Neighborhood Deprivation and Rurality Associated With Patient-Reported Outcomes and Survival in Men With Prostate Cancer in NRG RTOG 0415

Abstract

Purpose:

Rurality and neighborhood deprivation can contribute to poor patient-reported outcomes, which have not been systematically evaluated in patients with specific cancers in national trials. Our objective was to examine the effect of rurality and neighborhood socioeconomic and environmental deprivation on patient-reported outcomes and survival in men with prostate cancer in NRG RTOG 0415.

Methods and Materials:

Data from men with prostate cancer in trial NRG RTOG 0415 were analyzed; 1092 men were randomized to receive conventional radiation therapy or hypofractionated radiation therapy. Rurality was categorized as urban or rural. Neighborhood deprivation was assessed using the area deprivation index and air pollution indicators (nitrogen dioxide and particulate matter with a diameter less than 2.5 micrometers) via patient ZIP codes. Expanded Prostate Cancer Index Composite measured cancer-specific quality of life. The Hopkins symptom checklist measured anxiety and depression. EuroQoL−5 Dimension assessed general health.

Results:

We analyzed 751 patients in trial NRG RTOG 0415. At baseline, patients from the most deprived neighborhoods had worse bowel (P = .011), worse sexual (P = .042), and worse hormonal (P = .015) scores; patients from the most deprived areas had worse self-care (P = .04) and more pain (P = .047); and patients from rural areas had worse urinary (P = .03) and sexual (P = .003) scores versus patients from urban areas. Longitudinal analyses showed that the 25% most deprived areas (P = .004) and rural areas (P = .002) were associated with worse EuroQoL−5 Dimension visual analog scale score. Patients from urban areas (hazard ratio, 1.81; P = .033) and the 75% less-deprived neighborhoods (hazard ratio, 0.68; P = .053) showed relative decrease in risk of recurrence or death (disease-free survival).

Conclusions:

Patients with prostate cancer from the most deprived neighborhoods and rural areas had low quality of life at baseline, poor general health longitudinally, and worse disease-free survival. Interventions should screen populations from deprived neighborhoods and rural areas to improve patient access to supportive care services.

Introduction

Prostate cancer is the most common visceral cancer in men in the United States (US).¹ With advances in radiation therapy (RT), traditional outcomes of survival are hypothesized to be similar across treatment modalities. Patient-reported outcomes (PROs) of adverse events, symptoms, and health-related quality of life (QOL) thus play a significant role in patient, clinician, and policy considerations in cancer clinical trials (CCTs).² Assessing PROs has been shown to enhance provider-patient communication and decision-making, as well as increasing survival rates.³ Adverse socioeconomic status at both the individual and the neighborhood levels is associated with poor PROs among various cancers, including prostate cancer.^4–6

Neighborhood socioeconomic and environmental deprivation characterized by concentration of impoverished, less educated people and by poor living conditions⁷ can contribute to poor PROs. Patients with cancer from rural areas tend to have worse outcomes than their urban counterparts⁸; living in a neighborhood with limited resources or a toxic environment is associated with poor physical functioning, high body pain, depression, and cognitive dysfunction in general populations.^9,10 Studies have also shown that neighborhood environmental factors, such as air pollution, are associated with greater odds of cancers^7,11 and cognitive dysfunction.¹² Interventions targeting rurality, neighborhood socioeconomic, and environmental factors can be devoted to improving cancer outcomes.¹³ If specific rural, socioeconomic, and environmentally deprived areas with poor survival and PROs are identified, supportive care interventions can be tailored to enhance cancer care delivery from an area-level perspective.^14,15

The effect of neighborhood socioeconomic and environmental deprivation on PROs and survival has not been systematically evaluated in specific cancers. The direction and magnitude of associations between neighborhood socioeconomic and environmental deprivation and cancer outcomes vary because of differences in study populations, geographic regions, choice of neighborhood measures, and geographic scales.⁷ Additionally, existing studies analyzed small samples at the regional scale with limited generalibility.⁶ Using 1 large trial NRG RTOG 0415, this study aimed to examine the associations of rurality (rural vs urban) and neighborhood socioeconomic (area deprivation index [ADI]) and environmental (particulate matter with a diameter less than 2.5 micrometers [PM_2.5] and nitrogen dioxide [NO₂]) deprivation with cancer treatment-related PROs (QOL, symptoms, and general health) and survival, adjusting for demographic, disease, and treatment-related factors.

