Effects of Family and Neighborhood Risks on Glycemic Control Among Young Black Adolescents with Type 1 Diabetes: Findings from a Multi-center Study

Abstract

Objective

While individual and family risk factors that contribute to health disparities in children with type 1 diabetes have been identified, studies on the effects of neighborhood risk factors on glycemic control are limited, particularly in minority samples.

Method

This cross-sectional study tested associations between family conflict, neighborhood adversity and glycemic outcomes (HbA1c) in a sample of urban, young Black adolescents with type 1 diabetes(mean age=13.4 + 1.7), as well as whether neighborhood adversity moderated the relationship between family conflict and HbA1c. Participants (N=128) were recruited from five pediatric diabetes clinics in two major metropolitan US cities. Diabetes-related family conflict was measured via self-report questionnaire (Diabetes Family Conflict Scale; DFCS). Neighborhood adversity was calculated at the census block group level based on US census data. Indictors of adversity were used to calculate a neighborhood adversity index (NAI) for each participant. Median family income was $25,000, suggesting a low SES sample.

Results

In multiple regression analyses, DFCS and NAI both had significant, independent effects on glycemic control (β=.174, P=.034 and β=.226 P=.013, respectively) after controlling for child age, family socioeconomic status and insulin management regimen. Tests of effects of the NAI and DFCS interaction on HbA1c found no significant moderating effects of neighborhood adversity.

Conclusions

Even within contexts of significant socioeconomic disadvantage, variability in degree of neighborhood adversity predicts diabetes-related health outcomes in young Black adolescents with type 1 diabetes. Providers should assess social determinants of health such as neighborhood resources that may impact adolescents’ ability to maintain optimal glycemic control.

Introduction

Black youth with type 1 diabetes are at heightened risk for suboptimal glycemic control (1,2). In turn, suboptimal glycemic control has been shown to result in higher rates of diabetes complications in this population (3). A variety of individual and family-level risk factors have been shown to be associated with disparities in health outcomes among Black youth with type 1 diabetes. Individual-level risk factors include those associated with clinical management, such as lower rates of insulin pump use (4) and psychological factors such as fewer coping skills (5). Family-level risk factors include lower SES and higher rates of single-parent families (6). Studies of family relationship characteristics have also shown that caregivers of minority youth with type 1 diabetes engage in lower rates of supervision of daily diabetes care (7) and report higher rates of diabetes-related family conflict (8) than White caregivers.

While individual and family risk factors that contribute to disparities in health outcomes among youth with type 1 diabetes have been explored, studies on the effects of neighborhood risk factors on glycemic control are relatively limited. Neighborhoods are understood to be an important context of adversity and stress in daily life (9). In the broader child development literature, the detrimental effects of poor neighborhood quality, including inadequate housing stock (10) and high rates of crime and violence (11), on children’s health are well documented. While the effects of neighborhood characteristics such as neighborhood SES (12), stability of neighborhood income (13), population density (14) and rurality (15) have been investigated in large national and international epidemiological studies on type 1 diabetes incidence, such factors have been largely unexplored in studies on predictors of health outcomes in youth with type 1 diabetes. Of the few existing studies that have directly tested the effects of neighborhood factors on health outcomes such as glycemic control, most have focused on White, high SES youth likely to be residing in neighborhoods with low levels of adversity. For example, one recent study investigated the relationship between neighborhood disorder and HbA1c in a mostly White, U.S. sample of older adolescents and young adults with type 1 diabetes (16). Higher neighborhood disorder was found to be associated with higher HbA1c. A Canadian study reported that higher neighborhood equity was associated with lower HbA1c in sample of youth with type 1 diabetes (17). However, information regarding the sample’s race/ ethnicity was not reported. A third study that enrolled a subset of minority youth found associations between neighborhood disadvantage and HbA1c for Black, but not White, youth aged five to twenty years (18). However, the relatively small sample of Black participants (n=33) limits the generalizability of the study findings to the broader population of Black youth with type 1 diabetes.

