Associations of area-level and individual-level sociodemographic factors with adolescent phenotypes linked to subsequent cancer risk

Abstract

Background:

Indices of Concentration at the Extremes (ICEs) serve as joint proxies for social and environmental hazards for health disparities. We tested the hypothesis that ICEs at birth predict timing of menarche and adolescent overweight and investigated whether ICE associations are mediated by individual-level variables.

Methods:

In this prospective study, a subset of offspring from the Child Health and Development Studies, born 1959–1967 in Oakland, CA were assessed for early age at menarche (<12 vs. >=12) and overweight body mass index, (BMI, ≥25 kg/m²) at ages 15–17 years. ICEs characterizing neighborhood of birth were calculated using 1960 census tract data for race, income, education, and combined income/race. Associations between ICEs and maternal serum DDT levels were estimated using generalized linear models. Log-linear models estimated associations between ICEs with early menarche and adolescent overweight, adjusted for individual factors including race, socioeconomic status and perinatal serum DDTs and family clustering.

Results:

Address at birth was geocoded for 1749 (86.6 %); 42 % had overweight BMI; 15.6 % had early menarche. ICEs at birth predicted timing of menarche and adolescent overweight and were correlated with perinatal DDTs. Most ICE associations with adolescent outcomes were partly mediated by individual level socioeconomic variables, except for the income ICE association with early menarche [adjusted odds ratio: 1.6, 95 %CI:(1.0–2.6)].

Conclusions:

ICEs are accessible metrics of area-level spatial and social segregation that were associated with adolescent overweight, early menarche and DDT exposure. ICEs are useful indicators of high-risk neighborhoods that can be targeted for individual and community-level prevention beginning at birth.

1. Introduction

Ever-increasing evidence demonstrates that systemic and structural factors producing social injustice negatively affect health and wellness across the life course. Social determinants of health have been linked to inequality in cancer screening [1,2], cancer survival [3], and cancer burden [4,5]. While the adverse health effects of disadvantage and inequality are clear, questions remain about how best to measure these complex concepts in quantitative health research, and how to tease out the temporal dimension of sustained and acute stressors which contribute to chronic disease risk over the course of years and decades.

To address the complexities of measuring structural disadvantage, some researchers have called for the use of multi-category indices that incorporate factors across a range of sociodemographic areas [6,7]. Such indices are powerful indicators of composite disadvantage but can be challenging to interpret if a single component variable within the index is particularly salient in producing disparities for a given health outcome. Conversely, utilizing area-level demographic characteristics without transformation (e.g. percent of population below poverty line within a census tract) does directly measure disadvantage but may not capture the polarization that arises from spatial heterogeneity. As a middle ground, some researchers have called for the use of indices of concentration at the extremes (ICEs) [8–12]. ICEs transform area-level sociodemographic variables to incorporate a measure of disparity, while still allowing researchers to focus on the particular social or structural characteristics that are relevant for a given research question.

Epidemiological research using ICEs has shown that ICEs can be better predictors of adverse environmental exposures than more traditional individual or neighborhood measures of socioeconomic position such as individual income or neighborhood poverty level [11]. In this study, we investigate ICEs relationships with two recognized cancer risk factors: overweight/obesity and early age at menarche.

Obesity is a ubiquitous risk factor for many cancers [13] that correlates with environmental chemical exposures [14]. Since early life overweight and obesity in males and females track to adulthood [15,16], exploring relationships between ICE metrics in early life and the development of obesity may illuminate early risk factors for obesity-associated cancers. ICE indices can capture socio-economic neighborhood characteristics related to children’s health [17], underscoring the potential for ICEs to serve as proxies for disparities in risk.

Toxic environmental exposures are particularly relevant to breast cancer, and understanding the relationship between pollutants and neighborhood segregation provides opportunities to more completely elucidate complex and interrelated exposures. Some, but not all, components of air pollution are associated with risk of breast cancer, with stronger associations with estrogen receptor positive/progesterone receptive positive tumors [18]. Air pollution also associates with breast density [19], a risk factor for breast cancer. Environmental chemicals have been linked with both breast cancer incidence and with early age at menarche [20], a strong risk factor for breast cancer [21–23]. In the same cohort as the present study, maternal perinatal p,p’-DDT was associated with maternal breast cancer before age 50 in birth cohorts first exposed before and during puberty [23], and with breast cancer diagnosed between ages 50–54 in birth cohorts first exposed after infancy [24]. In offspring in this same paired parent-child cohort, in utero exposure to o,p’-DDT was associated with daughter’s breast cancer with stronger associations for human epidermal growth factor receptor 2 positive breast cancer and late-stage disease [25]. Findings illustrate that in utero exposure is a sensitive period and that the vulnerability of the breast differs according to exposure timing. However, research is lacking that directly measures relationships between neighborhood socioeconomic characteristics, chemical environmental exposures, and health outcomes that are relevant for public health planning. Addressing this research gap is crucial in understanding how ICE metrics may capture or correlate with multiple dimensions of residential neighborhood stressors: not only segregation along demographic or socioeconomic lines, but also exposure to hazardous environmental contaminants that produce life course health inequities.