Methods and Materials

Study design and population

A secondary data analysis was conducted in men with prostate cancer from the trial NRG RTOG 0415. We analyzed NRG RTOG 0415, a randomized phase 3 trial that sought to determine whether hypofractionated RT (HRT) of 70 Gy in 28 fractions was noninferior to conventional RT (CRT) of 73.8 Gy in 41 fractions with respect to disease-free survival (DFS). A total of 1092 randomized men (Fig. E1) were eligible for analysis with a median follow-up of 5.8 years.¹⁶ Eligible patients had organ-confined, biopsy-proven Gleason scores of 2 to 6 and prostate-specific antigen (PSA) of less than 10 ng/mL. Patients were ineligible if they had metastatic disease, had a Zubrod performance status score greater than 1, used antiandrogens, or had prior therapy for prostate cancer. In our analysis, patients from the Canadian sites were excluded because of the lack of ZIP code data similar to those in the US. All patient-reported instruments were administered at baseline (before start of therapy) and 6, 12, 24, and 60 months after the start of therapy.

Main outcomes and measures

Rural residence was defined using 2003 Rural-Urban Continuum Codes (RUCCs).¹⁷ RUCCs were matched with patient ZIP codes using 2010 US Postal Service ZIP Code Crosswalk Files from the US Department of Housing and Urban Development.¹⁸ The primary analysis divided the 9 RUCCs into urban (RUCCs 1–3) versus rural (RUCCs 4–9).^19–21

Neighborhood deprivation included both socioeconomic and environmental deprivation. Neighborhood socioeconomic deprivation was assessed using the ADI,^22–24 a factor-based index using 17 US Census income, education, employment, and housing indicators. The 17 ADI subscores were calculated based on the American Community Survey 5-year summary data.²⁵ A higher neighborhood percentile or decile indicates a more disadvantaged neighborhood. The ADI measure was split into quartiles based on previous work, and the patients considered the most deprived were defined as those in the highest ADI quartile (76%−100%).^26–28 Neighborhood environmental deprivation focused on the National Aeronautics and Space Administration satellite-based air pollution measures (PM_2.5 and NO₂). Both continuous variables were obtained from the National Aeronautics and Space Administration Socioeconomic Data and Applications Center, and the spatial resolution was 100 km. The 3-year average estimates were used for the clinical trial data set based on patient ZIP codes. Higher estimates of PM_2.5 and NO₂ indicate greater neighborhood environmental deprivation.

Prostate cancer−specific QOL was measured by the reliable and valid patient-reported Expanded Prostate Cancer Index Composite (EPIC).²⁹ EPIC is a 50-item measure designed to evaluate patient function after prostate cancer treatment using a Likert-like scale with responses transformed to a scale of 0 to 100. Higher scores indicate better QOL. The instrument includes 4 domains: urinary, bowel, sexual, and hormonal. Anxiety and depression were assessed with the 25-item Hopkins Symptom Checklist (HSCL).³⁰ The total score of the HSCL-25 ranges from 25 to 100, with higher scores indicating worse symptoms. A validated cutoff score of ≥44 was used to indicate a positive screen for anxiety or depression.^29,31 General health status was measured by the EuroQoL−5 Dimension questionnaire (EQ-5D). The EQ-5D is a 2-part questionnaire: the first part includes 5 items covering 5 dimensions: mobility, self-care, usual activities, pain or discomfort, and anxiety/depression. Each dimension is scored on 3 levels: no problems, moderate problems, and extreme problems. The second part is a 20-cm, 10-point visual analog scale (VAS) assessing current health state. The worst imaginable health state is scored as 0 at the bottom of the scale, and best imaginable health state is scored as 100.³²

Overall survival (OS) and DFS were obtained from the NRG RTOG 0415 data set. OS was measured from the date of randomization to the date of death; DFS was measured from the date of randomization to the date of first occurrence of local progression, distant progression, biochemical failure, or death from any cause.

