Identifying priority populations for lung cancer screening intervention using neighborhood-level factors and cancer registry data

Abstract

To evaluate the association of neighborhood level economic, environmental, and social indicators with lung cancer (LC) incidence and mortality.

Data for adult incident LC cases in Allegheny County, Pennsylvania, diagnosed between 2015-2019 were obtained from Pennsylvania cancer registry. Cases were summarized at census-tract level. Publicly available data on neighborhood deprivation index (NDI), built environment, and racial isolation at census-tracts were linked to cases. Poisson regression was used to compute relative risk (RR) for LC incidence and mortality, adjusting for covariates.

A total of 3256 LC cases were included in the analyses. About 68% were ≥65 years, 54% female, 14% Black or African American, and 63% deceased. Results of the multivariable model found that increasing quintiles (Q) of NDI were significantly associated with increasing risk of LC incidence and mortality. The RRs (95% confidence interval) of LC incidence for Q2, Q3, Q4 and Q5 were 1.36 (1.21-1.52), 1.55 (1.40-1.72), 1.68 (1.51-1.87), 2.08 (1.82-2.38), respectively, compared with Q1 (P trend <0.01). The corresponding RRs for LC mortality were 1.46 (1.27-1.68), 1.63 (1.42-1.88), 1.74 (1.51-2.01), 2.04 (2.02-2.88) (P trend <0.01). Targeted interventions for LC prevention and early detection in high NDI neighborhoods may be more effective to reduce LC health disparities.

INTRODUCTION

Lung cancer is the leading cause of cancer mortality in the United States (U.S.) affecting both men and women.¹ However, significant disparities exist by race and ethnicity, socioeconomic status, and geographic residence across the cancer continuum from prevention to survivorship.^2-4 While smoking remains a major risk factor for lung cancer with an attributable risk of 85%,⁵ environmental exposures combined with social determinants of health (SDoH) at individual and structural levels may contribute to the higher burden of lung cancer observed among health disparity populations.^4,6 Despite having lower smoking rates, Black men and women have higher incidence and mortality rates for lung cancer compared to White men and women, are more likely to present with advanced stage cancer at initial diagnosis, occur at younger age, and have lower survival rates after diagnosis.⁷ Similarly, incidence and mortality of lung cancer are also higher among people living in poverty, and those uninsured.⁸ The Healthy People 2020 defines SDoH as “conditions in the environments in which people are born, live, learn, work, play, worship, and age that affect a wide range of health, functioning, and quality-of-life outcomes and risks.” ⁹ The SDoH encompasses 5 key domains including the (1) economic stability which relates to issues such as poverty, unemployment, food insecurity and housing instability; (2) education domain which relates to literacy, educational attainment; (3) health and health care access which includes key issues such as access to healthcare, health literacy, (4) the neighborhood and build environment which includes issues such as access to healthy food, quality of housing, crimes rates, and environmental conditions; and (5) social and community domain which includes social cohesion, racial isolation, discrimination etc.

SDoH such as residence in an economically and socially segregated environment, poorly built environment, exposure to pollutants due to industry, increased psychosocial stress, poor access to health care and quality education have been associated with increased cancer risk.¹⁰ The SDoH at neighborhood level may contribute to lifestyle choices including smoking.^11,12 For example, studies have demonstrated that dynamics in neighborhood with violence, higher crime rates and poverty contribute to stress which has been utilized by tobacco companies to promote cigarette sales in these communities.^13-15 The tobacco companies have also used targeted campaigns to promote tobacco use among minority populations including the LGBTQ+ communities as well as youths of racial minorities. Vulnerable populations in these communities including children, pregnant people and elderly are often exposed to secondhand smoke resulting from household members smoking in homes and vehicles. Further, built environments such as crowding, lack of central air conditioning, or residential proximity to sources of environmental pollution may result in long-term exposure to environmental pollutants that increase risk of lung cancer in these population groups.¹⁶

Integrating SDoH into research on lung cancer etiology and assessing their interactions with environmental exposures can provide insights into disease pathways, identify vulnerable populations, and generate evidence that can impact translational research and policy. Long-term large cohort studies are needed to understand the etiologic pathway of how SDoH increased risk of lung; however, access to such cohorts with adequate data collection on SDoH elements is limited. Publicly available data sources on neighborhood level SDoH measures linked with population-based cancer registries can provide continuous, robust, and unbiased population estimates to generate hypothesis and identify at-risk populations. Our study used this approach to (1) to evaluate the relationship between neighborhood-level SDoH and risk of lung cancer incidence and mortality; and (2) to identify priority populations that are at higher risk of lung cancer. The University of Pittsburgh Institutional Review Board provided ethical approval for the study. A Waiver of Consent was obtained for the individuals included in the analyses.

