Neighborhood effect and obesity in adult survivors of pediatric cancer: A report from the St. Jude Lifetime Cohort Study

Abstract

Survivors of childhood cancer are at risk for obesity, a condition potentially modifiable if dietary intake and physical activity are optimized. These health behaviors are likely influenced by neighborhood of residence, a determinant of access to healthy, affordable food and safe and easy exercise opportunities. We examined associations between neighborhood level factors and obesity among survivors in the St. Jude Lifetime cohort and community comparison group members. Persons with residential addresses available for geocoding were eligible for analysis [N=2265, mean age 32.5 (SD 9.1) years, 46% female, 85% white]. Survivors completed questionnaires regarding individual behaviors; percent body fat was assessed via dual x-ray absorptiometry (obesity: ≥25% males; ≥35% females); neighborhood effect was characterized using census tract of residence (e.g. neighborhood socioeconomic status (SES), rurality). Structural equation modeling was used to determine associations between neighborhood effect, physical activity, diet, smoking, treatment exposures, and obesity. Obese survivors (n=1420, 62.7%) were more likely to live in neighborhoods with lower SES (RR:1.23, 95% CI: 1.10–1.38) and rural areas (RR:1.22, 95% CI: 1.07–1.39) compared to survivors with normal percent body fat. Resource poor neighborhoods (standardized effect: 0.06, p<0.001) and cranial radiation (0.16, p<0.001) had direct effects on percent body fat. Associations between neighborhood of residence and percent body fat were increased (0.01, p=0.04) among individuals with a poor diet. Neighborhoods where survivors reside as an adult is associated with obesity. Interventions targeting survivors should incorporate strategies that address environmental influences on obesity.

INTRODUCTION

Improvements in treatment for pediatric cancer have resulted in 83% of children diagnosed with a malignancy surviving at least 5 years.¹ As these survivors age, they are at risk for chronic conditions including obesity², diabetes³, and metabolic syndrome.² Getting adequate amounts of physical activity and consuming a healthy diet may either prevent or ameliorate these conditions in childhood cancer survivors. In fact, meeting the US Centers for Disease Control and Prevention (CDC) physical activity guidelines substantially reduces risk of obesity⁴, incidence of cardiovascular events⁵ and mortality⁶ among survivors. Unfortunately, childhood cancer survivors are generally inactive⁷, with 52% reporting not meeting guidelines.⁸ Studies also indicate that survivors have poor adherence to recommended dietary guidelines, including those endorsed by the American Cancer Society (ACS)² and the American Institute for Cancer Research (AICR).⁹ Poor adherence to these guidelines is associated with increased risk for metabolic syndrome,⁹ while better diet quality is associated with less obesity¹⁰ in survivors.

Environmental factors, also referred to as ‘neighborhood effect’,¹¹ influence obesity in the general population,¹² where a growing body of evidence indicates an association between access to exercise facilities, safe spaces for physical activity,¹³ and healthy food options,¹⁴ and health behaviors. Thus, the term ‘obesogenic environment’ is used to describe neighborhood attributes that foster obesity in specific geographic areas.¹⁵ For example, living in a neighborhood in close proximity to parks and green space,^13,16,17 and/or that have fitness and recreational facilities^16,17 promotes physical activity, while living in neighborhoods with limited access to farmers’ markets or grocery stores¹⁸ or that have many fast food restaurants¹⁹ promotes suboptimal dietary intake and increased adiposity. Recent work in female adult breast cancer survivors has established an association between built environment factors, neighborhood socio-economic status (nSES), and obesity,^20,21 with rural cancer survivors reporting more inactivity when compared to urban cancer survivors.²²

To date, no study has looked specifically at how neighborhood factors (both social and ecological) contribute to obesity in survivors of childhood cancer. Therefore, we sought to explore associations between neighborhood effect, individual health behaviors (e.g. physical activity, diet, smoking), treatment and obesity utilizing a clinically assessed cohort of adult survivors of childhood cancer and non-cancer community comparison group members (hereafter referred to as controls). First, we aimed to describe differences in obesity, assessed using percent body fat, in childhood cancer survivors across demographic, treatment, health behavior and neighborhood level factors. Second, we aimed to assess the direct and indirect effects of treatment, individual health behaviors and neighborhood effect on percent body fat in survivors and controls. We hypothesized that treatment and health behaviors would have direct effects on percent body fat while neighborhood factors would have indirect and direct effects in survivors.

