy,
which leads to fewer elderly AI patients.16Our results showing that AI patients are
substantially more likely to be admitted for nonelective procedures
are consistent with the trend toward urgent admission status observed
among AI patients undergoing coronary artery bypass grafting.14 In contrast to our findings for cancer
surgery, AI patients have been reported to have significantly higher
in-hospital mortality after coronary artery bypass grafting14 and higher 30-day mortality after major
noncardiac surgical procedures.11 These
differences may reflect features specific to cancer surgery and differences
in the factors that underlie in-hospital mortality and 30-day mortality.
In answer to this study's central question, differences in short-term
operative mortality do not appear to contribute to the overall poor
cancer survival rates among AIs, with the possible exception of rectal
and esophageal cancer. Ethnic disparities in the delivery of cancer
care exist at multiple points along the pathway from entry into the
health care system to definitive management (Figure 2). American Indians receive cancer diagnoses when
the disease is at more advanced stages,3 have significant delays from diagnosis to treatment,5 and are less likely to receive appropriate
therapy.4 However, the long-term
survival deficit has been reported to persist even after controlling
for tumor grade and stage at diagnosis and the type of treatment delivered.17 Our data contribute to an understanding
of poor survival among AIs with cancer by demonstrating that the remaining
variance in long-term survival is not accounted for by in-hospital
mortality after surgical procedures. Differences at other points along
the continuum of cancer care remain potential contributors to the
AI cancer survival disparity, including the delivery of adjuvant therapies
and long-term follow-up care. Alternately, AIs may fare worse among
patients receiving nonsurgical therapies, such as those whose disease
severity precludes operative management.
The present results highlight the challenge of delivering oncologic
services to isolated populations in the era of centralized surgical
care. Proximity to the place of treatment has been shown to affect
cancer therapy modality,18 and rural
location has been shown to limit access to other forms of surgical
care.19 Trends in the national referral
system for cancer surgery toward the redistribution of resources out
of rural areas may adversely affect access to services for those who
already face geographic barriers to care because travel distance increases
when care is shifted to high-volume centers.20 Significantly, because most low-volume
cancer surgery occurs in urban areas, reports suggest that centralization
efforts can exempt rural hospitals and still achieve their objectives.21 Policy makers should also be aware that
validated strategies to reduce the distance rural patients travel
for cancer therapy are available22 and that telemedicine may have an emerging role in nonoperative
oncologic management in remote areas.23
Efforts to reverse the discouraging trend toward a broadening
gap in cancer mortality between AI and NHW patients should target
the unique needs of rural and urban AI patients. Although rural patients
face these challenges and traditionally have been thought to present
with later-stage cancer,24 a developing
body of literature suggests that for some cancers, residents of urban
areas receive diagnoses at later stages.25 Among residents of urban areas, AIs may be particularly vulnerable
because the Indian Health Service (IHS) is best equipped to serve
rural populations,26 although urban
IHS programs are evolving and improving.16
Of 5.2 million people in the United States who belong to 1 of
565 federally recognized American Indian and Alaskan native tribes,
approximately 2 million receive health care through the IHS. The IHS
is an agency within the Department of Health and Human Services with
an annual budget of approximately $4.1 billion. The agency provides
health care through federally administered programs, services administered
via tribal governments, and the Urban Indian Health Programs, including
a total of 45 hospitals and 582 smaller health centers and other facilities.
In addition, funds are available to purchase services that would otherwise
be unavailable, such as complex surgical care, through contracts with
outside providers. These limited funds are allocated annually based
on the acuity of patients' needs and reflect broader scarcity within
the system.16 Although AIs bear a
disproportionate burden of preventable disease, the per capita personal
health expenditure for IHS users ($2741) is less than half that of
the US population in general.27 Despite
these challenges, IHS initiatives have contributed to a reduction
in the life expectancy disparity between AIs and the NHW population
from 8 to 5 years during the past several decades.16 Although federally administered IHS hospitals
are not included in the NIS, the IHS provides much of its surgical
cancer care through contracts with outside hospitals that would fall
within the NIS sampling frame.