METHODS.

Using the SDoH framework, we identified area-level measures that represent one or more of the five domains of SDoH. We also included potential risk factors associated with overall cancer risk or chronic diseases that share risk factors with lung cancer, to adjust for confounding. The description of risk factors, outcome measures and their data sources are discussed below.

Study design and population.

We used a cross-sectional design to evaluate the independent effects of socioeconomic (neighborhood deprivation), social and community-cultural (racial isolation), physical/built environment (air pollution, Radon, cancer toxic risk, green space), and behavioral (smoking), and incidence and mortality of lung cancer among people residing in Allegheny County, PA. Allegheny County is the most populous county in western PA, with a population of about 1.25 million persons, constituting about 10% of all population in the state of PA. The county is predominantly White (79%). It has the highest concentration of minority populations in western PA with 13.5% of Blacks or African American, 4.5% Asian, and 2.5% Hispanic or Latino. Approximately 11.3% of people are below poverty and 9.4% are uninsured. The UPMC Hillman Cancer Center (HCC), founded in 1985 and first designated as a Comprehensive Cancer Center by the National Cancer Institute (NCI) in 1990, is the only NCI-designated Comprehensive Cancer Center in western PA. Allegheny County, where the HCC is located, includes the City of Pittsburgh which was a major industrial center for coal and steel manufacturing since 1880’s.

Data sources for area level measures.

1. Neighborhood deprivation index (NDI). Using the approach developed by Diez Roux et al¹⁷ , we computed NDI for all census tracts in Allegheny County. The method uses factor analysis to identify key variables from 13 measures in the following dimensions of socioeconomic status: wealth and income, education, occupation, and housing conditions. The measures included were: (a) wealth and income: median household income (dollars), percent of households receiving dividends, interest, or rental income, percent of households receiving public assistance, median home value (dollars), and percent of families with incomes below the poverty level; (b) education: percent with a high school degree or higher, percent with a college degree or higher; (c) occupation: percent in a management, business, science, or arts occupation, percent unemployed; (c) housing conditions: percent of households that are female headed with any children under 18, percent of housing units that owner occupied, percent of household without a telephone, and percent of households without complete plumbing facilities. When applicable, some measures that were percentages, were reverse coded to reflect higher deprivation. NDI ranged between −3.6 to +2.8 with higher values indicating greater neighborhood deprivation. The NDI scores were also categorized into 5 levels using quintiles weighted by tract population. For our analyses, we used the NDI R Package to compute NDI using American Community Survey (ACS) 5-year estimates for years 2015-2019 corresponding to the study period.^18,19 Unlike the original study that used only 10 of the 13 indicator variables, our analyses, retained all the 13 measures in the computation of the NDI.

2. Racial isolation. A census geography (and its neighbors) that has nearly all of its population who identify with the specified race/ethnicity subgroup(s) (e.g., non-Hispanic or Latino, Black or African American alone) will have a racial isolation value close to 1.²⁰ In contrast, a census geography (and its neighbors) that has nearly none of its population who identify with the specified race/ethnicity subgroup(s) (e.g., not non-Hispanic or Latino, Black or African American alone) will have a racial isolation value close to 0.” Thus, in our study, a racial isolation value of 1 or close to 1, referred to neighborhoods with Black or African American population, and a value of racial isolation of 0 or close to 0 referred to neighborhoods that had 100% or close 100% White population.

3. Radon exposure. Radon measurements were taken from various locations within each census tract in Allegheny County from 1988 through 2020. On average, the number of measurements taken for each tract were 266 (standard deviation ± 250). Our examination of the data indicated that we had at least a minimum of 20 years of data for each of the census tracts. Since Radon levels do not vary significantly over time, we took an average of all measurements within a census tract to assign one value of Radon level per tract. We then dichotomized tracts into low and high based whether the average Radon measurement was below 4 pCi/L (picocuries per liter), the EPA recommended level, or equal to or greater than 4 pCi/L (picocuries per liter), respectively, for ease of interpretation.