MATERIALS AND METHODS

Study Population

Participants were members of the St. Jude Lifetime Cohort (SJLIFE), a study designed to characterize health outcomes among survivors of pediatric malignancies. Study design, assessments, and cohort characteristics have been previously described.²³ Briefly, participants undergo an on campus medical assessment and complete comprehensive questionnaires that assess health behaviors, psychosocial functioning, and demographic factors following detailed abstraction of treatment information from individual medical records. Participants do not differ markedly from those who decline participation in terms of socio-demographic factors and cancer related variables such as age at diagnosis, cancer type, and survival time.²⁴ Survivors eligible for this analysis included those diagnosed with a childhood cancer for which they received treatment at St. Jude Children’s Research Hospital (SJCRH) between 1962 and 1999; survived ≥10 years from diagnosis; were age ≥18 years at time of assessment; had informative addresses that could be geocoded to the census tract level; and were not missing any important outcome (e.g. obesity) or predictor data (e.g. individual health behaviors, neighborhood level data). Controls were volunteers from the community or non-first-degree relatives or friends of SJCRH patients who had full health evaluations on campus; and who met the same inclusion criteria as survivors except they had no history of childhood cancer. This study was reviewed and approved by the SJCRH institutional review board (IRB) and all survivor and control participants provided informed consent prior to participation.

Obesity Outcome of Interest

Due to the superiority of dual x-ray absorptiometry (DEXA) derived body composition values in classifying obesity in our childhood cancer survivor population²⁵, obesity was measured using DEXA percent body fat assessed in the total body scanning mode. The scanner was calibrated weekly with known phantoms to minimize machine drift. Body regions were isolated from each other using regional computer-generated default lines, with manual adjustment as described by Kim et al.²⁶ To compare obesity status between survivors, percent body fat was dichotomized based on established cut points in the literature: men ≥25% body fat and women ≥35%²⁷ were considered obese. Percent body fat was used as a continuous variable in structural equation modeling, adjusting for sex.

Predictor Variables

Treatment, Demographic and Questionnaire Data

Detailed treatment exposures were available on all participants including information on radiation, chemotherapy doses, and surgical procedures. All participants completed detailed questionnaires collecting demographic information and health behaviors. Clinical and treatment factors associated with increased risk of obesity (age at diagnosis, cranial radiation, glucocorticoid exposure, and presence of a tumor in the hypothalamic-pituitary-adrenal (HPA) region of the brain) and a decreased risk of obesity (pelvic/abdomen radiation exposure) were considered for this analysis. Dexamethasone dose was converted to a prednisone equivalent dose (multiplied by 6.67)²⁸ and summed with prednisone to create an overall cumulative glucocorticoid dose.

Individual Factors

Physical activity was assessed by asking participants how many days (and duration in minutes) in a typical week they engaged in either moderate or vigorous activity. For moderate physical activity, participants were assigned one minute for each reported minute and for vigorous activity, participants were assigned 1.67 minutes for each reported minute.⁹ Moderate and vigorous values were summed to derive weekly minutes of moderate to vigorous physical activity (weekly MVPA minutes). Individuals were also classified based on meeting the Centers for Disease Control physical activity guidelines (≥150 minutes of MVPA/week vs. <150 minutes MVPA/week).

Diet was assessed using the 110 item full length Block Food Frequency Questionnaire²⁹ to estimate customary intake of nutrients and food groups. The data are processed using a food list from NHANES dietary recalls and a nutrient database from the USDA Food and Nutrient Database. Diet quality was characterized with the Healthy Eating Index (HEI) which maps reported dietary intake to 13 components reflecting food groups and key recommendations in the 2015–2020 Dietary Guidelines for Americans.³⁰ The score ranges from 0–100 with 100 indicating complete adherence.

Smoking was assessed using values from self-reported questionnaires. If a participant endorsed smoking > 100 cigarettes in their lifetime, they were asked three follow up questions to calculate pack-years: age started smoking, age stopped smoking (if quit), and how many cigarettes smoked per day on average.

Depressive symptoms were assessed using the depression subscale of the Brief Symptom Inventory-18³¹, a scale that measures psychological distress using items that evaluate symptoms over the previous 7 days and has been previously validated in childhood cancer survivors.³² Raw scores were converted to T-scores based on community normative data, with higher T-scores indicating more depressive symptoms.

Neighborhood Effect

Neighborhood effect was characterized using publicly available data sources of neighborhood factors that could be linked geographically to a participant’s place of residence and contained data that spanned the conterminous United States due to the wide distribution of our study population. Neighborhood factors considered were access to exercise opportunities, access to healthy food options, nSES and rurality. Once linked, neighborhood factors were categorized into quintiles, with higher quintiles indicating a ‘resource poor’ neighborhood environment compared to lower quintiles.

Access to exercise opportunities was assessed using a metric developed by Roubal et al³³ that measured the percentage of individuals in a county that had access to exercise locations such as parks and recreational facilities. This metric is used in the County Rankings & Roadmaps program³⁴, a publicly available resource provided through a collaboration between the Robert Wood Johnson Foundation and the University of Wisconsin Population Health Institute.

