ABSTRACT
Introduction
Counties in persistent poverty (PP) have higher cancer‐related mortality rates. Intersectionality with race and rurality is present; individuals identifying as Black residing in rural PP counties have the highest cancer‐related mortality rates. However, heterogeneity exists throughout counties (e.g., demographics). To date, few studies evaluated the effect of small‐area PP residence on individual‐level survival and intersectionality with race and rurality.
Methods
This retrospective cohort utilized Alabama Statewide Cancer Registry data of adults with breast, prostate, colorectal, or lung cancer. The exposure was NCI‐designated 2010 census tract‐level PP (tracts with ≥ 20% population below poverty line for ≥ 30 years) at diagnosis and the outcome was overall survival from date of diagnosis to death or end of follow‐up. Associations were estimated via sequential Cox proportional hazards models first with PP as the only predictor then adjusting for demographics (age, race, sex [colorectal, lung]), demographics plus clinical characteristics (stage, subtype [breast], and first course treatment), and finally demographics, clinical characteristics, and social determinants (rurality, insurance status). We also performed a decomposition analysis of the association between PP and survival through race and rurality.
Results
In the fully adjusted model, living in a PP‐tract was associated with higher risk of death among those with breast (hazard ratio [95% CI]: 1.18 [1.09–1.27]), prostate (1.21 [1.11–1.31]), and lung cancer (1.09 [1.05–1.14]) compared to living in a non‐PP‐tract. Race and rurality were independently associated with higher risk of death across cancers. Individuals identifying as Black residing in a rural PP‐tract had the highest magnitude of risk for all cancers.
Conclusions
Living in a PP‐tract is associated with higher risk of death among patients with breast, prostate, and lung cancer in Alabama. This association is complicated by race and rurality. Multilevel interventions tailored to local context may help overcome adverse effects of social characteristics on cancer outcomes.
Keywords: cancer, intersectionality, persistent poverty, social determinants of health, survival
1. Introduction
Cancer is the second leading cause of death in the United States (U.S.). The leading causes of cancer‐related mortality are lung, breast, prostate, and colorectal [1]. Although mortality from these cancers has declined over several decades [1], certain population subgroups remain at higher risk of cancer‐related mortality [2]. One such subgroup includes individuals experiencing poverty. However, “current poverty” (i.e., ≥ 20% of population living in poverty) and “persistent poverty” (PP) (i.e., ≥ 20% of population living in poverty for ≥ 30 years) may influence cancer outcomes differently [3]. Most prior studies examined current rather than PP. Of limited existing data, across the U.S., cancer mortality rates were higher in PP than non‐PP counties [4]. Examining current poverty compounded by PP, mortality rates were higher compared to current, non‐PP counties, suggesting that individuals in counties experiencing PP have worse outcomes than current poverty only [3].
Compounding mortality disparities between PP and non‐PP counties is rurality and race. Studies in colorectal cancer (CRC) revealed intersectionality between race and rurality with each other and PP status [5, 6]. Analysis of mortality disparities related to PP at the intersection of race and rurality showed Black residents in rural PP counties had the highest mortality rates relative to other race groups [4]. However, county‐level measures ignore heterogeneity within counties. Examining census tract‐level PP, which more closely approximates neighborhood, 5‐year survival rates from cancer were higher in non‐PP, higher socioeconomic status, urban, and primarily non‐Hispanic White census tracts [7].
Prior studies provided insight into the role of PP in cancer‐related mortality across the U.S., but local context is not well‐understood, limiting ability to mitigate disparities between PP and non‐PP areas. Moreover, many studies examined county‐level data, which do not account for heterogeneity in demographics and cancer outcomes within counties. Additionally, previous studies focused on rates without adjusting for covariates and there are few studies examining intersectionality of rurality, race, and PP. Finally, certain states may be more subject to potential intersectionality due to population demographics and/or socioeconomics. For example, Alabama has a large Black population which faces significant health disparities in cancer outcomes, a large rural population, and 10%–15% of the state living in PP [8, 9, 10, 11, 12]. Therefore, the goal of this study was to evaluate the multivariable‐adjusted association between census tract‐level PP and overall survival (OS) among adults with cancer in Alabama, accounting for race and rurality.
2. Materials & Methods
2.1. Study Population
This retrospective, population‐based cohort utilized data from the Alabama Statewide Cancer Registry (ASCR) [13]. ASCR is a statewide, population‐based cancer registry. We included all incident cases of breast, prostate, CRC, and lung cancers among adults (≥ 18 years) in Alabama diagnosed between January 1, 2010, and December 31, 2019. We included only cases where the cancer was the only primary or first of multiple primary cancers diagnosed in the participant's lifetime including in situ cases. This study was approved by the Institutional Review Boards of the University of Alabama at Birmingham and the Alabama Department of Public Health (ADPH). All procedures were conducted in accordance with ethical standards and principles of the Declaration of Helsinki. Data supporting these findings are available upon request to ASCR at ADPH. These data were used under license for this study, as restrictions apply to their availability.
2.2. Exposure
The primary exposure was living in a PP census tract at diagnosis collected from ASCR. The National Cancer Institute (NCI) published a list of PP census tracts across the U.S., updated in December 2021 [14]. NCI based the PP definition as census tracts with ≥ 20% of the population in poverty at the 1990 and 2000 decennial censuses and in the 2007–2011 and 2015–2019 American Community Survey 5‐year estimates. Census tract boundaries are redefined at each decennial census. Thus, poverty estimates from 1990 and 2000 must be placed on a consistent geography for the four time points to describe the same area. NCI used Integrated Public Use Microdata Series National Historic Geographic Information System (IPUMS NHGIS) standardized time series data, which assigns the earlier decennial counts to 2010 census tract boundaries. All four periods therefore refer to 2010 tract definitions, which is the geography to which our sample was geocoded. We set Alabama census tracts from the list as PP and all others as non‐PP.
2.3. Outcome
The primary outcome was OS. ASCR collected death data using passive surveillance via yearly linkage to the National Death Index and quarterly linkage to the Alabama State Vital Statistics. Thus, there was a lag in collection. We set the end of follow‐up to December 31, 2021, and calculated follow‐up time as time in months from date of diagnosis to date of death, or end of follow‐up. Cause of death was not available; thus, the outcome was based on death from any cause.
2.4. Covariates
We included several baseline covariates: age at diagnosis, self‐reported sex (for colorectal, lung cancer) and race, SEER (Surveillance, Epidemiology, and End Results) stage at diagnosis, rural–urban status, and subtype (for breast cancer). We centered age at diagnosis by subtracting the mean. Race was categorized as White, Black, or other (regardless of ethnicity). We categorized SEER stage as early (i.e., in situ or localized) or late (i.e., regional or distant involvement). We linked census tract at diagnosis to 2010 Rural–Urban Commuting Area (RUCA) codes, and categorized rural–urban status via Categorization D from the University of Washington Rural Health Research Center, leading to a dichotomous variable of rural or urban [15, 16, 17]. We defined breast cancer subtype based on hormone receptor (HR) status (estrogen or progesterone), and human epidermal growth factor‐2 receptor (HER2) status. We considered HR as positive if either estrogen or progesterone receptor was positive. We defined subtypes as HR+/HER2+, HR+/HER2‐, HR‐/HER2+, or triple‐negative breast cancer (TNBC). We selected covariates a priori based on literature [1, 18, 19, 20]. We also examined first course treatment status. This variable is limited by what is reported from facilities to ASCR. Thus, some patients may have received treatments not reported to ASCR. We classified chemotherapy as having received or not having received chemotherapy regardless of the drug or reason for lack of receipt. We followed the same pattern for radiation receipt. We created a combined treatment variable with four levels: chemotherapy only, radiation only, both chemotherapy and radiation, and neither. Treatment with hormone modulating therapies were not available. We considered primary payer at diagnosis as an indicator of patient insurance status and classified at the following levels: Medicare, Medicaid, non‐Medicare/non‐Medicaid public insurance (e.g., TRICARE), private insurance, insured, not otherwise specified (NOS), and uninsured.
2.5. Effect Modifiers
We further examined race and rural–urban status as effect modifiers via interaction terms with the exposure.
2.6. Statistical Analysis
We performed a complete case analysis of data not missing covariates. We stratified all analyses by cancer type. We examined bivariate descriptive statistics for outcomes and covariates by PP status using Mann–Whitney U and χ 2 tests for continuous and categorical variables, respectively. We examined differences in breast cancer subtype by PP status stratified by race using χ 2 tests.