While prior studies have established the utility of ICE measures for understanding the impact of structural factors on health, these metrics have not been applied to examine near-term outcomes associated with future cancer risk. Leveraging a prospective, multi-generational birth cohort, we examine relationships between ICE measures at birth, perinatal chemical exposures, and adolescent outcomes associated with future cancer risk. This case study focuses on relationships between neighborhood-level ICE measures of socioeconomic disparity across four domains (race/ethnicity, income, education, and combined race/ethnicity and income) and individual risk of two adolescent outcomes: age at menarche and adolescent overweight. These two outcomes are significant risk factors of future breast cancer risk [26–34], while obesity is also a strong risk factor for many adult cancers [13]. Their relationships with ICE metrics may elucidate relationships by which neighborhood deprivation or privilege in childhood shapes both differential exposure to environmental hazards as well as life course cancer risk. In order to better understand relationships between environmental exposures and ICE metrics as neighborhood-level indices of disparity, we further examine relationships between individual maternal serum measurements of three DDTs, which were a ubiquitous nation-wide chemical exposure during the study period [23], in analyses of offspring outcomes and ICEs. Taken together, these analyses provide evidence for the salience of ICE measures in representing neighborhood chemical exposures and illustrate the complex interplay between environmental exposures, structural disadvantage, and individual-level sociodemographic characteristics.

2. Methods

1. Study population

   This study is set within the Child Health and Development Studies (CHDS) multi-generational cohort. The CHDS has been described in detail elsewhere [35]. Briefly, the CHDS prospectively enrolled 15, 528 pregnant people, with 20,754 pregnancies, enrolled in Kaiser Foundation Health Plan in Northern California from 1959 to 1967 and followed the index pregnant people (F0 generation) as well as the children (F1 generation) born from those pregnancies. The present study leverages a sample of F1 participants (N = 2020) who completed a follow-up study during adolescence when aged 15–17 (hereafter referred to as adolescent study). The adolescent study (Table 1) comprised F1 participants born between 1960 and 1963, who were residing in the Bay Area and remained under observation (54 %). Figure S1 provides an illustration of follow-up starting with enrollment of the CHDS cohort () at baseline, n =20,754 pregnancies (1959–1967), through participation in the Adolescent Study, n = 2020 (1977–1979). For this analysis, adolescent study participants were excluded if their address at birth was missing or could not be geocoded (N = 226, 11 %); the final analytical sample containing N = 1749 adolescents is presented in Fig. 1, which illustrates sample attrition due to missing information on adjustment covariates.

2. Outcome ascertainment

   Adolescent cohort participants completed a survey and participated in an office visit between 15 and 17 years of age. In the office visit, height and weight were directly measured; adolescent BMI was calculated from these anthropometrics. Age at menarche was self-reported during the adolescent exam (between age 15–17), minimizing recall bias by being close to the occurrence of menarche. A prior study that demonstrated that 59 % of a longitudinal sample of girls over a mean interval of 430 days recalled the exact month and year of menarche, with higher accuracy observed for recall at a time closer to the time of menarche [36].

   BMI was dichotomized as not overweight (<25), versus overweight or obese (>= 25). Age at menarche was dichotomized as early menarche (< age 12) versus non-early menarche (>= age 12).

3. Exposure assessment

   The primary exposure metric was the index of concentration at the extreme (ICE). The general form of the ICE metric is: [(population with structural advantage/privilege) – (population with structural disadvantage/deprivation) / (total population)]. The ICE metric was calculated at the census tract level using data from the 1960 census, and defined across four domains: income, education, race, and the combination of race and income. The 1960 census was selected because it provided the closest match to participant birth years (1960–1963). Table S1 shows census variables and specific calculations for each of the four ICE domains. Data from the 1960 census were obtained from the National Historical GIS Program [37].

   Census-tract-level ICE metrics were linked with individual participants based on their geocoded address at birth. Geocoding was performed as follows: addresses at birth were abstracted from CHDS data. A first round of geocoding was performed using the Google Geocoding API via the ggmap packing in R;[38] addresses that were geocoded with greater than 90 % certainty and which had complete information for street address, city, state, and zip code (N = 1709) were considered correctly geocoded. Those that did not match in the first round of geocoding were examined by hand, given the historical nature of the data. These addresses were searched in city records and historic maps to attempt to further ascertain address locations, given that the addresses were from 1960. Failure to match in the initial geocoding could have occurred due to changing urban development or changing street names; this approach allowed us to geocode an additional 40 addresses resulting in 1749 individuals in the sample with geocoded addresses.

   Personal levels of maternal DDT congeners (p,p’-DDT, p,p’-DDE and o,p’-DDT) were assayed as previously described [23,25,39] from archived frozen serum samples drawn during pregnancy (1 % from trimester 1, 2 % from trimester 2, 5 % from trimester 3) with the majority from the early postpartum 1–3 days after birth (92 %). DDT levels were available for a convenience sub-sample derived from the CHDS assay library of accumulated organochlorine measures over the course of multiple studies [39]. Assay levels were available for the sub-sample shown in Fig. 1.

4. Statistical approach

   We implemented log-linear models (PROC GENMOD) executed in SAS 9.4 to estimate associations using the repeated option to adjust the covariance matrix for correlation between siblings. The “repeated” option within the GENMOD procedure invokes the Generalized Estimating Equations (GEEs) approach and allows specification of the covariance structure for GEE model fitting. By specifying a plausible working correlation structure to account for within-subject correlations (i.e. within siblings), the GEE method estimates model parameters by iteratively solving a system of equations based on quasi-likelihood distributional assumptions [40]. (https://www.lexjansen.com/wuss/2006/tutorials/TUT-Smith.pdf ).