Our results should be interpreted in light of several limitations
related to the use of administrative data. First, our data set does
not include patients who did not undergo surgery, such as those with
disseminated disease, those who declined operative management, and
those whose comorbid conditions precluded surgery. Second, some states
do not participate in the NIS, including Alaska. Therefore, our results
should not be extrapolated to Alaskan natives, whose patterns of cancer
incidence3 and mortality2 have been shown to be distinct from those
of AIs. Third, the NIS does not contain information on tumor characteristics
or stage. Finally, although the 2009 edition of the NIS provides the
most comprehensive reporting of ethnicity to date, it is still missing
this information for 15% of discharges, limiting the generalizability
of our results in those states that do not report ethnicity data.
Despite these limitations, the large, inclusive nature of the NIS
makes it uniquely suited to address the important but underexplored
issue of surgical outcomes among this group.
The present study also has several significant strengths. As
the largest study of AI surgical patients, it provides an important
perspective on the AI surgical experience.Furthermore, given that
more than half of AIs and Alaskan natives do not receive care through
the IHS, the all-payer nature of the NIS database is an important
strength that provides data on AI patients from diverse backgrounds.16 Finally, by considering AIs separately
from Alaskan natives, our data may provide more precise information
than do studies that lump together these 2 very different groups.
Our findings suggest that the observed deficit in long-term
survival among AIs cannot be explained by excess perioperative mortality.
Other factors in the continuum of cancer care might account for the
previously reported unexplained variance in long-term outcomes. In
particular, differences in the use of adjuvant chemotherapy or radiotherapy
may contribute to disparities in survival and merit further investigation
as potential points of intervention. This possibility is supported
by research showing that AI patients treated for breast cancer have
different patterns of adjuvant therapy than NHW patients,28 and that radiotherapy and medical oncology
services are even more difficult for rural patients to access than
surgical care.29 Future research
should delineate clearly the significance of these potential contributors
to poor outcomes and further explore outcomes specific to malignant
disease. Researchers and policy makers should also expand work to
redress established disparities, such as ongoing IHS innovations that
have increased rates of colonoscopy.16
To our knowledge, this study is first to examine the effect
of AI ethnicity on patterns and outcomes of cancer surgery in the
United States. This relatively young cohort of surgical patients is
not subject to excess short-term operative mortality. Future research
should uncover other factors in the continuum of cancer care that
may contribute to poor long-term cancer survival for AIs, including
the delivery of perioperative therapies and follow-up care, while
simultaneously developing interventions to target known shortcomings.
Correspondence: Waddah B. Al-Refaie,
MD, Department of Surgery, Lombardi Comprehensive Cancer Center, Georgetown
University Hospital, 3800 Reservoir Rd NW, 4 PHC, Washington, DC 20007
Accepted for Publication: July 10,
Author Contributions: Mr Markin and
Dr Al-Refaie had full access to all the data in the study and take
responsibility for the integrity of the data and the accuracy of the
data analysis. Study concept and design:
Markin, Habermann, Vickers, and Al-Refaie. Acquisition
of data: Abraham. Analysis and interpretation
of data: Markin, Habermann, Zhu, Ahluwalia, and Al-Refaie. Drafting of the manuscript: Markin and Al-Refaie. Critical revision of the manuscript for important intellectual
content: All authors. Statistical analysis: Markin, Habermann, Zhu, and Al-Refaie. Obtained
funding: Vickers. Administrative, technical,
and material support: Ahluwalia. Study supervision: Habermann, Abraham, Ahluwalia, and Al-Refaie.
Conflict of Interest Disclosures: None
Funding/Support: This study was supported
by Enhancing Minority Participation in Clinical Trials grant 5RC2MD004797-02
and by the National Institute for Minority Health Disparities grant
1P60MD003422 from the Center for Health Equity.
Previous Presentation: This work was
presented in part at the 2012 Academic Surgical Congress; February
16, 2012; Las Vegas, Nevada.
Additional Contributions: Binyam Muluneh,
MD, contributed to an earlier version of this project and Crystal
Cunningham, MD, provided insightful perspectives on surgical oncology
services provided by the IHS.