Advertisement

Racial Disparities in Lung Cancer Survival: The Contribution of Stage, Treatment, and Ancestry

Open ArchivePublished:June 06, 2018DOI:https://doi.org/10.1016/j.jtho.2018.05.032

      Abstract

      Introduction

      Lung cancer is a leading cause of cancer-related death worldwide. Racial disparities in lung cancer survival exist between blacks and whites, yet they are limited by categorical definitions of race. We sought to examine the impact of African ancestry on overall survival among blacks and whites with NSCLC cases.

      Methods

      Incident cases of NSCLC in blacks and whites from the prospective Southern Community Cohort Study (N = 425) were identified through linkage with state cancer registries in 12 southern states. Vital status was determined by linkage with the National Death Index and Social Security Administration. We evaluated the impact of African ancestry (as estimated by using genome-wide ancestry-informative markers) on overall survival by calculating the time-dependent area under the curve (AUC) for Cox proportional hazards models, adjusting for relevant covariates such as stage and treatment. We replicated our findings in an independent population of NSCLC cases in blacks.

      Results

      Global African ancestry was not significantly associated with overall survival among NSCLC cases. There was no change in model performance when Cox proportional hazards models with and without African ancestry were compared (AUC = 0.79 for each model). Removal of stage and treatment reduced the average time-dependent AUC from 0.79 to 0.65. Similar findings were observed in our replication study.

      Conclusions

      Stage and treatment are more important predictors of survival than African ancestry is. These findings suggest that racial disparities in lung cancer survival may disappear with similar early detection efforts for blacks and whites.

      Keywords

      Introduction

      Lung cancer is the leading cause of cancer death among both men and women in the United States, with a 5-year relative survival rate of 18%.
      • Henley S.J.
      • Singh S.D.
      • King J.
      • et al.
      Invasive cancer incidence and survival–United States, 2011.

      National Institutes of Health. National Cancer Institute. Surveillance, Epidemiology, and End Results Program. Previous version: SEER Cancer Statistics Review, 1975-2013. http://seer.cancer.gov/csr/1975_2013/. Accessed April 12, 2018.

      Although lung cancer mortality has decreased in recent years (in large part because of greater smoking cessation efforts), a racial disparity exists such that blacks experience poorer survival than whites do.
      • Henley S.J.
      • Singh S.D.
      • King J.
      • et al.
      Invasive cancer incidence and survival–United States, 2011.
      • Aizer A.A.
      • Wilhite T.J.
      • Chen M.H.
      • et al.
      Lack of reduction in racial disparities in cancer-specific mortality over a 20-year period.
      • American Cancer Society
      Cancer Facts & Figures 2017.
      • Moolgavkar S.H.
      • Holford T.R.
      • Levy D.T.
      • et al.
      Impact of reduced tobacco smoking on lung cancer mortality in the United States during 1975-2000.
      Specifically, the national 5-year survival rate is 18% among whites and 15% among blacks.

      National Institutes of Health. National Cancer Institute. Surveillance, Epidemiology, and End Results Program. Previous version: SEER Cancer Statistics Review, 1975-2013. http://seer.cancer.gov/csr/1975_2013/. Accessed April 12, 2018.

      In blacks the disease is diagnosed at a late stage more frequently than in whites, and blacks are less likely to receive the recommended course of treatment based on disease stage.

      National Institutes of Health. National Cancer Institute. Surveillance, Epidemiology, and End Results Program. Previous version: SEER Cancer Statistics Review, 1975-2013. http://seer.cancer.gov/csr/1975_2013/. Accessed April 12, 2018.