Table 1.

Characteristics of study sample by timing of data collection:Child Health and Development Studies, Adolescent Follow-up, n = 1749.

	Mean (SD) or N (%)	Range (Min - Max)	Percentile
			25th	50th	75th
Characteristics at offspring adolescence
Year of birth	1961.2 (0.8)	(1960 – 1963)	1961	1961	1962
Year of study	1978.1 (0.7)	(1977 – 1979)	1978	1978	1979
Age (y) at study	16.6 (0.6)	(15 – 18)	16	17	17
Sex at birth
Female	865 (49.5 %)	———	———	———	———
Male	884 (50.5 %)	———	———	———	———
Age (y) at menarche (continuous)	13.0 (1.3)	(9 – 17.6)	12.0	13.0	13.8
Age at menarche (dichotomized)
<age 12 y	135 (15.7 %)	———	———	———	———
≥ age 11 y	723 (84.3 %)	———	———	———	———
Weight (kg)	63.2 (12.3)	(37.0 – 159.0)	55.0	61.0	69.0
Height (m)	1.7 (0.1)	(1.3 – 2.0)	1.6	1.7	1.8
BMI (continuous kg/m2)	22.0 (3.7)	(14.4 – 47.8)	19.8	21.3	23.3
BMI (dichotomized)
< 25 kg/m2	1476 (85.5 %)	———	———	———	———
≥ 25 kg/m2	250 (14.5 %)	———	———	———	———
Index of Concentration at the Extremes (ICE) for neighborhood of birth
Education ICEa	−0.36 (0.28)	(−0.84 – 0.59)	−0.56	−0.39	−0.25
Race ICEb	0.68 (0.51)	(−0.93 – 1.00)	0.60	0.96	0.98
Income ICEc	0.02 (0.22)	(−0.67 – 0.67)	−0.11	0.01	0.14
Income and Race ICEd	0.15 (0.22)	(−0.67 – 0.74)	0.05	0.17	0.25
Maternal characteristics at offspring birth
Age (y)	28.7 (5.9)	(15 – 47)	24	28	33
Age (y) at menarchee	12.6 (1.4)	(9 – 19)	12	13	13
Weight (kg)	60.8 (10.5)	(37.7 – 119.8)	54.0	59.1	65.7
Height (m)	1.6 (0.1)	(1.4 – 1.9)	1.6	1.6	1.7
BMI (kg/m2)	22.9 (3.7)	(14.2 – 43.8)	20.5	22.2	24.4
Race (dichotomized)
Black individuals	339 (19.4 %)	———	———	———	———
Non-Black individuals	1410 (80.6 %)	———	———	———	———
Parity (dichotomized)
Primiparous	437 (25.0 %)	———	———	———	———
Multiparous	1312 (75.0 %)	———	———	———	———
Highest education level (dichotomized)
<High School	200 (11.5 %)	———	———	———	———
High School	741 (42.4 %)	———	———	———	———
>High School	806 (46.1 %)	———	———	———	———
Total family income (dichotomized)
< 1960 census median	607 (38.5 %)	———	———	———	———
≥ 1960 census median	971 (61.5 %)	———	———	———	———

BMI, Body Mass Index

^aEducation ICE defined using 1960 census data for geocoded residential address calculated as: (Completed college - Less than High School) / Education Total.

^bRace ICE defined using 1960 census data for geocoded residential address calculated as: (White - Black Race) / Total Population.

^cIncome ICE defined using 1960 census data for geocoded residential address calculated as: ([Family Income > 10,000] – [Family income < 3999]) / All Family Income.

^dIncome and race ICE defined using 1960 census data for geocoded residential address calculated as: ([White family Income > 10,000] – [Non-white family income < 3999]) / All Family Income.

^eReported by mothers (F0) at a mean age of 26 years at time of entry into the study.

Fig. 1.

Flow chart showing available sample for study variables (n = 1749), starting from total participation in the Adolescent Study (n = 2020) To view the derivation of the Adolescent Study sample starting from baseline enrollment of CHDS pregnancies (n = 20,754), refer to Figure S1.

2.1. Analysis sample for obesity outcome by neighborhood

Models estimating associations of adolescent overweight with neighborhood characteristics at birth were based on n = 1726 study participants (males and females); n = 294 were excluded from this analysis due to missing information on birth address (n = 271) and BMI (n = 23).

2.2. Analysis sample for menarche outcome by neighborhood

Models estimating associations of early age at menarche with neighborhood characteristics at birth were based on n = 858 female study participants; n = 145 were excluded from this analysis due to missing information on birth address (n = 138) and BMI (n = 7).

2.3. Analysis sample for joint prenatal DDT and neighborhood associations with adolescent obesity and age at menarche

As noted above and illustrated in Fig. 1, prenatal maternal DDT assays were previously completed for a subsample of this study and constitute a convenience sample in the current study. Models estimating associations of adolescent overweight concurrently with neighborhood of birth and prenatal DDTs were based on n = 963, excluding n = 763 participants without measured DDT levels. Models estimating associations of early menarche concurrently with neighborhood of birth and prenatal DDTs were based on n = 574, excluding n = 284 participants without measure DDT levels.