Demographic variables included age at diagnosis, race, ethnicity, health insurance status, educational attainment, and marital status. Clinical variables included stages of prostate cancer diagnosis, treatment arm, PSA level, and Gleason score.

Statistical analysis

All analyses were performed using SAS statistical software (SAS Institute Inc, Cary, NC). The rurality, neighborhood socioeconomic (ADI), and environmental (PM_2.5 and NO₂ concentration) deprivation variables were mapped using ArcMap 10.7 based on patient ZIP codes. The associations of demographic and clinical variables with rurality (rural vs urban), ADI (25% most disadvantaged vs remaining 75%),^26–28 and environmental variables (PM_2.5 and NO₂ concentration) were examined using t test and χ² test.

Mixed effects models, using full information maximum likelihood estimation, were used to analyze the EPIC, HSCL-25, and EQ-5D continuous scores because of the presence of missing data. Generalized estimating equations were used for HSCL-25 (cutoff score ≥44 vs <44) and EQ-5D categories (no problems vs problems). Independent variables included rurality level, ADI, PM_2.5, NO₂, age at diagnosis, education status, marital status, race, ethnicity, and treatment arm. Collinearity diagnostics were conducted by means of the variance inflation factor for each independent variable. For significant effects, between group differences at each timepoint were conducted within the context of the model using t test. Reduced models were obtained by using a stepwise procedure at a significance level of 0.10.

Cox proportional hazards regression was used to determine whether rurality level (rural vs urban), ADI (25% most disadvantaged vs the remaining 75%), and environmental values were associated with OS and DFS, adjusting for age at study entry, education status, marital status, race, ethnicity, and treatment arm. Hazard ratios (HRs) and corresponding 95% confidence intervals (CIs) were presented to estimate associations of rurality and neighborhood socioeconomic and environmental deprivation with OS and DFS. Patient and clinical variables were tested separately in models with either ADI, rurality, or PM_2.5 and NO₂ and, if significant at the 0.10 level, were included in multivariable models with treatment arm and either ADI, rurality, or PM_2.5 and NO₂.

Results

Study population

A total of 751 patients with available US ZIP codes were analyzed after excluding participants from Canadian sites or with missing patient ZIP codes (Fig. 1). Most patients (92.7%) had a Zubrod performance status of 0 and 20.8% of participants had PSA less than 4 ng/mL. No significant differences were found in rurality, ADI scores, and baseline EPIC scores between CRT and HRT groups (Table E1). More patients randomized to receive HRT were above the threshold for anxiety/depression at baseline (10.8% vs 5.9%; P = .03). Patients randomized to HRT had a lower baseline EQ-5D VAS score compared with patients randomized to CRT (mean [standard deviation, SD] = 79.8 [15.8] vs mean [SD] = 82.2 [13.9]; P = .042) (Table E1).

Fig. 1.

CONSORT (Consolidated Standards of Reporting Trials) diagram.

Patients from the 25% most disadvantaged neighborhoods defined by ADI, compared with the remaining patients, were more likely to be non-White (39.1% vs 19.6%; P < .0001), with at most a high school degree or a General Educational Development diploma (51.7% vs 36.2%; P = .0003), not married (33.4% vs 22.7%; P = .0025), living in a rural area (25.9% vs 2.6%; P < .0001), with lower mean NO₂ concentration (mean [SD] = 2.6 [2.1] vs mean [SD] = 3.7 [3.0]; P < .0001) and lower mean PM_2.5 concentration (mean [SD] = 10.4 [2.4] vs mean [SD] = 10.8 [2.5]; P = .046) (Fig. E2). More patients from rural areas, compared with those in urban areas, were White (90.0% vs 72.5%; P = .007), of Zubrod status 1 (16.7% vs 6.5%; P = .004), and from the ADI top 25% (75.0% vs 18.7%; P < .0001). Rural areas experienced lower NO₂ and PM_2.5 concentrations versus urban areas (Table E2).