      • Efird J.T.
      • Landrine H.
      • Shiue K.Y.
      • et al.
      Race, insurance type, and stage of presentation among lung cancer patients.
      • Hardy D.
      • Liu C.C.
      • Xia R.
      • et al.
      Racial disparities and treatment trends in a large cohort of elderly black and white patients with nonsmall cell lung cancer.
      • Halpern M.T.
      • Ward E.M.
      • Pavluck A.L.
      • Schrag N.M.
      • Bian J.
      • Chen A.Y.
      Association of insurance status and ethnicity with cancer stage at diagnosis for 12 cancer sites: a retrospective analysis.
      Several recent studies have suggested that controlling for differential access to health care results in no difference in survival outcomes among blacks and whites.
      • Caposole M.Z.
      • Miller K.
      • Kim J.N.
      • Steward N.A.
      • Bauer T.L.
      Elimination of socioeconomic and racial disparities related to lung cancer: closing the gap at a high volume community cancer center.
      • Ganti A.K.
      • Subbiah S.P.
      • Kessinger A.
      • Gonsalves W.I.
      • Silberstein P.T.
      • Loberiza Jr., F.R.
      Association between race and survival of patients with non–small-cell lung cancer in the United States veterans affairs population.
      • Zheng L.
      • Enewold L.
      • Zahm S.H.
      • et al.
      Lung cancer survival among black and white patients in an equal access health system.
      We and others have demonstrated that blacks and whites experience no difference in lung cancer survival after controlling for stage and socioeconomic factors.
      • Hua X.
      • Ward K.C.
      • Gillespie T.W.
      • Lipscomb J.
      • Goodman M.
      Non-small cell lung cancer treatment receipt and survival among African-Americans and whites in a rural area.
      • Aldrich M.C.
      • Grogan E.L.
      • Munro H.M.
      • Signorello L.B.
      • Blot W.J.
      Stage-adjusted lung cancer survival does not differe between low-income blacks and whites.
      A recent analysis of Surveillance, Epidemiology, and End Results (SEER) program data also suggests that blacks have lung cancer survival similar to that of whites.
      • Zeng C.
      • Wen W.
      • Morgans A.K.
      • Pao W.
      • Shu X.O.
      • Zheng W.
      Disparities by race, age, and sex in the improvement of survival for major cancers: results from the National Cancer Institute Surveillance, Epidemiology, and End Results (SEER) program in the United States, 1990 to 2010.
      However, blacks are an admixed population with varying proportions of African ancestry
      • Bryc K.
      • Auton A.
      • Nelson M.R.
      • et al.
      Genome-wide patterns of population structure and admixture in West Africans and African Americans.
      • Tishkoff S.A.
      • Reed F.A.
      • Friedlaender F.R.
      • et al.
      The genetic structure and history of Africans and African Americans.
      and self-identified whites can carry African ancestry.
      • Bryc K.
      • Durand E.Y.
      • Macpherson J.M.
      • et al.
      The genetic ancestry of African Americans, Latinos, and European Americans across the United States.
      Identification of ancestry-informative markers, which are genetic variants that differ in frequency between ancestral populations, allows us to distinguish individual-level ancestral origins at the genetic level (i.e., genetic ancestry). Prior studies have shown important associations between genetic ancestry and biomedical phenotypes
      • Dumitrescu L.
      • Restrepo N.A.
      • Goodloe R.
      • et al.
      Towards a phenome-wide catalog of human clinical traits impacted by genetic ancestry.
      • Goetz L.H.
      • Uribe-Bruce L.
      • Quarless D.
      • Libiger O.
      • Schork N.J.
      Admixture and clinical phenotypic variation.
      • Peralta C.A.
      • Risch N.
      • Lin F.
      • et al.
      The Association of African Ancestry and elevated creatinine in the Coronary Artery Risk Development in Young Adults (CARDIA) Study.
      such as lung function
      • Kumar R.
      • Seibold M.A.
      • Aldrich M.C.
      • et al.
      Genetic ancestry in lung-function predictions.
      • Moreno-Estrada A.
      • Gignoux C.R.
      • Fernandez-Lopez J.C.
      • et al.
      Human genetics. The genetics of Mexico recapitulates Native American substructure and affects biomedical traits.
      • Salari K.
      • Choudhry S.
      • Tang H.
      • et al.
      Genetic admixture and asthma-related phenotypes in Mexican American and Puerto Rican asthmatics.
      and breast cancer risk
      • Fejerman L.
      • Hu D.
      • Huntsman S.
      • et al.
      Genetic ancestry and risk of mortality among U.S. Latinas with breast cancer.
      • Fejerman L.
      • John E.M.
      • Huntsman S.
      • et al.
      Genetic ancestry and risk of breast cancer among U.S. Latinas.
      ; however, the association between genetic ancestry and survival after a diagnosis of lung cancer has yet to be examined. We examined the effect of African ancestry on lung cancer survival in blacks and whites with NSCLC in the Southern Community Cohort Study (SCCS), which is the cohort with the largest representation of blacks in the United States. Black and white SCCS participants were primarily recruited from community health centers across the Southeast and thus have similar access to health care. Analyses were replicated in a population of black individuals with lung cancer that was ascertained from the population-based Metropolitan Detroit Cancer Surveillance System.

      Methods

      Study Population

      Study participants were selected from the SCCS, which is a prospective cohort study of approximately 86,000 adults age 40 to 79 years. Participants were enrolled between March 2002 and September 2009 from a 12-state region across the southeastern United States (Alabama, Arkansas, Florida, Georgia, Kentucky, Louisiana, Mississippi, North Carolina, South Carolina, Tennessee, Virginia, and West Virginia). Approximately 15% of participants were recruited through mail-in questionnaires that were sent to a random subset of adults across the 12-state region. The remaining 85% of participants were enrolled at community health centers throughout the region. Individuals were eligible to participate if they were between the ages of 40 and 79 years. Demographic characteristics, family history of disease, insurance coverage, tobacco use, and other information were collected through in-person interviews by a trained interviewer upon enrollment at the health centers and by completion of the same questionnaire for the recruits from the general population. Individuals self-reported race/ethnicity by selecting any of the following investigator-defined racial/ethnic groups: white, black/African American, Hispanic/Latino, Asian or Pacific Islander, American Indian or Alaska Native, or other racial or ethnic group. Approximately two-thirds of participants self-identified as black/African American. Upon enrollment, all individuals were asked to donate a biologic specimen (blood, urine, saliva, or buccal cell), to which approximately 90% of participants agreed. A detailed description of the study design and recruitment has been previously published.
      • Signorello L.B.
      • Hargreaves M.K.
      • Steinwandel M.D.
      • et al.
      Southern Community Cohort Study: establishing a cohort to investigate health disparities.
      • Signorello L.B.
      • Hargreaves M.K.
      • Blot W.J.
      The Southern Community Cohort Study: investigating health disparities.
      The SCCS was approved by institutional review boards at Vanderbilt University and Meharry Medical College. Written informed consent was obtained from all participants.

      Case Identification and Mortality Assessment

      All incident NSCLC cases occurring within the SCCS between 2002 and 2010 were identified through linkage with the 12 state cancer registries. Individuals with a diagnosis of lung cancer before study enrollment were excluded. Histologic type, stage at diagnosis, and treatment information were obtained from the individual state cancer registries. Stage was derived by using the American Joint Committee on Cancer TNM System staging guidelines (sixth and seventh editions). Because of the small sample size, we combined individuals with stage II and III disease. For individuals missing stage information, we used the SEER Summary Stage guidelines, assuming that local disease was equivalent to stage I, regional disease was equivalent to stage II/III, and distant disease was equivalent to stage IV. Treatment information describing the administration of chemotherapy, radiation therapy, hormone therapy, immunotherapy, surgery, or other cancer-directed treatment was summarized into a design variable with five levels: no treatment, chemotherapy only, radiation only, surgery only, and multimodality treatment (patients receiving any combination of the aforementioned treatment options). Participants were followed for all-cause mortality. Vital status was determined at the end of follow-up (December 31, 2011) through linkage with the Social Security Administration or the National Death Index. Survival time was defined as the time from the date of diagnosis to the date of death, loss to follow-up, or censoring.

