- Email: [email protected]
Interactive effects of individual and neighborhood race and ethnicity on rates of high-grade cervical lesions
Cancer Epidemiology 38 (2014) 248–252
Contents lists available at ScienceDirect
Cancer Epidemiology The International Journal of Cancer Epidemiology, Detection, and Prevention journal homepage: www.cancerepidemiology.net
Interactive effects of individual and neighborhood race and ethnicity on rates of high-grade cervical lesions Christina Waggaman a, Pamela Julian b, Linda M. Niccolai a,b,* a b
Yale School of Public Health, Department of Epidemiology of Microbial Diseases, 60 College Street, New Haven, CT 06520, USA Yale School of Public Health, Connecticut Emerging Infections Program, One Church Street, New Haven, CT 06520, USA
A R T I C L E I N F O
A B S T R A C T
Article history: Received 18 September 2013 Received in revised form 27 February 2014 Accepted 10 March 2014 Available online 4 April 2014
We estimated the main and interactive effects of individual race/ethnicity (black, Hispanic, white) and area race, ethnicity, and poverty (proportions of the female population black, Hispanic, and living below the federal poverty level at the census tract level, respectively) on rates of high-grade cervical lesions among young women. Using data from a statewide surveillance system during 2008–2011, we found a marginally significant interaction (P < 0.05) between individual race/ethnicity and area race, with black and Hispanic women living in areas with 20% of the female population black having elevated rates compared to those living in areas with <20% of the female population black. These findings indicate a possible synergistic effect between individual race/ethnicity and racial composition in neighborhoods on precancerous cervical lesions. ß 2014 Elsevier Ltd. All rights reserved.
Keywords: Health disparities Race Ethnicity High grade cervical lesions Interaction Synergy
1. Introduction Disparities in cervical cancer incidence and mortality have persisted in the United States despite overall declines in recent decades [1,2]. High-grade cervical lesions (HGCL) are precursors to invasive cervical carcinoma and an important public health problem for several additional reasons including their high burden among young women, associated health care costs, and psychological distress [3,4]. Over half of HGCL will persist or progress in the absence of treatment which typically involves an outpatient excisional surgical procedure to remove abnormal tissue [5]. Previous work has shown disparities in the burden of HGCL by area measures of race and poverty [6], but we are not aware of any studies that have assessed possible synergy between area and individual measures of race/ethnicity on cervical cancer or its precursors. Combining census data with geocoded surveillance data can be used to explore geographic differences in health outcomes to better elucidate socioeconomic disparities [7]. Specifically, analyses done at the neighborhood level (e.g., census tracts) can inform targeted public health interventions for specific communities [8,9].
* Corresponding author at: Yale School of Public Health, Department of Epidemiology of Microbial Diseases, 60 College Street, New Haven, CT 06520, USA. Tel.: +1 203 785 7834. E-mail address: [email protected] (L.M. Niccolai). http://dx.doi.org/10.1016/j.canep.2014.03.004 1877-7821/ß 2014 Elsevier Ltd. All rights reserved.
Area characteristics such as the physical environment, quality of housing, availability of services, socio-cultural factors, and perceived reputation by outsiders can lead to disparities in health outcomes for resident individuals [10]. However, studies that solely assess risk at the area-level are unable to disentangle health risks caused by the area itself from those caused by the characteristics of the individuals who inhabit it. As such, multilevel analysis has increasingly become a valuable method for analyzing geographic health differences, as it allows for the amount of risk ‘‘explained’’ by area characteristics to be measured while controlling for individual level risk factors [11]. Previous research using surveillance data in multilevel models has shown neighborhood socioeconomic variables to be risk factors for health outcomes such as all-cause mortality, obesity, risky sexual behavior, and cancer survival after controlling for individual risk factors [12–17]. Although less researched, interactive effects between area and individual measures may also be present and are potentially important when considering public health interventions; significant interaction between individual demographic variables and neighborhood socioeconomic variables has been shown to be associated with risk of low birth-weight and ischemic stroke [18,19]. While no research thus far has explored synergy between individual and area-level race in development of HGCL, studies have used multilevel models to demonstrate that both area and individual levels of race/ethnicity and socioeconomic status are
C. Waggaman et al. / Cancer Epidemiology 38 (2014) 248–252
associated with cervical cancer mortality. One study discovered that the measure of proportion of black residents was associated with mortality after controlling for individual race/ethnicity, age at diagnosis, cancer stage, and surgery [20]; another study showed that neighborhood deprivation was a significant risk factor for cervical cancer mortality after controlling for individual socioeconomic variables [21]. The goal of the present analysis was to examine the interaction between individual race/ethnicity and area measures of racial, ethnic, and income composition at the census tract level specifically on the rates of HGCL. 2. Methods Methods for this project have been previously described [6]. Briefly, the Connecticut Department of Public Health added HGCL including cervical intraepithelial neoplasia grades 2 and 3 and adenocarcinoma in situ to the list of mandatory reportable diseases in 2008. All pathology laboratories in the state report diagnostic information for all cases as well as patient demographics including residential address using a one-page case report form. Cases were geocoded to the census tract level and linked to 2010 US Census data. Census tracts are widely used as proxies for neighborhood, and were thus used as the area unit of analysis [22]. Enhanced surveillance activities for women residing in New Haven County between the ages of 18 and 39 years include brief telephone interviews and medical record reviews to collect missing demographic information and vaccination histories. All data undergo extensive quality control procedures. Surveillance data were analyzed from January 1, 2008, through December 31, 2011, years for which reporting is complete, for women aged 20–39 years in New Haven County. Women ages 18 and 19 were excluded because different cervical cancer screening guidelines may result in different rates of detection, and women aged 20 years were included to create age categories consistent with surveillance reports as described below. Data were restricted to cases from New Haven County because of more rigorous data collection (patient interviews and medical record reviews) for demographic information. New Haven County has a total population of 862,477 including 14% black and 15% Hispanic residents. Individual race/ethnicity measures obtained from surveillance reports, medical record reviews, and interviews were combined to form a single race/ethnicity variable, which consisted of three categories: non-Hispanic black (subsequently referred to as ‘black’), Hispanic, and non-Hispanic white (subsequently referred to as ‘white’). We used a hierarchical classification scheme when race/ethnicity was available from more than one data source that prioritized interview responses over medical records over surveillance reports. Of women with non-missing race/ethnicity, interview data were most complete (64%) followed by medical records (53%) and surveillance reports (23%). Concordance between each pair of the three data sets was >90%. Age, an important determinant of rates of HGCL [3], was obtained from surveillance reports and grouped into the following five-year categories: 20–24, 25–29, 30–34, and 35–39 years. Area measures of race, ethnicity, and poverty at the census tract level were obtained from 2010 US Census data and included percentage of the female population black (20% black or <20% black), percentage of the female population Hispanic (20% Hispanic or <20% Hispanic), and percentage of the female population living below the federal poverty level (20% in poverty or <20% in poverty), respectively. These cut-points were chosen based on the Public Health Disparities Geocoding Project with the three lower levels combined to maximize differences for comparison [19]. Of the 189 census tracts in the county, 23% had 20% black and 29% had 20% Hispanic, and 20% had 20% in poverty.