4. The Air Toxics Screening Assessment (AirToxScreen). The U.S. Environmental Protection Agency (EPA) developed the AirToxScreen as a screening tool for state, local and tribal air agencies to help identify which pollutants, emission sources and places may pose possible risks to public health from air toxics.²¹ The AirToxScreen provides a snapshot of outdoor air quality with respect to emissions of air toxics which are air pollutants known to cause cancer amongst other health impacts. AirToxScreen estimates the cancer risks from inhalation of air toxics over many years. Thus, for a given tract, the AirToxScreen cancer risk estimates the number of people who may develop cancer for every 1 million people who are continuously exposed to a certain level of a pollutant over 70 years. Data for AirToxScreen were available for years 2017, 2018 and 2019. For our study, we used data from year 2017 since it was the midpoint between 2015 and 2019. We divided the AirToxScreen data for cancer risk into four categories using quartile cutoff. Each quartile represented the number of people who may develop cancer if 1 million people were continuously exposed to the pollutants over 70 years. Thus, the lowest quartile (Q1) represented lowest risk while highest quartile or quartile 4 represented highest risk of developing cancer. The quartile cutoffs were as follows: Q1= 30–39 people per million at risk of developing cancer; Q2= 40-49 people per million at risk for developing cancer; Q3=50–69 people per million at risk for developing cancer; and Q4 =70-200 people per million at risk for developing cancer.

5. Green space. Urban green space has been found to have a protective effect on various health outcomes including some cancers such as breast and lung cancer through mitigation of air pollution and increasing social cohesion in an urban landscape. However, the evidence is still scarce and inconsistent. Data on proportion of green space was obtained from US EPA Enviroatlas.²² We further categorized green space per capita (square meters) into four categories using quartile cutoffs. The categories are as follows: Q1 (17.4-269.0 per capita sq. meters); Q2 (269.1-542.2 per capita sq. meters); Q3 (542.3-1326.1 per capita sq. meters); and Q4 (1326.2-19,695 per capita sq. meters)

6. Particulate matter of size 2.5μm (PM_2.5). Daily tract-level estimates of PM_2.5 were downloaded for the period 2012 through 2019 from U.S. EPA .²³ According to the revised National Ambient Air Quality Standards for Particle Pollution, an area meets the standard if the three-year average of its annual average PM_2.5 concentration (at each monitoring site in the area) is less than or equal to 12.0 μg/m³.²⁴ We evaluated the PM_2.5 data in our study area computing three-year average prior to each year of diagnosis. For example, for all cases diagnosed in 2015 in a given census tract, a 3-year average PM_2.5 was computed using data from 2012-2014 for the census tract. We then used the 3-year average to identify tracts that exceed the standards. However, we found that measurements for all tracts averaged to below 12.0 μg/m³. These findings contrast with the NAAQS designation of non-attainment status for some parts of Allegheny County for all years of the study period (2012-2019). The variation from the NAAQS designation could be due to our approach of using tract-level estimates and not the actual estimates from the monitoring sites. Since there were no major differences between 2012 and 2015 years in annual average estimates of PM_2.5, for analyses, we used the continuous measurements from a midpoint period between 2015-2017 for all the cases diagnosed between 2015 and 2019.

7. Smoking status. Tract-level estimates of ever-smoked vs. never smoked were obtained from the Centers for Disease Control and Prevention (CDC)’s PLACES-Local Data for Better Health dataset for the study area.²⁵ This dataset uses a multilevel modeling of several indicator measures from the ACS and the Behavioral Risk Factor Surveillance System (BRFSS) to compute small area estimates for various geographical units such as census tracts, blocks and ZIP code. As a result, any indices such as the NDI that depend on the ACS data for computation correlate highly with the PLACES data.

Data sources for individual level measures.

• (8)Outcome measures. Incident lung cancer cases diagnosed between 2015 and 2019 in Allegheny County were identified from the Pennsylvania Cancer Registry. Data on demographic (age, race, sex, address, insurance payer), cancer diagnostics (date of diagnosis, stage, grade), and vital status (alive, dead) were obtained from the registry. The date of last follow up for vital status determination was December 31, 2019. Based on their residential address at the time of diagnosis, cases were geocoded at the census tract level. We then computed the total number cases per census tract stratified by age, sex, and race populations. Similarly, using data from vital status, counts of deceased were computed for each census tract stratified by age, sex, and race. Age was divided into two categories: lung cancer cases aged <65 years or aged 65 years or above at the time of diagnosis. Sex variable was divided into two categories: male or female. For race, we excluded lung cancer cases from racial groups other than White or Black or African American due to very small numbers in these groups [50 (1.4%)]. Age-, race- and sex- stratified population denominators for each tract were computed from the 5-year population estimates of ACS for year 2017 as it was a midpoint of our study period.

Data management and analyses.

Data manipulations were conducted as described above to summarize both independent variables and outcome measures at tract level and to create categorical variables for selected variables (NDI, Radon exposure). Descriptive statistics were computed to describe the lung cancer cases stratified by their vital status (dead vs. alive). Bivariate analyses compared the distribution of economic, environmental, and social indicators across vital status (dead vs. alive) using Chi-square test or Fisher’s exact test statistic for categorical variable or Kruskal-Wallis test for non-normal continuous variables. We used Spearman’s correlation to examine collinearity between the independent variables prior to inclusion in the model. Variables with correlation coefficient >0.7 were included in the model one at a time.