      Genotyping and Quality Control

      Individuals in the SCCS were genotyped with the Illumina HumanExome BeadChip v1.1 (Illumina, San Diego, CA), which contains a panel of more than 3000 ancestry-informative markers for distinguishing between African and European ancestries. To remove technical artifacts and ensure high-quality data for ancestry analysis, we performed standard quality control of all genotyping data. Briefly, quality control removed individuals with sex inconsistencies, less than 98% genotyping efficiency, or self-reported race other than black/African American or White. Relatedness among individuals was also examined, and the individual with the lowest call rate in each relationship pair was removed. Variants were removed during quality control if they were nonautosomal, had less than 98% genotyping efficiency, or less than 5% minor allele frequency. Mendelian errors were examined by using HapMap trio controls. Variants were pruned on the basis of linkage disequilibrium (a window size of 50, step size of 10, and r2 > 0.4) for ancestry estimation in blacks and whites separately. All quality control measures described were applied by using PLINK
      • Purcell S.
      • Neale B.
      • Todd-Brown K.
      • et al.
      PLINK: a tool set for whole-genome association and population-based linkage analyses.
      (version 1.07).

      Ancestry Estimation

      Global ancestry estimates describe the proportion of an individual’s total genome inherited from each contributing ancestral population. To distinguish European from African ancestry, we used a panel of ancestry-informative markers. The numbers of markers available for global ancestry estimation were 553 for whites and 1137 for blacks. Supervised admixture analysis was performed by using the software program ADMIXTURE
      • Alexander D.H.
      • Novembre J.
      • Lange K.
      Fast model-based estimation of ancestry in unrelated individuals.
      to estimate individual ancestry proportions by assuming two ancestral populations with CEU (Utah residents with ancestry from northern and western Europe) and YRI (Yoruban in Ibadan, Nigeria) HapMap
      International HapMap Consortium
      The International HapMap Project.
      populations as representative ancestral populations. The resulting output contained the estimated proportions of African and European ancestry for each individual; the proportion of African ancestry was then converted to a percentage and used for analyses (hereafter referred to as African ancestry).

      Statistical Analysis

      Chi-square tests and t tests were used to assess racial differences in categorical and continuous descriptive characteristics, respectively. The impact of African ancestry on overall mortality among individuals with lung cancer who self-reported as black and among those who self-reported as white was examined by using a Cox proportional hazards model. Individuals with less than 30 days' survival time were excluded from analysis to remove potential bias related to treatment effects. Model fit statistics for the Cox model are well established, so we used a time-dependent area under the curve (AUC) receiver operating characteristic (ROC) curve for the purpose of describing the model’s discriminative ability over time. At every observable time, subjects were classified as alive or dead (censoring was handled as described by Heagerty et al.
      • Heagerty P.J.
      • Lumley T.
      • Pepe M.S.
      Time-dependent ROC curves for censored survival data and a diagnostic marker.
      ) and an ROC curve was constructed. The AUC was then graphed as a function of time, so the ability of the model to correctly classify subjects was assessed as a function of time. The AUC for an unadjusted Cox proportional hazards model with African ancestry only was estimated first. Then, covariates selected on the basis of a priori knowledge were added to the model and the time-dependent predictive ability of the new model was estimated. The model examined overall mortality as a function of African ancestry, age at diagnosis, sex, body mass index (kg/m2), number of cigarettes smoked per day, stage at diagnosis, treatment, highest education level achieved, and family history of lung cancer (hereafter called the main effects model). Through use of a flexible parametric additive model, it was determined that self-reported race could be predicted from African ancestry and the other model covariates, so self-reported race was removed from all Cox regression models. Continuous variables, including African ancestry, were modeled with restricted cubic splines using three knots, which allows for a nonlinear relationship between a continuous variable and the outcome. Missing data was multiply imputed 10 times by using predictive mean matching among the eight potential confounding variables. The Cox proportional hazards models were fit with each of the 10 completed data sets, and the results were pooled using the Rubin rules.
      • Rubin D.B.
      Multiple Imputation for Nonresponse in Surveys.
      A time-dependent ROC curve
      • Heagerty P.J.
      • Lumley T.
      • Pepe M.S.
      Time-dependent ROC curves for censored survival data and a diagnostic marker.
      and AUC were calculated for every time point between 30 days and 4 years to estimate how well the model predicts mortality. The time-dependent AUCs were then averaged over the time interval to obtain the average time-dependent AUC. We then compared the time-dependent AUC of our main effects model with an overfit Cox proportional hazards model with two-way interaction terms (i.e., African ancestry × treatment, African ancestry × stage, African ancestry × education, education × treatment, education × stage, treatment × stage, sex × age, sex × body mass index, and sex × cigarettes per day) to assess how well our main effects model performed. We examined the impact of African ancestry, stage, and treatment by removing these variables from the interaction and main effects models and compared the time-dependent AUC with and without these variables. Because the model did not assume a linear effect of African ancestry (i.e., splines were used), there is not a single p value or hazard ratio associated with African ancestry. The effect of African ancestry was assessed by using a likelihood ratio test to estimate nonconstant hazard ratios for mortality and corresponding 95% confidence intervals (CIs). Unadjusted survival models for race, stage, and interaction of race with stage and stratified survival curves were fit to evaluate the individual effects of these variables on survival. An analysis was performed only among self-reported black individuals to verify results. All statistical analyses were performed in R software (version 3.2.2) with the packages Hmisc and survivalROC.

      Replication

      Incident cases of NSCLC in blacks were identified from three lung cancer case-control studies (Family Health Study III; Women’s Epidemiology of Lung Disease Study; and Exploring Health, Ancestry, and Lung Epidemiology Study) that were conducted at the Barbara Ann Karmanos Cancer Institute, which is affiliated with Wayne State University in Detroit, Michigan. These studies have been previously described.
      • Schwartz A.G.
      • Cote M.L.
      • Wenzlaff A.S.
      • Land S.
      • Amos C.
      Racial differences in the association between SNPs on 15q25.1, smoking behavior, and risk of non-small cell lung cancer.
      Rapid case ascertainment was used to identify cases in the population-based Metropolitan Detroit Cancer Surveillance System, which is a National Cancer Institute–funded SEER registry. The institutional review board at Wayne State University approved this study, and written informed consent was provided by all participants. Stage, treatment, and vital status were obtained through linkage with the Detroit SEER registry. Treatment and stage variables were summarized in the same manner as the SCCS with use of both American Joint Committee on Cancer and SEER staging information. Individuals were previously genotyped on the Illumina 1M-Duo BeadChip (Illumina). Supervised analysis (K = 2) was performed with the use of genome-wide single nucleotide polymorphisms, CEU and YRI reference populations, and the software program ADMIXTURE. Time-dependent AUCs were estimated by using the same methods described earlier in this section.