249
The data were fit to a multilevel Poisson random effects model with the following variables: individual race/ethnicity, area race, area ethnicity, area poverty, individual age, and census tract. Age was included to adjust for potential confounding and census tract was included as a random effect to control for correlation between strata in the same census tract. The four age and three individual racial/ethnic groups were stratified so that each of the 189 census tracts was divided into 12 strata. Stratum numerators were the counts of cases of HGCL for each age and race/ethnicity combination. Stratum denominators were the total female population in each census tract for each age and race/ethnicity combination (the 12 strata) and multiplied by four, the number of years during which surveillance data were collected. To assess the main effect of individual and area measures, the following model was fit (Model 1): logðli j Þ ¼ logðhi j Þ þ b0 þ bi ðageÞ þ bi ðrace=ethnicityÞ þ b j ðCT raceÞ þ b j ðCT ethnicityÞ þ b j ðCT povertyÞ þ m j where i represents stratum membership, j represents variables at the census tract-level, hi represents the person-time in each stratum, and yj represents the random effect due to each census tract. To assess interaction between individual race/ethnicity and area race, ethnicity, and poverty, three interaction terms were added and the model was fit as (Model 2):
logðli j Þ ¼ logðhi j Þ þ b0 þ bi ðageÞ þ bi ðrace=ethnicityÞ þ b j ðCT raceÞ þ b j ðCT ethnicityÞ þ b j ðCT povertyÞ þ bi j ðrace=ethnicityÞðCT raceÞ þ bi j ðrace=ethnicityÞðCT ethnicityÞ þ bi j ðrace=ethnicityÞðCT povertyÞ þ m j All of the statistical analysis was conducted using SAS v 9.2 (SAS Institute Inc., Cary, NC). 3. Results During 2008–2011, 2338 cases of HGCL were reported among women aged 20–39 years in New Haven County and considered for inclusion in this analysis. Of those, 1359 (58.1%) had non-missing individual race/ethnicity information; cases with missing race/ ethnicity were more likely to be from census tracts with low percentages of the female population black and Hispanic but were not different regarding area poverty and individual age. The analyzed sample included 21.7% (n = 295) black, 29.1% (n = 395) Hispanic, and 49.2% (n = 669) white cases. All cases had complete information on census tract residence; 37%, 39%, and 26% of cases were residents of census tracts with 20% black, Hispanic, and in poverty, respectively. Black women had the highest rates (475 cases per 100,000) followed by Hispanic women (456 cases per 100,000); white women had the lowest rates (253 cases per 100,000). Rates for race/ethnicity groups by area measures of race, ethnicity, and poverty are presented in Table 1. In the analysis of main effects (Table 2), black and Hispanic women had significantly higher rates of HGCL as estimated by the rate ratio (RR) than white women (RR = 1.70, 95% confidence interval (CI): 1.42, 1.99 and RR = 1.64, 95% CI: 1.40, 1.87, respectively). Women residing in census tracts with 20% black had significantly higher rates compared to women in census tracts with <20% black (RR = 1.23, 95% CI: 1.02, 1.47). Rates did not differ significantly by area ethnicity (RR = 1.07, 95% CI: 0.91, 1.27) or area poverty (RR = 0.90, 95% CI: 0.74, 1.09).
C. Waggaman et al. / Cancer Epidemiology 38 (2014) 248–252
250
Table 1 Annual rates of high-grade cervical lesions per 100,000 women by individual and area (census tract) measures, New Haven County, Connecticut, 2008–2011. Area measures
with 20% black (RR = 2.38, 95% CI: 1.82, 3.04) compared to all other areas. The interactions between individual race/ethnicity and area ethnicity (P = 0.64) and area poverty (P = 0.40) were not significant.
Individual measures Black
Hispanic
White
Total
20% Black <20% Black
531.4 361.5
538.8 388.5
239.4 254.6
449.7 283.2
20% Hispanic <20% Hispanic
483.5 466.5
497.9 382.7
270.6 249.4
419.8 288.4
20% In poverty <20% In poverty
502.3 454.6
523.6 411.4
199.8 258.1
423.7 304.2
Total
474.7
456.1
252.7
328.7
Table 2 Annual adjusteda rate ratios (RR) of high-grade cervical lesions for main effects of individual and area measures, New Haven County, Connecticut, 2008–2011 (Model 1).
Individual race/ethnicity White (Reference) Black Hispanic Area race <20% black (Reference) 20% black Area ethnicity <20% Hispanic (Reference) 20% Hispanic Area poverty <20% in poverty (Reference) 20% in poverty
RR
95% CI
1.00 1.70** 1.64**
– (1.42, 1.99) (1.40, 1.87)
1.00 1.23*
– (1.02, 1.47)
1.00 1.07
– (0.91, 1.27)
1.00 0.90
– (0.74, 1.09)
a All models adjusted for each individual and area measure in the table and individual age and census tract. * p-value < 0.05. ** p-value < 001.