Neighborhood socioeconomic deprivation and rurality associated with baseline PROs

At baseline, patients from the top 25% of disadvantaged neighborhoods, compared with those from other areas, had lower bowel score (mean [SD] = 91.9 [10.7] vs mean [SD] = 93.9 [8.1]; P = .011), lower sexual scores (mean [SD] = 44.8 [25.4] vs mean [SD] = 49.9 [26.9]; P = .042), and lower hormonal scores (mean [SD] = 88.9 [12.9] vs mean [SD] = 91.4 [10.6]; P = .015) (Fig. 2A). There was no difference in urinary and anxiety/depression scores based on the ADI. Patients from the 25% most disadvantaged neighborhoods, compared with those from other areas, had more problems with self-care (4.5% vs 1.5%; P = .04) and more problems with pain (39.4% vs 32.5%; P = .047), but no differences for mobility and usual activities scores (Table 1).

Fig. 2.

Expanded Prostate Cancer Index Composite (EPIC) scores at baseline by neighborhood socioeconomic deprivation and rurality status. A indicates mean scores of EPIC by area deprivation index; B indicates mean scores of EPIC by rurality.

Table 1.

EQ-5D at baseline by neighborhood socioeconomic deprivation and rurality status

	ADI 75% (n = 577)	ADI top 25% (n = 174)		Total (n = 751)		P value*
Mobility						.07
I have no problems in walking about	428 (82.0%)	116 (75.3%)		544 (80.5%)
I have some problems in walking about	94 (18.0%)	38 (24.7%)		132 (19.5%)
Self-care						.04 †
I have no problems with self-care	513 (98.5%)	148 (95.5%)		661 (97.8%)
I have some problems washing or dressing myself	7 (1.3%)	7 (4.5%)		14 (2.1%)
I am unable to wash or dress myself	1 (0.2%)	0 (0.0%)		1 (0.1%)
Activity						.09†
I have no problems with performing my usual activities	460 (88.1%)	127 (81.9%)		587 (86.7%)
I have some problems with performing my usual activities	60 (11.5%)	27 (17.4%)		87 (12.9%)
I am unable to perform my usual activities	2 (0.4%)	1 (0.6%)		3 (0.4%)
Pain						.047
I have no pain or discomfort	353 (67.5%)	94 (60.6%)		447 (65.9%)
I have moderate pain or discomfort	160 (30.6%)	53 (34.2%)		213 (31.4%)
I have extreme pain or discomfort	10 (1.9%)	8 (5.2%)		18 (2.7%)
Anxiety						.76†
I am not anxious or depressed	408 (78.3%)	123 (79.9%)		531 (78.7%)
I am moderately anxious or depressed	109 (20.9%)	31 (20.1%)		140 (20.7%)
I am extremely anxious or depressed	4 (0.8%)	0 (0.0%)		4 (0.6%)
VAS score						.51‡
Mean (SD)	81.2 (15.2)	80.3 (13.9)		81.0 (14.9)
Index score						.09‡
Mean (SD)	0.9 (0.1)	0.9 (0.2)		0.9 (0.1)
	Rural (n = 60)	Urban (n = 691)		Total (n = 751)		P value*
Mobility						.25
I have no problems in walking about	41 (74.5%)	503 (81.0%)		544 (80.5%)
I have some problems in walking about	14 (25.5%)	118 (19.0%)		132 (19.5%)
Self-care						.37†
I have no problems with self-care	53 (96.4%)	608 (97.9%)		661 (97.8%)
I have some problems washing or dressing myself	2 (3.6%)	12 (1.9%)		14 (2.1%)
I am unable to wash or dress myself	0 (0.0%)	1 (0.2%)		1 (0.1%)
Activity						.64
I have no problems with performing my usual activities	46 (83.6%)	541 (87.0%)		587 (86.7%)
I have some problems with performing my usual activities	9 (16.4%)	78 (12.5%)		87 (12.9%)
I am unable to perform my usual activities	0 (0.0%)	3 (0.5%)		3 (0.4%)
Pain						.89
I have no pain or discomfort	36 (65.5%)	411 (66.0%)		447 (65.9%)
I have moderate pain or discomfort	17 (30.9%)	196 (31.5%)		213 (31.4%)
I have extreme pain or discomfort	2 (3.6%)	16 (2.6%)		18 (2.7%)
Anxiety						.34†
I am not anxious or depressed	44 (80.0%)	487 (78.5%)		531 (78.7%)
I am moderately anxious or depressed	10 (18.2%)	130 (21.0%)		140 (20.7%)
I am extremely anxious or depressed	1 (1.8%)	3 (0.5%)		4 (0.6%)
VAS score						.23‡
Mean (SD)	78.6 (12.5)	81.2 (15.1)		81.0 (14.9)
Index score						.75‡
Mean (SD)	0.9 (0.2)	0.9 (0.1)		0.9 (0.1)