2.4. Sensitivity analysis addressing missing data on DDT and timing of blood draw

We compared F0 maternal characteristics for F1 who participated in the Adolescent Study (n = 2020) vs. eligible F1 who did not (n = 9333) and for participants with and without available F0 maternal prenatal DDT measures and investigated impact of adjustment for trimester of blood draw.

2.5. Modeling

Models were run separately for each ICE metric and then adjusted for the individual characteristic(s) that corresponded to the particular ICE measure, e.g. the income ICE measure model was run unadjusted, then adjusted for family income. Education, and income models estimating associations for outcomes were then additionally adjusted for maternal race. Models estimating the age at menarche associations were also adjusted for adolescent overweight.

We examined models for associations for ICE measures categorized in quartiles. Final model representation of ICE metrics was chosen based on categorization that maximized statistical power to estimate associations as recommended previously [41] and shown in Table S2. This approach is justified in the absence of prior theoretical or observational information to suggest the presence of monotonic dose response. As a result, quartile 1 was compared to a combination of quartiles 2,3, and 4 (see Table S2).

In models estimating adolescent overweight, interactions between each ICE measure and sex at birth were tested using product terms between the ICE measures and sex. In models estimating early menarche, interactions between each ICE measure and adolescent overweight were tested using product terms between the ICE measures and overweight. In models testing both outcomes interactions between individual race and the ICE measures, income and race, were tested using product terms between each ICE measure and maternal Black race.

In the assayed sub-sample, we estimated associations between continuous ICE measures and each of the three DDTs (p,p’-DDT, p,p’-DDE and o,p’-DDT) as both log-transformed and untransformed continuous measures, using generalized linear univariate models. To examine whether DDTs influenced the observed associations of adolescent outcomes with the ICE measures, we added all three DDTs concurrently to the log-linear models estimating the early menarche and adolescent overweight associations with each ICE measure.

3. Results

3.1. Descriptive analysis

Among 2020 adolescents in the analytical cohort, 263 (14.2 %) had an overweight BMI of >= 25. Among the female participants (N = 1003), 155 (15.6 %) had early menarche onset before age 12. Across four calculated ICE domains, participants had more heterogeneity in the education, income, and combined income-race domains; the race ICE domain had a smaller range and interquartile range in values. Table 1 shows full outcome, exposure, and covariate characteristics of the study sample. Mothers of Adolescent Study participants were slightly older and had slightly higher parity, education and income than non-participants (Table S3). It is unlikely that the slight differences between these two groups would explain the observed associations with obesity and early menarche. The proportion of primiparous mothers remained fairly constant for all study groups: 25 % of all Adolescent Study participants (n = 2020); 25 % of those with available residence of birth; 25 % of those with available prenatal DDT assay; and 29 % of non-participants.

3.2. Age at menarche

Models estimating associations for young age at menarche with all four ICE indices, assessed as quartiles, support the use of classifying the exposure of interest and risk as quartile 1 vs. quartiles 2–4 (Table S2).

In unadjusted analyses, residence at birth in a neighborhood with greater sociodemographic segregation as measured by ICEs was associated with a substantial increased risk of early menarche, across four ICE domains (Table 2). In analyses where the model for each neighborhood-level ICE domain was adjusted for the corresponding individual-level characteristics, large associations persisted for the education, income, and race-income domains, but effect size for the race ICE domain was attenuated by adjustment for maternal Black race (Table 2). Additional adjustment of the education and income ICE models further attenuated associations with age at menarche (Table 2), suggesting that only the income ICE remained a significant area-level predictor of early menarche after accounting for individual level characteristics.

Table 2.

Associations of adolescent phenotypes with birth neighborhood measures as characterized by Indices of Concentration at the Extremes (ICEs) for education, race, income and income-by-race: Child Health and Development Studies, Adolescent Follow-up, n = 1749.

	Young age at menarche (<12 y vs. ≥12 y)
	Unadjusted				Individual-Level Adjustede				Individual-Level Adjusted plus Maternal Raceg
ICE measures for neighborhood of birth	Odds Ratio	95 % Confidence Limits		P-value	Odds Ratio	95 % Confidence Limits		P-value	Odds Ratio	95 % Confidence Limits		P-value
		Lower	Upper			Lower	Upper			Lower	Upper
Neighborhood with education segregationa	1.51	1.01	2.25	0.04	1.52	1.02	2.28	0.04	1.32	0.83	2.09	0.24
Neighborhood with racial segregationb	1.62	1.10	2.39	0.01	1.46	0.82	2.61	0.20	N/A	N/A	N/A	N/A
Neighborhood with income segregationc	1.67	1.13	2.48	0.01	1.80	1.19	2.70	< 0.01	1.59	0.96	2.61	0.07
Neighborhood with joint income-race segregationd	1.78	1.21	2.61	< 0.01	1.76	0.99	3.13	0.05	N/A	N/A	N/A	N/A
	Adolescent overweight (BMI ≥25 kg/m2 vs <25 kg/m2)
	Unadjusted				Individual-Level Adjustede				Individual-Level Adjusted plus Maternal Raceg
ICE measures for neighborhood of birth	Odds Ratio	95 % Confidence Limits		P-value	Odds Ratio	95 % Confidence Limits		P-value	Odds Ratio	95 % Confidence Limits		P-value
		Lower	Upper			Lower	Upper			Lower	Upper
Neighborhood with education segregationa	1.82	1.37	2.43	< 0.01	1.66	1.23	2.24	< 0.01	1.25	0.88	1.80	0.22
Neighborhood with racial segregationb	2.00	1.50	2.66	< 0.01	1.38	0.93	2.06	0.11	N/A	N/A	N/A	N/A
Neighborhood with income segregationc	1.79	1.34	2.38	< 0.01	1.67	1.22	2.29	< 0.01	1.19	0.81	1.75	0.37
Neighborhood with joint income-race segregationd,f	1.87	1.41	2.49	< 0.01	1.20	0.76	1.88	0.43	N/A	N/A	N/A	N/A

BMI, Body Mass Index, N/A is not applicable because it is characteristic is included in Individual-Level Adjustment models

^aEducation ICE defined using 1960 census data for geocoded residential address calculated as: (Completed college - Less than High School) / Education Total. Dichotomized as quartile 1 vs. quartiles 2–4.