There was a marginally significant interaction between individual race/ethnicity and area race (P = .049). For black women, those residing in census tracts where 20% black had 56% higher rates (RR = 1.56, 95% CI: 1.14, 2.12) of HGCL compared to black women who lived in census tracts where <20% black (Table 3). A similar but less strong pattern that was marginally non-significant existed for Hispanic women: those residing in census tracts where 20% black had 27% higher rates (RR = 1.27, 95% CI: 0.98, 1.64) of HGCL compared to Hispanic women who lived in census tracts where <20% black. There was not a significant difference for white women in these two areas. Table 4 shows the disparities between blacks and Hispanics compared to whites for each type of census tract. The disparity between black and white women was strongest in areas
4. Discussion These results reveal a possible synergistic effect between individual race/ethnicity and residence in areas with a higher percentage of the female population black on rates of HGCL. In other words, there is a disproportionate burden of HGCL, an important precursor to invasive cervical cancer, among black women residing in areas with a higher percentage of black female residents. A similar but slightly weaker pattern also existed for Hispanic women. The disparity between black and white women was also strongest in areas with a higher percentage of black female residents (more than 2-fold). It is important to note that these findings are from multivariable models adjusted for area poverty, suggesting that these associations are independent of the relationship between individual race/ethnicity or area race and measures of neighborhood poverty. There are several possible explanations for these descriptive results that are worth mentioning. First, it is important to note that black women have been observed to have the highest prevalence of human papillomavirus (HPV), a necessary cause of HGCL, though we are not aware of any reports of HPV prevalence by area measures [23]. Areas with concentrated minorities may have a greater burden of disease than expected based on individual racial/ ethnic composition alone because of cultural, historical, peer norm, or network characteristics that influence the rates of HGCL. For cancer in general and cervical cancer in particular, health disparities may be related to social inequalities that relate to area-based measures such as geographic or social context factors more so than individual measures [24,25]. For example, it has been noted that discrimination, racism, and residential segregation have played an important role in health disparities among black Americans [26]. Another possible explanation is the availability, access, and quality of health care services for cervical cancer prevention in neighborhoods. For example, differences in screening for cervical cancer and access to timely and high-quality follow-up care for abnormal Pap tests may play a role in these observed associations. HPV vaccination coverage is another such factor. The importance of sexual networks in perpetuating health disparities for sexually transmitted infections has also been empirically demonstrated [27]. Other studies for a range of health conditions including preterm birth, depression, gonorrhea, hypertension, and lung cancer have shown that neighborhood racial composition can increase risk for poor health outcomes among black residents relative to others [28–33].
Table 3 Annual adjusteda rate ratios (RR) of high-grade cervical lesions for area measures within individual race/ethnic groups, New Haven County, Connecticut, 2008–2011 (Model 2). Area measures
Individual measures Hispanic
Black RR
95% CI
RR
(1.14, 2.12)
1.0 1.27
White 95% CI
RR
95% CI
(0.98, 1.64)
1.0 0.96
(0.72, 1.27)
<20% black (reference) 20% black
1.0 1.56*
<20% Hispanic (reference) 20% Hispanic
1.0 0.98
(0.75, 1.29)
1.0 1.13
(0.87, 1.46)
1.0 1.14
(0.90, 1.45)
<20% in poverty (reference) 20% in poverty
1.0 0.87
(0.65, 1.17)
1.0 0.99
(0.76, 1.31)
1.0 0.75
(0.53, 1.06)
a *
All models adjusted for each individual and area measure in the table and individual age and census tract. p-value < 0.01.
C. Waggaman et al. / Cancer Epidemiology 38 (2014) 248–252
251
Table 4 Annual adjusteda rate ratios (RR) of high-grade cervical lesions for individual race/ethnicity within area measures, New Haven County, Connecticut, 2008–2011 (Model 2). Individual measures
Area measures 20% Black RR
White (Reference) Black Hispanic a * **
1.00 2.38** 2.19**
<20% Black
20% Hispanic
<20% Hispanic
20% In poverty
<20% In poverty
95% CI
RR
95% CI
RR
95% CI
RR
95% CI
RR
95% CI
RR
95% CI
(1.82, 3.04) (1.65, 2.90)
1.00 1.46* 1.65**
(1.07, 1.99) (1.28, 2.11)
1.00 1.73** 1.89**
(1.31, 2.29) (1.48, 2.41)
1.00 2.01** 1.91**
(1.54, 2.60) (1.47, 2.49)
1.00 2.01** 2.19**
(1.39, 2.91) (1.54, 3.11)
1.00 1.73** 1.65**
(1.39, 2.14) (1.34, 2.02)
All models adjusted for each individual and area measure in the table and individual age and census tract. p-value < 0.05. p-value < 001.
Results of this study have implications for clinical care and public health programs. In addition to ensuring access to access to high quality prevention (e.g., HPV vaccination), screening (e.g., Pap and HPV testing), and follow-up care (e.g., timely referral to treatment as indicated) at the individual-level for women of color, area-based interventions at the local level may also be warranted. High prevalence areas can be targeted for the most efficient and effective use of prevention resources. For example, local venuebased programs, targeted public awareness campaigns, and social network interventions for promoting the individual-level clinical interventions previously mentioned may be important and impactful. This study has some limitations. Missing demographic information resulted in approximately 40% of cases being excluded from the analysis cases with missing race/ethnicity were more likely to be from census tracts with low percentage black female residents; a significant factor in our results. However, this would only bias our findings if these women also had different rates of HGCL than those with the same race/ ethnicity who were included and this is not known. While we cannot rule out the possibility of bias, we believe that this type of differential inclusion was unlikely to have occurred in this crosssectional analysis. Another potential limitation was that we could not include a measure of individual poverty in the model. Including age (4 categories) and race/ethnicity (3 categories) was possible because this model uses population counts for each of these 12 strata in the denominators of rate calculations for each census tract; these counts are readily available. To include any individual poverty measure would have required population counts further stratified by a census measure of poverty, but these data are not available in public use census files. It is also possible that our findings for individual-level black and Hispanic measures reflect associations with related measures of poverty. Individual socioeconomic factors such as income, employment, and education that could not be included in this analysis may partly explain the observed associations. Furthermore, data were from a single county in Connecticut and results may not be generalizable to other regions. However, a major strength of this analysis is the use of population-based surveillance data with high case ascertainment thereby strengthening the internal validity of our findings. In conclusion, we have demonstrated higher rates of HGCL among black and Hispanic women residing in areas with higher percentages of black female residents. These findings call for additional research into the possible mechanisms for this troubling finding as well as targeted public health interventions for these women to prevent and detect disease early so that negative health outcomes such as cervical cancer can be avoided. Conflict of interest The authors declare that there is no conflict of interest.