The generalized estimating equation (GEE) with Poisson distribution was used to evaluate associations between area-level indicators and lung cancer incidence using a data matrix of cases determined for each age strata (age levels), sex (female or male), race (White or Black), and census tract. We chose year 2017 as a midpoint to obtain the base population estimates by age, sex, and race, for each census tract in Allegheny County, from the ACS 5-year estimates and then multiplied it by 5 to account for the 5 year lung cancer data.²⁶ The GEE models were adjusted for age, sex, race, and population offset (log-transformed), and relative risk (RR)and 95% confidence intervals (CI) were extracted from the models. To adjust for potential correlation of patients clustered within a census tract, we included the census tract id as a repeated ID in the model. P-values were determined using the Wald test. Similarly, GEE with Poisson regression was used to compute unadjusted and adjusted RR and 95% CI for area-level indicators and lung cancer mortality. All statistical analyses were conducted in R and R Studio (https://cran.r-project.org/). The tidyverse package was used for the bulk of data management. The tidycensus package was used to import census data directly to our environment. Geepack was utilized when modeling generalized estimating equations. We used the compareGroups and table1 packages to run bivariate analyses and output tables of the analyses.

Table 1.

Demographic characteristics of lung cancer patients between 2015-2019 in Allegheny County, PA

	All N=3256	Dead N=2047	Alive N=1209	P-value*†
	N(%)	N(%)	N(%)
Age Group:				<0.0001
18-64 years	1033 (32)	582 (28)	451 (37)
> 64 years	2223 (68)	1465 (73)	758 (63)
Sex:				<0.0001
Female	1754 (54)	1012 (49)	742 (61)
Male	1502 (46)	1035 (51)	467 (39)
Race:				0.7064
White	2787 (86)	1748 (85)	1039 (86)
Black	469 (14)	299 (15)	170 (14)
Payer:				<0.001
Medicaid	371 (11)	229 (11)	142 (12)
Medicare	2086 (64)	1359 (66)	727 (60)
Private	799 (25)	459 (22)	340 (28)

^*P-value significant α=0.05

^†P-value associated with Chi-Square test statistic.

Based on the results of the model, we generated a bivariate heat map displaying selected SDoH (NDI) measure in contrast with spatial distribution of age-adjusted mortality rates (AMR) of lung cancer to identify neighborhoods with disproportionate burden of lung cancer. To allow for easy interpretation of the map, we grouped areas with lung cancer mortality rates between quintile 1 and quintile 3 cut offs as areas with low mortality (AMR range: 3.4 per 100,000 population – 43.9 per 100,000 population) and quintiles 4 and 5 as areas with high lung cancer mortality (AMR range: 44.0 per 100,000 population – 200.0 per 100,000 population). Similarly, we categorized areas with NDI between quintiles 1 to 3 as neighborhoods with low deprivation and those within quintiles 4 and 5 as neighborhood with higher deprivation. The bivariate map was thus divided into 4 levels: Level 1 (Low-Low) with lung cancer mortality rates in quintiles 1-3 – NDI quintiles 1-3; Level 2 (High-Low): Lung cancer mortality rates in quintiles 4,5 – NDI quintiles 1-3; (level 4) Low-High: Lung cancer mortality rates in quintiles 1-3 – NDI quintiles 4,5; (level 4) High-High: lung cancer mortality rates in quintiles 4,5 – NDI quintiles 4,5.

RESULTS

A total of 3256 people diagnosed with lung cancer between 2015 and 2019 in Allegheny County were identified and included in our analyses (Table 1). Almost 68% of the cases were 65 years or older, 54% were female, and 86% identified racially as White and 14% identified as Black or African American. About 25% had private insurance, 64% had Medicare and 11% had Medicaid as primary insurance provider. Of the total, 23% were diagnosed with localized or early-stage while 77% were diagnosed with late-stage or regional and distant lung cancer. About 63% had vital status as deceased when registry data was obtained. Comparison of demographic characteristics by vital status and stage of diagnosis revealed significant differences by age categories (p<0.01), sex (p<0.01) and type of insurance (p<0.01). Results of the bivariate analyses found significant differences in median levels of PM_2.5 for those deceased vs. those alive (p<0.01) (Table 2). Differences were also noted by air toxic cancer risk quartiles (p=0.0795) and population who currently smoke (0.0605) but were not statistically significant at α=0.05. Prior to model fitting, we assessed the results of the correlation between the independent variables and did not find high correlation between any factors except NDI and smoking status (supplementary figure 1)

Table 2.