      Results

      In the SCCS, 450 incident NSCLC cases occurred, with 425 (286 in blacks and 139 in whites) remaining after quality control procedures. The individuals in these cases had a median survival time of 0.7 years (range 0.003-8.6 years), during which 359 deaths occurred (248 in blacks and 111 in whites). A total of 42 individuals (32 black and 10 white) were excluded for having survival times less than 30 days. Forty-seven percent of blacks with lung cancer had less than 12 years of education compared with 35% of whites (Table 1). The mean age at lung cancer diagnosis was 60 years for blacks and 63 years for whites. Among males, lung cancer was diagnosed more among blacks than among whites (in 60% versus 42%). Smoking status (current, former, or never) did not differ between blacks and whites, with 94% of blacks and 96% of whites having smoked cigarettes. Twenty-three percent of whites reported a positive family history of lung cancer compared with 9% of blacks. The percentage of cases diagnosed at stage IV was higher in blacks than in whites (52% versus 43%). Although similar numbers of blacks and whites had their lung cancer diagnosed at stage I (16% versus 21%), almost twice as many whites than blacks received a surgery-only course of treatment (11% versus 21%) (Supplementary Fig. 1). The median percentage of African ancestry for self-reported blacks was 85.6% versus 1.3% for self-reported whites (Fig. 1 and see also Table 1).
      Table 1Descriptive Characteristics of Incident NSCLC Cases in the Southern Community Cohort Study by Race
      CharacteristicBlacks (n = 286), n (%)Whites (n = 139), n (%)Total (N = 425), n (%)p Value
      Chi-square and t test p values reported for categorical and continuous variables, respectively.
      Sex6.7 × 10−4
       Male171 (59.8)58 (41.7)229 (53.9)
       Female115 (40.2)81 (58.3)196 (46.1)
      Vital status0.09
       Alive38 (13.3)28 (20.1)66 (15.5)
       Dead248 (86.7)111 (79.9)359 (84.5)
      Median percentage of African ancestry (range)85.6 (<0.01–98.7)1.3 (<0.01–91.1)80.1 (<0.01–98.7)2.2 × 10−16
      Lung cancer stage at diagnosis0.15
       I44 (15.7)29 (21.3)73 (17.5)
       II/III90 (32.0)49 (36.0)139 (33.3)
       IV147 (52.3)58 (42.6)205 (49.2)
       Unknown538
      Treatment0.10
       No treatment75 (26.8)36 (27.3)111 (26.9)
       Surgery only31 (11.1)27 (20.5)58 (14.1)
       Chemotherapy only51 (18.2)18 (13.6)69 (16.7)
       Radiation only36 (12.9)12 (9.1)48 (11.7)
       Multimodality87 (31.1)39 (29.5)126 (30.6)
       Unknown6713
      Histologic type0.74
      Chi-square test conducted with category multiple histologic types excluded on account of small cell count.
       Adenocarcinoma113 (39.5)51 (36.7)164 (38.6)
       NSCLC NOS78 (27.3)34 (24.5)112 (26.4)
       Squamous72 (25.2)41 (29.5)113 (26.6)
       Other NSCLC22 (7.7)12 (8.6)34 (8.0)
       Multiple histologic types1 (0.3)1 (0.7)2 (0.5)
      Mean age at enrollment, y (SD)56.5 (9.0)60.4 (8.6)57.8 (9.1)2.7 × 10−5
      Mean age at diagnosis, y (SD)59.6 (9.1)62.8 (8.7)60.6 (9.1)4.5 × 10−4
      Median observed duration of disease among those who died, y (range)0.50 (0.003–8.61)0.54 (0.01–5.4)0.52 (0.003–8.61)0.43
      Median observed duration of disease among those alive at last follow-up, y (range)3.8 (1.5–8.2)4.0 (1.5–7.7)3.5 (1.5–8.2)0.54
      Highest education level, y0.02
       <12134 (47.2)48 (34.8)182 (43.1)
       ≥12150 (52.8)90 (65.2)240 (56.9)
       Unknown213
      Household income in past year0.53
       <$15,000190 (67.6)87 (64.0)277 (66.4)
       ≥$15,00091 (32.4)49 (36.0)140 (33.6)
       Unknown538
      Smoking status at cohort entry0.29
       Current206 (72.8)93 (68.4)299 (71.4)
       Former59 (20.8)37 (27.2)96 (22.9)
       Never18 (6.4)6 (4.4)24 (5.7)
       Unknown336
      Mean cigarettes per day (SD)15.7 (13.1)23.2 (14.7)18.1 (14.1)8.7 × 10−7
      Smokes menthol cigarettes2.2 × 10−16
       Yes182 (69.2)23 (17.8)205 (52.3)
       No81 (30.8)106 (82.2)187 (47.7)
       Unknown231033
      Self-reported doctor diagnosis of emphysema or chronic bronchitis5.3 × 10−7
       Yes28 (9.9)41 (29.7)69 (16.4)
       No255 (90.1)97 (70.3)352 (83.6)
       Unknown314
      First-degree relative with lung cancer0.001
       Yes21 (9.3)26 (22.6)47 (13.8)
       No205 (90.7)89 (77.4)294 (86.2)
       Unknown602484
      Mean BMI, kg/m2 (SD)26.8 (6.1)27.1 (6.0)26.9 (6.1)0.59
      Health insurance status0.06
       Yes174 (61.7)98 (71.5)272 (64.9)
       No108 (38.3)39 (28.5)147 (35.1)
       Unknown426
      Enrollment source1.6 × 10−4
       Community health center268 (93.7)113 (81.3)381 (89.6)
       General population18 (6.3)26 (18.7)44 (10.4)
      NOS, not otherwise specified; BMI, body mass index.
      a Chi-square and t test p values reported for categorical and continuous variables, respectively.
      b Chi-square test conducted with category multiple histologic types excluded on account of small cell count.
      Figure thumbnail gr1
      Figure 1Genetic ancestry estimates for blacks (n = 286) and whites (n = 139) with NSCLC in the Southern Community Cohort Study. Global ancestry was estimated with ADMIXTURE software and ancestry-informative markers (n =1137 in blacks and n =553 in whites) by using a supervised method and including CEU (European) and YRI (African) reference populations from the International HapMap Project. Individuals are plotted along the x axis, and the percentage of African ancestry (dark blue) and percentage of European ancestry (light blue) for each individual are plotted along the y axis.