Abbreviations: ADI = area deprivation index; EQ-5D = EuroQoL−5 Dimension; SD = standard deviation; VAS = visual analog scale. Bolded numbers indicate p-values < 0.05.

^*P value from χ² test.

^†P value from Fisher exact test.

^‡P value from 2-sided t test.

Patients from rural areas, compared with those from urban areas, had lower urinary scores (mean [SD] = 84.1 [13.6] vs mean [SD] = 87.7 [11.6]; P = .03) and sexual scores (mean [SD] = 38.2 [26.7] vs mean [SD] = 49.7 [26.4]; P = .003), but no difference in bowel and hormonal scores (Fig. 2B). There was no difference in anxiety/depression (7.8% vs 9.7%; P = .47) or any score from the EQ-5D (Table E2).

Rurality and neighborhood socioeconomic and environmental deprivation associated with longitudinal PROs

Longitudinal modeling showed no effect of rurality, ADI, or neighborhood PM_2.5 and NO₂ concentrations on any of the EPIC domain scores. There was also no difference in rurality, ADI, or air pollution values with respect to anxiety/depression. Both ADI and rurality were significantly associated with EQ-5D VAS score while PM_2.5 and NO₂ were not (Table 2). Because of collinearity among rurality, ADI, PM_2.5 and NO₂ concentrations, longitudinal models were run separately for each of these variables.

Table 2.

Reduced EQ-5D VAS mixed effects models

Model	Effect	Estimate	Standard error	P value
ADI	Intercept	40.73	2.89	<.001
	Time	−0.04	0.01	.0014
	Treatment arm (3D-CRT/IMRT 73.8 Gy)	0.21	0.95	.82
	Baseline EQ-5D VAS score	0.47	0.032	<.001
	ADI (75% vs top 25%)	3.23	1.13	.0043
Rurality	Intercept	42.46	2.88	<.001
	Time	−0.04	0.01	.0014
	Treatment arm (3D-CRT/IMRT 73.8 Gy)	−0.16	0.97	.87
	Baseline EQ-5D VAS score	0.46	0.03	<.001
	Marital status (married vs other)	2.55	1.12	.023
	Rurality 2013 (rural vs urban)	−5.41	1.77	.0024
Pollution	Intercept	42.53	3.52	<.001
	Time	−0.04	0.01	.0012
	Treatment arm (3D-CRT/IMRT 73.8 Gy)	0.036	0.98	.97
	Baseline EQ-5D VAS score	0.47	0.03	<.001
	Marital status (married vs other)	2.49	1.13	.03
	Mean PM2.5	−0.06	0.24	.79
	Mean NO2	−0.04	0.23	.87

Reference level is bolded in parentheses. Separate models were run for ADI, rurality, and neighborhood environmental deprivation (pollution) because of collinearity of these variables. Bolded numbers indicate p-values < 0.05.

ADI/rurality/pollution variables, along with time, treatment arm, and baseline EQ-5D VAS score were forced into the models.

Abbreviations: 3D-CRT = 3-dimensional conventional radiation therapy; ADI = area deprivation index; EQ-5D = EuroQoL−5 Dimension; IMRT = intensity modulated radiation therapy; NO₂ = nitrogen dioxide; PM_2.5 = particulate matter with a diameter less than 2.5 micrometers; VAS = visual analog scale.

