The study comorbidities are detailed in Table 1.
Tammemagi CM, Nerenz D, Neslund-Dudas C, Feldkamp C, Nathanson D. Comorbidity and Survival Disparities Among Black and White Patients
With Breast Cancer. JAMA. 2005;294(14):1765-1772. doi:10.1001/jama.294.14.1765
Author Affiliations: Department of Community
Health Sciences, Brock University, St Catharines, Ontario (Dr Tammemagi);
and Center for Health Services Research (Dr Nerenz), Josephine Ford Cancer
Center (Ms Neslund-Dudas), Department of Pathology and Laboratory Medicine
(Dr Feldkamp), and Department of General Surgery (Dr Nathanson), Henry Ford
Health System, Detroit, Mich.
Context Reasons for the shorter survival of black breast cancer patients compared
with their white counterparts are not completely understood.
Objective To evaluate the role of comorbidity in this racial disparity among breast
Design, Setting, and Patients Historical cohort from the Henry Ford Health System (a large comprehensive
health system in Detroit, Mich) followed up for a median of 10 years. Patients
(n = 906) included 264 black (29.1%) and 642 white (70.9%) women
diagnosed as having breast cancer between 1985 and 1990. Detailed comorbidity
data (268 comorbidities) and study data were abstracted from medical records
and institutional, Surveillance, Epidemiology, and End Results, and Michigan
State registries. Associations were analyzed with logistic and Cox regression.
Main Outcome Measures Breast cancer recurrence/progression and survival to death from all,
breast cancer, and competing (non–breast cancer) causes.
Results Of blacks, 64 (24.9%) died of breast cancer and 95 (37.0%) died of competing
causes. Comparable data for whites were 115 (18.3%) and 202 (32.1%). Blacks
had worse all-cause survival (hazard ratio [HR], 1.34; 95% confidence interval
[CI], 1.11-1.62), breast cancer–specific survival (HR, 1.47; 95% CI,
1.08-2.00), and competing-causes survival (HR, 1.27; 95% CI, 1.00-1.63). A
total of 77 adverse comorbidities were associated with reduced survival. Adverse
comorbidity count was associated with all-cause (adjusted HR, 1.29; 95% CI,
1.19-1.40) and competing-causes survival but was not associated with recurrence/progression
or breast cancer–specific survival. At least 1 adverse comorbidity was
observed in 221 (86.0%) blacks and 407 (65.7%) whites (odds ratio, 3.20; 95%
CI, 2.17-4.72). Comparisons of unadjusted and comorbidity-adjusted HRs indicated
that adverse comorbidity explained 49.1% of all-cause and 76.7% of competing-causes
survival disparity. Diabetes and hypertension were particularly important
in explaining disparity.
Conclusions More black breast cancer patients die of competing causes than of breast
cancer. Effective control of comorbidity in black breast cancer patients should
help improve life expectancy and lead to a reduction in survival disparities.
Although breast cancer survival has improved over the last 30 years,
disparities in breast cancer survival between blacks and whites have not declined
and remain sizeable.1 The 5-year US survival
rates in 1995-2000 for black and white breast cancer patients were 75% and
89%, respectively.2 Although several causes
have been identified, such as advanced cancer stage, lack of access to medical
care, inferior treatment, and lower socioeconomic status (SES), not all reasons
for this disparity are understood.3- 17
Numerous studies have found that comorbidity is an independent predictor
of survival in breast cancer patients.18- 24 The
extent to which racial/ethnic differences in comorbidity explain disparities
in breast cancer survival has not been well studied. Thus, in a group of breast
cancer patients we evaluated the associations between adverse comorbidities
and the following outcomes: survival to death due to all causes, breast cancer
recurrence/progression and survival to breast cancer–specific death,
and survival to competing (non–breast cancer) causes of death.
A cohort of incident cases of breast cancer (N = 924, 1985-1990
inclusive) was identified from the Henry Ford Health System Tumor Registry,
an American College of Surgeons, Commission on Cancer–certified registry.
This health system is a large, comprehensive, nonprofit system that annually
provides medical care for more than 500 000 people, approximately 30%
of whom are black. In 1997, the Henry Ford Health System patient population
distribution in 10 age, 2 race, and 2 sex categories (40 strata) differed
from the metropolitan Detroit (Wayne, Oakland, and Macomb counties, 1990 census)
distribution by 5.3% or less in all strata. These observations suggest that
the Henry Ford Health System’s patient population is representative
of the community it serves. The study was approved by the Henry Ford Health
System institutional review board. Because the study examined patient medical
records only, the review board waived the need to obtain patient consent.