      Cox proportional hazards models were implemented to determine the impact of African ancestry on overall mortality. The unadjusted Cox model for percentage of African ancestry had an average time-dependent AUC of 0.54 (Table 2). In the main effects model, African ancestry was not associated with overall mortality, with or without stage and treatment (Fig. 2A and B), although at smaller values of African ancestry a reduction in mortality was observed. We then estimated the AUC to assess the predictive ability of each model. With an average time-dependent AUC of 0.79, the main effects multivariable model examining the association between African ancestry and overall mortality performed dramatically better than the univariate model with African ancestry alone (Fig. 3). The inclusion of interaction terms to the main effects model to create an “overfit” model increased the average time-dependent AUC slightly (to 0.83), indicating that the main effects model had high predictive ability. Removing African ancestry had no impact on the average time-dependent AUC for either the main effects model (see Fig. 3 and Table 2 ) or the interactions model (see Table 2). Removing stage and treatment from the main effects and interaction models substantially decreased the average time-dependent AUC to 0.65 and 0.67, respectively (see Table 2). Further removal of African ancestry from the main effects and interaction models without stage and treatment resulted in little change in the average time-dependent AUC (see Fig. 3 and Table 2). We then removed whites from the SCCS and examined each of the Cox proportional hazards models among blacks only. The observations were similar to those for the overall NSCLC population and are presented in Supplementary Figure 2 and Supplementary Table 1. The unadjusted survival curves stratified by race, stage, and interaction of race and stage (Fig 4) show that a clinically relevant difference in lung cancer survival between black and white individuals is not supported by the data. Instead, stage and treatment appear to play a more significant role in survival.
      Table 2Average Time-Dependent AUCs for Cox Proportional Hazards Models in the Southern Community Cohort Study
      ModelAverage Time-Dependent AUC
      African ancestry only (unadjusted)0.54
      Main effects
      Main effects model is a Cox proportional hazards model examining the association between African ancestry and lung cancer mortality with adjustment for age at diagnosis, sex, body mass index (kg/m2), cigarettes per day, stage at diagnosis, treatment, highest education level, and family history of lung cancer.
      0.79
      Interactions
      Interaction model is a Cox proportional hazards model examining the association between African ancestry and lung cancer mortality with adjustment for the same variables in the main effects model but also including the following two-way interactions (proportion of African ancestry × treatment, African ancestry × stage, African ancestry × education, education × treatment, education × stage, treatment × stage, sex × age, sex × body mass index, and sex × cigarettes per day).
      0.83
      Main effects without African ancestry0.79
      Interactions without African ancestry0.82
      Main effects without stage and treatment0.65
      Interactions without stage and treatment0.67
      Main effects without stage, treatment, and African ancestry0.63
      Interactions without stage, treatment, and African ancestry0.66
      AUC, area under the curve.
      a Main effects model is a Cox proportional hazards model examining the association between African ancestry and lung cancer mortality with adjustment for age at diagnosis, sex, body mass index (kg/m2), cigarettes per day, stage at diagnosis, treatment, highest education level, and family history of lung cancer.
      b Interaction model is a Cox proportional hazards model examining the association between African ancestry and lung cancer mortality with adjustment for the same variables in the main effects model but also including the following two-way interactions (proportion of African ancestry × treatment, African ancestry × stage, African ancestry × education, education × treatment, education × stage, treatment × stage, sex × age, sex × body mass index, and sex × cigarettes per day).
      Figure thumbnail gr2
      Figure 2African ancestry is not associated with lung cancer mortality, with or without stage and treatment included in the model. Hazard ratios and 95% confidence intervals are plotted on the y axis for the association between splined percentage of African ancestry (x axis) and lung cancer mortality in the Southern Community Cohort Study for the main effects model (A) and the main effects model without stage and treatment (B) (see ). Median African ancestry (80%) is the referent (vertical gray line). The main effects model is a Cox proportional hazards model examining the association between African ancestry and lung cancer mortality, with adjustment for age at diagnosis, sex, body mass index (kg/m2), cigarettes per day, stage at diagnosis, treatment, highest education level, and family history of lung cancer.
      Figure thumbnail gr3
      Figure 3Time-dependent areas under the curve (AUCs) for each Cox proportional hazards model in the Southern Community Cohort Study (SCCS). Removal of stage and treatment, not African ancestry, resulted in a reduction in the predictive ability of the main effects model. African ancestry had no effect on the time-dependent AUC of the main effects model, with or without stage and treatment. The Kaplan-Meier survival probability estimates below the x axis represent the probability of surviving to the indicated 6-month intervals as estimated from the main effects model (black text) and from the main effects model without stage or treatment (blue text).
      Figure thumbnail gr4
      Figure 4Probability of survival stratified by race (A), stage (B), and race and stage (C).
      We evaluated the impact of African ancestry on overall mortality in a population of 316 blacks with NSCLC ascertained from the Karmanos Cancer Institute at Wayne State University. Briefly, the mean age of diagnosis was 60 years and 41% of lung cancer cases were in males. Thirty-four percent of cases were diagnosed at stage I, 39% were diagnosed at stage II/III, and 26% were diagnosed at stage IV (Supplementary Table 2 and see also Supplementary Fig. 1). The median percentage of African ancestry was 83.3% (Supplementary Fig. 3). Additional descriptive characteristics of this population are provided in Supplementary Table 2. Similar findings for the impact of African ancestry on overall mortality were observed in the blacks in the SCCS (Supplementary Fig. 4 and see also Supplementary Table 1). Removal of African ancestry from both the main effects model and the interaction model had a negligible effect on the average time-dependent AUC values (0.74 and 0.76, respectively [see Supplementary Table 1]). As in the SCCS, the removal of stage and treatment from the main effects model resulted in a dramatic decrease in the average time-dependent AUC (0.63) and removal of African ancestry in addition to stage and treatment had little impact (average time-dependent AUC = 0.61).