Access to healthy food options was assessed using a metric from the USDA Food Access Research Atlas³⁵ that describes the percent of people with low access to supermarkets in a census tract. This was defined as the percentage of people in a tract living at least one mile from the nearest supermarket, supercenter, or large grocery store for urban areas and living within at least 10 miles for rural tracts.

The nSES was measured using the Yost SES Index³⁶, which utilizes seven components from census data³⁷ linked to the census tract level: education index³⁸, percent persons living 200% below poverty line, median household income, median house value, median rent, percent blue collar workers, and percent older than 16 in the workforce without a job. The Yost SES Index has been used in recent analyses that examined associations between body composition and neighborhood level factors in breast cancer survivors.^21,39 Principal component analysis was used to develop a weighted linear combination of variables from the study data to create a value for each census tract.

Rurality was determined using the USDA Rural-Urban Commuting Area (RUCA) codes⁴⁰ and categorized as metropolitan area, micropolitan area (10,000 to 49,999 population), small town (2500 to 9,999 population), and rural area.

Data Availability

The data that support the findings of this study are available from the corresponding author upon reasonable request.

Analytical Approach

Descriptive statistics were used to characterize the study population, with characteristics of the participants and controls compared using chi-square or two sample t-tests as appropriate. Survivors were categorized based on obesity status, and log-binomial regression was used to evaluate the unadjusted associations between demographic, treatment, health behaviors and neighborhood level factors and high body fat status.

Access to exercise opportunities, nSES, rurality and access to healthy food outlets were evaluated with exploratory factor analysis to create the latent construct of neighborhood. This approach allows inclusion of correlated variables that represent the same construct and accounts for measurement error in each indicator. Meaningful factors were identified using the principal factor method and a scree test. Neighborhood factors with a loading greater than 0.40 were retained to represent the latent construct of neighborhood.

Structural equation modeling (SEM) was used to evaluate the effects of demographic, treatment, individual and neighborhood factors on obesity. Our initial theoretical model is shown in Figure 1 for both the survivor and control populations, with “neighborhood effect” treated as a latent variable in the models. The initial theoretical model was selected a priori based on known risk factors for obesity in childhood cancer survivors, and on known associations between area level social and environmental factors and obesity in the general population. Models were fit separately for each of the participant groups (survivors and controls) with SAS 9.4 software, using a twostep process⁴¹. In the first step, confirmatory factor analysis was used to test the relationship between the measured and latent variables to develop the measurement model. In the second step, the measurement model was modified to a structural equation model which tests the relationship between the independent and dependent variables.

Figure 1.

Initial theoretical model for childhood cancer survivors indicating the hypothesized influence of neighborhood level factors, individual health behaviors, and treatment exposure on increases in percent body fat. Note that the model tested for the control group is identical to this model, but without treatment exposures included.

Confirmatory factor analysis (CFA) was used to determine the best measurement model for percent body fat for the survivors and controls to assure that the model adequately fit the data. The measurement models were modified as necessary based on the psychometric properties discovered in CFA to achieve an adequate fit. SEM was then used to test the association between individual physical activity, individual diet score, pack-years of smoking, neighborhood effect, age at assessment, depression, sex, treatment (survivors only) and percent body fat for both survivors and controls. Model fit statistics were compared between the initial and final models and revised using maximum likelihood, fit statistics, and standardized factor loadings until a meaningful and statistically acceptable model was achieved.

For goodness-of-fit indices, we used an absolute index (SRMR: standardized root mean square residual), two incremental indexes (CFI: comparative fit index and the NFI: Bentler-Bonnett normed fit index) and a parsimony index (RMSEA: root mean square error of approximation), to determine the overall fit of the model, the improvement in fit over the null model, and the fit adjusted for parsimony⁴². For SRMR and RMSEA, values ≤ 0.055 are considered ideal, with a narrow 90% confidence interval for the RMSEA⁴³. For NFI and CFI, values ≥ 0.94 suggest excellent fit.

RESULTS

Of the 3,491 survivors who had informative addresses that could be geocoded to the tract level, 56 were lost to follow up, 589 refused participation, 194 completed surveys only, and 387 had missing outcome or important indicator data, leaving 2,265 survivors for analysis (Supplementary Figure 1). Of the 360 controls with informative addresses, 42 were missing outcome or important indicator data, leaving 318 controls for this analysis. Survivors and controls differed across all demographic, individual and neighborhood level characteristics except access to healthy food outlets, body mass index (BMI), pack-years of smoking and weekly MVPA (Table 1). Comparisons of survivor participants to survivor non-participants can be found in Supplementary Table 1.

Table 1.