Sociodemographic, exposure, and clinicopathologic data, including comorbidity
data, were abstracted directly from medical records. Treatment data included
surgery, chemotherapy, radiation, and hormone therapy as dichotomous variables
(received or not). Additionally, information on type of surgery was collected.
For chemotherapy and radiation therapy, data were not available on treatment
completion or dose reduction. Survival and cause-of-death data were obtained
from the Henry Ford Health System and Metropolitan Detroit Surveillance, Epidemiology,
and End Results (SEER) tumor registries and from Michigan Department of Vital
Statistics death certificate data. The last date of follow-up was May 1, 2002.
Data on failure to control cancer, either as local recurrence following
resection or progression reflected in local expansion or regional or distant
spread of cancer, were abstracted from medical records. Socioeconomic status
was estimated by area-based socioeconomic measures taken from patients’
addresses and 1990 census data at the block group level, and included median
household income, proportion living below poverty level, and proportion not
completing high school. Race was classified according to self-report on registration
forms. Analysis was limited to blacks and whites; Asians, Pacific Islanders,
Native Americans, and others were excluded because the numbers were too small
to permit meaningful analysis. There were only 4 patients who were identified
as Hispanic and they were included in the analysis according to the race category
they chose (3 white, 1 black).
Comorbidity data were abstracted directly from medical records from
3 years prior to breast cancer diagnosis up until first breast cancer treatment
or 6 months following diagnosis if no treatment was administered. In this
study, all comorbidity data were collected and coded into 259 essentially
mutually exclusive diagnostic categories of the Clinical
Classification Software,25 developed
by the US Agency for Healthcare Research and Quality to facilitate health
research by producing a manageable number of clinically meaningful disease
categories from the more than 12 000 codes in the International Classification of Diseases, 9th Revision, Clinical Modifications text.26 This comorbidity list was supplemented
by comorbidities suggested to be important in our previous studies of comorbidity
and lung cancer outcomes27,28 to
yield 268 comorbidities in 16 comorbidity categories.
This study focuses on the impact of adverse comorbidities on outcomes,
with adverse comorbidities being defined as those that had significantly elevated
Cox regression hazard ratios (all-cause survival) regardless of effect magnitudes,
that had hazard ratios (HRs) greater than 1.20 regardless of statistical significance,
or that were deemed to be adverse a priori based on clinical knowledge and/or
past research (eg, HIV/AIDS). Seventy-seven comorbidities were classified
as adverse (Table 1) and in this report
mention of comorbidity refers specifically to these 77 comorbidities. The
abstraction form and table of HRs and racial distributions for comorbidities
are available online at http://www.brocku.ca/communhealthsci/jama.html.
To facilitate interpretation of effect estimates and to evaluate dose-response
relationships, comorbidity counts were divided into approximate quintiles
of comparable size.
For comparative purposes the Charlson Comorbidity Index29 was
evaluated. The index was developed using hospital emergency department admissions
with 1-year mortality as the outcome and was validated in breast cancer patients.
It is computed by weighting and summing 19 comorbidities. Of all comorbidity
indices to date, the Charlson has been most extensively studied and has been
deemed to be a valid and reliable method of measuring comorbidity for clinical
Cancer stage, based on pathological stage and in its absence on clinical
stage, was analyzed according to the American Joint Committee on Cancer TNM
staging system.31 Histotype was based on World
Health Organization categories.32
Contingency table analyses were carried out and null hypotheses were
evaluated using the Fisher exact test and nonparametric tests for trend33 when applied to ordinal data. Logistic regression
odds ratios (ORs) and 95% confidence intervals (CIs) were used to evaluate
associations between predictors and dichotomous outcomes.
Univariate and multivariate survival analyses were carried out using
Kaplan-Meier, life table, and Cox proportional hazards regression analyses.34,35 Modeling proceeded from univariate
to multivariate and preparation of parsimonious multivariable models was guided
by a priori considerations36 and was aided
by backward stepwise elimination. Death was considered to be breast cancer–specific
if either the SEER Registry or Michigan death certificate data indicated the
cause of death was breast cancer. Classification of survival status into the
categories alive, breast cancer death, or competing causes of death in the
2 registries had 91.9% agreement. Where cause of death data were present in
both registries agreement was high: the κ statistic37 was
0.98 (lower confidence limit, 0.80) for blacks and 0.96 (lower confidence
limit, 0.83) for whites.