      Discussion

      We found that African ancestry was not associated with NSCLC mortality in the SCCS when we used a Cox proportional hazard model adjusted for stage and treatment. Because African ancestry did not have a linear relationship with mortality, we chose to model the predictor as a restricted cubic spline. As such, there is no single hazard ratio or p value to describe the magnitude and significance of the association. Instead, we have presented the hazard ratio and confidence interval as a function of African ancestry in Figure 2, with use of the median African ancestry value as the referent. It is worth noting that it is impossible to measure evidence of no effect by using a p value.
      • Goodman S.N.
      Toward evidence-based medical statistics. 1: the p value fallacy.
      Hence, the display of the 95% CI for hazard ratios over African ancestry is critically important. We found the 95% CI to be tight around the estimated hazard ratio of approximately 1.0 for African ancestry, providing evidence to support the null hypothesis of no association. Furthermore, we examined the impact of African ancestry on mortality by comparing the average time-dependent AUC for Cox proportional hazards models with and without African ancestry or stage and treatment. This allows us to examine the clinical impact of such predictors rather than simply the strength of the association. Using these methods, we found that stage and treatment are strongly predictive of overall mortality and are more important predictors of mortality than African ancestry. This result was recapitulated in an independent population of blacks with NSCLC from Karmanos Cancer Institute at Wayne State University, providing further evidence of the null association between African ancestry and overall mortality.
      In this study we used genetic ancestry as a continuous proxy for race. Although genetic ancestry and race are highly correlated, race can be viewed as a social construct that captures both genetic and nongenetic factors such as culture and social perception. Given the high variability of genetic ancestry that is not captured by categorically defined race, adjusting for race alone while examining diseases with established racial disparities may not entirely account for underlying population substructure.
      • Bryc K.
      • Auton A.
      • Nelson M.R.
      • et al.
      Genome-wide patterns of population structure and admixture in West Africans and African Americans.
      • Burchard E.G.
      • Ziv E.
      • Coyle N.
      • et al.
      The importance of race and ethnic background in biomedical research and clinical practice.
      Here, the utilization of African ancestry instead of race allowed us to attempt to disentangle the genetic and social disparities associated with lung cancer survival.
      • Aldrich M.C.
      • Selvin S.
      • Wrensch M.R.
      • et al.
      Socioeconomic status and lung cancer: unraveling the contribution of genetic admixture.
      • Gonzalez Burchard E.
      • Borrell L.N.
      • Choudhry S.
      • et al.
      Latino populations: a unique opportunity for the study of race, genetics, and social environment in epidemiological research.
      • Tang H.
      • Quertermous T.
      • Rodriguez B.
      • et al.
      Genetic structure, self-identified race/ethnicity, and confounding in case-control association studies.
      We found that African ancestry is not associated with overall mortality when adjusting for stage and treatment (Figure 2A), with hazard ratios and confidence intervals encompassing 1.0 for all values of African ancestry. When stage and treatment were removed from the model (see Fig. 2B), we observed a slightly, although not statistically significant, decreased mortality for individuals with smaller proportions of African ancestry in the SCCS. We hypothesize that this observed reduction in mortality is due to the influence of treatment, social, environmental, or other factors that are not captured by genetic ancestry alone. It is possible that these observed differences are the result of cultural differences in the perception of disease and the willingness to seek treatment.
      • Lathan C.S.
      • Okechukwu C.
      • Drake B.F.
      Bennett GGal. Racial differences in the perception of lung cancer: the 2005 Health Information National Trends Survey.
      • Margolis M.L.
      • Christie J.D.
      • Silvestri G.A.
      • Kaiser L.
      • Santiago S.
      • Hansen-Flaschen J.
      Racial differences pertaining to a belief about lung cancer surgery: results of a multicenter survey.
      Herein we have observed a striking decrease in the predictive ability of the model when stage and treatment are excluded from Cox models for lung cancer. Prior work has shown that racial disparities in lung cancer survival disappear when controlling for stage at diagnosis or treatment.
      • Ganti A.K.
      • Subbiah S.P.
      • Kessinger A.
      • Gonsalves W.I.
      • Silberstein P.T.
      • Loberiza Jr., F.R.
      Association between race and survival of patients with non–small-cell lung cancer in the United States veterans affairs population.
      • Zheng L.
      • Enewold L.
      • Zahm S.H.
      • et al.
      Lung cancer survival among black and white patients in an equal access health system.
      • Aldrich M.C.
      • Grogan E.L.
      • Munro H.M.
      • Signorello L.B.
      • Blot W.J.
      Stage-adjusted lung cancer survival does not differe between low-income blacks and whites.
      Together, these studies along with the present analysis suggest that the observed disparity in survival between blacks and whites can be attributed to differences in stage at diagnosis or receipt of treatment rather than to race. It is important to note that although this study shows no association between genetic ancestry and overall mortality, it does not eliminate the potential for a genetic contribution to lung cancer survival or the possibility of race-specific genetic risk factors.
      To our knowledge, this is the first study to examine the relationship between genetic ancestry and lung cancer mortality in blacks and whites with NSCLC. By utilizing the unique SCCS as a resource, we were able to control for multiple factors potentially influencing lung cancer survival, including socioeconomic status, family history of lung cancer, cigarette smoking, disease stage, and treatment received. Despite our limited sample size and the different characteristics between our discovery and replication study populations, including stage at diagnosis, our confidence intervals are narrow and the consistency of our findings between two independent study populations emphasizes the robustness of our findings. Furthermore, we acknowledge that the staging and treatment information obtained through linkage with cancer registries may not depict the most accurate and up-to-date clinical and staging information. Additionally, these data do not capture recent advances in lung cancer treatment, which include targeted therapeutics and immunotherapies. Future studies should be conducted in a clinical population with carefully annotated clinical information for examining racial differences in lung cancer survival.
      In summary, we find that African ancestry is not associated with NSCLC mortality and that stage and treatment are robust predictors of lung cancer mortality. These findings suggest that efforts to increase the early detection of lung cancer will improve lung cancer outcomes for both blacks and whites.