Comparison of survivor participants and controls by continuous and categorical measures

	Participants n=2265		Controls n=318		P-value
	Mean	SD	Mean	SD
Age at assessment	32.5	9.1	35.1	10.2	<0.001
Obesity measures
Percent body fat	32.00	9.23	27.42	9.63	<0.001
Waist to height ratio (WHtR)	0.54	0.10	0.52	0.11	0.06
Body mass index kg/m2	28.6	7.2	28.6	7.7	0.96
Individual health behaviors
Weekly minutes MVPAa	416.4	699.2	375.6	50.10	0.20
Healthy eating index	58.05	12.15	60.70	12.41	0.0003
Smoking pack years	2.52	6.59	2.77	6.94	0.54
Depressive symptomsb	49.73	10.28	47.12	8.38	<0.001
Sex	N	%	N	%
Male	1217	53.7	141	44.3	0.002
Female	1048	46.3	177	55.7
Race / Ethnicity
Non-Hispanic white	1899	83.8	282	88.7	<0.001
Non-Hispanic black	317	14.0	19	5.9
Non-Hispanic other	23	1.0	6	1.9
Hispanic	26	1.2	11	3.5
Health Behaviors
Met CDC guidelines for physical activityc
No	1080	47.68	119	37.42	<0.001
Yes	1185	52.32	199	62.58
Healthy eating indexd
Quartile 1	582	25.70	63	19.81	0.02
Quartile 2	575	25.39	71	22.33
Quartile 3	560	24.72	86	27.04
Quartile 4	548	24.19	98	30.82
Smoking packyears
0	1596	70.46	221	69.50	0.81
0.1–19.9	592	26.14	84	26.42
20.0+	77	3.40	13	4.09
Depressive symptomse
No	1939	85.61	298	93.71	<0.001
Yes	326	14.39	20	6.29
Neighborhood Effects
Percent access to exercise opportunitiesf
Quintile 5	471	20.8	48	15.1	<0.001
Quintile 4	469	20.7	52	16.4
Quintile 3	459	20.3	49	15.4
Quintile 2	483	21.3	129	40.6
Quintile 1	383	16.9	40	12.6
Percent access to supermarketsg
Quintile 5	440	19.4	76	23.9	0.37
Quintile 4	461	20.4	57	17.9
Quintile 3	455	20.1	60	18.9
Quintile 2	205	9.1	25	7.9
Quintile 1	704	31.0	100	31.5
Urbanicityh
Rural	80	3.5	4	1.3	<0.001
Small town	211	9.4	19	5.9
Micropolitan	313	13.8	24	7.6
Metropolitan	1661	73.3	271	85.2
Yost Socioeconomic Indexi
Quintile 5	468	20.7	46	14.5	<0.001
Quintile 4	471	20.8	43	13.5
Quintile 3	466	20.6	51	16.0
Quintile 2	456	20.1	63	19.8
Quintile 1	404	17.8	115	36.2

^a:MVPA: moderate to vigorous physical activity

^b:Measured using the depression subscale of the Brief Symptom Inventory-18. Represented as a T-score, with a mean of 50 and a standard deviation of 10.

^c:CDC: Centers for Disease Prevention and Control. Meeting guidelines defined as ≥150 min/week of moderate or vigorous activity.

^d:Higher quartiles indicate better diet quality.

^e:Measured using the depression subscale of the Brief Symptom Inventory-18. Represented as a T-score, with a mean of 50 and a standard deviation of 10.

^f:Percent in county of survivor’s residence who have access to an exercise venue such as a park, forest, or recreational facility. Based on a measure developed by Roubal et al and used in the County Rankings and Roadmaps project (www.countyrhealthankings.org). Higher quintile indicates less access.

^g:Based on a measure developed by the United States Department of Agriculture (USDA) and represents the percentage of people who have access to a supermarket in survivor’s census tract of residence. Higher quintile indicates less access.

^h:Defined using the USDA’s Rural-Urban Commuting Code classifications.

^i:Developed using seven census variables that are combined to create a weighted linear combination. Higher quintiles indicate poorer socio-economic status in the survivor’s place of residence.

The comparison of host factors, treatment, health behaviors and neighborhood effect for survivors by percent body fat status are presented in Table 2. Figure 2 shows the geographical distribution for survivors by percent body fat status. The risk for high percent body fat was increased among those age ≥ 30 years compared to those <30, and those who received cranial radiation, exposure to glucocorticoids, and had a tumor in the HPA region. Survivors who did not meet the CDC physical activity guidelines (<150 minutes MVPA/week) were at increased risk of higher percent body fat. Survivors who lived in a rural area and had lower nSES were also at an increased risk of having higher percent body fat.

Table 2.