Acknowledgements We gratefully acknowledge James Meek MPH and Lynn Sosa MD for guidance with project design and implementation. We thank the Centers for Disease Control and Prevention 5U50CK000195-03 for partial funding of this work. References [1] Watson M, Saraiya M, Benard V, Coughlin SS, Flowers L, Cokkinides V, et al. Burden of cervical cancer in the US, 1998–2003. Cancer 2008;113(10 Suppl.):2855–64. [2] Singh GK, Miller BA, Hankey BF, Edwards BK. Persistent area socioeconomic disparities in US incidence of cervical cancer, mortality, stage, and survival, 1975–2000. Cancer 2004;101:1051–7. [3] Insinga RP, Glass AG, Rush BB. The health care costs of cervical human papillomavirus-related disease. Am J Obstet Gynecol 2004;191:114–20. [4] Rogstad KE. The psychological impact of abnormal cytology and colposcopy. BJOG 2002;109:364–8. [5] Wheeler CM. Natural history of human papillomavirus infections, cytologic and histologic abnormalities, and cancer. Obstet Gynecol Clin North Am 2008;35:519–36. [6] Niccolai LM, Julian PJ, Bilinski A, Mehta NR, Meek JI, Zelterman D, et al. Geographic poverty and racial/ethnic disparities in cervical cancer precursor rates in Connecticut, 2008–2009. Am J Public Health 2013;103(1):156–63. [7] Krieger N, Williams DR, Moss NE. Measuring social class in US public health research: concepts, methodologies, and guidelines. Ann Rev Public Health 1997;18(16):341–78. [8] Roux AVD. Investigating neighborhood and area effects on health. Am J Public Health 2001;19(11):1783–9. [9] Sampson RJ. The neighborhood context of well-being. Perspect Biol Med 2003;46(3):S53–64. [10] MacIntyre S, Maciver S, Sooman A. Area, class and health: should we be focusing on places or people? J Soc Policy 1993;22(2):213–34. [11] Roux AVD, Aiello A. Multilevel analysis of infectious diseases. J Infect Dis 2005;191:S25–33. [12] Subramanian SV, Chen JT, Rehkopf DH, Waterman PD, Krieger N. Racial disparities in context: a multilevel analysis of neighborhood variations in poverty and excess mortality among black populations in Massachusetts. Am J Public Health 2005;95(2):260–5. [13] Chen JT, Rehkopf DH, Waterman PD, Subramanian SV, Coull BA, Cohen B, et al. Mapping and measuring social disparities in premature mortality: the impact of census tract poverty within and across Boston neighborhoods, 1999–2001. J Urban Health: Bull N Y Acad Med 2006;83(6):1063–84. [14] Corral I, Landrine H, Hao Y, Zhao L, Mellerson JL, Cooper DL. Residential segregation, health behavior and overweight/obesity among a national sample of African American adults. J Health Psychol 2012;17(3):371–8. [15] Lopez RP. Neighborhood risk factors for obesity. Obesity 2007;15(8):2111–9. [16] Chen AC-C, Thompson EA, Morrison-Beedy D. Multi-system influences on adolescent risky sexual behavior. Res Nurs Health 2010;33(6):512–27. [17] Russell EF, Kramer MR, Cooper HLF, Gabram-Mendola S, Senior-Crosby D, Jacob Arriola KR. Metropolitan area racial residential segregation, neighborhood racial composition, and breast cancer mortality. Cancer Causes Control 2012;23(9):1519–27. [18] O’Campo P, Xue X, Wang MC, Caughy M. Neighborhood risk factors for low birthweight in Baltimore: a multilevel analysis. Am J Public Health 1997; 87(7):1113–8. [19] Lisabeth LD, Diez-Roux AV, Escobar JD, Smith MA, Morgenstern LB. Neighborhood environment and risk of ischemic stroke: the Brain Attack Surveillance in Corpus Christi (BASIC) project. Am J Epidemiol 2007;165(3):279–89. [20] Lim J, Ashing-Giwa KT. Examining the effect of minority status and neighborhood characteristics on cervical cancer survival outcomes. Gyn Oncol 2011;121(1):87–93. [21] Li X, Sundquist J, Calling S, Zoller B, Sundquist K. Neighborhood deprivation and risk of cervical cancer morbidity and mortality: a multilevel analysis from Sweden. Gyn Oncol 2012;127(2):283–9.
252
C. Waggaman et al. / Cancer Epidemiology 38 (2014) 248–252
[22] Krieger N, Chen JT, Waterman PD, Rehkopf DH, Subramanian SV. Painting a truer picture of US socioeconomic and racial/ethnic health inequalities: the public health disparities geocoding project. Am J Public Health 2005;95(2):312–23. [23] Hariri S, Unger ER, Sternberg M, Dunne EF, Swan D, Patel S, et al. Prevalence of genital human papillomavirus among females in the United States, the National Health and Nutrition Examination Survey, 2003–2006. J Infect Dis 2011;204:566–73. [24] Newmann SJ, Garner EO. Social inequities along the cervical cancer continuum: a structured review. Cancer Causes Control 2005;16:63–70. [25] Krieger N. Defining and investigating social disparities in cancer: critical issues. Cancer Causes Control 2005;16:5–14. [26] Adimora AA, Schoenbach VJ. Social context, sexual networks, and racial disparities in rates of sexually transmitted infections. J Infect Dis 2005;191: S115–22. [27] Laumann E, Youm Y. Race/ethnic group differences in the prevalence of sexually. transmitted diseases in the United States: a network explanation. Sex Transm Dis 1999;66:250–61.
[28] Mason SM, Messer LC, Laraia BA, Mendola P. Segregation and preterm birth: the effects of neighborhood racial composition in North Carolina. Health Place 2009;15:1–9. [29] Osypuk TL, Acevedo-Garcia D. Are racial disparities in preterm birth larger in hypersegregated areas? Am J Epidemiol 2008;167(11):1295–304. [30] Mair C, Diez Roux AV, Osypuk TL, Rapp SR, Seeman T, Watson KE. Is neighborhood racial/ethnic composition associated with depressive symptoms? The multi-ethnic study of atherosclerosis. Soc Sci Med 2010;71:541– 50. [31] Biello KB, Kershaw T, Nelson R, Hogben M, Ickovics J, Niccolai L. Racial residential segregation and rates of gonorrhea in the United States, 2003– 2007. Am J Public Health 2012;102(7):1370–7. [32] Kershaw KN, Diez Roux AV, Burgard SA, Lisabeth LD, Mujahid MS, Schulz AJ. Metropolitan-level racial residential segregation and black-white disparities in hypertension. Am J Epidemiol 2011;174(5):537–45. [33] Hayanga AJ, Zeliadt SB, Backhus LM. Residential segregation and lung cancer mortality in the United States. JAMA Surg 2013;148(1):37–42.