When death due to competing causes was analyzed, breast cancer–specific
deaths were censored. The reverse-censoring Kaplan-Meier method, which eliminates
bias introduced by differential death rates, was used to compare follow-up
between groups.38 The c statistic
was used to measure the predictive ability of Cox models.39 The c statistic is analogous to the area under the receiver
operating characteristic curve and can be thought of as follows: considering
all possible combinations of paired individuals under study with differing
survival times, the c statistic represents the proportion
for which the regression model correctly predicts the survival order.39,40
The amount of racial survival disparity explained by comorbidity was
estimated by the proportion decline in HR, black vs white, comparing the comorbidity-adjusted
with the unadjusted model.
In survival regression analysis, proportional hazard assumptions were
tested graphically and statistically and were met for all presented models.
In logistic regression analysis, modeling of collinear variables was avoided,
regression diagnostics were carried out, and conventional diagnostic standards
were met for all models presented. In multivariate models interaction terms
were considered. The α error was set at .05 and all reported P values, except for the κ statistic, are 2-sided. Stata version
7.0 software (Stata Corporation, College Station, Tex) was used to prepare
Through the tumor registry, 924 individuals were identified as having
breast cancer. Analyses were restricted to the 906 individuals who were black
(n = 264, 29.1%) or white (n = 642, 70.9%). Follow-up
data were missing for 7 blacks (2.7%) and 13 whites (2.0%) (P = .62). Of 886 individuals for whom survival data were
available, loss to follow-up occurred prior to 10 years in 56 individuals
(6.3%). Cox regression analysis demonstrated that time to loss to follow-up
was associated with younger age (HR per 10 years, 0.62; 95% CI, 0.52-0.74)
but not with race or comorbidity; adjusted for age, the HR for loss to follow-up
for race was 0.63 (95% CI, 0.29-1.37) and the HR for comorbidity was 1.07
(95% CI, 0.56-2.07). Comorbidity data were missing for 7 blacks (2.7%) and
23 whites (3.6%) (P = .55). The most complex
multivariate model included race, age, comorbidity, estrogen receptor status,
tumor stage, surgery, chemotherapy, and radiation therapy as predictors and
was based on 227 of 264 blacks (86.0%) and 555 of 642 whites (86.4%) (P = .83).
The distribution of baseline characteristics for selected variables
by race are presented in Table 2. Blacks
were generally older than whites. Significantly fewer blacks were married.
Blacks had significantly lower SES as measured by median household income,
poverty level, or education (all 3, rank sum test P<.001).
Blacks were diagnosed at a higher tumor stage: 52 blacks (21.5%) and 88 whites
(14.8%) had stage III or IV disease (OR for stage III-IV vs lower stages,
1.57; 95% CI, 1.07-2.30). In univariate analysis, blacks tended to have fewer
estrogen receptor–positive tumors (Table
2). Adjusted for stage and age, this association was significant
(OR for black vs white, 0.62; 95% CI, 0.40-0.97).
Overall, the median follow-up was 10.0 years (range, 0.04-17.8 years).
Of those who survived, median follow-up did not differ significantly between
blacks (12.8 years) and whites (12.7 years), indicating comparable follow-up
data quality. A total of 159 blacks (61.9%) and 317 whites (50.4%) died (P = .002) (Table
2). Overall, 62.4% of deaths were attributed to competing causes.
Proportionately more blacks than whites died of breast cancer (64 [24.9%]
vs 115 [18.3%], P = .03) and of competing
causes (95 [37.0%] vs 202 [32.1%], P = .18).
For those with comorbidity data, 28.3% had no comorbidities. Patients
had a mean of 2.02 (median, 1; range, 0-13) comorbidities. The univariate
HR for comorbidity count as a single 5-level variable (0, 1, 2, 3, and 4-13
comorbidities) was 1.40 (95% CI, 1.31-1.49). Following adjustment for age,
tumor stage, estrogen receptor positivity, surgery, chemotherapy, and radiation
therapy, the HR for comorbidity count (5-level) was 1.29 (95% CI, 1.19-1.40).
A dose-response relationship was indicated by a monotonic increase in HRs
with increasing numbers of comorbidities (Table
3 and Figure 1). The effect
of comorbidity on survival did not differ by age; the comorbidity × age
interaction P value was .46.