Distribution of socioeconomic, socio-cultural and environmental risk factors by vital status.

	All N=3256	Dead N=2047	Alive N=1209	P- value*‡
Neighborhood Deprivation Index (Quintiles) (N,%)				0.1167
1-Least	460 (14)	272 (13)	188 (16)
2-BelowAvg	728 (22)	467 (23)	261 (22)
3-Avg	692 (21)	426 (21)	266 (22)
4-AboveAvg	683 (21)	423 (21)	260 (22)
5-Most	693 (21)	459 (22)	234 (19)
Residential Isolation Index (Median, IQR)	0.16 [0.10;0.34]	0.16 [0.10;0.35]	0.16 [0.10;0.34]	0.4746
Cancer risk quartiles (AirToxScreen)†				0.0795
Quartile 1 (30-39)	38 (1.2)	23 (1.1)	15 (1.2)
Quartile 2 (40-49)	1013 (31)	607 (30)	406 (34)
Quartile 3 (50-69)	1263 (39)	800 (39)	463 (38)
Quartile 4 (70-200)	942 (29)	617 (30)	325 (27)
Green Space Quartiles (in sq. meters)				0.2217
1 (17.4-269.0)	803 (25)	497 (24)	306 (25)
2 (269.0-542.2)	810 (25)	534 (26)	276 (23)
3 (542.2-1326.1)	813 (25)	500 (24)	313 (26)
4 (1326.1-19695.4)	830 (26)	516 (25)	314 (26)
Radon Levels (in picocuries per liter(pCi/L)				0.4937
Low (<4 pCi/L)	1516 (47)	963 (47)	553 (46)
High (≥4 pCi/L)	1740 (53)	1084 (53)	656 (54)
Particulate matter (2.5μg/m3), Median (IQR)	11.0 [10.7;11.2]	11.1 [10.8;11.2]	10.9 [10.5;11.1]	<0.0001
Population who currently smoke (%), Median (IQR)	16.4 [13.5;20.0]	16.4 [13.5;20.3]	16.4 [13.4;19.2]	0.0605

^*P-value significant α=0.05

^†Number of people per million at risk for developing cancer

^‡P-value associated with Chi-Square test statistic for categorical variables and Kruskal Wallis test statistic for non-normal continuous variables comparing median and interquartile range.

Results of the GEE model found that associations with cumulative lung cancer incidence were statistically significant for the following variables in the unadjusted models – older age, Black race, male sex, NDI quintiles, continuous PM_2.5 and residential isolation. In the multivariable model, older age, Black or African American race, male sex, and NDI quintiles persisted their association with cumulative lung cancer incidence (Table 3). The RRs (95% CIs) of lung cancer incidence was 5.13 (4.80-5.48) for persons of age 65 years or older compared with those of 64 years or younger, 1.33 (1.17-1.51) for Black or African American compared with White race, and 1.13 (1.06-1.20) for male sex as compared to female sex. The RRs (95% CIs) of lung cancer incidence for NDI quintiles 2, 3, 4, and 5 were 1.36 (1.21-1.52), 1.55 (1.40-1.72), 1.68 (1.51-1.87), and 2.08 (1.82-2.38), respectively, compared with quintile 1 (P trend <0.001). PM_2.5 and racial isolation index were no longer associated with risk of lung cancer incidence in the adjusted model, with the magnitude for relative risks closer to the null. Results of the GEE model evaluating relationship between area-level measures and cumulative lung cancer mortality were similar to cumulative incidence. Older age, Black race, male sex, NDI quintiles, PM_2.5 and racial isolation were significantly associated with risk for lung cancer mortality in unadjusted model. In the multivariable model, only age, race, sex and NDI were positively associated with cumulative lung cancer mortality. Compared to age 64 years or younger, the RR (95% CI) for those 65 years or older was 6.19 (5.66-6.77), compared to people who identified as White, the RR for lung cancer among people who identified as Black or African American was 1.33 (95% CI 1.11-1.59), and compared to females, males had a higher RR of 1.36 (95% CI, 1.25-1.48). For NDI, compared with the lowest quintile of NDI, the relative risk (95% CIs) of lung cancer mortality for NDI quintiles 2, 3, 4, and 5 were 1.46 (1.27-1.68), 1.63 (1.42-1.88), 1.74 (1.51-2.01), and 2.41 (2.02-2.88), respectively (Table 4) (P trend <0.001). The magnitude of relative risks for lung cancer mortality were slightly higher compared to that of cumulative lung cancer incidence. High Radon (≥ 4 pCi/L) was not found to be associated with lung cancer risk in both univariate and multivariate models.

Table 3.