      Acknowledgments

      This work was supported by a Department of Defense Early Investigator Synergistic Idea Award to Dr. Aldrich (grant W81XWH-12-1-0547) and Dr. Grogan (grant W81XWH-12-1-0544); the National Cancer Institute at the National Institutes of Health (grant K07 CA172294) to Dr. Aldrich; a Department of Veterans Affairs Career Development Award (grant 10-024) to Dr. Grogan; the National Institutes of Health to Dr. Schwartz (grants R01CA60691, R01CA87895 and HHSN261201300011I) and Dr. Blot (grants R01CA092447 and U01202979); and a National Institute of General Medical Sciences at the National Institutes of Health training grant to Dr. Jones (grant 5T32GM080178 [principal investigator Nancy Cox]). The funding bodies had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication. The data on cancer cases in the Southern Community Cohort Study that were used in this publication were provided by the Alabama Statewide Cancer Registry; Kentucky Cancer Registry; Tennessee Department of Health, Office of Cancer Surveillance; Florida Cancer Data System; North Carolina Central Cancer Registry, North Carolina Division of Public Health; Georgia Comprehensive Cancer Registry; Louisiana Tumor Registry; Mississippi Cancer Registry; South Carolina Central Cancer Registry; Virginia Department of Health, Virginia Cancer Registry; and Arkansas Department of Health, Cancer Registry. The Arkansas Central Cancer Registry is fully funded by a grant from National Program of Cancer Registries, U.S. Centers for Disease Control and Prevention (CDC). Data on cancer cases from Mississippi that were part of the SCCS were collected by the Mississippi Cancer Registry, which participates in the National Program of Cancer Registries of the CDC. The contents of this publication are solely the responsibility of the authors and do not necessarily represent the official views of the CDC or the Mississippi Cancer Registry.

      Supplementary Data

      References

        • Henley S.J.
        • Singh S.D.
        • King J.
        • et al.
        Invasive cancer incidence and survival–United States, 2011.
        MMWR Morbid Mortal Wkly Rep. 2015; 64: 237-242
      1. National Institutes of Health. National Cancer Institute. Surveillance, Epidemiology, and End Results Program. Previous version: SEER Cancer Statistics Review, 1975-2013. http://seer.cancer.gov/csr/1975_2013/. Accessed April 12, 2018.