Patients from the bottom 75% of ADI scores had higher EQ-5D VAS scores (ie, better general health) compared with patients from the top 25% more deprived neighborhoods (estimate = 3.23; standard error [SE] = 1.13; P = .0043). Tests of EQ-5D VAS score showed significance between ADI group differences at 6 months (mean difference = 2.95; SE = 2.88; P < .0095) and 12 months (mean difference = 2.67; SE = 1.14; P = .019) but not at 24 and 60 months (P > .05).

Rurality was significantly associated with EQ-5D VAS score (estimate = −5.41; SE = 1.77; P = .0024) longitudinally. Compared with patients from urban areas, patients from rural areas were significantly associated with lower EQ-5D VAS scores at 6 months (mean difference = −5.63; SE = 1.78; P = .0017), 12 months (mean difference = −5.90; SE = 1.78; P = .0011), 24 months (mean difference = −6.45; SE = 1.80; P = .0004), and 60 months (mean difference = −8.09; SE = 1.97; P < .0001).

Neighborhood environmental variables, NO₂ and PM_2.5, were not significantly associated with PROs longitudinally.

Associations of neighborhood socioeconomic and environmental deprivation and rurality with survival

Cox proportional hazards models were run separately for rurality, neighborhood ADI, and neighborhood environmental values because of collinearity. Zubrod of 0 had improved OS compared with Zubrod of 1 for each of the 3 models (P < .01; Table 3). Both rurality in favor of urban areas (HR, 1.81; 95% CI, 1.05–3.12; P = .033; Table 4, Fig. E3) and ethnicity in favor of non-Hispanics (HR, 0.45; 95% CI, 0.21–0.97; P = .042) were significantly associated with improved DFS; ADI (HR, 0.68; 95% CI, 0.56–1.01; P = .053) showed a marginal association with improved DFS (Table 4). PM_2.5 (HR, 1.06; 95% CI, 0.96–1.16; P = .25) was not associated with DFS but NO₂ was (HR, 0.90; 95% CI, 0.82–1.00; P = .047).

Table 3.

Multivariate reduced Cox proportional hazards models for overall survival

Outcome	Variable	Hazard ratio	95% CI	P value
Model 1: ADI	Zubrod (0 vs 1)	2.56	1.33–4.90	.0047
	ADI (75% vs top 25%)	0.88	0.51–1.51	.64
	Treatment arm (HRT vs CRT)	0.80	0.49–1.30	.36
Model 2: Rurality	Zubrod (0 vs 1)	2.51	1.30–4.86	.0064
	T stage (2 vs 1)	0.52	0.25–1.09	.081
	Rurality (rural vs urban)	1.52	0.74–3.11	.26
	Treatment arm (HRT vs CRT)	0.78	0.48–1.27	.33
Model 3: Pollution	Zubrod (0 vs 1)	2.60	1.36–4.99	.004
	T stage (2 vs 1)	0.51	0.24–1.07	.077
	Mean PM2.5 concentration	0.99	0.88–1.21	.91
	Mean NO2 concentration	0.93	0.82–1.06	.26
	Treatment arm (HRT vs CRT)	0.77	0.48–1.25	.29
	Patients deceased/total: 67/751

Reference level is bolded in parentheses. Because of missing values for ethnicity, models including ethnicity have fewer patients. All models were built using the following process: (1) models with either ADI, rurality, or pollution and 1 clinical/pretreatment characteristic were run; (2) those characteristics significant at the 0.10 level were included in the multivariable model. Bolded numbers indicate p-values < 0.05.

Abbreviations: ADI = area deprivation index; CI = confidence interval; CRT = conventional radiation therapy; HRT = hypofractionated radiation therapy; NO₂ = nitrogen dioxide; PM_2.5 = particulate matter with a diameter less than 2.5 micrometers; PSA = prostate-specific antigen.

Table 4.