Unadjusted comparisons between survivors by body fat status

	Survivors, n=2265
	Normal body fat	High body fata
	n=845	n=1420
	Mean (SD)	Mean (SD)	p-value
Age at assessment
Treatment
Maximum cranial radiation dose, Gray	6.46 (15.45)	12.28 (19.69)	<0.001
Maximum pelvic dose, Gray	4.04 (10.44)	4.64 (10.98)	0.20
Maximum abdomen dose, Gray	4.38 (10.13)	5.13 (10.80)	0.09
Cumulative glucocorticoids dose, mg/m2 b	2549.6 (4631.2)	2865.8 (4670.5)	0.12
Health behaviors
Weekly minutes MVPAc	552.0 (806.5)	335.7 (613.0)	<0.001
Healthy eating index	58.6 (12.88)	57.72 (11.68)	0.09
Smoking pack years	2.40 (6.16)	2.60 (6.83)	0.48
Depressive symptomsd	48.98 (9.96)	50.19 (10.43)	0.007
			Relative Risk	95% CI
Individual level	N (%)	N (%)
Age at diagnosis
0–4 years	329 (38.9)	505 (35.6)	1.14	0.73–1.80
5–9 years	187 (22.1)	317 (22.3)	1.19	0.76–1.87
10–14 years	194 (23.0)	351 (24.7)	1.22	0.77–1.91
15–20 years	127 (15.0)	238 (16.8)	1.23	0.78–1.94
20+	8 (1.0)	9 (0.6)	Reference
Age at assessment
18–29	441 (52.2)	575 (40.5)	Reference
30–39	257 (30.4)	494 (34.8)	1.16	1.08–1.25
40–49	116 (13.7)	291 (20.5)	1.26	1.16–1.37
50–59	31 (3.7)	60 (4.2)	1.17	1.00–1.36
Sex
Male	465 (55.0)	752 (52.9)	Reference
Female	380 (45.0)	668 (47.1)	1.03	0.97–1.10
Race / Ethnicity
Non-Hispanic White	695 (82.3)	1204 (84.8)	Reference
Non-Hispanic Black	126 (14.9)	191 (13.4)	0.95	0.86–1.05
Non-Hispanic Other	12 (1.4)	11 (0.8)	0.75	0.49–1.16
Hispanic	12 (1.4)	14 (1.0)	0.85	0.59–1.21
Cranial radiation dosimetry
None	661 (78.2)	858 (60.4)	Reference
< 20 Gray	75 (8.9)	186 (13.1)	1.26	1.15–1.38
≥ 20 Gray	109 (12.9)	376 (26.5)	1.37	1.29–1.46
Pelvic or abdomen radiation exposure
No	665 (78.7)	1079 (76.0)	Reference
Yes	180 (21.3)	341 (24.0)	1.06	0.98–1.14
Glucocorticoids exposure
No	505 (59.8)	741 (52.2)	Reference
Yes	340 (40.2)	679 (47.8)	1.12	1.05–1.19
Tumor in HPA axise
No	840 (99.4)	1369 (96.4)	Reference
Yes	5 (0.6)	51 (3.6)	1.47	1.35–1.61
Met CDC guidelines for physical activityf
No	309 (36.6)	771 (54.3)	1.30	1.22–1.39
Yes	536 (63.4)	649 (45.7)	Reference
Healthy eating indexg
Quartile 1	219 (25.92)	363 (25.56)	1.07	0.98–1.18
Quartile 2	199 (23.55)	376 (26.48)	1.13	1.03–1.24
Quartile 3	197 (23.31)	363 (25.56)	1.12	1.02–1.23
Quartile 4	230 (27.22)	319 (22.39)	Reference
Smoking packyears
0	569 (67.3)	1027 (72.3)	Reference
0.1 – 19.9	254 (30.1)	338 (23.8)	0.89	0.82–0.96
20.0+	22 (2.6)	55 (3.9)	1.11	0.96–1.28
Depressive symptomsh
No	736 (87.1)	1203 (84.7)	Reference
Yes	109 (12.9)	217 (15.3)	1.07	0.99–1.17
Neighborhood level
Percent access to exercise opportunitiesi
Quintile 5	169 (20.0)	302 (21.3)	1.09	0.98–1.21
Quintile 4	159 (18.8)	310 (21.8)	1.12	1.01–1.25
Quintile 3	184 (21.8)	275 (19.4)	1.02	0.91–1.14
Quintile 2	176 (20.8)	307 (21.6)	1.07	0.97–1.20
Quintile 1	157 (18.6)	226 (15.9)	Reference
Percent access to supermarketsj
Quintile 5	154 (18.2)	286 (20.1)	1.07	0.98–1.17
Quintile 4	165 (19.5)	296 (20.9)	1.06	0.97–1.16
Quintile 3	164 (19.4)	291 (20.5)	1.05	0.96–1.16
Quintile 2	85 (10.1)	120 (8.5)	0.97	0.85–1.10
Quintile 1	277 (32.8)	427 (30.0)	Reference
Urbanicityk
Rural	20 (2.4)	60 (4.2)	1.22	1.07–1.39
Small town	69 (8.2)	142 (10.0)	1.09	0.99–1.21
Micropolitan	118 (13.9)	195 (13.7)	1.01	0.92–1.11
Metropolitan	638 (75.5)	1023 (72.0)	Reference
Socioeconomic Indexl
Quintile 5	158 (18.7)	310 (21.8)	1.23	1.10–1.38
Quintile 4	170 (20.1)	301 (21.2)	1.19	1.06–1.33
Quintile 3	165 (19.5)	301 (21.2)	1.21	1.07–1.65
Quintile 2	165 (19.5)	291 (20.5)	1.18	1.06–1.33
Quintile 1	187 (22.1)	217 (15.3)	Reference