^bRace ICE defined using 1960 census data for geocoded residential address calculated as: (White - Black Race) / Total Population. Dichotomized as quartile 1 vs. quartiles 2–4.

^cIncome ICE defined using 1960 census data for geocoded residential address calculated as: ([Family Income > 10,000] – [Family income < 3999]) / All Family Income. Dichotomized as quartile 1 vs. quartiles 2–4.

^dIncome and race ICE defined using 1960 census data for geocoded residential address calculated as: ([White family Income > 10,000] – [Non-white family income < 3999]) / All Family Income. Dichotomized as quartile 1 vs. quartiles 2–4.

^eAdjusted models include ICE measure and individual characteristics that correspond to the specific ICE measure. The Education ICE model was adjusted for maternal F1 education at birth, classified as less than a high school education, and a high school degree vs. all higher education. The Race ICE model was adjusted for maternal Black race vs. non-Black race. The Income ICE model was adjusted for family income at F1 birth, classified as below the 1960 Bay Area Census median family income vs. equal to or greater than the census median. The Income by Race ICE model was adjusted for maternal Black (vs. non-Black) race and low family income (<median vs. ≥1960 Bay Area census median) at F1 birth.

^fAdjustment of this model with only F1 family income did not affect the association between adolescent overweight and the income-race ICE (family income-adjusted OR=1.79l OR=1.31–2.44). Adjustment with maternal Black race was the factor responsible for explaining this association (maternal race-adjusted OR=1.18; 95% CI=0.77–1.80)

^gModels are adjusted for individual level characteristics corresponding to each ICE measure plus maternal race, classified as Black vs. non-Black.

3.3. Adolescent overweight

Models estimating associations for adolescent overweight with all four ICE indices, assessed as quartiles, support the use of classifying the exposure of interest and risk as quartile 1 vs. quartiles 2–4 (Table S2).

Residence at birth in a neighborhood with greater sociodemographic polarization was associated with increased risk of adolescent overweight across all four ICE domains (Table 2). Participants in neighborhoods with the greatest polarization in the race domain were twice as likely to have adolescent overweight, and risk was similarly elevated for the income, combined income-race and education ICE indices in unadjusted models (Table 2). Adjustment for individual level characteristics attenuated but did not explain associations for the education and income ICE indices; however, adjustment for maternal Black race did explain the associations for the race and combined income-race ICE indices (Table 2). Additional adjustment of the education and income ICE models further attenuated associations with adolescent overweight (Table 2), suggesting that area-level predictors were not as effective in predicting overweight in adolescence as individual-level characteristics.

3.4. Assayed sub-sample

DDTs were significantly negatively associated with ICE measures (Table 3), indicating that higher levels of all three DDT congeners were associated with greater social and economic segregation, as expected. Log-transformation of DDTs produced similar results. Adjusting associations between adolescent outcomes and ICE measures with personal perinatal DDT levels demonstrated little impact (Table 4). The associations were slightly attenuated primarily due to the smaller assayed sample size but most remained robust and significant (Table 4). In models co-adjusted for ICE measures and DDTs, p,p’-DDE remained a significant independent predictor of early menarche. In co-adjusted models predicting adolescent overweight, only o,p’-DDT remained marginally significant (p = 0.08 in models with Education ICE and p < 0.2 models with all other ICEs). Table S5 shows the effect of adjusting for parity and timing of blood draw on associations of adolescent phenotypes with ICE metrics. Adjusting for these covariates had little impact on the findings reported in Table 4.

Table 3.

Associations of ICE measures with personal pregnancy levels of DDT compounds in CHDS Adolescent Study assayed sub-sample estimated from univariate general linearized models:Illustrates correspondence between neighborhood measures and individual exposures.

	p,p’-DDE*
ICE measures for neighborhood of birth (continuous)	Beta	95 % Confidence Limits		P-value
		Lower	Upper
Neighborhood with education segregation1	−0.0008	−0.0016	0.0001	0.0865
Neighborhood with racial segregation2	−0.0022	−0.0036	−0.0008	0.0025
Neighborhood with income segregation3	−0.0007	−0.0013	−0.0000	0.0499
Neighborhood with joint income-race segregation4	−0.0008	−0.0014	−0.0001	0.0236
	p,p’-DDT*
ICE measures for neighborhood of birth (continuous)	Beta	95 % Confidence Limits		P-value
		Lower	Upper
Neighborhood with education segregation1	−0.0023	−0.0044	−0.0002	0.0347
Neighborhood with racial segregation2	−0.0096	−0.0133	−0.0058	< 0.0001
Neighborhood with income segregation3	−0.0033	−0.0050	−0.0015	0.0003
Neighborhood with joint income-race segregation4	−0.0035	−0.0053	−0.0018	< 0.0001
	o,p’-DDT*
ICE measures for neighborhood of birth (continuous)	Beta	95 % Confidence Limits		P-value
		Lower	Upper
Neighborhood with education segregation1	−0.0610	−0.0875	−0.0346	< 0.0001
Neighborhood with racial segregation2	−0.2118	−0.2692	−0.1543	< 0.0001
Neighborhood with income segregation3	−0.0747	0.0966	−0.0529	< 0.0001
Neighborhood with joint income-race segregation4	−0.0800	−0.1028	−0.0572	< 0.0001

^*Tests of associations with log-transformed DDT congeners produced similar results.