Multivariate reduced Cox proportional hazards models for disease-free survival

Outcome	Variable		Hazard ratio	95% CI		P value
Model 1: ADI	PSA (<4 ng/mL vs 4-<10 ng/mL)		1.59	0.96–2.62		.072
	Ethnicity (white vs others)		0.46	0.22–1.00		.049
	ADI (75% vs top 25%)		0.68	0.56–1.01		.053
	Treatment arm (HRT vs CRT)		0.80	0.56–1.42		.21
Model 2: Rurality	PSA (<4 ng/mL vs 4-<10 ng/mL)		1.58	0.96–2.61		.074
	Ethnicity (white vs others)		0.44	0.20–0.94		.034
	Rurality (rural vs urban)		1.81	1.05–3.12		.033
	Treatment arm (HRT vs CRT)		0.80	0.56–1.15		.23
Model 3: Pollution	PSA (<4 ng/mL vs 4-<10 ng/mL)		1.60	0.97–2.64		.068
	Ethnicity (white vs others)		0.43	0.20–0.94		.033
	Mean PM2.5 concentration		1.06	0.96–1.16		.25
	Mean NO2 concentration		0.90	0.82–1.00		.047
	Treatment arm (HRT vs CRT)		0.78	0.54–1.12		.17
	Patients progressed or deceased/total: 122/683

Reference level is bolded in parentheses. Because of missing values for ethnicity, models including ethnicity have fewer patients. All models were built using the following process: (1) models with either ADI, rurality, or pollution and 1 clinical/pretreatment characteristic were run; (2) those characteristics significant at the 0.10 level were included in the multivariable model. Bolded numbers indicate p-values < 0.05.

Abbreviations: ADI = area deprivation index; CI = confidence interval; CRT = conventional radiation therapy; HRT = hypofractionated radiation therapy; NO₂ = nitrogen dioxide; PM_2.5 = particulate matter with a diameter less than 2.5 micrometers; PSA = prostate-specific antigen.

Discussion

Rural residence and neighborhood socioeconomic/environmental deprivation were found to be associated with worse PROs and DFS in men with prostate cancer from a national CCT. These findings demonstrated geographic disparities in PROs (eg, self-care, pain, and general health) and DFS among patients with low-risk prostate cancer. Our findings provide promising indicators of cancer care provision, particularly for those living in rural areas or socioeconomically deprived neighborhoods.

In this study, men with low-risk prostate cancer from rural areas had worse QOL (eg, urinary and sexual scores) pretreatment, lower general health across the CCT, and lower DFS (but no difference on OS) than those from urban areas. As reported in previous literature, patients from rural areas reported high rates of cancer-related mortality and other negative treatment outcomes.³³ Therefore, our results are concordant with prior work that documented worse survival for patients with cancer from rural areas.^20,34 Until now, the association between the rural-urban residence and PROs has been understudied. Depta et al³⁵ compared patient PROs by rural-urban residence in patients with head and neck cancer and found that patients from rural areas had more anxiety about their financial situation and expressed more loneliness. In this study, men with prostate cancer from rural areas, compared with their counterparts from urban areas, had worse QOL, particularly for baseline urinary and sexual domain scores, before treatment and longitudinally. Compared with urban areas, rural areas potentially have limited access to medical and oncology providers, long travel times, high out-of-pocket cancer costs, and accentuating disparities in access to health care and prevention services.^8,20,34 These barriers potentially explained the longitudinal effect of rural-urban residence on QOL across the trial. One recent study further showed that rural and urban patients receiving uniform access to cancer care through participation in a CCT had similar outcomes,⁸ suggesting that standardized access to cancer care can reduce the rural and urban disparities in cancer care outcomes. Regarding the worse QOL (eg, urinary and sexual scores) among rural patients before treatment, it is possibly related to privacy and confidentially concerns, lack of health insurance and poverty, and limited access to pharmacotherapy. More work is needed to confirm this finding and understand the rural-urban differences in QOL before treatment.