We examined the association between PP status and OS using Cox proportional hazards regression models adjusted for age at diagnosis, race, sex (where applicable), stage at diagnosis, subtype (breast cancer), and rural–urban status. We performed four models sequentially adjusting for covariates using complete case analysis. First, to determine the baseline, unadjusted association between PP and OS, we performed an unadjusted model with PP as the only predictor. Next, to evaluate the effect of demographics on the association between PP and OS, we performed a demographics‐only adjusted model with age, race, and sex (where applicable). Next, to understand the additional effect of downstream clinical characteristics on the association between PP and OS, we performed a demographics and clinical characteristics‐adjusted model with demographics plus subtype (breast cancer), stage at diagnosis, and first course treatment. Finally, to evaluate the added effect of upstream social determinants, we performed a fully‐adjusted model with all previous variables plus rural–urban status and insurance status. For each of the above‐mentioned models, we reported Akaike Information Criteria (AIC) values for each model to determine the improvement of model fit by addition of each set of covariates. Attenuation across these models is descriptive and the final model should not be interpreted as the total effect. We conducted a sensitivity analysis excluding subtype from the modeling in the breast cancer sample due to a large amount of missingness in the subtype variable. We also conducted a sensitivity analysis to examine receipt of surgery as a covariate (ASCR only collects date of surgery if it is reported to the Registry; thus, we assumed that if a date of surgery was present, surgery occurred and if not, surgery did not occur; this may misclassify some participants). We did not report AIC for the sensitivity analyses.
Additionally, to evaluate the intersection of PP with both race and rurality, we evaluated race and rural–urban status as effect modifiers via two‐way interactions with PP status in the fully‐adjusted model. Finally, to provide practical application, we estimated the linear combinations for an individual 65 years old across each unique combination of PP status, race, and rural–urban status (e.g., White, PP, urban, and so on) with all covariates held at reference values.
For all Cox models, we included a squared term for centered age at diagnosis due to observed violation of proportional hazards for the linear term and to account for potential non‐linear effects. All models also incorporated cluster‐robust standard errors to account for census tract clustering.
Finally, to determine how racial and rurality distributions influence the relationship between PP and OS, we employed a regression‐based mediation method on the hazard scale implemented in the CMAVerse package in R [21]. This method produces model‐based standardized hazard ratios (HRs) comprising the total association of PP with OS, the portion of the PP HR not accounted for by race or rurality, the portion accounted for by race or rurality, and the proportion of the total association decomposed, calculated as the ratio of log‐HRs. This results in a decomposition similar to the Oaxaca‐Blinder decomposition, applied here on the hazard scale [22]. We included covariates from the fully‐adjusted models detailed above. We focused on race and rurality as our main explanatory variables rather than any of the other covariates included in our models because they have a well‐documented association with PP, they have a well‐documented association with our outcomes, and because they all represent social factors.
To simplify this analysis, we excluded those not identifying in White or Black racial categories; given that there are very few patients identifying as “Other” in the sample, convergence of mediation models was challenging. We did not examine decomposition by race among patients with CRC because the association between PP and survival was not statistically significant in Cox models. Additionally, we only examined rurality as a decomposer in prostate cancer due to its lack of statistical significance in breast and lung cancer Cox modeling. We calculated 95% confidence intervals using 1000 bootstrap resamples. We conducted decomposition analyses using RStudio Version 2025.09.1 [21]. We performed all other analyses using SAS Version 9.4 (SAS Institute Inc., Cary, NC, USA). Statistical significance was set at α = 0.05.
3. Results
A flow diagram of how the final analytic sample was obtained for each cancer type is presented in Figure 1. In breast cancer, those in a PP tract were more likely than those in a non‐PP tract to identify as Black (64.2% vs. 19.6%, p < 0.001), have HR‐/HER2+ or TNBC (HR‐/HER2+: 5.8% vs. 5.1%; p < 0.001; TNBC: 18.4% vs. 14.3%; p < 0.001), be diagnosed at late stage (42.1% vs. 34.2%; p < 0.001), receive chemotherapy‐only (27.5% vs. 24.2%; p < 0.001), live in a rural census tract (31.0% vs. 21.2%; p < 0.001), be on Medicare (45.7% vs. 41.6%; p < 0.001), Medicaid (12.5% vs. 5.5%; p < 0.001), or uninsured (3.3% vs. 2.1%; p < 0.001), die (29.4% vs. 21.7%; p < 0.001), and have shorter follow‐up time (63 vs. 66 months; p < 0.001; Table 1). Among Black women, there was a higher rate of HR+/HER2‐ breast cancer among those in a PP census tract compared to those in a non‐PP census tract (Table S1); among White women, no difference was observed in breast cancer subtype by PP status. In prostate cancer, results were similar for race, rurality, Medicaid, uninsured, death, and follow‐up time. Additionally, those in PP census tracts were more likely to be on radiation‐only treatment (29.8% vs. 25.8%; p < 0.001) and have public insurance (8.1% vs. 5.2%; p < 0.001; Table 1).
FIGURE 1.
Flow diagram of participants to arrive at final analytic sample. This figure shows the flow of participants and exclusions made to arrive at the final analytic sample.
TABLE 1.
Participant characteristics by persistent poverty (PP) status by cancer type. a
| Variable | Breast (n = 26,434) | Prostate (n = 27,623) | Colorectal (n = 19,478) | Lung (n = 27,924) | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Total | PP | Non‐PP | p | Total | PP | Non‐PP | p | Total | PP | Non‐PP | p | Total | PP | Non‐PP | p | |
| Median age at diagnosis, years, (IQR) | 62 (52, 71) | 62 (53, 71) | 62 (52, 71) | 0.20 | 65 (59, 71) | 65 (59, 70) | 65 (59, 71) | < 0.001 | 65 (55, 74) | 65 (56, 74) | 65 (55, 74) | 0.62 | 67 (60, 74) | 66 (59, 73) | 68 (60, 74) | < 0.001 |
| Race, No. (%) | < 0.001 | < 0.001 | < 0.001 | < 0.001 | ||||||||||||
| White | 19,466 (73.6) | 1180 (35.2) | 18,286 (79.2) | 18,450 (66.8) | 1223 (29.5) | 17,227 (73.4) | 14,407 (74.0) | 1022 (35.2) | 13,385 (80.7) | 22,369 (80.1) | 1618 (42.4) | 20,751 (86.1) | ||||
| Black | 6683 (25.3) | 2154 (64.2) | 4529 (19.6) | 8950 (32.4) | 2897 (69.9) | 6053 (25.8) | 4862 (25.0) | 1853 (63.9) | 3009 (18.2) | 5366 (19.2) | 2186 (57.3) | 3180 (13.2) | ||||
| Other | 285 (1.1) | 19 (0.6) | 266 (1.2) | 223 (0.8) | 22 (0.5) | 201 (0.9) | 209 (1.1) | 26 (0.9) | 183 (1.1) | 189 (0.7) | 12 (0.3) | 177 (0.7) | ||||
| Sex, male, No. (%) | NA | NA | NA | NA | NA | NA | 10,324 (53.0) | 1458 (50.3) | 8866 (53.5) | 0.001 | 12,011 (43.0) | 1571 (41.2) | 10,440 (43.3) | 0.01 | ||
| Subtype, No. (%) b | < 0.001 | |||||||||||||||
| HR+/HER2+ | 3225 (12.2) | 394 (11.8) | 2831 (12.3) | NA | NA | NA | NA | NA | NA | NA | NA | NA | ||||
| HR+/HER2− | 17,942 (67.9) | 2150 (64.1) | 15,792 (68.4) | NA | NA | NA | NA | NA | NA | NA | NA | NA | ||||
| HR−/HER2+ | 1359 (5.1) | 193 (5.8) | 1166 (5.1) | NA | NA | NA | NA | NA | NA | NA | NA | NA | ||||
| TNBC | 3908 (14.8) | 616 (18.4) | 3292 (14.3) | NA | NA | NA | NA | NA | NA | NA | NA | NA | ||||
| Stage, No. (%) | < 0.001 | 0.84 | 0.24 | 0.06 | ||||||||||||
| Early | 17,122 (64.8) | 1940 (57.9) | 15,182 (65.8) | 22,879 (82.8) | 3426 (82.7) | 19,453 (82.9) | 8217 (42.2) | 1195 (41.2) | 7022 (42.4) | 6170 (22.1) | 798 (20.9) | 5372 (22.3) | ||||
| Late | 9312 (35.2) | 1413 (42.1) | 7899 (34.2) | 4744 (17.2) | 716 (17.3) | 4028 (17.2) | 11,261 (57.8) | 1706 (58.8) | 9555 (57.6) | 21,754 (77.9) | 3018 (79.1) | 18,736 (77.7) | ||||
| First course treatment, No. (%) | < 0.001 | < 0.001 | 0.07 | < 0.001 | ||||||||||||