Compared with whites, blacks had shorter overall survival (HR, 1.34;
95% CI, 1.11-1.62; Figure 2). One or
more comorbidities were reported in 221 blacks (86%) and 407 whites (65.7%)
(OR, 3.20; 95% CI, 2.17-4.72). The distributions of comorbidities by
race are presented in Table 2 and Figure 3. Comorbidity count evaluated as a single
5-level variable explained 49.1% of the racial disparity in overall survival
(comorbidity-adjusted HR, 1.17; 95% CI, 0.96-1.43 (Table 4). Adjusted for age, tumor stage, estrogen receptor status,
surgery, chemotherapy, and radiation therapy, the HR for black vs white was
1.14 (95% CI, 0.92-1.40), and when additionally adjusted for comorbidity count
(5 levels) the HR was 1.02 (95% CI, 0.83-1.27). These data indicate that comorbidity
explains disparity in all-cause survival in addition to other prognostic factors.
The effect of comorbidity on survival did not differ by race (P value for interaction = .99).
A total of 280 patients (30.9%) were older than 70 years. Racial disparity
in survival was evident in patients younger than 70 years (HR, 1.23;
95% CI, 0.94-1.61) and in those 70 years or older (HR, 1.25; 95% CI,
0.95-1.65). The effect of race on all-cause survival did not differ by age
(P for interaction = .89). In the group
aged younger than 70 years, at least 1 comorbidity was present in 78.7% of
blacks and 56.9% of whites (P<.001); in the group
aged 70 years or older, at least 1 comorbidity was present in 97.1% of blacks
and 89.3% of whites (P = .02). Adjusted
for comorbidities, the HR for black vs white was 1.13 (95% CI, 0.85-1.50)
in the younger group and 1.18 (95% CI, 0.89-1.57) in the older group, declines
of 43.4% and 27.0%, respectively.
We also evaluated the impacts of all specific comorbidities with elevated
frequencies in blacks compared with whites. Two of the most important comorbidities
explaining survival disparity were diabetes and hypertension. The HR associated
with diabetes was 1.85 (95% CI, 1.47-2.32) and the prevalence of diabetes
in blacks and whites was 68 (26.4%) and 59 (9.5%), respectively (OR for black
vs white, 3.41; 95% CI, 2.32-5.02). The HR associated with hypertension was
1.65 (95% CI, 1.37-1.99) and the prevalence of hypertension in blacks and
whites was 163 (63.4%) and 220 (35.5%), respectively (OR, 3.14; 95% CI, 2.32-4.26).
Adjusted for these 2 comorbidities, the HR for black vs white was 1.09 (95%
CI, 0.89-1.35), a decline of 72.4% from the unadjusted estimate. Most of the
complications of diabetes and hypertension that led to patient deaths developed
well after the breast cancer diagnosis and outside the record abstraction
period, and thus these data were unavailable for analysis. At the time of
diagnosis, 8 black (3.1%) and 7 white (1.1%) patients (OR, 2.81, 95% CI, 1.01-7.83)
had diabetic complications, which included circulatory problems, retinopathy,
neuropathy, and ketoacidosis in both groups. Coronary disease occurred in
31 blacks (12.1%) and 51 whites (8.2%) (OR, 1.53; 95% CI, 0.95-2.45)
and was associated with increased hazard (HR, 1.78; 95% CI, 1.35-2.35).
Blacks had significantly more breast cancer recurrence/progression than
did whites (35.8% vs 27.6%; OR, 1.47; 95% CI, 1.06-2.03). Breast cancer recurrence/progression
was strongly predictive of reduced breast cancer–specific survival (HR,
24.40; 95% CI, 15.41-38.62) and compared with whites, blacks experienced shorter
breast cancer–specific survival (HR, 1.47; 95% CI, 1.08-2.00). Comorbidity
(5 levels) was not associated with breast cancer recurrence/progression (OR, 0.98;
95% CI, 0.88-1.08) or with breast cancer–specific survival (HR, 1.01;
95% CI, 0.91-1.12) and did not explain racial disparity in recurrence/progression
or breast cancer–specific survival. Adjusted for comorbidities (5 levels),
the recurrence/progression OR for black vs white was 1.47 (95% CI, 1.06-2.03)
and breast cancer–specific HR was 1.50 (95% CI, 1.09-2.05).