Results of GEE model comparing relationship between area level socioeconomic, socio-cultural, and environmental factors and cumulative lung cancer incidence

Variables	Unadjusted		Adjusted‡
	RR	95% CI	RR	95% CI
Age (≥ 65 years)	5.79	(5.32, 6.30)**	5.13	(4.80, 5.48)**
Race (Black)	1.37	(1.17, 1.61)**	1.33	(1.17, 1.51)**
Sex (Male)	1.03	(0.90, 1.17)	1.13	(1.06, 1.20)**
Neighborhood Deprivation Index
Quintile 2	1.29	(1.03, 1.62)*	1.36	(1.21, 1.52)**
Quintile 3	1.37	(1.09, 1.71)**	1.55	(1.40, 1.72)**
Quintile 4	1.48	(1.20, 1.84)**	1.68	(1.51, 1.87)**
Quintile 5	1.80	(1.45, 2.23)**	2.08	(1.82, 2.38)**
Cancer risk quartiles (AirToxScreen)†
Quartile 2	0.58	(0.36, 0.94)*	0.86	(0.67, 1.11)
Quartile 3	0.74	(0.46, 1.20)	0.90	(0.69, 1.18)
Quartile 4	0.75	(0.47, 1.21)	0.83	(0.63, 1.10)
Greenspace
Quartile 2	1.13	(0.94, 1.36)	0.97	(0.87, 1.07)
Quartile 3	1.04	(0.87, 1.26)	0.94	(0.84, 1.06)
Quartile 4	0.86	(0.71, 1.03)	0.87	(0.77, 0.99)*
Radon (High≥ 4 pCi/L)	0.92	(0.81, 1.05)	1.06	(0.98, 1.15)
PM2.5 (μg/m3)	1.70	(1.29, 2.24)**	0.95	(0.77, 1.18)
Racial Isolation Index	1.37	(1.58, 2.90)**	0.94	(0.70, 1.25)

^†Number of people per million at risk for developing cancer

^‡All variables were included in the adjusted models

^*P-value <0.05, associated with Wald statistic

^**P-value <0.001, associated with Wald statistic

RR = Relative Risk; CI= Confidence Intervals

Table 4.

Results of GEE model comparing relationship between area level socioeconomic, socio-cultural, and environmental factors and lung cancer mortality

Variables	Unadjusted		Adjusted‡
	RR	95% CI	RR	95% CI
Age (≥ 65 years)	5.70	(5.17, 6.29)**	6.19	(5.66, 6.77)**
Race (Black)	1.45	(1.23, 1.72)**	1.33	(1.11, 1.59)**
Sex (Male)	1.17	(1.02, 1.34)*	1.36	(1.25, 1.48)**
Neighborhood Deprivation Index
Quintile 2	1.38	(1.09, 1.74)**	1.46	(1.27, 1.68)**
Quintile 3	1.44	(1.14, 1.82)**	1.63	(1.42, 1.88)**
Quintile 4	1.58	(1.26, 1.98)**	1.74	(1.51, 2.01)**
Quintile 5	2.03	(1.62, 2.55)**	2.41	(2.02, 2.88)**
Cancer risk quartiles (AirToxScreen)†
Quartile 2	0.64	(0.38, 1.09)	0.83	(0.58, 1.18)
Quartile 3	0.83	(0.49, 1.40)	0.92	(0.64, 1.33)
Quartile 4	0.87	(0.51, 1.46)	0.87	(0.59, 1.27)
Greenspace
Quartile 2	1.13	(0.94, 1.37)	0.99	(0.86, 1.13)
Quartile 3	1.04	(0.86, 1.26)	0.93	(0.81, 1.08)
Quartile 4	0.82	(0.67, 0.99)*	0.84	(0.72, 0.99)*
Radon (High≥ 4 pCi/L)	0.90	(0.79, 1.03)	1.09	(0.98, 1.21)
PM2.5 (μg/m3)	1.85	(1.38, 2.47)**	0.94	(0.72, 1.22)
Racial Isolation Index	2.36	(1.71, 3.24)**	0.90	(0.61, 1.33)

^†Number of people per million at risk for developing cancer

^‡All variables were included in the adjusted models

^*P-value <0.05, associated with Wald statistic

^**P-value <0.001, associated with Wald statistic

RR = Relative Risk; CI= Confidence Intervals

Visual inspections of bivariate heat map (Figure 1) displaying areas with highest lung cancer age-adjusted mortality rates (quintile 4 and 5, rates greater than or equal to 43.9 per 100,000 population) contrasted with highest neighborhood deprivation (areas classified under quintile 4 and quintile 5), highlight neighborhoods along the banks of the two major rivers in Allegheny County. These neighborhoods in central and eastern parts of the county have historically been the sites of heavy industries (coal, steel, glass).