While recent studies have begun to demonstrate the importance of neighborhood characteristics for understanding health outcomes in youth with type 1 diabetes, no studies have investigated whether neighborhood risk and protective factors may interact with family-level risk and protective factors such as family relationships to predict glycemic control. Socio-ecological models of child development such as the Family Stress Model (19) suggest that family relationship quality is particularly important for adolescents’ adjustment and health when families reside in neighborhoods with high levels of adversity. Such families experience lower-quality resources, higher rates of crime, and other chronic psychological stressors (20) and family relationships may play an even more important role in such neighborhoods. Consistent with predictions from such models, studies using samples of adolescents without chronic health conditions demonstrate that in high-risk neighborhoods, the effects of various components of family relationships such as parenting quality, affect/warmth and amount of parental supervision on youth psychological adjustment and health outcomes are more pronounced than they are in low-risk neighborhoods (21–25).

In the pediatric diabetes literature, the Diabetes Resilience Model (26) proposes that individual, family and neighborhood/community risk and protective factors may interact to predict diabetes health outcomes. However, studies testing the interaction of neighborhood risk factors with individual and family risk factors in samples of youth with type 1 diabetes are lacking to date. Family conflict has been identified as a robust predictor of glycemic control in multiple studies of youth with type 1 diabetes (27, 28). Therefore, the potential for neighborhood adversity to moderate the relationship between family conflict and glycemic control warrants further study. For example, under conditions where neighborhood adversity-and related stressors such as violence, crime and noise- is higher, the association between family conflict, which can also cause youth stress, and glycemic control might be stronger than under conditions where neighborhood adversity is lower. Explorations of such moderator effects within low SES and minority samples of youth with type 1 diabetes are also of importance because protective factors in such samples may differ from those found among high SES or White samples (29).

The objectives of the present study were 1) to test associations between family conflict, neighborhood adversity and health outcomes in a sample of young Black adolescents with type 1 diabetes and 2) to determine if neighborhood adversity moderated the relationship between family conflict and glycemic control. An additional objective was to use improved methods to capture variability in neighborhood adversity by using a more precise geographic unit of analysis, the census block group, to represent neighborhoods. Existing studies described above that have investigated the relationship between neighborhood characteristics and diabetes health in youth with type 1 diabetes used census tract level indicators to characterize neighborhoods rather than the smaller census block group unit of analysis. Census block groups are the geographical areas that compose census tracts, and their smaller geographic and population size represent a more sociologically meaningful representation of the neighborhood concept and are therefore a better geographic unit of analysis for the measurement of neighborhood attributes; conversely, census tract level measures are more likely to obscure neighborhood variability and also to be less reliable (30).

Methods

Data for the present study were drawn from a clinical trial investigating the effectiveness of an eHealth intervention to promote optimal glycemic control in young Black adolescents. The trial was registered in clinicaltrials.gov (registration number NCT03168867). Data used in the analyses were drawn from the participant’s baseline data collection prior to study randomization or delivery of the intervention. Data from the first 129 participants to enroll in the clinical trial were selected for use in the present study.

Participants were recruited from two pediatric diabetes clinics located in the greater metropolitan Detroit area and three located in Chicago between 2017 and 2019. All clinics were staffed by pediatric endocrinologists and followed American Diabetes Association (ADA) guidelines for the care of children with diabetes. In order to be eligible for the clinical trial, adolescent participants had to be between ten and fourteen years of age, be diagnosed with type 1 diabetes for at least six months, self-identify as Black/African American and be residing with a caregiver who was willing to participate in the study. No child psychiatric diagnoses were exclusionary, with the exception of moderate or severe cognitive impairment, suicidal ideation and psychosis. Families were also excluded if they were not English speaking, could not complete study measures in English or if the child had a medical diagnosis leading to atypical diabetes management (e.g. cystic fibrosis). Potential participants were either 1) recruited through letters describing the study followed by phone calls from research staff to assess interest or 2) approached in person at the time of a regularly scheduled clinic visit. 27% of eligible participants who could be contacted refused participation. The research was approved by the IRB of the first author’s university using a single IRB agreement. All participants provided informed consent and assent to participate. Demographic characteristics of the participants are shown in Table 1. Mean yearly family income was $33,100, corresponding to approximately 125% of the US 2020 poverty line for a family of four; median family income was $25,000 (<$10,000=16%;$10,000-$19,999=21%;$20,000-$29,999=17%;$30,000-$39,000=10%; $40,000-$49,999=11%;$50,000-$59,999=10%;≥$60,000=15%). Mean hemoglobin A1c was 11.5% (102 mmol/mol), suggesting that the sample was in suboptimal glycemic control. Community samples of youth with type 1 diabetes typically demonstrate suboptimal control in comparison with guidelines for HbA1c to be maintained at or below 7.0% in adolescents (31); the suboptimal glycemic control demonstrated in the present sample is also consistent with known disparities in health outcomes for Black youth with type 1 diabetes. For example the mean HbA1c of Black youth in the T1D Exchange was reported to be 9.6% in one recent study (4) and was reported to be 12.5% in another study focusing on health outcomes in Black youth with diabetes conducted in two US cities (32).