Comorbidity may lead to breast cancer recurrence/progression and reduced
breast cancer−specific survival by causing less aggressive or no treatment.
The single most protective cancer treatment was surgery (HR, 0.29; 95% CI,
0.17-0.50), and comorbidity (5 levels) predicted nonreceipt of surgery (OR,
0.72, 95% CI, 0.58-0.88). Adjusted for marital status, SES (block group poverty
level), and tumor stage, comorbidity remained predictive of nonreceipt of
surgery (OR, 0.71; 95% CI, 0.54-0.91). In univariate analysis age was
associated with nonreceipt of surgery (OR per 10 years, 0.70; 95% CI, 0.55-0.89).
Age and comorbidity were collinear (OR for ≥1 vs 0 comorbidity per 10 years,
2.37; 95% CI, 2.06-2.71) and in multivariate analysis comorbidity was more
predictive of surgery than age was. However, in this study population, regardless
of the higher frequency of comorbidities in blacks, similarly high proportions
underwent surgery: 239 blacks (94.1%) and 584 whites (95.6%) (P = .39). This latter observation in part explains why comorbidity
did not account for the racial disparity in recurrence/progression and breast
For competing-causes survival, the HR for black vs white was 1.27 (95%
CI, 1.00-1.63) and for comorbidity (5 levels) the HR was 1.69 (95% CI, 1.55-1.84).
Adjusted for comorbidity, the HR for black vs white was 1.06 (95% CI, 0.83-1.36),
a decline of 76.7% from the unadjusted analysis (Table 4).
Body mass index (calculated as weight in kilograms divided by height
in meters squared) of 25 or higher (ie, overweight) was observed in 72% of
blacks and 49.7% of whites (P<.001) and had the
greatest impact on competing-causes survival. Compared with a BMI of 18.5
to less than 25, the univariate HRs (competing-causes survival) for BMI were
as follows: for BMI of 25 to less than 30: 1.12 (95% CI, 0.84-1.50); for BMI
of 30 to less than 35: 1.16 (95% CI, 0.81-1.65); and for BMI of 35 or higher:
1.35 (95% CI, 0.91-2.00). Adjusted for diabetes, hypertension, or comorbidity,
all of these HRs approached the null, suggesting that adverse effects of obesity
are mediated through comorbidity. Low BMI carried significant risk (HR for ≤18.5
vs >18.5 to <25, 2.34; 95% CI, 1.22-4.50). However, low BMI occurred less
frequently in blacks than in whites (0.8% vs 4.0%, P = .02)
and did not explain disparity in competing-causes or breast cancer–specific
survival (low BMI–adjusted HR for black vs white, 1.26; 95% CI, 0.98-1.63
and HR, 1.49; 95% CI, 1.07-2.07, respectively).
In the current study, comorbidity was an important predictor of survival
and explained important amounts of survival disparity. Comorbidity is complex
and optimal measurement methods that are robust across different outcomes
and populations have not been established. Our study comorbidities demonstrated
criterion (concurrent) validity41 because they
correlated well with the established Charlson Comorbidity Index (r2 = 0.47, P<.001).
Charlson index score greater than 0 occurred in 51.7% of blacks and 40.4%
of whites (P for trend = .002) (Table 2). The Charlson index demonstrated a dose
response with all-cause survival (Table 3). Table 4 presents a comparison of the ability
of the study comorbidities and Charlson index to predict survival to all-cause
and competing-causes death and explain disparity in these outcomes. The predictive
ability as measured by the c statistic for both outcomes
was consistently greater for study comorbidities than for the Charlson index,
and this held true whether comorbidity was modeled as a single 5-level variable
or multiple categorical variables. Similarly, the study comorbidities explained
more survival disparity for all-cause and competing-causes survival (Table 4). These statistics suggest that the Charlson
index failed to capture some relevant information present in the study comorbidities,
which were based on a more extensive comorbidity inventory.
The current study found that black breast cancer patients have more
cancer recurrence/progression and worse all-cause, breast cancer–specific,
and competing-causes survival. Comorbidity explained approximately half of
the overall survival disparity and the majority of competing-causes survival
disparity, which accounted for the preponderance of deaths in black patients.