Contents lists available at ScienceDirect
Cancer Epidemiology The International Journal of Cancer Epidemiology, Detection, and Prevention journal homepage: www.cancerepidemiology.net
Interactive effects of individual and neighborhood race and ethnicity on rates of high-grade cervical lesions Christina Waggaman a, Pamela Julian b, Linda M. Niccolai a,b,* a b
Yale School of Public Health, Department of Epidemiology of Microbial Diseases, 60 College Street, New Haven, CT 06520, USA Yale School of Public Health, Connecticut Emerging Infections Program, One Church Street, New Haven, CT 06520, USA
A R T I C L E I N F O
A B S T R A C T
Article history: Received 18 September 2013 Received in revised form 27 February 2014 Accepted 10 March 2014 Available online 4 April 2014
We estimated the main and interactive effects of individual race/ethnicity (black, Hispanic, white) and area race, ethnicity, and poverty (proportions of the female population black, Hispanic, and living below the federal poverty level at the census tract level, respectively) on rates of high-grade cervical lesions among young women. Using data from a statewide surveillance system during 2008–2011, we found a marginally significant interaction (P < 0.05) between individual race/ethnicity and area race, with black and Hispanic women living in areas with 20% of the female population black having elevated rates compared to those living in areas with <20% of the female population black. These findings indicate a possible synergistic effect between individual race/ethnicity and racial composition in neighborhoods on precancerous cervical lesions. ß 2014 Elsevier Ltd. All rights reserved.
Keywords: Health disparities Race Ethnicity High grade cervical lesions Interaction Synergy
1. Introduction Disparities in cervical cancer incidence and mortality have persisted in the United States despite overall declines in recent decades [1,2]. High-grade cervical lesions (HGCL) are precursors to invasive cervical carcinoma and an important public health problem for several additional reasons including their high burden among young women, associated health care costs, and psychological distress [3,4]. Over half of HGCL will persist or progress in the absence of treatment which typically involves an outpatient excisional surgical procedure to remove abnormal tissue [5]. Previous work has shown disparities in the burden of HGCL by area measures of race and poverty [6], but we are not aware of any studies that have assessed possible synergy between area and individual measures of race/ethnicity on cervical cancer or its precursors. Combining census data with geocoded surveillance data can be used to explore geographic differences in health outcomes to better elucidate socioeconomic disparities [7]. Specifically, analyses done at the neighborhood level (e.g., census tracts) can inform targeted public health interventions for specific communities [8,9].
* Corresponding author at: Yale School of Public Health, Department of Epidemiology of Microbial Diseases, 60 College Street, New Haven, CT 06520, USA. Tel.: +1 203 785 7834. E-mail address: [email protected] (L.M. Niccolai). http://dx.doi.org/10.1016/j.canep.2014.03.004 1877-7821/ß 2014 Elsevier Ltd. All rights reserved.
Area characteristics such as the physical environment, quality of housing, availability of services, socio-cultural factors, and perceived reputation by outsiders can lead to disparities in health outcomes for resident individuals [10]. However, studies that solely assess risk at the area-level are unable to disentangle health risks caused by the area itself from those caused by the characteristics of the individuals who inhabit it. As such, multilevel analysis has increasingly become a valuable method for analyzing geographic health differences, as it allows for the amount of risk ‘‘explained’’ by area characteristics to be measured while controlling for individual level risk factors [11]. Previous research using surveillance data in multilevel models has shown neighborhood socioeconomic variables to be risk factors for health outcomes such as all-cause mortality, obesity, risky sexual behavior, and cancer survival after controlling for individual risk factors [12–17]. Although less researched, interactive effects between area and individual measures may also be present and are potentially important when considering public health interventions; significant interaction between individual demographic variables and neighborhood socioeconomic variables has been shown to be associated with risk of low birth-weight and ischemic stroke [18,19]. While no research thus far has explored synergy between individual and area-level race in development of HGCL, studies have used multilevel models to demonstrate that both area and individual levels of race/ethnicity and socioeconomic status are
C. Waggaman et al. / Cancer Epidemiology 38 (2014) 248–252
associated with cervical cancer mortality. One study discovered that the measure of proportion of black residents was associated with mortality after controlling for individual race/ethnicity, age at diagnosis, cancer stage, and surgery [20]; another study showed that neighborhood deprivation was a significant risk factor for cervical cancer mortality after controlling for individual socioeconomic variables [21]. The goal of the present analysis was to examine the interaction between individual race/ethnicity and area measures of racial, ethnic, and income composition at the census tract level specifically on the rates of HGCL. 2. Methods Methods for this project have been previously described [6]. Briefly, the Connecticut Department of Public Health added HGCL including cervical intraepithelial neoplasia grades 2 and 3 and adenocarcinoma in situ to the list of mandatory reportable diseases in 2008. All pathology laboratories in the state report diagnostic information for all cases as well as patient demographics including residential address using a one-page case report form. Cases were geocoded to the census tract level and linked to 2010 US Census data. Census tracts are widely used as proxies for neighborhood, and were thus used as the area unit of analysis [22]. Enhanced surveillance activities for women residing in New Haven County between the ages of 18 and 39 years include brief telephone interviews and medical record reviews to collect missing demographic information and vaccination histories. All data undergo extensive quality control procedures. Surveillance data were analyzed from January 1, 2008, through December 31, 2011, years for which reporting is complete, for women aged 20–39 years in New Haven County. Women ages 18 and 19 were excluded because different cervical cancer screening guidelines may result in different rates of detection, and women aged 20 years were included to create age categories consistent with surveillance reports as described below. Data were restricted to cases from New Haven County because of more rigorous data collection (patient interviews and medical record reviews) for demographic information. New Haven County has a total population of 862,477 including 14% black and 15% Hispanic residents. Individual race/ethnicity measures obtained from surveillance reports, medical record reviews, and interviews were combined to form a single race/ethnicity variable, which consisted of three categories: non-Hispanic black (subsequently referred to as ‘black’), Hispanic, and non-Hispanic white (subsequently referred to as ‘white’). We used a hierarchical classification scheme when race/ethnicity was available from more than one data source that prioritized interview responses over medical records over surveillance reports. Of women with non-missing race/ethnicity, interview data were most complete (64%) followed by medical records (53%) and surveillance reports (23%). Concordance between each pair of the three data sets was >90%. Age, an important determinant of rates of HGCL [3], was obtained from surveillance reports and grouped into the following five-year categories: 20–24, 25–29, 30–34, and 35–39 years. Area measures of race, ethnicity, and poverty at the census tract level were obtained from 2010 US Census data and included percentage of the female population black (20% black or <20% black), percentage of the female population Hispanic (20% Hispanic or <20% Hispanic), and percentage of the female population living below the federal poverty level (20% in poverty or <20% in poverty), respectively. These cut-points were chosen based on the Public Health Disparities Geocoding Project with the three lower levels combined to maximize differences for comparison [19]. Of the 189 census tracts in the county, 23% had 20% black and 29% had 20% Hispanic, and 20% had 20% in poverty.