Table 4.

Associations of adolescent outcomes with ICE measures based on neighborhood of birth, adjusted for personal perinatal levels of p,p’-DDT, p,p’-DDE and o,p’-DDT, Child Health and Development Studies, Adolescent Follow-up, n = 1749.

Young age at menarche (<12 y vs. ≥12 y)
	Unadjusted full study sample, n = 858				Unadjusted assayed sub-sample, n = 574				Adjusted for p,p’-DDT, p,p’-DDE, and o, p’-DDT, n = 574
ICE measures for neighborhood of birth	Beta	95 % Confidence Limits		P-value	Beta	95 % Confidence Limits		P-value	Beta	95 % Confidence Limits		P-value
		Lower	Upper			Lower	Upper			Lower	Upper
Education segregationa	0.41	0.01	0.81	0.04	0.32	−0.17	0.82	0.20	0.26	−0.26	0.78	0.32
Racial segregationb	0.48	0.10	0.87	0.01	0.35	−0.14	0.84	0.16	0.31	−0.20	0.83	0.23
Income segregationc	0.52	0.12	0.91	0.01	0.54	0.07	1.02	0.02	0.56	0.08	1.04	0.02
Joint income-race segregationd	0.58	0.19	0.96	< 0.01	0.51	0.04	0.99	0.03	0.51	0.01	1.02	0.05
Adolescent overweight (BMI ≥25 kg/m2 vs <25 kg/m2)
	Unadjusted full study sample, n = 1726				Unadjusted assayed sub-sample, n = 963				Adjusted for p,p’-DDT, p,p’-DDE, and o, p’-DDT, n = 963
ICE measures for neighborhood of birth	Beta	95 % Confidence Limits		P-value	Beta	95 % Confidence Limits		P-value	Beta	95 % Confidence Limits		P-value
		Lower	Upper			Lower	Upper			Lower	Upper
Education segregationa	0.60	0.31	0.89	< 0.01	0.52	0.15	0.90	0.01	0.42	0.03	0.82	0.04
Racial segregationb	0.69	0.41	0.98	< 0.01	0.62	0.24	0.99	< 0.01	0.54	0.14	0.93	0.01
Income segregationc	0.58	0.29	0.87	< 0.01	0.55	0.16	0.94	< 0.01	0.45	0.04	0.85	0.03
Joint income-race segregationd	0.63	0.34	0.91	< 0.01	0.54	0.16	0.91	< 0.01	0.45	0.05	0.85	0.03

^aEducation ICE defined using 1960 census data for geocoded residential address calculated as: (Completed college - Less than High School) / Education Total. Dichotomized as quartile 1 vs. quartiles 2–4.

^bRace ICE defined using 1960 census data for geocoded residential address calculated as: (White - Black Race) / Total Population. Dichotomized as quartile 1 vs. quartiles 2–4.

^cIncome ICE defined using 1960 census data for geocoded residential address calculated as: ([Family Income > 10,000] – [Family income < 3999]) / All Family Income. Dichotomized as quartile 1 vs. quartiles 2–4.

^dIncome and race ICE defined using 1960 census data for geocoded residential address calculated as: ([White family Income > 10,000] – [Non-white family income < 3999]) / All Family Income. Dichotomized as quartile 1 vs. quartiles 2–4.

3.5. Sensitivity analyses

ICE measures did not interact with adolescent overweight in estimating associations for early menarche, nor did being overweight explain young age at menarche associations with ICE measures. In models estimating associations for early menarche, tests of interaction between the race ICE and maternal race (p-value=0.13) and between the combined income-race ICE and maternal race (p = 0.20) were not statistically significant. However, in models stratified by maternal race there was evidence of differential risk for Black vs. non-Black adolescents. Among non-Black adolescents there were strong and statistically significant associations with both the race ICE (OR=1.78; 95 % CI=1.01–3.14) and the income-race ICE (OR=2.04; 95 % CI=1.16–3.60) indices. But among Black adolescents, associations were null (OR=0.72; 95 % CI=0.27–1.89 for race ICE and OR=0.90; 95 % CI=0.32–2.55 for income-race ICE).

Similarly, none of the ICE measures predicting adolescent overweight interacted with biological sex at birth, nor did sex explain overweight associations. Adjusting for age in continuous months had essentially no impact on the estimated associations, consistent with the fact that there was a very narrow range of age at interview because the cohort members were born within 3 years of each other.