        • Aizer A.A.
        • Wilhite T.J.
        • Chen M.H.
        • et al.
        Lack of reduction in racial disparities in cancer-specific mortality over a 20-year period.
        Cancer. 2014; 120: 1532-1539
        • American Cancer Society
        Cancer Facts & Figures 2017.
        American Cancer Society, Atlanta, GA2017
        • Moolgavkar S.H.
        • Holford T.R.
        • Levy D.T.
        • et al.
        Impact of reduced tobacco smoking on lung cancer mortality in the United States during 1975-2000.
        J Natl Cancer Inst. 2012; 104: 541-548
        • Efird J.T.
        • Landrine H.
        • Shiue K.Y.
        • et al.
        Race, insurance type, and stage of presentation among lung cancer patients.
        SpringerPlus. 2014; 3: 710
        • Hardy D.
        • Liu C.C.
        • Xia R.
        • et al.
        Racial disparities and treatment trends in a large cohort of elderly black and white patients with nonsmall cell lung cancer.
        Cancer. 2009; 115: 2199-2211
        • Halpern M.T.
        • Ward E.M.
        • Pavluck A.L.
        • Schrag N.M.
        • Bian J.
        • Chen A.Y.
        Association of insurance status and ethnicity with cancer stage at diagnosis for 12 cancer sites: a retrospective analysis.
        Lancet Oncol. 2008; 9: 222-231
        • Caposole M.Z.
        • Miller K.
        • Kim J.N.
        • Steward N.A.
        • Bauer T.L.
        Elimination of socioeconomic and racial disparities related to lung cancer: closing the gap at a high volume community cancer center.
        Surg Oncol. 2014; 23: 46-52
        • Ganti A.K.
        • Subbiah S.P.
        • Kessinger A.
        • Gonsalves W.I.
        • Silberstein P.T.
        • Loberiza Jr., F.R.
        Association between race and survival of patients with non–small-cell lung cancer in the United States veterans affairs population.
        Clin Lung Cancer. 2014; 15: 152-158
        • Zheng L.
        • Enewold L.
        • Zahm S.H.
        • et al.
        Lung cancer survival among black and white patients in an equal access health system.
        Cancer Epidemiol Biomarkers Prev. 2012; 21: 1841-1847
        • Hua X.
        • Ward K.C.
        • Gillespie T.W.
        • Lipscomb J.
        • Goodman M.
        Non-small cell lung cancer treatment receipt and survival among African-Americans and whites in a rural area.
        J Commun Health. 2014; 39: 696-705
        • Aldrich M.C.
        • Grogan E.L.
        • Munro H.M.
        • Signorello L.B.
        • Blot W.J.
        Stage-adjusted lung cancer survival does not differe between low-income blacks and whites.
        J Thorac Oncol. 2013; 8: 1248-1454
        • Zeng C.
        • Wen W.
        • Morgans A.K.
        • Pao W.
        • Shu X.O.
        • Zheng W.
        Disparities by race, age, and sex in the improvement of survival for major cancers: results from the National Cancer Institute Surveillance, Epidemiology, and End Results (SEER) program in the United States, 1990 to 2010.
        JAMA Oncol. 2015; 1: 88-96
        • Bryc K.
        • Auton A.
        • Nelson M.R.
        • et al.
        Genome-wide patterns of population structure and admixture in West Africans and African Americans.
        Proc Natl Acad Sci U S A. 2010; 107: 786-791
        • Tishkoff S.A.
        • Reed F.A.
        • Friedlaender F.R.
        • et al.
        The genetic structure and history of Africans and African Americans.
        Science. 2009; 324: 1035-1044
        • Bryc K.
        • Durand E.Y.
        • Macpherson J.M.
        • et al.
        The genetic ancestry of African Americans, Latinos, and European Americans across the United States.
        Am J Hum Genet. 2015; 96: 37-53
        • Dumitrescu L.
        • Restrepo N.A.
        • Goodloe R.
        • et al.
        Towards a phenome-wide catalog of human clinical traits impacted by genetic ancestry.
        BioData Min. 2015; 8: 35
        • Goetz L.H.
        • Uribe-Bruce L.
        • Quarless D.
        • Libiger O.
        • Schork N.J.
        Admixture and clinical phenotypic variation.
        Hum Hered. 2014; 77: 73-86
        • Peralta C.A.
        • Risch N.
        • Lin F.
        • et al.
        The Association of African Ancestry and elevated creatinine in the Coronary Artery Risk Development in Young Adults (CARDIA) Study.
        Am J Nephrol. 2010; 31: 202-208
        • Kumar R.
        • Seibold M.A.
        • Aldrich M.C.
        • et al.
        Genetic ancestry in lung-function predictions.
        N Engl J Med. 2010; 363: 321-330
        • Moreno-Estrada A.
        • Gignoux C.R.
        • Fernandez-Lopez J.C.
        • et al.
        Human genetics. The genetics of Mexico recapitulates Native American substructure and affects biomedical traits.
        Science. 2014; 344: 1280-1285
        • Salari K.
        • Choudhry S.
        • Tang H.
        • et al.
        Genetic admixture and asthma-related phenotypes in Mexican American and Puerto Rican asthmatics.
        Genet Epidemiol. 2005; 29: 76-86
        • Fejerman L.
        • Hu D.
        • Huntsman S.
        • et al.
        Genetic ancestry and risk of mortality among U.S. Latinas with breast cancer.
        Cancer Res. 2013; 73: 7243-7253
        • Fejerman L.
        • John E.M.
        • Huntsman S.
        • et al.
        Genetic ancestry and risk of breast cancer among U.S. Latinas.
        Cancer Res. 2008; 68: 9723-9728
        • Signorello L.B.
        • Hargreaves M.K.
        • Steinwandel M.D.
        • et al.
        Southern Community Cohort Study: establishing a cohort to investigate health disparities.
        J Natl Med Assoc. 2005; 97: 972-979
        • Signorello L.B.
        • Hargreaves M.K.
        • Blot W.J.
        The Southern Community Cohort Study: investigating health disparities.
        J Health Care Poor Underserved. 2010; 21: 26-37
        • Purcell S.
        • Neale B.
        • Todd-Brown K.
        • et al.
        PLINK: a tool set for whole-genome association and population-based linkage analyses.
        Am J Hum Genet. 2007; 81: 559-575
        • Alexander D.H.
        • Novembre J.
        • Lange K.
        Fast model-based estimation of ancestry in unrelated individuals.
        Genome Res. 2009; 19: 1655-1664
        • International HapMap Consortium
        The International HapMap Project.
        Nature. 2003; 426: 789-796
        • Heagerty P.J.
        • Lumley T.
        • Pepe M.S.
        Time-dependent ROC curves for censored survival data and a diagnostic marker.
        Biometrics. 2000; 56: 337-344
        • Rubin D.B.
        Multiple Imputation for Nonresponse in Surveys.
        John Wiley & Sons, Hoboken, NJ2009
        • Schwartz A.G.
        • Cote M.L.
        • Wenzlaff A.S.
        • Land S.
        • Amos C.
        Racial differences in the association between SNPs on 15q25.1, smoking behavior, and risk of non-small cell lung cancer.
        J Thorac Oncol. 2009; 4: 1195-1201
        • Goodman S.N.
        Toward evidence-based medical statistics. 1: the p value fallacy.
        Ann Inter Med. 1999; 130: 995-1004
        • Burchard E.G.
        • Ziv E.
        • Coyle N.
        • et al.
        The importance of race and ethnic background in biomedical research and clinical practice.
        N Engl J Med. 2003; 348: 1170-1175
        • Aldrich M.C.
        • Selvin S.
        • Wrensch M.R.
        • et al.
        Socioeconomic status and lung cancer: unraveling the contribution of genetic admixture.
        Am J Public Health. 2013; 103: e73-e80
        • Gonzalez Burchard E.
        • Borrell L.N.
        • Choudhry S.
        • et al.
        Latino populations: a unique opportunity for the study of race, genetics, and social environment in epidemiological research.
        Am J Public Health. 2005; 95: 2161-2168
        • Tang H.
        • Quertermous T.
        • Rodriguez B.
        • et al.
        Genetic structure, self-identified race/ethnicity, and confounding in case-control association studies.
        Am J Hum Genet. 2005; 76: 268-275
        • Lathan C.S.
        • Okechukwu C.
        • Drake B.F.
        Bennett GGal. Racial differences in the perception of lung cancer: the 2005 Health Information National Trends Survey.
        Cancer. 2010; 116: 1981-1986
        • Margolis M.L.
        • Christie J.D.
        • Silvestri G.A.
        • Kaiser L.
        • Santiago S.
        • Hansen-Flaschen J.
        Racial differences pertaining to a belief about lung cancer surgery: results of a multicenter survey.
        Ann Intern Med. 2003; 139: 558-563