^a:≥ 35% for females and ≥ 25% for males

^b:Dexamethasone dose was converted to a prednisone equivalent dose (multiplied by 6.67) and summed with prednisone to create an overall cumulative glucocorticoid dose.

^c:MVPA: moderate to vigorous physical activity.

^d:Measured using the depression subscale of the Brief Symptom Inventory-18. Represented as a T-score, with a mean of 50 and a standard deviation of 10. Note that the relative risk of using depression medication was 1.02 (95% CI 0.99–1.05) for survivors with high body fat versus normal body fat.

^e:Hypothalamic-pituitary-adrenal axis

^f:CDC: Centers for Disease Prevention and Control. Meeting guidelines defined as ≥150 min/week of moderate or vigorous activity.

^g:Higher quintiles indicate better diet quality.

^h:Measured using the depression subscale of the Brief Symptom Inventory-18. A T-score ≥ 63 indicates high depressive symptoms.

^i:Percent in county of survivor’s residence who have access to an exercise venue such as a park, forest, or recreational facility. Based on a measure developed by Roubal et al and used in the County Rankings and Roadmaps project (www.countyrhealthankings.org). Lower quintile indicates more access.

^j:Based on a measure developed by the United States Department of Agriculture (USDA) and represents the percentage of people who have access to a supermarket in survivor’s census tract of residence. Lower quintile indicates more access.

^k:Defined using the USDA’s Rural-Urban Commuting Code classifications.

^l:Developed using seven census variables that are combined to create a weighted linear combination. Lower quintiles indicate better socio-economic status in the survivor’s place of residence.

Figure 2.

The geographical distribution of childhood cancer survivors by obesity status. Survivors with high body fat are indicated in blue while those with normal body fat are indicated in yellow. High body fat was defined as ≥25% for males and ≥35% for females.

In exploratory factor analysis, access to exercise opportunities, nSES and rurality were related to the neighborhood effect construct. Access to healthy food outlets did not contribute to the neighborhood effect construct and was excluded from further analysis. Standardized factor loadings, standardized path coefficients and composite reliabilities for both survivor and control percent body fat models are presented in Supplementary Table 2. The composite reliabilities were similar across both survivor (0.73) and control model (0.73) and reflected adequate internal consistency of the indicators measuring the neighborhood effect construct (>0.69).

Figure 3 presents the estimated direct effects of neighborhood and individual level factors on percent body fat for survivors. CFA indicated that pack-years of smoking, and pelvic/abdomen radiation did not contribute to model fit and that improved fit was obtained by removing estimates between neighborhood and physical activity and between depressive symptoms and physical activity. Five revisions were made to obtain the best fitting model. Fit indices are shown in Table 3. The model indicated that neighborhood, sex, cranial radiation dose, glucocorticoid dose, a tumor located in the HPA region, diet, weekly MVPA, and age had direct effects on percent body fat. The associations between neighborhood and percent body fat and depression and percent body fat were mediated by diet. The associations between age and percent body fat and sex and percent body fat were mediated by weekly MVPA and diet. Standardized direct effect sizes are indicated in Figure 3; standardized direct and indirect effects can also be found in Supplementary Table 3. All significant effects were in the expected direction and model fit indices indicated an excellent fit of our data to the final model (Table 3).

Figure 3.

Estimated direct effects of neighborhood factors, individual level factors and treatment exposures on percent body fat in childhood cancer survivors in the St. Jude Lifetime Cohort study (N=2,265).

Table 3.