| Chemotherapy only | 5592 (24.6) | 801 (27.5) | 4791 (24.2) | 243 (1.1) | 39 (1.2) | 204 (1.1) | 4486 (27.1) | 653 (26.6) | 3833 (27.2) | 4723 (19.1) | 566 (16.8) | 4157 (19.5) | ||||
| Radiation only | 5288 (23.3) | 583 (20.0) | 4705 (23.8) | 5892 (26.4) | 991 (29.8) | 4901 (25.8) | 208 (1.3) | 40 (1.6) | 168 (1.2) | 3313 (13.4) | 484 (14.4) | 2829 (13.3) | ||||
| Chemotherapy and radiation | 3734 (16.5) | 496 (17.1) | 3238 (16.4) | 143 (0.6) | 19 (0.6) | 124 (0.7) | 1918 (11.6) | 257 (10.5) | 1661 (11.8) | 6404 (25.9) | 854 (25.3) | 5550 (26.0) | ||||
| Neither chemotherapy nor radiation | 8091 (35.6) | 1029 (35.4) | 7062 (35.7) | 16,019 (71.8) | 2274 (68.4) | 13,745 (72.4) | 9955 (60.1) | 1502 (61.3) | 8453 (59.9) | 10,270 (41.6) | 1468 (43.5) | 8802 (41.3) | ||||
| Rural, No. (%) | 5929 (22.4) | 1039 (31.0) | 4890 (21.2) | < 0.001 | 6647 (24.1) | 1238 (29.9) | 5409 (23.0) | < 0.001 | 5285 (27.1) | 937 (32.3) | 4348 (26.2) | < 0.001 | 7845 (28.1) | 1176 (30.8) | 6669 (27.7) | < 0.001 |
| Insurance status, No. (%) c | < 0.001 | < 0.001 | < 0.001 | < 0.001 | ||||||||||||
| Medicare | 10,976 (42.1) | 1503 (45.7) | 9473 (41.6) | 11,648 (47.4) | 1785 (47.9) | 9863 (47.3) | 9693 (51.5) | 1525 (54.4) | 8168 (50.9) | 16,950 (62.0) | 2254 (60.5) | 14,696 (62.2) | ||||
| Medicaid | 1656 (6.4) | 412 (12.5) | 1244 (5.5) | 528 (2.2) | 188 (5.1) | 340 (1.6) | 849 (4.5) | 239 (8.5) | 610 (3.8) | 1728 (6.3) | 398 (10.7) | 1330 (5.6) | ||||
| Public (non‐Medicare/non‐Medicaid) | 581 (2.2) | 51 (1.6) | 530 (2.3) | 1397 (5.7) | 303 (8.1) | 1094 (5.2) | 471 (2.5) | 64 (2.3) | 407 (2.5) | 883 (3.2) | 137 (3.7) | 746 (3.2) | ||||
| Private | 9601 (36.8) | 932 (28.3) | 8669 (38.1) | 8356 (34.0) | 987 (26.5) | 7369 (35.3) | 5282 (28.0) | 603 (21.5) | 4679 (29.2) | 5061 (18.5) | 515 (13.8) | 4546 (19.3) | ||||
| Insured, NOS | 2671 (10.3) | 284 (8.6) | 2387 (10.5) | 2135 (8.7) | 300 (8.1) | 1835 (8.8) | 1463 (7.8) | 169 (6.0) | 1294 (8.1) | 1172 (4.3) | 166 (4.5) | 1006 (4.3) | ||||
| Uninsured | 573 (2.2) | 107 (3.3) | 466 (2.1) | 524 (2.1) | 162 (4.4) | 362 (1.7) | 1081 (5.7) | 201 (7.2) | 880 (5.5) | 1544 (5.7) | 256 (6.9) | 1288 (5.5) | ||||
| Died, No. (%) | 5986 (22.7) | 985 (29.4) | 5001 (21.7) | < 0.001 | 6209 (22.5) | 1138 (27.5) | 5071 (21.6) | < 0.001 | 9493 (48.7) | 1513 (52.2) | 7980 (48.1) | < 0.001 | 23,359 (83.7) | 3246 (85.1) | 20,113 (83.4) | 0.01 |
| Median follow‐up, months (IQR) | 66 (38, 99) | 63 (35, 95) | 66 (39, 99) | < 0.001 | 69 (41, 104) | 67 (40, 102) | 69 (42, 104) | 0.002 | 45 (20, 81) | 43 (17, 81) | 45 (21, 81) | 0.02 | 9 (2, 30) | 8 (2, 27) | 10 (2, 31) | < 0.001 |
Estimated using Mann Whitney U and χ 2 tests for continuous and categorical variables, respectively.
HR indicates hormone receptor status (either estrogen receptor or progesterone receptor); HER2 indicates Human Epidermal Growth Factor‐2 Receptor status.
Denominators for treatment and insurance status do not equal total sample size. The final analytic sample was a complete case analysis of those not missing all variables in the table except treatment and insurance status as these variables were added after primary analyses were completed.
In CRC, results were similar for race, rurality, Medicare, Medicaid, death, and follow‐up time. Those in a PP tract were less likely to be male (50.3% vs. 53.5%; p = 0.001; Table 1). In lung cancer, results were consistent for race, sex, rurality, Medicare, death, and follow‐up time. Those in a PP tract were more likely to receive neither chemotherapy nor radiation (43.5% vs. 41.3%, p < 0.001) and have private insurance (18.5% vs. 13.8%; p < 0.001; Table 1).
In unadjusted models, PP was significantly associated with OS across all cancer types examined (hazard ratio [95% CI]‐breast: 1.42 [1.33, 1.52]; prostate: 1.31 [1.22, 1.41]; CRC: 1.11 [1.05, 1.18]; lung: 1.08 [1.04, 1.12]). After adjusting for demographics, these associations were attenuated in breast and prostate cancer; lost statistical significance in CRC, and remained unchanged in lung cancer (Table 2). After further adjustment for downstream clinical characteristics, the association between PP and OS remained relatively unchanged compared to the demographics‐adjusted model (Table 2). After full adjustment for downstream social determinants, results were slightly attenuated except for lung cancer. Living in a PP tract was associated with higher adjusted hazard of death compared to living in a non‐PP tract (hazard ratio [95% CI]: breast: 1.13 [1.04–1.23]; prostate: 1.18 [1.08, 1.29]; lung: 1.07 [1.02, 1.12]; Table 2). Among adults with CRC, living in a PP tract was associated with 4% higher hazard of death; this association did not reach statistical significance (hazard ratio: 1.04, 95% CI: 0.96, 1.11; Table 2). Race and rurality were independently associated with OS across cancer types with Black race associated with higher hazard of death and identifying as non‐Black or non‐White associated with lower hazard of death compared to identifying as White (Table 2). Among adults with lung cancer, there was no association between identifying as Black and hazard of death, but there was a significant 23% lower hazard of death among those identifying as another racial group compared to White (hazard ratio: 0.77, 95% CI: 0.66, 0.93; Table 2). Living in a rural tract was associated with significant higher hazard of death compared to living in an urban tract among Alabamians with prostate cancer (Table 2). Attenuation across sequential models is only descriptive and model 4 should not be interpreted as the total effect. On sensitivity analysis in the breast cancer sample excluding subtype as a covariate, results did not change (Table S2). On sensitivity analysis adding receipt of surgery as an additional covariate, results for PP were slightly attenuated across all cancer types, and lost statistical significance in breast and lung cancer. However, it is possible that adding surgery induced multi‐collinearity between surgery and systemic treatment status, surgery and PP, and/or surgery and insurance status, which increased standard error and reduced statistical power (Table S3). Figure 2 displays Kaplan–Meier curves for each cancer type stratified by PP status. AIC values decreased across all cancer types with sequential adjustment indicating that the fully‐adjusted model was the best fit. The greatest amount of variance was explained when adding clinical characteristics (Breast Unadjusted: 116801.38, Demographics: 114611.58, Clinical Characteristics: 95016.149, Full: 93550.712; Prostate Unadjusted: 120862.90, Demographics: 117243.56, Clinical Characteristics: 89214.634, Full: 82295.149; Colorectal Unadjusted: 178788.07, Demographics: 177031.64, Clinical Characteristics: 149304.57, Full: 145754.92; Lung Unadjusted: 446319.12, Demographics: 445468.29, Clinical Characteristics: 388899.50, Full: 381799.20; Table 2). Two‐way interactions between rural–urban status and PP were not statistically significant except for lung cancer (breast cancer: p = 0.70; prostate: p = 0.35; CRC: p = 0.37; lung: p = 0.01). Two‐way interactions between race and PP were not statistically significant (breast cancer: p = 0.70; prostate: p = 0.76; CRC: p = 0.08; lung: p = 0.58). We examined HRs and 95% CIs of the risk of all‐cause mortality for a 65‐year‐old individual across all possible combinations of race, rural–urban, and PP status to assess effect modification. Generally, across all cancer types, stratum‐specific estimates had higher magnitude for strata representing PP areas, rural areas, and individuals identifying as Black race. The highest stratum‐specific association was among those identifying as Black living in a rural PP tract (hazard ratio [95% CI]: breast: 1.73 [1.53–1.95]; prostate: 1.73 [1.53–1.95]; male colorectal: 1.44 [1.30–1.60]; female colorectal: 1.19 [1.10–1.31]; male lung: 1.25 [1.17–1.33]; female lung: 1.02 [0.96–1.8]; Table 3). The 95% CIs overlap across rural–urban status among those with breast, colorectal, and lung cancer. The 95% CIs slightly overlap across White and Black racial groups among males with colorectal cancer and heavily overlap across White and Black racial groups among those with lung cancer. Finally, 95% CIs overlap across PP versus non‐PP status among those with colorectal cancer and lung cancer (Table 3).