249
The data were fit to a multilevel Poisson random effects model with the following variables: individual race/ethnicity, area race, area ethnicity, area poverty, individual age, and census tract. Age was included to adjust for potential confounding and census tract was included as a random effect to control for correlation between strata in the same census tract. The four age and three individual racial/ethnic groups were stratified so that each of the 189 census tracts was divided into 12 strata. Stratum numerators were the counts of cases of HGCL for each age and race/ethnicity combination. Stratum denominators were the total female population in each census tract for each age and race/ethnicity combination (the 12 strata) and multiplied by four, the number of years during which surveillance data were collected. To assess the main effect of individual and area measures, the following model was fit (Model 1): logðli j Þ ¼ logðhi j Þ þ b0 þ bi ðageÞ þ bi ðrace=ethnicityÞ þ b j ðCT raceÞ þ b j ðCT ethnicityÞ þ b j ðCT povertyÞ þ m j where i represents stratum membership, j represents variables at the census tract-level, hi represents the person-time in each stratum, and yj represents the random effect due to each census tract. To assess interaction between individual race/ethnicity and area race, ethnicity, and poverty, three interaction terms were added and the model was fit as (Model 2):
logðli j Þ ¼ logðhi j Þ þ b0 þ bi ðageÞ þ bi ðrace=ethnicityÞ þ b j ðCT raceÞ þ b j ðCT ethnicityÞ þ b j ðCT povertyÞ þ bi j ðrace=ethnicityÞðCT raceÞ þ bi j ðrace=ethnicityÞðCT ethnicityÞ þ bi j ðrace=ethnicityÞðCT povertyÞ þ m j All of the statistical analysis was conducted using SAS v 9.2 (SAS Institute Inc., Cary, NC). 3. Results During 2008–2011, 2338 cases of HGCL were reported among women aged 20–39 years in New Haven County and considered for inclusion in this analysis. Of those, 1359 (58.1%) had non-missing individual race/ethnicity information; cases with missing race/ ethnicity were more likely to be from census tracts with low percentages of the female population black and Hispanic but were not different regarding area poverty and individual age. The analyzed sample included 21.7% (n = 295) black, 29.1% (n = 395) Hispanic, and 49.2% (n = 669) white cases. All cases had complete information on census tract residence; 37%, 39%, and 26% of cases were residents of census tracts with 20% black, Hispanic, and in poverty, respectively. Black women had the highest rates (475 cases per 100,000) followed by Hispanic women (456 cases per 100,000); white women had the lowest rates (253 cases per 100,000). Rates for race/ethnicity groups by area measures of race, ethnicity, and poverty are presented in Table 1. In the analysis of main effects (Table 2), black and Hispanic women had significantly higher rates of HGCL as estimated by the rate ratio (RR) than white women (RR = 1.70, 95% confidence interval (CI): 1.42, 1.99 and RR = 1.64, 95% CI: 1.40, 1.87, respectively). Women residing in census tracts with 20% black had significantly higher rates compared to women in census tracts with <20% black (RR = 1.23, 95% CI: 1.02, 1.47). Rates did not differ significantly by area ethnicity (RR = 1.07, 95% CI: 0.91, 1.27) or area poverty (RR = 0.90, 95% CI: 0.74, 1.09).
C. Waggaman et al. / Cancer Epidemiology 38 (2014) 248–252
250
Table 1 Annual rates of high-grade cervical lesions per 100,000 women by individual and area (census tract) measures, New Haven County, Connecticut, 2008–2011. Area measures
with 20% black (RR = 2.38, 95% CI: 1.82, 3.04) compared to all other areas. The interactions between individual race/ethnicity and area ethnicity (P = 0.64) and area poverty (P = 0.40) were not significant.
Individual measures Black
Hispanic
White
Total
20% Black <20% Black
531.4 361.5
538.8 388.5
239.4 254.6
449.7 283.2
20% Hispanic <20% Hispanic
483.5 466.5
497.9 382.7
270.6 249.4
419.8 288.4
20% In poverty <20% In poverty
502.3 454.6
523.6 411.4
199.8 258.1
423.7 304.2
Total
474.7
456.1
252.7
328.7
Table 2 Annual adjusteda rate ratios (RR) of high-grade cervical lesions for main effects of individual and area measures, New Haven County, Connecticut, 2008–2011 (Model 1).
Individual race/ethnicity White (Reference) Black Hispanic Area race <20% black (Reference) 20% black Area ethnicity <20% Hispanic (Reference) 20% Hispanic Area poverty <20% in poverty (Reference) 20% in poverty
RR
95% CI
1.00 1.70** 1.64**
– (1.42, 1.99) (1.40, 1.87)
1.00 1.23*
– (1.02, 1.47)
1.00 1.07
– (0.91, 1.27)
1.00 0.90
– (0.74, 1.09)
a All models adjusted for each individual and area measure in the table and individual age and census tract. * p-value < 0.05. ** p-value < 001.