To examine the impact of possible correlation between siblings on associations of adolescent phenotypes with ICE metrics we estimated these associations in the subset excluding siblings. This subset included all participants who had no siblings (n = 1476) – most of the study participants – and the first observed sibling from sibling pairs (n = 254) and from sibling trios (n = 12). Among the n = 254 sibling pairs, n = 242 had available BMI measures for both siblings: the corresponding BMI correlation between sibling pairs as calculated using the Pearson correlation method was, ρ= 0.43, p-value< 0.01; and as calculated using the Spearman correlation method was, ρ= 0.36, P-value< 0.01. Among the n = 63 sister pairs, n = 62 had available data for age at menarche for both sisters: the corresponding age at menarche correlation between sisters as calculated using the Pearson correlation method was, ρ= 0.13, P-value= 0.32; and as calculated using the Spearman correlation method was, ρ= 0.29, p-value= 0.02. The results of this sensitivity analysis are presented in Table S4 and show that the associations in the subset excluding siblings are not different from those estimated in the full study sample.

4. Discussion

This investigation addresses prenatal social determinants of overweight/obesity in adolescence, timing of menarche and measured maternal prenatal DDT exposure in the Child Health and Development Studies. The design combined area -level and individual-level measures of social determinants with individual-level measures of prenatal exposures to predict subsequent adolescent outcomes known to be associated with risk of cancer over the life-course: early age at menarche and overweight/obesity in males and females. We found that area-level demographic and economic segregation at birth, captured by four domains of ICE (income, education, race, and combined race/income) was associated with age at menarche, adolescent overweight and maternal prenatal DDT. Our findings are consistent with studies of obesity by neighborhood racial segregation [42], social determinants of early puberty [43], ICE associations with cancer mortality of multiple sites [44], and prenatal DDT associations with menarche timing, obesity, and cancer [45–50].

4.1. Influence of ICEs in combination with Individual-Level Social Determinants

Clarifying the influence of area-level and individual-level social and additionally with environmental determinants of cancer risk factors is necessary for public health surveillance and cancer risk evaluation. ICE metrics distinctly capture area-level segregation compared to other measures, because they take into account the number of people who experience both structural disadvantage and structural privilege, identifying spatial areas with a high rate of disadvantage in a certain domain as well as areas with a high degree of heterogeneity or disadvantage within a given domain (Figs. 2 and 3, and Figures S2 and S3). Because different sociodemographic characteristics may have very different spatial distributions, ICEs therefore capture different degrees of polarization depending on the spatial and social distribution of certain characteristics. The difference between the race ICE and the income ICE indices illustrates this well (Figs. 2 and 3): the lowest quartile of the race ICE metric includes areas with predominantly Black residents, as well as more racially heterogeneous neighborhoods. In contrast, the lowest quartile of the income ICE encompasses almost exclusively census tracts with majority low-income households.

Fig. 2.

Illustration of the relationship between the two components of the racial segregation ICE numerator and the resulting calculated racial segregation ICE score. The x-axis gives calculated values of the Race ICE score. The y-axis gives the percent of each subgroup that corresponds to the value of the Race ICE score. This figure presents the percent Black, the percent non-Black and the corresponding calculated value of racial segregation ICE for all neighborhoods of birth for F1 in the Adolescent Study and shows that: a.) 1960 neighborhoods were clearly segregated by race. b.) The lowest quartile of the Race ICE includes neighborhoods that with predominantly Black residents, and neighborhoods that had heterogeneous Black and non-Black racial composition. c.) Quartiles 2, 3 and 4 of the Race ICE include neighborhoods with predominantly non-Black residents.

Fig. 3.

Illustration of the relationship between the two components of the Income ICE numerator and the resulting calculated Income ICE score. The x-axis gives calculated values of the Income ICE score. The y-axis gives the percent of each subgroup that corresponds to the value of the Income ICE score. This figure presents the percent low family income, the percent high income, and the corresponding calculated value of the Income ICE score for all neighborhoods of birth for F1 in the Adolescent Study and shows that: a.) The distribution of high and low income was more homogeneous and not as extreme as the racial composition of 1960 neighborhoods. b.) Quartile 1 of the Income ICE largely encompasses only low-income neighborhoods, suggesting it is a sensitive threshold for identifying income disadvantage.

Fig. 2 dramatically illustrates the racial segregation of residential neighborhoods in the 1960’s that has been well documented, particularly in the East Bay Area [51–53], and persists today [54]. Although early menarche associations with the race ICE and the income ICE were not statistically different by maternal race, associations were stronger for non-Black individuals in structurally disadvantaged neighborhoods (see Result Section 5. Sensitivity Analyses). This finding suggests that the first quartile of the race ICE may be capturing a vulnerability for non-Black residents in majority-Black and mixed-race neighborhoods, compared to a high level of sustained risk for Black individuals in all neighborhoods. This interpretation is also consistent with the attenuation of the early menarche association with race ICE after adjustment for individual level maternal race (seen in Table 2).

A comparison of menarche associations with the race ICE (Table 2 and Fig. 2) and those with the income ICE (Table 2 and Fig. 3) suggest that these indices are capturing different aspects of polarization and disadvantage. Individual-level race attenuated many of the unadjusted effects of area-level measures of segregation. Race profoundly shapes access to resources due to systemic racism, which manifests in socioeconomic inequalities, residential segregation, and differential access to healthcare, significantly contributing to poorer health outcomes among historically disadvantaged racial groups. The adverse health outcomes associated with race are not a consequence of inherent biological differences but rather the result of the social, economic, and political contexts in which racial groups are situated [55–57].

4.2. Relevance to cancer

Age at Menarche.