TABLE 2.
Hazard ratio (HR) and 95% confidence interval (95% CI) of the association between persistent poverty and all‐cause mortality by cancer type with sequential adjustment for covariates. a
| Model 1: unadjusted | Model 2: demographics adjusted | Model 3: demographics + clinical characteristics adjusted | Model 4: full adjustment | |
|---|---|---|---|---|
| Breast | ||||
| PP | 1.42 (1.33, 1.52) | 1.22 (1.13, 1.31) | 1.17 (1.07, 1.27) | 1.13 (1.04, 1.23) |
| Age | 1.95 (1.88, 2.01) | 1.96 (1.89, 2.04) | 1.78 (1.71, 1.85) | |
| Race | ||||
| White | 1.0 (Ref) | 1.0 (Ref) | 1.0 (Ref) | |
| Black | 1.36 (1.27, 1.44) | 1.21 (1.13, 1.30) | 1.16 (1.08, 1.24) | |
| Other | 0.47 (0.33, 0.69) | 0.45 (0.31, 0.67) | 0.41 (0.28, 0.62) | |
| Stage at diagnosis | ||||
| Early | 1.0 (Ref) | 1.0 (Ref) | ||
| Late | 3.10 (2.93, 3.29) | 3.04 (2.87, 3.22) | ||
| Subtype b | ||||
| HR+/HER2+ | 1.0 (Ref) | 1.0 (Ref) | ||
| HR+/HER2− | 0.94 (0.86, 1.03) | 0.95 (0.87, 1.04) | ||
| HR−/HER2+ | 1.28 (1.11, 1.47) | 1.30 (1.13, 1.50) | ||
| TNBC | 1.72 (1.55, 1.91) | 1.70 (1.53, 1.89) | ||
| First course treatment | ||||
| Neither chemotherapy nor radiation | 1.0 (Ref) | 1.0 (Ref) | ||
| Chemotherapy only | 0.80 (0.73, 0.87) | 0.80 (0.74, 0.88) | ||
| Radiation only | 0.61 (0.56, 0.66) | 0.61 (0.56, 0.67) | ||
| Chemotherapy and radiation | 0.66 (0.60, 0.73) | 0.67 (0.61, 0.75) | ||
| Rural‐urban RUCA | ||||
| Urban | 1.0 (Ref) | |||
| Rural | 0.99 (0.92, 1.06) | |||
| Insurance status | ||||
| Private | 1.0 (Ref) | |||
| Public | 1.13 (0.89, 1.43) | |||
| Medicare | 1.68 (1.52, 1.84) | |||
| Medicaid | 2.13 (1.87, 2.43) | |||
| Insured, NOS | 1.03 (0.91, 1.16) | |||
| Uninsured | 2.69 (2.26, 3.19) | |||
| Prostate | ||||
| PP | 1.31 (1.22, 1.41) | 1.21 (1.12, 1.31) | 1.25 (1.14, 1.36) | 1.18 (1.08, 1.29) |
| Age | 2.14 (2.08, 2.20) | 2.18 (2.11, 2.25) | 2.09 (2.01, 2.18) | |
| Race | ||||
| White | 1.0 (Ref) | 1.0 (Ref) | 1.0 (Ref) | |
| Black | 1.32 (1.25, 1.40) | 1.35 (1.26, 1.44) | 1.31 (1.22, 1.40) | |
| Other | 0.73 (0.51, 1.04) | 0.73 (0.50, 1.07) | 0.73 (0.49, 1.09) | |
| Stage at diagnosis | ||||
| Early | 1.0 (Ref) | 1.0 (Ref) | ||
| Late | 2.56 (2.40, 2.74) | 2.54 (2.37, 2.72) | ||
| First course treatment | ||||
| Neither chemotherapy nor radiation | 1.0 (Ref) | 1.0 (Ref) | ||
| Chemotherapy only | 2.76 (2.25, 3.37) | 2.72 (2.19, 3.38) | ||
| Radiation only | 1.16 (1.09, 1.25) | 1.15 (1.08, 1.24) | ||
| Chemotherapy and radiation | 2.28 (1.70, 3.06) | 3.01 (2.04, 4.42) | ||
| Rural‐urban RUCA | ||||
| Urban | 1.0 (Ref) | |||
| Rural | 1.13 (1.06, 1.22) | |||
| Insurance status | ||||
| Private | 1.0 (Ref) | |||
| Public | 1.42 (1.22, 1.64) | |||
| Medicare | 1.45 (1.32, 1.59) | |||
| Medicaid | 2.72 (2.18, 3.40) | |||
| Insured, NOS | 1.07 (0.93, 1.25) | |||
| Uninsured | 3.18 (2.63, 3.85) | |||
| Colorectal | ||||
| PP | 1.11 (1.05, 1.18) | 1.05 (0.99, 1.12) | 1.06 (0.99, 1.14) | 1.04 (0.96, 1.11) |
| Age | 1.72 (1.68, 1.76) | 1.74 (1.69, 1.79) | 1.67 (1.61, 1.73) | |
| Race | ||||
| White | 1.0 (Ref) | 1.0 (Ref) | 1.0 (Ref) | |
| Black | 1.15 (1.10, 1.21) | 1.13 (1.07, 1.19) | 1.10 (1.04, 1.16) | |
| Other | 0.69 (0.54, 0.86) | 0.67 (0.52, 0.85) | 0.65 (0.51, 0.85) | |
| Sex | ||||
| Female | 1.0 (Ref) | 1.0 (Ref) | 1.0 (Ref) | |
| Male | 1.21 (1.16, 1.26) | 1.20 (1.14, 1.25) | 1.21 (1.15, 1.27) | |
| Stage at diagnosis | ||||
| Early | 1.0 (Ref) | 1.0 (Ref) | ||
| Late | 2.64 (2.49, 2.79) | 2.60 (2.46, 2.75) | ||
| First course treatment | ||||
| Neither chemotherapy nor radiation | 1.0 (Ref) | 1.0 (Ref) | ||
| Chemotherapy only | 0.98 (0.92, 1.04) | 0.99 (0.93, 1.05) | ||
| Radiation only | 1.22 (1.02, 1.47) | 1.20 (1.00, 1.44) | ||
| Chemotherapy and radiation | 0.84 (0.78, 0.91) | 0.85 (0.78, 0.92) | ||
| Rural‐urban RUCA | ||||
| Urban | 1.0 (Ref) | |||
| Rural | 1.03 (0.98, 1.09) | |||
| Insurance status | ||||
| Private | 1.0 (Ref) | |||
| Public | 1.15 (0.98, 1.34) | |||
| Medicare | 1.33 (1.24, 1.43) | |||
| Medicaid | 1.96 (1.75, 2.19) | |||
| Insured, NOS | 1.01 (0.92, 1.12) | |||
| Uninsured | 1.66 (1.50, 1.84) | |||
| Lung | ||||
| PP | 1.08 (1.04, 1.12) | 1.08 (1.04, 1.12) | 1.09 (1.04, 1.13) | 1.07 (1.02, 1.12) |
| Age | 1.22 (1.20, 1.24) | 1.19 (1.17, 1.22) | 1.20 (1.18, 1.23) | |
| Race | ||||
| White | 1.0 (Ref) | 1.0 (Ref) | 1.0 (Ref) | |
| Black | 1.03 (0.99, 1.07) | 0.99 (0.96, 1.03) | 0.98 (0.94, 1.02) | |
| Other | 0.79 (0.66, 0.95) | 0.81 (0.66, 0.98) | 0.77 (0.63, 0.95) | |
| Sex | ||||
| Female | 1.0 (Ref) | 1.0 (Ref) | 1.0 (Ref) | |
| Male | 1.27 (1.24, 1.31) | 1.22 (1.19, 1.26) | 1.22 (1.19, 1.26) | |
| Stage at diagnosis | ||||
| Early | 1.0 (Ref) | 1.0 (Ref) | ||
| Late | 3.00 (2.87, 3.13) | 3.00 (2.86, 3.13) | ||
| First course treatment | ||||
| Neither chemotherapy nor radiation | 1.0 (Ref) | 1.0 (Ref) | ||
| Chemotherapy only | 0.72 (0.69, 0.75) | 0.72 (0.69, 0.75) | ||
| Radiation only | 1.01 (0.96, 1.06) | 1.00 (0.96, 1.05) | ||
| Chemotherapy and radiation | 0.65 (0.62, 0.67) | 0.65 (0.62, 0.68) | ||
| Rural‐urban RUCA | ||||
| Urban | 1.0 (Ref) | |||
| Rural | 1.01 (0.97, 1.04) | |||
| Insurance status | ||||
| Private | 1.0 (Ref) | |||
| Public | 1.00 (0.92, 1.08) | |||
| Medicare | 1.10 (1.05, 1.15) | |||
| Medicaid | 1.37 (1.28, 1.47) | |||
| Insured, NOS | 1.04 (0.96, 1.13) | |||
| Uninsured | 1.39 (1.29, 1.49) | |||
Note: Akaike Information Criteria (AIC) Values for each model: Breast Unadjusted: 116801.38; Demographics: 114611.58; Clinical Characteristics: 95016.149; Full: 93550.712; Prostate Unadjusted: 120862.90; Demographics: 117243.56; Clinical Characteristics: 89214.634; Full: 82295.149; Colorectal Unadjusted: 178788.07; Demographics: 177031.64; Clinical Characteristics: 149304.57; Full: 145754.92; Lung Unadjusted: 446319.12; Demographics: 445468.29; Clinical Characteristics: 388899.50; Full: 381799.20.