There was a marginally significant interaction between individual race/ethnicity and area race (P = .049). For black women, those residing in census tracts where 20% black had 56% higher rates (RR = 1.56, 95% CI: 1.14, 2.12) of HGCL compared to black women who lived in census tracts where <20% black (Table 3). A similar but less strong pattern that was marginally non-significant existed for Hispanic women: those residing in census tracts where 20% black had 27% higher rates (RR = 1.27, 95% CI: 0.98, 1.64) of HGCL compared to Hispanic women who lived in census tracts where <20% black. There was not a significant difference for white women in these two areas. Table 4 shows the disparities between blacks and Hispanics compared to whites for each type of census tract. The disparity between black and white women was strongest in areas
4. Discussion These results reveal a possible synergistic effect between individual race/ethnicity and residence in areas with a higher percentage of the female population black on rates of HGCL. In other words, there is a disproportionate burden of HGCL, an important precursor to invasive cervical cancer, among black women residing in areas with a higher percentage of black female residents. A similar but slightly weaker pattern also existed for Hispanic women. The disparity between black and white women was also strongest in areas with a higher percentage of black female residents (more than 2-fold). It is important to note that these findings are from multivariable models adjusted for area poverty, suggesting that these associations are independent of the relationship between individual race/ethnicity or area race and measures of neighborhood poverty. There are several possible explanations for these descriptive results that are worth mentioning. First, it is important to note that black women have been observed to have the highest prevalence of human papillomavirus (HPV), a necessary cause of HGCL, though we are not aware of any reports of HPV prevalence by area measures [23]. Areas with concentrated minorities may have a greater burden of disease than expected based on individual racial/ ethnic composition alone because of cultural, historical, peer norm, or network characteristics that influence the rates of HGCL. For cancer in general and cervical cancer in particular, health disparities may be related to social inequalities that relate to area-based measures such as geographic or social context factors more so than individual measures [24,25]. For example, it has been noted that discrimination, racism, and residential segregation have played an important role in health disparities among black Americans [26]. Another possible explanation is the availability, access, and quality of health care services for cervical cancer prevention in neighborhoods. For example, differences in screening for cervical cancer and access to timely and high-quality follow-up care for abnormal Pap tests may play a role in these observed associations. HPV vaccination coverage is another such factor. The importance of sexual networks in perpetuating health disparities for sexually transmitted infections has also been empirically demonstrated [27]. Other studies for a range of health conditions including preterm birth, depression, gonorrhea, hypertension, and lung cancer have shown that neighborhood racial composition can increase risk for poor health outcomes among black residents relative to others [28–33].
Table 3 Annual adjusteda rate ratios (RR) of high-grade cervical lesions for area measures within individual race/ethnic groups, New Haven County, Connecticut, 2008–2011 (Model 2). Area measures
Individual measures Hispanic
Black RR
95% CI
RR
(1.14, 2.12)
1.0 1.27
White 95% CI
RR
95% CI
(0.98, 1.64)
1.0 0.96
(0.72, 1.27)
<20% black (reference) 20% black
1.0 1.56*
<20% Hispanic (reference) 20% Hispanic
1.0 0.98
(0.75, 1.29)
1.0 1.13
(0.87, 1.46)
1.0 1.14
(0.90, 1.45)
<20% in poverty (reference) 20% in poverty
1.0 0.87
(0.65, 1.17)
1.0 0.99
(0.76, 1.31)
1.0 0.75
(0.53, 1.06)
a *
All models adjusted for each individual and area measure in the table and individual age and census tract. p-value < 0.01.
C. Waggaman et al. / Cancer Epidemiology 38 (2014) 248–252
251
Table 4 Annual adjusteda rate ratios (RR) of high-grade cervical lesions for individual race/ethnicity within area measures, New Haven County, Connecticut, 2008–2011 (Model 2). Individual measures
Area measures 20% Black RR
White (Reference) Black Hispanic a * **
1.00 2.38** 2.19**
<20% Black
20% Hispanic
<20% Hispanic
20% In poverty
<20% In poverty
95% CI
RR
95% CI
RR
95% CI
RR
95% CI
RR
95% CI
RR
95% CI
(1.82, 3.04) (1.65, 2.90)
1.00 1.46* 1.65**
(1.07, 1.99) (1.28, 2.11)
1.00 1.73** 1.89**
(1.31, 2.29) (1.48, 2.41)
1.00 2.01** 1.91**
(1.54, 2.60) (1.47, 2.49)
1.00 2.01** 2.19**
(1.39, 2.91) (1.54, 3.11)
1.00 1.73** 1.65**
(1.39, 2.14) (1.34, 2.02)
All models adjusted for each individual and area measure in the table and individual age and census tract. p-value < 0.05. p-value < 001.
Results of this study have implications for clinical care and public health programs. In addition to ensuring access to access to high quality prevention (e.g., HPV vaccination), screening (e.g., Pap and HPV testing), and follow-up care (e.g., timely referral to treatment as indicated) at the individual-level for women of color, area-based interventions at the local level may also be warranted. High prevalence areas can be targeted for the most efficient and effective use of prevention resources. For example, local venuebased programs, targeted public awareness campaigns, and social network interventions for promoting the individual-level clinical interventions previously mentioned may be important and impactful. This study has some limitations. Missing demographic information resulted in approximately 40% of cases being excluded from the analysis cases with missing race/ethnicity were more likely to be from census tracts with low percentage black female residents; a significant factor in our results. However, this would only bias our findings if these women also had different rates of HGCL than those with the same race/ ethnicity who were included and this is not known. While we cannot rule out the possibility of bias, we believe that this type of differential inclusion was unlikely to have occurred in this crosssectional analysis. Another potential limitation was that we could not include a measure of individual poverty in the model. Including age (4 categories) and race/ethnicity (3 categories) was possible because this model uses population counts for each of these 12 strata in the denominators of rate calculations for each census tract; these counts are readily available. To include any individual poverty measure would have required population counts further stratified by a census measure of poverty, but these data are not available in public use census files. It is also possible that our findings for individual-level black and Hispanic measures reflect associations with related measures of poverty. Individual socioeconomic factors such as income, employment, and education that could not be included in this analysis may partly explain the observed associations. Furthermore, data were from a single county in Connecticut and results may not be generalizable to other regions. However, a major strength of this analysis is the use of population-based surveillance data with high case ascertainment thereby strengthening the internal validity of our findings. In conclusion, we have demonstrated higher rates of HGCL among black and Hispanic women residing in areas with higher percentages of black female residents. These findings call for additional research into the possible mechanisms for this troubling finding as well as targeted public health interventions for these women to prevent and detect disease early so that negative health outcomes such as cervical cancer can be avoided. Conflict of interest The authors declare that there is no conflict of interest.