Figure 1.

Bivariate map of neighborhoods with highest annual age-adjusted lung cancer mortality per 100,000 population, categorized using quintiles cutoffs at 4 and 5 (between 44.0 to 200.0 per 100,000 population) contrasted with neighborhoods with quintiles 4 and 5 of neighborhood deprivation index. Categories are defined as follows: Low-Low: Lung cancer mortality rates in quintiles 1-3 (between 3.6 to 43.8 per 100,000 population) – NDI quintiles 1-3; High-Low: Lung cancer mortality rates in quintiles 4,5 (between 44.0 to 200.0 per 100,000 population) – NDI quintiles 1-3; Low-High: Lung cancer mortality rates in quintiles 1-3 (between 3.6 to 43.8 per 100,000 population) – NDI quintiles 4,5; High-High: Lung cancer mortality rates in quintiles 4,5 (between 44.0 to 200.0 per 100,000 population) – NDI quintiles 4,5

DISCUSSION

Our study found a significant increased risk of lung cancer incidence and mortality, associated with increasing quintiles of neighborhood deprivation, with results indicating a dose-response relationship. This association remained unchanged after adjusting for potential confounding factors such as age, race, and sex, racial isolation, Radon levels, air quality and green space.

Our findings complement and add to the growing body of evidence linking neighborhood deprivation with adverse cancer outcomes. Our results are consistent with results of a recent study conducted in Sweden²⁷ which found increased odds (odds ratio: 1.27 and 1.32) of both incidence and mortality rates of lung cancer, respectively, among individuals living in most deprived neighborhoods as compared to those living in affluent neighborhoods. The study used data from for the entire Swedish population aged over 50 (~3.2 million individuals) between 2000 and 2010 and adjusted socio-demographic factors, comorbidities and urban/rural status. The NDI used in the study was constructed from variables including education, income, unemployment and welfare assistance, which align with the indicators used to construct NDI in our study. Several recent studies in the U.S. have also reported strong association between neighborhood level deprivation and lung cancer incidence, with hazard ratios of 1.29 (95%CI 1.07-1.55)²⁸, and 1.30 (95 % CI: 1.04, 1.63, p = 0.023)²⁹.

The causal pathways between neighborhood deprivation and lung cancer are not fully understood. Several possible explanations can be made based on previous literature between neighborhood socioeconomic levels and other cancers that share similar risk factors with lung cancer.³⁰ A few studies have suggested that residing in relatively deprived neighborhoods may result in psychological stress which lies in the causal pathway of chronic diseases but may also lead adoption of harmful lifestyle choices such as smoking, lack of physical activity and alcohol consumption.^13-15 The NDI used in this study is a cumulative index created from 13 sociodemographic indicators including proportion of people uninsured in each neighborhood as well educational attainment of the population within the neighborhood. Research has found that neighborhood deprivation may be associated with increased smoking rates and higher levels of exposure to environmental pollutants. In our study, data on smoking (ever smoked vs. never) was highly correlated with NDI (r²= 0.92). However, since the smoking data was available for 2020 onwards, we chose to include NDI instead of smoking status, in our analyses. The high correlation does suggest a possible mediating effect of neighborhoods with higher economic deprivation to provide socio-cultural environments that may lead to adoption of smoking habits. Investigators have also found that neighborhood deprivation is associated with poor survival outcomes;³¹ thus, supporting our findings of higher NDI associated with increased lung cancer mortality. A cohort study of more than 95,000 patients with non-metastatic cancers found a hazard ratio of 1.14 (95% CI 1.09-1.20) for lung cancer associated with most deprived neighborhoods (area deprivation quintile 5) compared to affluent neighborhoods.³²

The indicators included in construction of NDI mainly relate to the economic domain of SDoH and to some extend to the educational domain. Our study also evaluated risk factors such as air quality measures and racial isolation, related including built environment, and socio-cultural domains of SDoH. However, we did not find any significant relationship between these factors and lung cancer incidence or mortality, in the adjusted models. Other studies investigating neighborhood ambient PM_2.5 along with neighborhood deprivation also reported similar findings where ambient PM_2.5 concentrations were not associated with increased risk of lung cancer.^29,33 Epidemiological studies measuring biomarkers of exposure to ambient air pollutants at individual levels are needed to characterize the relationship between ambient air, NDI and lung cancer.