Models estimated using Cox proportional hazards regression incorporating cluster‐robust standard errors to account for clustering by census tract.
HR indicates hormone receptor status (either estrogen receptor or progesterone receptor); HER2 indicates Human Epidermal Growth Factor‐2 Receptor status.
FIGURE 2.
Kaplan–Meier curve by persistent poverty status. This figure shows Kaplan–Meier survival curves by persistent poverty status across cancer types.
TABLE 3.
Hazard ratio (HR) and 95% confidence interval (95% CI) of the risk of all‐cause mortality for a 65 year‐old across all combinations of race, rural–urban status, and persistent poverty (PP) status by cancer type. a
| White | Black | Other | ||||
|---|---|---|---|---|---|---|
| Rural | Urban | Rural | Urban | Rural | Urban | |
| Breast [HR (95% CI)] | ||||||
| PP | 1.26 (1.13, 1.40) | 1.28 (1.17, 1.39) | 1.46 (1.30, 1.63) | 1.48 (1.36, 1.61) | 0.52 (0.34, 0.79) | 0.52 (0.34, 0.79) |
| Non‐PP | 1.12 (1.04,1.20) | 1.13 (1.12, 1.15) | 1.29 (1.17, 1.44) | 1.31 (1.22, 1.41) | 0.46 (0.31, 0.69) | 0.47 (0.31, 0.70) |
| Prostate [HR (95% CI)] | ||||||
| PP | 1.32 (1.18, 1.48) | 1.17 (1.07, 1.28) | 1.73 (1.53, 1.95) | 1.52 (1.38, 1.68) | 0.96 (0.63, 1.47) | 0.85 (0.56, 1.29) |
| Non‐PP | 1.12 (1.04, 1.21) | 0.99 (0.99, 0.99) | 1.46 (1.32, 1.62) | 1.29 (1.21, 1.38) | 0.82 (0.54, 1.24) | 0.72 (0.48, 1.08) |
| Colorectal: Male [HR (95% CI)] | ||||||
| PP | 1.31 (1.19, 1.45) | 1.27 (1.16, 1.38) | 1.44 (1.30, 1.60) | 1.39 (1.28, 1.52) | 0.86 (0.65, 1.13) | 0.83 (0.63, 1.08) |
| Non‐PP | 1.27 (1.18, 1.36) | 1.23 (1.17, 1.28) | 1.39 (1.27, 1.53) | 1.35 (1.25, 1.45) | 0.83 (0.64, 1.08) | 0.80 (0.62, 1.04) |
| Colorectal: Female [HR (95% CI)] | ||||||
| PP | 1.09 (0.99, 1.19) | 1.05 (0.98, 1.13) | 1.19 (1.09, 1.31) | 1.15 (1.08, 1.24) | 0.71 (0.54, 0.93) | 0.69 (0.53, 0.90) |
| Non‐PP | 1.05 (0.99, 1.11) | 1.01 (1.01, 1.02) | 1.15 (1.06, 1.25) | 1.11 (1.05, 1.18) | 0.69 (0.53, 0.89) | 0.66 (0.51, 0.86) |
| Lung: Male [HR (95% CI)] | ||||||
| PP | 1.27 (1.20, 1.35) | 1.26 (1.20, 1.33) | 1.25 (1.17, 1.33) | 1.24 (1.17, 1.31) | 0.98 (0.79, 1.22) | 0.97 (0.78, 1.21) |
| Non‐PP | 1.19 (1.14, 1.24) | 1.18 (1.15, 1.22) | 1.17 (1.10, 1.24) | 1.16 (1.10, 1.22) | 0.92 (0.74, 1.13) | 0.91 (0.74, 1.12) |
| Lung: Female [HR (95% CI)] | ||||||
| PP | 1.04 (0.99, 1.10) | 1.03 (0.99, 1.08) | 1.02 (0.96, 1.08) | 1.01 (0.97, 1.06) | 0.80 (0.65, 0.99) | 0.80 (0.64, 0.98) |
| Non‐PP | 0.97 (0.94, 1.01) | 0.97 (0.96, 0.97) | 0.96 (0.90, 1.01) | 0.95 (0.92, 0.99) | 0.75 (0.61, 0.92) | 0.74 (0.61, 0.92) |
Note: To provide practical application, we estimated the linear combinations for an individual 65 years old across each unique combination of PP status, race, and rural–urban status (e.g., White, PP, urban, and so on) with all covariates held at reference values. For example, the 1.26 HR for white, rural, PP can be interpreted as for a 65‐year old white woman with breast cancer living in a rural, PP census tract, she has a 1.26 HR of death, which significantly differs from 0.
Bold indicates statistical significance.
In decomposition analyses, there was statistical evidence that differing racial distribution across PP categories among individuals with breast and prostate cancer explains some of the differences in OS between PP and non‐PP. The total effect hazard of death at any time point in the study for PP area residents was 20% higher than for non‐PP area residents among those with breast cancer and 31% higher among those with prostate cancer (hazard ratio [95% CI]‐breast: 1.20 [1.11, 1.30]; prostate: 1.31 [1.21, 1.41]). The portion of the total effect explained by each decomposing variable (race or rurality) suggested differences in the racial distribution between PP and non‐PP explain some PP‐related differences in OS (hazard ratio [95% CI]‐breast: 1.07 [1.03, 1.10]; prostate: 1.11 [1.08, 1.14]). However, differences in the racial distribution do not completely account for the difference in OS between PP and non‐PP as the portion of the total effect explained by PP alone, not through the decomposing variable, were also statistically significant (hazard ratio [95% CI]‐breast: 1.13 [1.03, 1.22]; prostate: 1.18 [1.08, 1.28]) (Table 4). As described in Table 1, Black patients were more likely to reside in PP areas. The proportion of the total effect decomposed by race was 37% for breast cancer (95% CI: 18%–72%) and 42% for prostate cancer (95% CI: 29%–63%). In prostate cancer, the decomposition model with only rural–urban status demonstrated a total effect indicating a 19% higher hazard of death at any time point in the study for PP area residents compared to non‐PP area residents (hazard ratio [95% CI]: 1.19 [1.09, 1.30]). However, most of this effect was driven by the portion of the total effect of PP on OS rather than accounted for by rural–urban status as a decomposer (hazard ratio [95% CI]‐portion of total effect due to PP only: 1.18 [1.08, 1.28]; portion of total effect accounted for by decomposer: 1.01 [1.01, 1.02]). When adding both race and rural–urban status to the model, the portion of the total effect accounted for by rurality was similar to that observed in the race‐only model suggesting differences in the race distribution across PP and non‐PP rather than differences in the distribution of rural–urban residence account for more of the observed differences in OS between PP and non‐PP (Table 4).
TABLE 4.
Decomposition of the total association statistically accounted for by differences in racial composition and rurality. a
| Effect | Breast cancer: race as only decomposer | Prostate cancer: race as only decomposer | Prostate cancer: rural–urban status as only decomposer | Prostate cancer: race and rural–urban status as decomposers | Lung cancer: race as only decomposer |
|---|---|---|---|---|---|
| Total effect | 1.20 (1.11, 1.30) | 1.31 (1.21, 1.41) | 1.19 (1.09, 1.30) | 1.32 (1.21, 1.42) | 1.06 (1.02, 1.11) |
| Portion of total effect accounted for by PP | 1.13 (1.03, 1.22) | 1.18 (1.08, 1.28) | 1.18 (1.08, 1.28) | 1.18 (1.08, 1.28) | 1.07 (1.02, 1.12) |
| Portion of total effect accounted for by decomposer | 1.07 (1.03, 1.10) | 1.11 (1.08, 1.14) | 1.01 (1.01, 1.02) | 1.12 (1.09, 1.15) | 0.99 (0.97, 1.01) |
| Proportion decomposed | 0.37 (0.18, 0.72) | 0.42 (0.29, 0.63) | 0.07 (0.03, 0.19) | 0.44 (0.31, 0.63) | −0.16 (−0.64, 0.24) |
Note: We did not examine decomposition by race among patients with CRC because the association between PP and survival was not statistically significant in Cox models. We only examined rurality as a decomposer in prostate cancer due to its lack of statistical significance in breast and lung cancer Cox modeling.