This study found that neighborhood socioeconomic deprivation (eg, top 25% ADI) was associated with poor PROs, such as worse QOL and general health status (ie, less self-care and more pain). This finding is consistent with a previous report indicating that patients with head and neck cancer residing in more deprived neighborhoods had worse social-emotional functioning and overall QOL after adjustment for demographic and clinical factors.³⁶ Recently, Rosenzweig et al³⁷ reported that a higher ADI (more deprivation) was associated with higher levels of anxiety among patients with advanced cancers. We did not find a significant effect of ADI on anxiety and depression, potentially because of our analysis of men with low-risk prostate cancer rather than patients with advanced cancers. Socioeconomically deprived neighborhoods are reported to have higher levels of environmental pollutants, overcrowding, and violence, and less social cohesion and access to services.³⁸ All these negative factors are linked to chronic stress and inflammation,^7,39 ultimately leading to worse QOL and general health status. Although previous work has reported the effect of individual socioeconomic status on CCTs,⁴⁰ this study highlighted the role of area-level socioeconomic deprivation in PROs across the cancer care continuum.⁴¹ The adverse effect of neighborhood ADI on cancer PROs supports the need for appropriate screening and evaluation strategies to implement neighborhood-based cancer care services, even among those on standardized CCTs.

Individuals from highly socioeconomically deprived neighborhoods showed poorer OS.^42,43 We found a similar result for DFS but not OS, which is possibly explained as a small number of participants who died of early-stage prostate cancer in this trial. This study found that rurality and neighborhood socioeconomic deprivation were significantly associated with DFS when ethnicity was included in the analyzing model, indicating multilevel factors involved in geographic disparities in cancer survival.⁴⁴ Currently, several multilevel conceptual frameworks have been tested to guide analysis of rural cancer outcomes,^45,46 including individual, interpersonal, community, and societal factors. Multilevel determinants of PROs and cancer survival remain unknown. Future studies should focus on the research gap on geographic disparities in cancer survival and PROs from a multilevel perspective.

This study found a longitudinal effect of rurality and neighborhood socioeconomic deprivation (ie, ADI) on EQ-5D VAS, with worsening EQ-5D VAS over time for patients from rural and the top 25% socioeconomically deprived areas. Both geographic measures were associated with variations in access and use of prostate cancer-related services for low- to high-risk disease,⁴⁷ thus, men residing in rural and socioeconomically deprived areas may not receive adequate follow-up care after cancer treatment,⁴⁸ and ultimately this results in worse VAS longitudinally. This finding suggested that supportive care services and interventions are needed to increase longitudinal PROs among patients with prostate cancer in rural and neighborhood deprived areas via leveraging access to screening, treatment, and ancillary health care resources.

Neighborhood environmental deprivation (ie, air population factors) has been linked with carcinogenesis and survival in multiple cancers, including lung, breast, and bladder cancers.⁴⁹ Previous work in cancer has primarily focused on PM_2.5 and NO_x gases,^50,51 while limited evidence was reported for prostate cancer risk.^52,53 In this study, we found that PM_2.5 and NO₂ were not associated with PROs and cancer survival among men with low-risk prostate cancer. This finding needs to be further confirmed in other cancer sites and metastatic diseases.

This study has several limitations. Our findings may not represent patterns of outcomes for patients with cancer outside of a trial setting.⁵⁴ Clinical trial participants have generally been shown to have better outcomes than those treated in a nontrial setting, possibly because of differences in baseline comorbidities or insurance status. Additionally, for all the area-level factors, calculations were based on patient ZIP codes, the level of data obtainable from NRG RTOG 0415. This can cause analytical challenges, as ZIP code boundaries change more frequently than county or census tracts and because of spatial mismatch between ZIP codes and ZIP code tabulation areas. On the other hand, ZIP code level geographic data provide a balance of geographic granularity with having too many area-level units that may contribute to sparse cells in multilevel analyses even in large clinical trials. Another limitation to the conclusions of this study is the potential for false positive results. Because this was an exploratory, retrospective analysis, no formal multiplicity adjustment was implemented. These results need to be tested in prospective clinical trials. Lastly, the neighborhood variable (ie, ADI) defined at the area-level cannot reflect the circumstance of an individual patient. Using the ADI as a screening tool should be done with caution, particularly among individuals with higher education, income, and better jobs than their neighbors in the same ZIP code.

Conclusion

Patients with low-risk prostate cancer from the socioeconomically deprived neighborhoods and rural areas had low QOL before CCTs, poor general health, and low DFS. Future CCTs should assess area-level factors (ie, rurality and neighborhood deprivation), screen patients based on adverse geographic locations, and incorporate multilevel interventions to improve patient access to supportive care services and cancer care outcomes.