Numerous studies have found that women who experience menarche at a younger age have a higher risk of developing breast cancer [26–30,58]. Longer life-time estrogen exposure was considered a likely mechanism, but this hypothesis has now been challenged by consideration of multiple cancer-related molecular pathways enhanced in women with early menarche and observed unexpected concordance of associations in both estrogen positive and estrogen negative tumors [59]. Similarly, although evidence suggests that exposure to DDE and DDT isomers and to other environmental chemicals classified as endocrine disruptors are associated with early menarche [20,60,61], the explanation for these associations are is now thought to involve multiple intersecting but poorly understood mechanisms especially in humans [62]. In the same cohort as the present study (Child Health and Development Studies) we have previously reported on in utero exposure to DDT compounds and breast cancer in daughters. We found stronger associations for advanced stage and Her-2 positive breast cancers in daughters exposed to higher levels of the o,p’-DDT isomer [25], although data were not available to assess the contribution of daughter menarche timing. Findings of this current study suggest a plausible connection of daughter’s breast cancer risk to prenatal ICEs, prenatal DDT and associated early menarche in daughters in The Child Health and Development Studies cohort.

Adolescent BMI and obesity.

A number of studies have found that higher body weight during adolescence is associated with an increased risk of cancers in males and females later in life [63]. Several studies have linked DDT or DDE exposure to higher BMI in children [64]. In this same cohort perinatal DDT exposures has been linked to mid-life obesity in adult daughters [65] and early adulthood obesity in daughters and granddaughters [20]. Studies have also demonstrated that prenatal exposure to DDT and DDE was associated with increased BMI and waist circumference in childhood [66,67].

4.3. Proof of concept that ICEs can capture environmental exposure relevant to cancer: the case of DDT

When DDTs were considered in the pathway of ICE effects on early menarche, p,p’-DDE was a significant contributor for all ICE measures. For adolescent overweight, the DDT isomer, o,p’-DDT played a marginally significant role in all ICE models, especially for the education ICE measure. However, ICE measures were independent of prenatal DDT. This finding suggests that the area-level exposures captured by ICEs may contribute to cancer risk factors through multiple pathways, capturing both social and economic disparities as well as chemical and environmental toxicants.

5. Study strengths and limitations

This study examines associations between neighborhood-level measures of segregation as measured by ICE at birth, and age of menarche (females) and body weight (males and females), as captured during ages 15–17 among participants in the Child Health and Development mutli-generation cohort. While the majority of investigations establishing ICE as an area-level social determinant of health are cross-sectional in nature [68], this prospective study investigates prenatal exposure, a critical period of susceptibility to social and environmental risk factors. In addition, perinatal exposure to DDT and other organocholorines is well-established in the CHDS pregnancy cohort, which has contributed substantially to the understanding of breast cancer and other chronic disease determinants over the life course [20,22,23,69,65,70]. Moreover, the historical timing of the study in the East Bay Area of California permitted a robust analysis of the interplay of race, neighborhood segregation, and the patterning of age at menarche and body weight among Black and non-black adolescents. The East Bay Area in 1960 was largely racially binary [51].

Among the limitations of this study is the relative advantage of the cohort members. All were members of the Kaiser Permanente health plan, indicative of employment in subscribing organizations or plan purchase. Thus, the extreme disadvantage tail of the distribution of East Bay Area residents at the time is most likely not represented in these data. On the other hand, confounding by extreme poverty and lack of access to prenatal and early infant health care is unlikely in this population. Maternal prenatal serum DDT is a marker of fetal exposure which can lead to developmental toxicant effects and /or contaminant storage in the fetus’s body with longer term effects. We cannot distinguish between these potential explanations of our findings. We are unable to determine the impact of continuing or later exposure to DDT or other contaminants or to changes in neighborhood-level of individual social factors.

6. Conclusions

ICE measures at birth across four domains – education, race, income and combined income/race – distinctly captured social disadvantage, were correlated with individual prenatal DDT exposure and were strongly associated with adolescent outcomes that correlate with cancer risk over the life-course. As we show here, ICE analyses revealed the impact of neighborhood disparities in income that are independent of race and family household income and align with environmental exposures which play a potential substantial role in shaping cancer risk factors. Importantly, as area-level indices, ICEs provide an accessible and effective risk metric that identifies neighborhoods for implementing targeted intervention to reduce cancer risk on a large scale.

Supplementary Material

Appendix A. Supporting information

Supplementary data associated with this article can be found in the online version at doi:10.1016/j.reprotox.2025.108944.

Abbreviations:

ICE

Index of Concentration at the Extremes ICE

BMI

Body Mass IndexChild Health and Development Studies

p,p′-DDT

1,1,1-trichloro-2,2-bisp-chlorophenylethane

p,p′-DDE

1,1′-dichloro-2,2′-bisp-chlorophenylethylene

o,p′-DDT

1,1,1-trichloro-2-pchlorophenyl-2-o-chlorophenyl-ethane

AOR

Adjusted Odds Ratio

95 % CI/CL

95 % Confidence Interval/Confidence Limit

Data Availability

De-identified (anonymized) data are available upon request from Barbara A. Cohn, PhD, Director of the Child Health and Development Studies. Requests will be reviewed by Dr. Cohn, research staff, and the Institutional Review Board at the Public Health Institute. Approval of requests for de-identified (anonymized) data requires execution of a data use agreement.