Estimated using regression‐based mediation analysis of Cox proportional hazards model with 1000 bootstraps. Significance set at α = 0.05. Bold indicates statistical significance.
4. Discussion
Results suggest that living in a PP tract is associated with increased overall risk of death among patients with breast, prostate, and lung cancer in Alabama after adjusting for race and rurality. However, after full adjustment, race demonstrates significant independent effects. In general, the highest magnitude of effect was among those identifying as Black living in a rural PP tract (Table 3). However, the intersectionality of these effects differs by cancer type (Table 3). Moreover, among patients with breast or prostate cancer in Alabama, race partially decomposes the association between PP and OS (Table 4).
Our results are generally consistent with prior findings—individuals identifying as Black and living in rural or PP tracts have higher mortality and lower 5‐year survival rates [4, 7]—but rural–urban residence was not independently significant in fully‐adjusted models. There are a few methodological distinctions between current and previous studies. Our results report longitudinal OS rather than mortality or survival rates alone. We adjusted for clinical covariates such as stage, subtype (for breast cancer), and first course treatment, which can confound the association between PP, rurality, race, and OS. Taken together, prior and current results suggest PP is associated with poorer outcomes in cancer, with major overlap occurring with race. However, as all‐cause, not cancer‐specific, mortality was used, the association between PP and survival may partly reflect death from other causes, particularly for breast and prostate cancer.
Living in PP exposes individuals to chronic stress, dysregulating stress‐responses and disrupting multiple physiological systems, including impaired immune response and chronic inflammation which becomes biologically embedded over time contributing to poorer cancer outcomes [23, 24, 25]. Individuals living in PP areas may engage in unhealthy lifestyle behaviors (e.g., smoking, physical inactivity, poor diet) more than those living in non‐PP areas, which can exacerbate biological dysregulation [26, 27, 28, 29]. Systemic factors and policies that influence health behaviors are also important. For example, PP areas lack resources to promote healthy behaviors (e.g., access to healthy foods, physical activity, and screening). Policies such as Alabama's lack of Medicaid expansion may disproportionately limit access to preventive care for PP residents [27, 30, 31]. Studies show that PP and rural area residents have lower rates of screening for common cancers, resulting in later‐stage diagnosis and higher mortality [32, 33]. PP areas have a higher proportion of Black individuals and the paucity of resources in PP areas combined with higher rates of unhealthy behaviors observed in Black individuals likely drives racial disparities in mortality [27, 30, 34]. Moreover, in recent years, despite higher screening rates and lower incidence rates for some cancers (e.g., breast), Black individuals remain at higher risk of mortality than White individuals [35, 36]. Even if differences in lifestyle behaviors and screening are equalized across racial, geographic, and PP groups, upstream factors, including social determinants of health (e.g., access to care, health literacy, built environment), remain major contributors to health disparities [20, 37].
It is important to note differences between models. The demographics‐only adjusted model explained the least amount of variance across all cancer types examined indicating that non‐modifiable characteristics such as age at diagnosis, race, and sex have a smaller influence in explaining survival differences. The demographics plus clinical characteristics‐adjusted model explained a greater amount of variance, indicating potentially modifiable factors such as stage at diagnosis and first course treatment (both of which may be affected by social factors such as access to care) have moderate influence in explaining survival differences. However, the fully‐adjusted model including all previous covariates plus rural–urban status and insurance status explained the highest level of variance, indicating that socially‐influenced factors in addition to potentially modifiable clinical characteristics have a large influence in explaining survival differences. Rural–urban and insurance status, while modifiable in theory, are challenging in practice [38]. For example, individuals may choose to live in rural areas due to social support, lower cost of living, and others [39]. Similarly, most individuals are limited in insurance they can afford or qualify for due to employment and income level, high premium cost, and more [40]. Collectively, these results suggest socially‐modifiable factors explain a large amount of variation in survival. Multi‐level, community‐engaged public health and/or policy interventions may alleviate the role of these factors (e.g., providing greater access to screening and care through mobile units and/or telehealth interventions, improving cancer screening and prevention education using community‐engaged approaches, implementing navigation to cancer care services, and expanding Medicaid under the Affordable Care Act) [41, 42, 43, 44].
Our results suggest that in Black women, HR+ breast cancer may be more socially driven whereas TNBC may be more biologically driven (e.g., molecular differences occurring in response to increased stress from differential exposure to resource scarcity among Black individuals), but neither of these seems true for White women [18, 23, 24, 25, 45, 46]. This is further supported by our results with overlapping 95% CIs between Black and White individuals among those with CRC and lung cancer, but not breast and prostate and that race partially decomposes the association between PP and survival in only breast and prostate cancer, suggesting that cancers more often diagnosed at later stages with poorer 5‐year survival rates (e.g., CRC, lung) may not be as affected by social factors [1]. Further work is needed to tease apart social from biological factors among Black women with breast cancer and Black men with prostate cancer. For example, how molecular differences between Black versus White individuals, arising from increased stress or differential exposure to resource scarcity (in PP areas), account for both PP‐related and racial disparities in survival [23, 24, 25]. A deeper examination of the relationship between PP, race, and/or rurality with cancer‐specific mortality is needed. The associations identified in this study may be more reflective of disparities in all‐cause rather than cancer‐specific mortality in breast and prostate cancer, whereas among CRC and lung cancer patients it may be more reflective of cancer‐specific mortality. If so, the relationship between PP, race, and survival in this study among patients with breast and prostate cancer reflects complexities of living in a PP area and/or identifying as Black—higher comorbidity development/exacerbation and/or interaction between treatment and comorbidities leading to earlier death from non‐cancer causes.
This study has several strengths. It utilizes data from a population‐based cancer registry. We measured area‐level data at the census tract rather than county level, which allows for more specific inference and targeting of interventions to areas most in need. We controlled for clinical characteristics that may confound the associations under study. Finally, we examined longitudinal survival rather than mortality or survival rates.
This study has limitations. This study examines PP by census tract, which may be subject to ecological fallacy wherein area‐level associations do not reflect individual‐level experiences [47]; studies have consistently linked both area‐ and individual‐level poverty with worse health outcomes [3, 4, 7, 27, 30, 48, 49] and no objective measure of individual‐level PP currently exist. Even if these results are consistent only at the area level, they remain relevant, as PP areas face higher unemployment, poorer access to healthy foods, limited healthcare access, and other social risks [3]. We present these analyses as a decomposition analysis and do not make causal claims about the relationships between PP, our explanatory variables of race and rurality, and our outcomes. Race and rurality lie upstream of PP thus, our decomposition cannot be interpreted how much of the gap in outcomes between PP groups would remain if explanatory variables were equalized, but rather this is the amount of the association between PP and our outcomes left unexplained after we account for race and rurality holding our covariates constant. We could not adjust for health behaviors (e.g., smoking, physical activity, screening) and unhealthy lifestyles are likely more common in PP areas, potentially contributing to observed associations. We also could not adjust for residential history and individual socioeconomic status; PP area residents may not have lived there long, and residency alone does not confirm personal poverty. Adjustment for stage at diagnosis, first course treatment, and insurance status may constitute overadjustment if these variables lie on the pathway between persistent poverty tract residence and survival. Estimates from the fully adjusted model therefore reflect the association remaining after accounting for these factors and should not be interpreted as the total effect of persistent poverty on mortality; we are likely violating the rare event assumption of the decomposition analysis which means estimates are approximate not exact. HRs are non‐collapsible meaning the total association and its components are not strictly comparable across models and the multiplicative decomposition holds only approximately when the outcome is common, as it is here. However, this phenomenon reduces rather than inflates the portion of the association attributable to the decomposing variables, so the reported decompositions are likely conservative. If anything, we underestimated rather than over‐estimated the HR. Additionally, we were only able to examine all‐cause mortality, though this likely reflects cause‐specific mortality for lung cancer and CRC. Regardless, these results still reflect adverse effects of PP, rurality, and race on survival, and interventions are needed to alleviate these disparities. Some deaths may have been missed due to ASCR data collection limitations. To minimize this, we set end of follow‐up as the end of the year prior to most recent death data linkage. We were only able to adjust for SEER stage rather than the more precise AJCC stage, though SEER staging is standard in cancer studies. Treatment adjustment was limited to first course treatment, which captures an incomplete picture of an individual's full treatment history. Finally, although RUCA is not the only definition of rural–urban status, we have noted in our previous study that many federal definitions do not alter observed associations for research conducted in Alabama [20]. However, it is possible that results may change with more effective rural–urban definitions (e.g., based on resource proximity rather than commuting flow as in RUCA).