Fit indices for measurement and final theoretical models for survivors and controls – percent body fat as outcome

Models
	Chi-square	DF	CFI	SRMR	RMSEA	RMSEA LCL	RMSEA UCL	NFI
Survivors
Baseline model	3998.61	66
Uncorrelated model	708.55	65	0.84	0.05	0.06	0.06	0.07	0.82
Measurement model	202.74	37	0.96	0.03	0.04	0.04	0.05	0.95
Final model	106.21	27	0.98	0.02	0.04	0.03	0.04	0.97
Controls
Baseline model	514.96	45
Uncorrelated model	44.27	23	0.95	0.05	0.05	0.03	0.08	0.91
Measurement model	30.94	17	0.97	0.03	0.05	0.02	0.08	0.94
Final model	13.40	14	1.00	0.03	0.00	0.00	0.05	0.97

DF: Degrees of Freedom; CFI: Confirmatory Fit Index; SRMR: Standardized Root Mean Square Residual; RMSEA: Root Mean Square Error of Approximation; LCL: Lower Confidence Limit; UCL: Upper Confidence Limit; NFI: Normed Fit Index

The estimated direct effects of neighborhood and individual level factors on percent body fat for controls can be found in Supplementary Figure 2. CFA indicated that pack-years of smoking and depression did not contribute to model fit and that improved fit was obtained by removing estimates between neighborhood and weekly MVPA as well as between age and weekly MVPA. Four revisions were made to obtain the best fitting model. Fit indices are shown in Table 3. This model indicated that neighborhood, diet, weekly MVPA, and age had direct effects on percent body fat. Further, the association between neighborhood and percent body fat was mediated by diet, while the association between age and percent body fat were mediated by diet and the association between sex and percent body fat was mediated by weekly MVPA. Standardized direct effect sizes are indicated in Supplementary Figure 2; standardized direct and indirect effects can also be found in Supplementary Table 3. All significant effects were in the expected direction and model fit indices indicated an excellent fit of our data to the final model (Table 3).

DISCUSSION

Survivors of childhood cancer are at increased risk for obesity and associated chronic conditions as they age. Interventions to prevent and remediate obesity are needed. In this first study of neighborhood effects on obesity in survivors, we found that the neighborhood in which a childhood cancer survivor resides as an adult is associated with percent body fat. Comparing obese to non-obese survivors, we found that survivors who lived in a rural area and had lower nSES were at an increased risk of having higher percent body fat. Our final SEM indicated that a resource poor neighborhood, female sex, higher doses of cranial radiation or glucocorticoids, presence of a tumor in the HPA region, poorer diet scores, lower weekly MVPA, higher depressive symptoms and increasing age were associated with increases in percent body fat. These data provide useful information that should be considered in clinical and research programs promoting healthy weight maintenance and weight reduction during and after completion of childhood cancer therapy.

Results from our analysis examining differences in percent body fat across neighborhood level factors within pediatric cancer survivors are consistent with recent work in adult cancer survivor populations. Shariff-Marco et al²¹ found that, in a cohort of 4,354 female breast cancer survivors, survivors with a BMI ≥ 30 kg/m² were more likely to reside in a census block group with lower nSES compared to their normal/underweight counterparts (Odds ratio [OR] 2.32, 95% Confidence interval [95% CI]: 1.55–2.04). Earlier work from this group found that lower nSES was associated with an increased waist-to-hip ratio (OR: 2.54, 95% CI: 1.26–5.11)²⁰ in female breast cancer survivors. The influence of rurality on physical inactivity and obesity in adult cancer survivors has been previously documented²² while access to exercise opportunities has been associated with obesity^12,15 in the general population.

Our structural equation model results indicated that living in a resource poor neighborhood was associated with increases in percent body fat, even after accounting for known health behaviors, demographic and treatment exposures associated with obesity in this population. In addition, our final model found that individual diet accounted for indirect effects of neighborhood environment and female sex on percent body fat. Specifically, survivors who lived in resource poor neighborhoods had lower quality individual diets, which in turn influenced increases in percent body fat. For female survivors, we found that higher diet quality correlated with decreases in percent body fat. Interestingly, this association persisted despite the omission of the access to healthy food options measure from the latent construct of neighborhood. This measure was based on the distance between place of residence and supermarkets to define lack of access to healthy food; however, this access measure can vary based on neighborhood characteristics⁴⁴ and several studies show that opening new supermarkets in these areas did not improve healthy food availability or BMI.⁴⁵ Thus access does not necessarily translate into better individual diets.⁴⁵ Recent literature suggests that future investigations begin to look at ‘food swamps’ or areas with a high-density of stores selling unhealthy foods (i.e. fast food restaurant and convenience stores)⁴⁶ when examining associations between food environments and health outcomes. Nonetheless, the association between higher nSES and better diet quality has been described in the general population⁴⁷; to our knowledge, this is the first study to correlate neighborhood factors with diet quality in survivors of childhood cancer.