Acknowledgements We gratefully acknowledge James Meek MPH and Lynn Sosa MD for guidance with project design and implementation. We thank the Centers for Disease Control and Prevention 5U50CK000195-03 for partial funding of this work. References [1] Watson M, Saraiya M, Benard V, Coughlin SS, Flowers L, Cokkinides V, et al. Burden of cervical cancer in the US, 1998–2003. Cancer 2008;113(10 Suppl.):2855–64. [2] Singh GK, Miller BA, Hankey BF, Edwards BK. Persistent area socioeconomic disparities in US incidence of cervical cancer, mortality, stage, and survival, 1975–2000. Cancer 2004;101:1051–7. [3] Insinga RP, Glass AG, Rush BB. The health care costs of cervical human papillomavirus-related disease. Am J Obstet Gynecol 2004;191:114–20. [4] Rogstad KE. The psychological impact of abnormal cytology and colposcopy. BJOG 2002;109:364–8. [5] Wheeler CM. Natural history of human papillomavirus infections, cytologic and histologic abnormalities, and cancer. Obstet Gynecol Clin North Am 2008;35:519–36. [6] Niccolai LM, Julian PJ, Bilinski A, Mehta NR, Meek JI, Zelterman D, et al. Geographic poverty and racial/ethnic disparities in cervical cancer precursor rates in Connecticut, 2008–2009. Am J Public Health 2013;103(1):156–63. [7] Krieger N, Williams DR, Moss NE. Measuring social class in US public health research: concepts, methodologies, and guidelines. Ann Rev Public Health 1997;18(16):341–78. [8] Roux AVD. Investigating neighborhood and area effects on health. Am J Public Health 2001;19(11):1783–9. [9] Sampson RJ. The neighborhood context of well-being. Perspect Biol Med 2003;46(3):S53–64. [10] MacIntyre S, Maciver S, Sooman A. Area, class and health: should we be focusing on places or people? J Soc Policy 1993;22(2):213–34. [11] Roux AVD, Aiello A. Multilevel analysis of infectious diseases. J Infect Dis 2005;191:S25–33. [12] Subramanian SV, Chen JT, Rehkopf DH, Waterman PD, Krieger N. Racial disparities in context: a multilevel analysis of neighborhood variations in poverty and excess mortality among black populations in Massachusetts. Am J Public Health 2005;95(2):260–5. [13] Chen JT, Rehkopf DH, Waterman PD, Subramanian SV, Coull BA, Cohen B, et al. Mapping and measuring social disparities in premature mortality: the impact of census tract poverty within and across Boston neighborhoods, 1999–2001. J Urban Health: Bull N Y Acad Med 2006;83(6):1063–84. [14] Corral I, Landrine H, Hao Y, Zhao L, Mellerson JL, Cooper DL. Residential segregation, health behavior and overweight/obesity among a national sample of African American adults. J Health Psychol 2012;17(3):371–8. [15] Lopez RP. Neighborhood risk factors for obesity. Obesity 2007;15(8):2111–9. [16] Chen AC-C, Thompson EA, Morrison-Beedy D. Multi-system influences on adolescent risky sexual behavior. Res Nurs Health 2010;33(6):512–27. [17] Russell EF, Kramer MR, Cooper HLF, Gabram-Mendola S, Senior-Crosby D, Jacob Arriola KR. Metropolitan area racial residential segregation, neighborhood racial composition, and breast cancer mortality. Cancer Causes Control 2012;23(9):1519–27. [18] O’Campo P, Xue X, Wang MC, Caughy M. Neighborhood risk factors for low birthweight in Baltimore: a multilevel analysis. Am J Public Health 1997; 87(7):1113–8. [19] Lisabeth LD, Diez-Roux AV, Escobar JD, Smith MA, Morgenstern LB. Neighborhood environment and risk of ischemic stroke: the Brain Attack Surveillance in Corpus Christi (BASIC) project. Am J Epidemiol 2007;165(3):279–89. [20] Lim J, Ashing-Giwa KT. Examining the effect of minority status and neighborhood characteristics on cervical cancer survival outcomes. Gyn Oncol 2011;121(1):87–93. [21] Li X, Sundquist J, Calling S, Zoller B, Sundquist K. Neighborhood deprivation and risk of cervical cancer morbidity and mortality: a multilevel analysis from Sweden. Gyn Oncol 2012;127(2):283–9.
252
C. Waggaman et al. / Cancer Epidemiology 38 (2014) 248–252
[22] Krieger N, Chen JT, Waterman PD, Rehkopf DH, Subramanian SV. Painting a truer picture of US socioeconomic and racial/ethnic health inequalities: the public health disparities geocoding project. Am J Public Health 2005;95(2):312–23. [23] Hariri S, Unger ER, Sternberg M, Dunne EF, Swan D, Patel S, et al. Prevalence of genital human papillomavirus among females in the United States, the National Health and Nutrition Examination Survey, 2003–2006. J Infect Dis 2011;204:566–73. [24] Newmann SJ, Garner EO. Social inequities along the cervical cancer continuum: a structured review. Cancer Causes Control 2005;16:63–70. [25] Krieger N. Defining and investigating social disparities in cancer: critical issues. Cancer Causes Control 2005;16:5–14. [26] Adimora AA, Schoenbach VJ. Social context, sexual networks, and racial disparities in rates of sexually transmitted infections. J Infect Dis 2005;191: S115–22. [27] Laumann E, Youm Y. Race/ethnic group differences in the prevalence of sexually. transmitted diseases in the United States: a network explanation. Sex Transm Dis 1999;66:250–61.
[28] Mason SM, Messer LC, Laraia BA, Mendola P. Segregation and preterm birth: the effects of neighborhood racial composition in North Carolina. Health Place 2009;15:1–9. [29] Osypuk TL, Acevedo-Garcia D. Are racial disparities in preterm birth larger in hypersegregated areas? Am J Epidemiol 2008;167(11):1295–304. [30] Mair C, Diez Roux AV, Osypuk TL, Rapp SR, Seeman T, Watson KE. Is neighborhood racial/ethnic composition associated with depressive symptoms? The multi-ethnic study of atherosclerosis. Soc Sci Med 2010;71:541– 50. [31] Biello KB, Kershaw T, Nelson R, Hogben M, Ickovics J, Niccolai L. Racial residential segregation and rates of gonorrhea in the United States, 2003– 2007. Am J Public Health 2012;102(7):1370–7. [32] Kershaw KN, Diez Roux AV, Burgard SA, Lisabeth LD, Mujahid MS, Schulz AJ. Metropolitan-level racial residential segregation and black-white disparities in hypertension. Am J Epidemiol 2011;174(5):537–45. [33] Hayanga AJ, Zeliadt SB, Backhus LM. Residential segregation and lung cancer mortality in the United States. JAMA Surg 2013;148(1):37–42.