5. Conclusion
Future studies should examine the effects of lifestyle behaviors and screening on these survival disparities and tease apart the overlap between PP and race. Finally, identification of disparities is simply not enough. Strong mixed‐methods research is needed to scaffold multilevel interventions that are tailored to the local context and aimed at effectiveness and scalability.
Author Contributions
Mackenzie E. Fowler: conceptualization, methodology, data curation, investigation, validation, formal analysis, visualization, writing – original draft, writing – review and editing, software. Hayden D. Reeves: conceptualization, investigation, writing – original draft, writing – review and editing. Suzanne Judd: conceptualization, investigation, writing – review and editing. Kelley Pettee Gabriel: conceptualization, investigation, writing – review and editing. Geetanjali Saini: conceptualization, investigation, data curation, writing – review and editing. Mahak Bhargava: conceptualization, data curation, investigation, writing – review and editing. Timiya S. Nolan: conceptualization, investigation, writing – review and editing. Ritu Aneja: conceptualization, investigation, funding acquisition, writing – review and editing, supervision.
Funding
This work was supported by the National Cancer Institute at the National Institutes of Health (grant number R01CA239120 to [RA], U54CA280779); and the Breast Cancer Research Foundation of Alabama. Funders had no role in the design, data collection, data analysis, or interpretation of study findings.
Ethics Statement
This study was approved by the Institutional Review Boards of the University of Alabama at Birmingham and the Alabama Department of Public Health (ADPH). All procedures were conducted in accordance with ethical standards and principles of the Declaration of Helsinki.
Consent
A waiver of informed consent was granted by the Institutional Review Board at the University of Alabama at Birmingham. This retrospective cohort study included approximately 100,000 participants, and many may have deceased prior to start of this study so obtaining informed consent for all participants was not possible. Moreover, the data did not include any direct patient identifiers, so contacting participants was not possible.
Conflicts of Interest
The authors declare no conflicts of interest.
Supporting information
Table S1: Breast cancer subtypes by persistent poverty status (PP) stratified by race.
Table S2: Hazard ratio (HR) and 95% confidence interval (95% CI) of the association between persistent poverty and all‐cause mortality by cancer type with sequential adjustment for covariates (excluding subtype due to high missingness).
Table S3: Sensitivity analysis of hazard ratio (HR) and 95% confidence interval (95% CI) of the association between persistent poverty and all‐cause mortality by cancer type with receipt of surgery as an additional covariate.
Acknowledgements
The authors would like to acknowledge Justin T. George and the Alabama Statewide Cancer Registry at the Alabama Department of Public Health for their provision of data used in this study.
Data Availability Statement
The data that support the findings of this study are available from the Alabama Statewide Cancer Registry at the Alabama Department of Public Health. Restrictions apply to the availability of these data, which were used under license for this study.
References
- 1. American Cancer Society , “Cancer Facts & Figures 2025,” (2025), https://www.cancer.org/content/dam/cancer‐org/research/cancer‐facts‐and‐statistics/annual‐cancer‐facts‐and‐figures/2025/2025‐cancer‐facts‐and‐figures‐acs.pdf.
- 2. American Society for Cancer Research , AACR Cancer Disparities Progress Report 2024 (American Association for Cancer Research, 2024), https://cancerprogressreport.aacr.org/disparities/cdpr24‐contents/cdpr24‐the‐state‐of‐us‐cancer‐disparities‐in‐2024/. [Google Scholar]
- 3. Moss J. L., Pinto C. N., Srinivasan S., Cronin K. A., and Croyle R. T., “Persistent Poverty and Cancer Mortality Rates: An Analysis of County‐Level Poverty Designations,” Cancer Epidemiology, Biomarkers & Prevention 29, no. 10 (2020): 1949–1954, 10.1158/1055-9965.Epi-20-0007. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4. Moss J. L., Pinto C. N., Srinivasan S., Cronin K. A., and Croyle R. T., “Enduring Cancer Disparities by Persistent Poverty, Rurality, and Race: 1990–1992 to 2014–2018,” Journal of the National Cancer Institute 114, no. 6 (2022): 829–836, 10.1093/jnci/djac038. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5. Tsai M. H., Coughlin S. S., Cortes J., and Thompson C. A., “Intersection of Poverty and Rurality for Early‐Onset Colorectal Cancer Survival,” JAMA Network Open 7, no. 8 (2024): e2430615, 10.1001/jamanetworkopen.2024.30615. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6. Tobin E. C., Nolan N., Thompson S., Elmore M., and Richmond B. K., “The Intersection of Race and Rurality and Its Effect on Colorectal Cancer Survival,” American Surgeon 89, no. 7 (2023): 3163–3170, 10.1177/00031348231160833. [DOI] [PubMed] [Google Scholar]
- 7. Bhattacharya M., Cronin K. A., Farrigan T. L., Kennedy A. E., Yu M., and Srinivasan S., “Description of Census‐Tract‐Level Social Determinants of Health in Cancer Surveillance Data,” Journal of the National Cancer Institute Monographs 2024, no. 65 (2024): 152–161, 10.1093/jncimonographs/lgae027. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8. Benson C., Bishaw A., and Glassman B., “Persistent Poverty in Counties and Census Tracts,” American Community Survey Reports, (2023), https://www.census.gov/content/dam/Census/library/publications/2023/acs/acs‐51%20persistent%20poverty.pdf.
- 9. Alabama Department of Public Health, Alabama Statewide Cancer Registry , “Alabama Cancer Statistics 2021,” (2021), https://www.alabamapublichealth.gov/ASCR/assets/acs2021.pdf.
- 10. Alabama Department of Public Health , “Rural Health: At a Glance,” https://www.alabamapublichealth.gov/ruralhealth/at‐a‐glance.html.
- 11. Alabama Department of Public Health , “Office of Primary Care and Rural Health: At a Glance,” https://www.alabamapublichealth.gov/ruralhealth/at‐a‐glance.html.
- 12. United States Census Bureau , “Alabama,” (2026), https://data.census.gov/profile/Alabama?g=040XX00US01.
- 13. Alabama Statewide Cancer Registry , “Alabama Department of Public Health,” https://www.alabamapublichealth.gov/ascr/.
- 14. National Cancer Institute , “Census Tracts in Persistent Poverty,” https://cancercontrol.cancer.gov/sites/default/files/2021‐12/PP%20CT%20data‐final.pdf.
- 15. University of Washington School of Medicine , “RUCA Data: Using RUCA Data,” https://depts.washington.edu/uwruca/ruca‐uses.php.
- 16. University of Washington School of Medicine , “RUCA Data: General Methods,” https://depts.washington.edu/uwruca/ruca‐methodology.php.
- 17. University of Washington School of Medicine , “RUCA Data: Code Definitions,” https://depts.washington.edu/uwruca/ruca‐codes.php.