The control and survivor models were similar in that resource poor neighborhoods, female sex, increasing age, poor diet quality and weekly minutes of physical activity correlated with increases in percent body fat. They differed in that, among survivors, but not among controls, depression was associated with percent body fat. There is evidence that chemotherapy and radiation lead to dysregulation of the hypothalamic pituitary axis which can potentially manifest in depression.⁴⁸ Likewise, this dysregulation can lead to metabolic abnormalities and increases in body fat.⁴⁹ Thus, treatment exposures may be driving this association in the survivors in our study.

Although elucidating how neighborhoods mechanistically influence obesity is challenging, obesity likely results from an interplay between individual host factors and health behaviors, social influence and built environment.⁵⁰ Nonetheless, we believe that the real utility of examining characteristics of neighborhoods where survivors live, and how these characteristics influence obesity is to ascertain relevant information to guide tailored intervention efforts. For example, resource heavy interventions may be needed for survivors who live in resource poor neighborhoods (e.g. provide workout equipment, gym memberships, transportation to healthy food outlets), whereas interventions that focus on incorporating healthy behaviors into daily life (e.g. breaking up sedentary time at work/school) may be sufficient for survivors who live in resource rich neighborhoods. Further, researchers working with survivor populations can use publicly available resources to link daily environmental contexts/exposures with health outcomes to aid in study design, counseling, or intervention decisions. Lastly, in a clinical setting, practitioners should acknowledge the importance of social and environmental contexts in the management of obesity.

The results of this analysis should be examined in the context of several limitations. First, as with any geographical analysis, results vary based on units of analysis.⁵¹ We were restricted in our analysis due to the geographical spread of our survivors (across the United States), which required us to use publicly available datasets containing environmental variables for all U.S. census tracts. Zoning in on a specific catchment or regional area might afford an opportunity to use more granular spatial data, or pinpoint areas with survivors most in need of interventions. Second, we were unable to account for residential history (length of time at address) or changes in residential conditions over time.⁵² Yearly geocoding of the SJLIFE cohort will allow the ability to account for residential histories in future analyses. Third, our analysis only considered place of residence, even though survivors spend substantial amounts of time in other environments⁵³, including school and work locations, which have been shown to influence BMI in other populations.⁵⁴ Fourth, as this study was cross-sectional, we cannot draw inferences about causality; however, as our prospective cohort grows in longitudinal assessments, we should be well poised to examine longitudinal associations between environmental factors and clinical outcomes and identify opportunities for intervention efforts.⁵⁵

Current initiatives to integrate location of residence as correlates of health and health disparities highlight the need to incorporate geospatial approaches when examining cancer risk, survival and outcomes.⁵⁶ Our current work is the first to examine how the social and physical context of a childhood cancer survivor’s residence influences obesity outcomes. Importantly, the results of this analysis should inform future obesity intervention efforts in this population.

Supplementary Material

Supplemental Figure 1

Supplementary Figure 1. Participant flow for both childhood cancer survivor and control participants. Survivor participant flow is indicated in the left panel, with control participant flow in the right panel.

Supplemental Figure 2

Supplementary Figure 2. Estimated direct effects of neighborhood factors, individual level factors and treatment exposures on percent body fat in control participants in the St. Jude Lifetime Cohort study (N=318).

Novelty and Impact:

The neighborhood where a childhood cancer survivor resides may influence obesity and obesity-related health behaviors, in addition to known obesity related treatments. No study to date has examined the influence of neighborhood effect (e.g. socioeconomic status, access to healthy food options and places to exercise, rurality) on obesity in childhood cancer survivors. Results are important to guide future obesity intervention efforts in childhood cancer survivors.

ABBREVIATION LIST

SES

Socioeconomic status

CDC

Centers for Disease Control and Prevention

ACS

American Cancer Society

AICR

American Institute for Cancer Research

nSES

Neighborhood Socioeconomic Status

SJLIFE

St. Jude Lifetime Cohort

SJCRH

St. Jude Children’s Research Hospital

IRB

Institutional Review Board

DEXA

Dual X-Ray Absorptiometry

HPA

Hypothalamic-Pituitary-Adrenal

MVPA

Moderate to vigorous physical activity

NHANES

National Health and Nutrition Examination Survey

USDA

United States Department of Agriculture

HEI

Healthy Eating Index

RUCA

Rural-Urban Commuting Area

SEM

Structural Equation Modeling

CFA

Confirmatory Factor Analysis

SRMR

Standardized Root Mean Square Residual

CFI

Comparative Fit Index

NFI

Bentler-Bonnett Normed Fit Index

RMSEA

Root Mean Square Error of Approximation

BMI

Body Mass Index

OR

Odds Ratio