Comorbidity was not associated with recurrence/progression or breast cancer–specific
As in the current study, Eley et al5 found
that comorbidity was an important independent predictor of all-cause survival
but not of breast cancer–-specific survival, was significantly more
frequent in blacks, and explained 25% of all-cause survival disparity (vs
49% in this study). Eley and colleagues considered only 6 categories of comorbidity
and their comorbidity frequencies were substantially below those observed
in our study, suggesting that their comorbidity adjustment may have been incomplete.
Additionally, their follow-up was shorter than in the current study, which
led to a heavier weighting of breast cancer vs competing-causes deaths, thus
underemphasizing the impact of comorbidity on overall survival.
Analysis of all-cause survival may seem superfluous following analysis
of breast cancer–specific and competing-causes survival. However, the
misclassification that can occur between the latter 2 categories does not
occur with all-cause death. Also, all-cause survival to some extent captures
combined effects, because it is likely that in some cases comorbidity and
breast cancer are not mutually exclusive but in combination contribute to
shortened survival. Thus, survival to all-cause death serves as a useful outcome
for summarizing the overall impact of comorbidity on the study cohort.
Optimal methods for comorbidity measurement are under development. The
Charlson Comorbidity Index was validated, that is, found to significantly
predict survival in breast cancer patients, and has been shown to be comparably
predictive in black and white breast cancer patients.42 However,
demonstration of significant association with survival does not alone indicate
that a measure has high content validity and that it is optimized for studying
disparities. Our previous study of comorbidity and lung cancer survival found
that infrequent comorbidities in aggregate had an important impact on survival,
that the Charlson index omitted several important predictive comorbidities,
that the Charlson index’s weighting scheme did not correspond to HRs
for several comorbidities, and that the Charlson index’s explanatory
ability was at par with count of any comorbidity.27 Similarly,
this study’s findings suggest that the Charlson index might not be an
optimal scale for studies of breast cancer survival and disparity. The current
data set served as a test/validation set for the Charlson index, whereas it
was the training set for the current study’s comorbidity inventory.
However, study comorbidities appeared to be consistently better than the Charlson
index in predicting survival and explaining disparity, even though no weighting
of comorbidity effects was applied.
The current study had design features that overcame limitations of some
earlier studies. It included a relatively large, single-institution cohort
of breast cancer patients with a relatively lengthy follow-up. Comorbidity
data were collected systematically in detail from medical records, which are
generally considered superior to data collected from administrative databases.43- 46 Although
cancer treatment was controlled for in the analysis, data indicate that treatment
differences were not important in leading to survival disparity within this
health care system. This report presents surgery as a dichotomous variable.
Specific types of surgery (eg, lumpectomy, mastectomy) were also analyzed
but did not explain survival or disparity beyond the binary surgery variable
(data not shown). The generalizability of our study findings to other populations
needs to be established through further investigation. Important factors that
impede access to quality health care were not evaluated in this analysis but
must be considered in overall understanding of disparity.
Our findings indicate that control of comorbidity may be an important
way of improving the survival of black breast cancer patients and reducing
racial disparity. That comorbidity explained more than 40% of the survival
disparity in patients younger than 70 years indicates that effective management
of comorbidity has the potential to lead to a substantial increase in person-years
of life gained. Control of just 2 comorbidities, diabetes and hypertension,
could have a major beneficial impact.
Corresponding Author: C. Martin Tammemagi,
PhD, Department of Community Health Sciences, Brock University, 500 Glenridge
Ave, St Catharines, Ontario, Canada L2S 3A1 (email@example.com).
Author Contributions: Dr Tammemagi had full
access to all of the data in the study and takes responsibility for the integrity
of the data and the accuracy of the data analysis.
Study concept and design: Tammemagi.
Acquisition of data: Tammemagi, Neslund-Dudas,
Analysis and interpretation of data: Tammemagi,
Drafting of the manuscript: Tammemagi, Nathanson.
Critical revision of the manuscript for important
intellectual content: Tammemagi, Nerenz, Neslund-Dudas, Feldkamp, Nathanson.
Statistical analysis: Tammemagi.
Obtained funding: Tammemagi.
Administrative, technical, or material support:
Tammemagi, Neslund-Dudas, Feldkamp.
Study supervision: Tammemagi.
Financial Disclosures: None reported.
Funding/Support: This study was funded by US
Department of Defense (US Army Medical Research and Material Command) grant
Role of the Sponsor: The funding agency did
not have any control or influence over the design and conduct of the study
in any fashion, including collection, management, analysis, and interpretation
of the data; and preparation, review, or approval of the manuscript.