- 18. Luningham J. M., Seth G., Saini G., et al., “Association of Race and Area Deprivation With Breast Cancer Survival Among Black and White Women in the State of Georgia,” JAMA Network Open 5, no. 10 (2022): e2238183, 10.1001/jamanetworkopen.2022.38183. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19. Sopik V., Sun P., and Narod S. A., “The Prognostic Effect of Estrogen Receptor Status Differs for Younger Versus Older Breast Cancer Patients,” Breast Cancer Research and Treatment 165, no. 2 (2017): 391–402, 10.1007/s10549-017-4333-2. [DOI] [PubMed] [Google Scholar]
- 20. Fowler M. E., Reeves H. D., Smith M. J., Saini G., Bhargava M., and Aneja R., “Structural Determinants of Racial Disparities in Breast Cancer Survival in Alabama,” Cancer Epidemiology, Biomarkers & Prevention 34 (2025): 1740–1748, 10.1158/1055-9965.Epi-25-0195. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21. Shi B., Choirat C., Coull B. A., VanderWeele T. J., and Valeri L., “CMAverse: A Suite of Functions for Reproducible Causal Mediation Analyses,” Epidemiology 32, no. 5 (2021): e20–e22, 10.1097/ede.0000000000001378. [DOI] [PubMed] [Google Scholar]
- 22. Jackson J. W. and VanderWeele T. J., “Decomposition Analysis to Identify Intervention Targets for Reducing Disparities,” Epidemiology 29, no. 6 (2018): 825–835, 10.1097/ede.0000000000000901. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23. Krieger N., “Defining and Investigating Social Disparities in Cancer: Critical Issues,” Cancer Causes & Control 16, no. 1 (2005): 5–14, 10.1007/s10552-004-1251-5. [DOI] [PubMed] [Google Scholar]
- 24. Taylor S. E., Repetti R. L., and Seeman T., “Health Psychology: What Is an Unhealthy Environment and How Does It Get Under the Skin?,” Annual Review of Psychology 48 (1997): 411–447, 10.1146/annurev.psych.48.1.411. [DOI] [PubMed] [Google Scholar]
- 25. Adler N. E. and Newman K., “Socioeconomic Disparities in Health: Pathways and Policies,” Health Affairs (Millwood) 21, no. 2 (2002): 60–76, 10.1377/hlthaff.21.2.60. [DOI] [PubMed] [Google Scholar]
- 26. Bennett K. J., Probst J. C., and Pumkam C., “Obesity Among Working Age Adults: The Role of County‐Level Persistent Poverty in Rural Disparities,” Health & Place 17, no. 5 (2011): 1174–1181, 10.1016/j.healthplace.2011.05.012. [DOI] [PubMed] [Google Scholar]
- 27. Moss J. L., Pinto C. N., and Shen C., “Prevalence of Cancer Risk Behaviors by County‐Level Persistent Poverty,” Cancer Epidemiology 94 (2025): 102735, 10.1016/j.canep.2024.102735. [DOI] [PubMed] [Google Scholar]
- 28. Roberts M. E., Doogan N. J., Kurti A. N., et al., “Rural Tobacco Use Across the United States: How Rural and Urban Areas Differ, Broken Down by Census Regions and Divisions,” Health & Place 39 (2016): 153–159, 10.1016/j.healthplace.2016.04.001. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29. Abildso C. G., Daily S. M., Umstattd Meyer M. R., Perry C. K., and Eyler A., “Prevalence of Meeting Aerobic, Muscle‐Strengthening, and Combined Physical Activity Guidelines During Leisure Time Among Adults, by Rural‐Urban Classification and Region ‐ United States, 2020,” MMWR. Morbidity and Mortality Weekly Report 72, no. 4 (2023): 85–89, 10.15585/mmwr.mm7204a1. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30. Papageorge M. V., Woods A. P., de Geus S. W. L., et al., “The Persistence of Poverty and Its Impact on Cancer Diagnosis, Treatment and Survival,” Annals of Surgery 277, no. 6 (2023): 995–1001, 10.1097/sla.0000000000005455. [DOI] [PubMed] [Google Scholar]
- 31. Song S. and Kucik J. E., “Trends in the Impact of Medicaid Expansion on the Use of Clinical Preventive Services,” American Journal of Preventive Medicine 62, no. 5 (2022): 752–762, 10.1016/j.amepre.2021.11.002. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32. Lee H., Baeker Bispo J., Pal Choudhury P., Wiese D., Jemal A., and Islami F., “Factors Contributing to Differences in Cervical Cancer Screening in Rural and Urban Community Health Centers,” Cancer 130, no. 13 (2024): 2315–2324, 10.1002/cncr.35265. [DOI] [PubMed] [Google Scholar]
- 33. Sepassi A., Li M., Zell J. A., Chan A., Saunders I. M., and Mukamel D. B., “Rural‐Urban Disparities in Colorectal Cancer Screening, Diagnosis, Treatment, and Survivorship Care: A Systematic Review and Meta‐Analysis,” Oncologist 29, no. 4 (2024): e431–e446, 10.1093/oncolo/oyad347. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34. Cook W. K., Kerr W. C., Karriker‐Jaffe K. J., Li L., Lui C. K., and Greenfield T. K., “Racial/Ethnic Variations in Clustered Risk Behaviors in the U.S,” American Journal of Preventive Medicine 58, no. 1 (2020): e21–e29, 10.1016/j.amepre.2019.08.020. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35. Saka A. H., Giaquinto A. N., McCullough L. E., et al., “Cancer Statistics for African American and Black People, 2025,” CA: A Cancer Journal for Clinicians 75, no. 2 (2025): 111–140, 10.3322/caac.21874. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 36. Ford S., Tarraf W., Williams K. P., Roman L. A., and Leach R., “Differences in Cervical Cancer Screening and Follow‐Up for Black and White Women in the United States,” Gynecologic Oncology 160, no. 2 (2021): 369–374, 10.1016/j.ygyno.2020.11.027. [DOI] [PubMed] [Google Scholar]
- 37. Alcaraz K. I., Wiedt T. L., Daniels E. C., Yabroff K. R., Guerra C. E., and Wender R. C., “Understanding and Addressing Social Determinants to Advance Cancer Health Equity in the United States: A Blueprint for Practice, Research, and Policy,” CA: A Cancer Journal for Clinicians 70, no. 1 (2020): 31–46, 10.3322/caac.21586. [DOI] [PubMed] [Google Scholar]
- 38. Alwan N. A., Stannard S., Berrington A., et al., “Risk Factors for Ill Health: How Do We Specify What Is ‘modifiable’?,” PLOS Global Public Health 4, no. 3 (2024): e0002887, 10.1371/journal.pgph.0002887. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 39. Burkhart‐Kriesel C., Cantrell R., Hodur N., Narjes C., and Vogt R., “Marketing Rural Communities: A Qualitative Study on the Power of Community Images,” Community Development 45, no. 2 (2014): 181–194, 10.1080/15575330.2014.890632. [DOI] [Google Scholar]
- 40. Cha A. E. and Cohen R. A., “Reasons for Being Uninsured Among Adults Aged 18–64 in the United States, 2019,” NCHS Data Brief 382 (2020): 1–8. [PubMed] [Google Scholar]
- 41. Coughlin S. S., Clary C., Johnson J. A., et al., “Continuing Challenges in Rural Health in the United States,” Journal of Environmental Health Science 5, no. 2 (2019): 90–92. [PMC free article] [PubMed] [Google Scholar]
- 42. Matthews K. A., Spears K. S., and Anderson‐Lewis C., “Rural Health Disparities: Contemporary Solutions for Persistent Rural Public Health Challenges,” Preventing Chronic Disease 22 (2025): E27, 10.5888/pcd22.250202. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 43. Paskett E., Thompson B., Ammerman A. S., Ortega A. N., Marsteller J., and Richardson D., “Multilevel Interventions to Address Health Disparities Show Promise in Improving Population Health,” Health Affairs (Millwood) 35, no. 8 (2016): 1429–1434, 10.1377/hlthaff.2015.1360. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 44. O'Mara‐Eves A., Brunton G., Oliver S., Kavanagh J., Jamal F., and Thomas J., “The Effectiveness of Community Engagement in Public Health Interventions for Disadvantaged Groups: A Meta‐Analysis,” BMC Public Health 15 (2015): 129, 10.1186/s12889-015-1352-y. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 45. Saini G., Ogden A., McCullough L. E., Torres M., Rida P., and Aneja R., “Disadvantaged Neighborhoods and Racial Disparity in Breast Cancer Outcomes: The Biological Link,” Cancer Causes & Control 30, no. 7 (2019): 677–686, 10.1007/s10552-019-01180-4. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 46. Collin L. J., Jiang R., Ward K. C., et al., “Racial Disparities in Breast Cancer Outcomes in the Metropolitan Atlanta Area: New Insights and Approaches for Health Equity,” JNCI Cancer Spectrum 3, no. 3 (2019): pkz053, 10.1093/jncics/pkz053. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 47. Krieger N., Chen J. T., Waterman P. D., Soobader M. J., Subramanian S. V., and Carson R., “Geocoding and Monitoring of US Socioeconomic Inequalities in Mortality and Cancer Incidence: Does the Choice of Area‐Based Measure and Geographic Level Matter?: The Public Health Disparities Geocoding Project,” American Journal of Epidemiology 156, no. 5 (2002): 471–482, 10.1093/aje/kwf068. [DOI] [PubMed] [Google Scholar]
- 48. Bradley C. J., Stevens J. L., Enewold L., and Warren J. L., “Stage and Mortality of Low‐Income Patients With Cancer: Evidence From SEER‐Medicaid,” Cancer 127, no. 2 (2021): 229–238, 10.1002/cncr.33207. [DOI] [PubMed] [Google Scholar]
- 49. Singh G. K. and Jemal A., “Socioeconomic and Racial/Ethnic Disparities in Cancer Mortality, Incidence, and Survival in the United States, 1950–2014: Over Six Decades of Changing Patterns and Widening Inequalities,” Journal of Environmental and Public Health 2017 (2017): 2819372, 10.1155/2017/2819372. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Table S1: Breast cancer subtypes by persistent poverty status (PP) stratified by race.
Table S2: Hazard ratio (HR) and 95% confidence interval (95% CI) of the association between persistent poverty and all‐cause mortality by cancer type with sequential adjustment for covariates (excluding subtype due to high missingness).
Table S3: Sensitivity analysis of hazard ratio (HR) and 95% confidence interval (95% CI) of the association between persistent poverty and all‐cause mortality by cancer type with receipt of surgery as an additional covariate.
Data Availability Statement
The data that support the findings of this study are available from the Alabama Statewide Cancer Registry at the Alabama Department of Public Health. Restrictions apply to the availability of these data, which were used under license for this study.