The strongest evidence linking Radon with lung cancer incidence come from occupational studies of miners as well as Radon exposure among smokers. The lack of association between Radon exposure and lung cancer risk in our study may be accounted by misclassification of individual-level long-term Radon exposure as compared to area-level Radon exposure. A recent review of studies conducted in the US and Europe on domestic Radon exposure concluded that there is considerable uncertainty in the postulated risk estimates for residential Radon because of misclassification of smokers, Radon dosimetry inaccuracies, differences in histopathological types of lung cancer and co-exposure of Radon with smoking and other indoor air pollutants.³⁴ Environmental epidemiological studies with robust personal monitoring and biomarkers of exposures, are needed to assess impact of long-term residential Radon exposures on the risk of lung cancer.

In contrast to our study, other studies have found association between racial isolation or segregation and advanced state lung cancer diagnosis and mortality among Black patients, independent of neighborhood deprivation.^35,36 In our study higher index for racial isolation corresponding with higher proportion of Black populations in a given census tract, did not correlate highly with increasing levels of NDI (r²=0.42). The median index for racial isolation in our area was 0.16 (interquartile range: 0.10-0.34). Studies that found positive association between racial isolation and lung cancer had a mean Index of Dissimilarity at 0.62 (±0.09) which is much higher than that observed in our study area.³⁶ More studies are needed at smaller neighborhood level with larger population diversity to examine impact of racial isolation on the risk of lung cancer.

We created a bivariate heat map of neighborhoods that had higher deprivation as well as higher lung cancer mortality to identify populations disproportionately burdened in Allegheny County. These findings demonstrate the feasibility of using registry data and publicly sourced data from census.gov, U.S. EPA and CDC to identify at-risk populations. The same approach can be used to identify areas in our catchment area of 29 counties to identify priority populations at higher risk for lung cancer and direct resources for prevention, early detection and clinical trial participation. For example, the map will be used by leaders in the Office of Community Outreach and Engagement (COE) in the UPMC Hillman Cancer Center, to focus outreach and screening education and navigation efforts in communities at higher cancer risk, but lower screening rates. The heat maps can also inform recruitment for clinical trials and observational studies focused on lung cancer, increasing representation of those at highest risk into cancer research. Federally qualified cancer institutes, non-research cancer clinics and community organizations with limited resources and/or expertise in conducting large-scale epidemiological studies and those without access to local cancer registry data, can utilize the heat maps and associated data to identify populations at-risk for lung cancer within their catchment areas and tailor cancer prevention efforts, accordingly.

As with any cross-sectional design study, our findings suggest an association between neighborhood deprivation and lung cancer incidence and mortality but do not provide evidence to support causality. A major limitation is we did not adjust for individual-level smoking status. It is possible that the effect of neighborhood deprivation on lung cancer incidence and mortality may be explained by individual smoking status.³⁷ However, findings from a recent cohort investigation linking never-smoking Black women residing in disadvantaged neighborhoods with higher risk of developing lung cancer, supports independent association between NDI and lung cancer. Another limitation of our study was lack of information on residential history of the incident lung cancer cases. Due the long period of latency prior to diagnosis of lung cancer, there is a potential for misclassification of neighborhood level factors where the cases may have had varying residential history prior to their lung cancer diagnosis. The within-county and outside-county residential mobility of populations in Allegheny County is 8.9% and 1.5% respectively. These numbers are comparable with the national average of 7.8% and 2.8% respectively.²⁶ The numbers suggest that about 90% of the lung cancer cases may have lived in the same neighborhoods. However, without collecting additional data on residential history, no further conclusion can be made. The geographical area of our study limited to single county data which may limit the generalizability of the results to populations that have similar demographic and environmental variability. There are certainly some limitations of using large datasets such as cancer registries for epidemiological investigations; however, the strengths of having large sample size, diverse populations, longer periods of data and relatively easier access than cohort studies, far outweigh the limitations.

Our study examined association between area-level socioeconomic, environmental, and social risk factors and lung cancer incidence and mortality and found that in addition to older age and Black race, neighborhood deprivation, a key indicator of socioeconomic status, was significantly associated with increased risk of lung cancer incidence and mortality. The associations did not change after adjusting for known environmental risk factors such as PM_2.5 and Radon exposure, suggesting that neighborhood deprivation is independently associated with increased risk of lung cancer. Our study findings support the need for policies for ongoing investments in disadvantaged neighborhoods and low-income households, as well as tailored interventions for lung cancer prevention and early detection. With the growing body of evidence linking neighborhood deprivation with adverse cancer outcomes, further research is warranted in identifying the causal pathways and how individual-level behaviors such as tobacco use, diet and nutrition, and physical activity may interact with area-level factors to increase health risks. Further, it is critical that this data be accessible to all organizations seeking to reduce the cancer burden. Our study demonstrates the feasibility of using large databases such as cancer registries and publicly available data on SDoH to identify cancer-related health disparities in catchment populations and direct resources to improve outcomes.