Table 1.

Descriptive statistics for demographic variables, diabetes-related variables, family conflict and neighborhood adversity

Child age (years)	13.4 ± 1.7
Child gender
Male	58 (45)
Female	70 (55)
Duration of diabetes (years)	6.0 ± 3.9
HbA1c
%	11.5 ±2.7
mmol/mol	102 ± 30
Insulin regimen
Basal Bolus Therapy-Injection	90 (70)
Basal Bolus Therapy-Pump	30 (24)
Other	8 (6)
Recruitment Site
Detroit	76 (59)
Chicago	52 (41)
Caregiver age (years)	42.0 ± 9.3
Caregiver gender
Female	115 (90)
Male	13 (10)
Caregiver ethnicity
African American	120 (94)
Other	8 (6)
Caregiver education (years)	13.3 ± 2.1
Number of caregivers in the home
Two	63 (49)
One	64 (50)
Not reported	1 (1)
Yearly Median Income in US dollars	25,000
DFCS	32.4 ± 8.1
NAI	0.69±0.79

Note. Data are mean ±SD or n (%) except as noted for family income. DFCS= Diabetes Family Conflict Scale; NAI= Neighborhood Adversity Index.

Procedures

All measures were collected by a trained research assistant in the participants’ homes. Both the youth and the primary caregiver completed questionnaires. Families were provided $50 to compensate them for participating in the data collection session.

Measures

Neighborhood Adversity Index (NAI). Informed by an approach validated by Messer and colleagues (33), we used possible indicators of neighborhood adversity available in the Census Bureau’s American Community Survey 5-year estimates for 2017 census block groups, released in 2019. The census block groups included to create the NAI for the Detroit and Chicago metropolitan areas were those in the core counties of each area (three for Detroit, five for Chicago). Core counties were selected from a larger set of counties included in the city’s metropolitan statistical area (MSA) by an evaluation of both geographic proximity to the urban core and geographic continuity with each other as well as by excluding counties on the geographic perimeter where population densities exhibited a distinct decrease from the more densely populated counties. In the Detroit metro area, this decrease in density was indicated by a drop from 608 people per square mile to 129 people per square mile; in the Chicago metro area, the exclusion was suggested by a 810 to 512 shift in persons per square mile while also taking the other criteria into account. All but five families of those participating in the current study (3.9%) resided in one of these higher population density core counties.

Fifteen variables representing domains of education, employment, housing, occupation, poverty, and socio-demographics were included. The distribution of each was examined and transformations conducted where necessary to eliminate significant skew and kurtosis. First, the full set of indicators were included in separate factor analyses to determine the best indicators of neighborhood diversity for each metro area. For each metro area, one factor explained a high percentage of the overall variance, with all included indicators loading over 0.48. Factor solutions for each metro area were slightly different but nine indicators were common across both areas’ factor. The nine indicators included: median household income; percent persons in poverty; percent of households with no vehicle available; percent of persons with less than a 12^th grade education, no diploma; percent of households renter occupied; percent females in management occupations; percent males in management occupations; percent of housing units vacant; and percent female headed households. A second set of analyses forced a one-factor solution with the nine indicators common across both metro areas. The factor scores for each census block group were saved, representing the Neighborhood Adversity Index for each block group. Using this approach, the NAI scores were based on the distribution of adversity across each metro area but unique to each neighborhood, weighted for the contribution of each indicator, normally distributed, and standardized, with higher (positive) scores representing higher neighborhood adversity. The NAI differed from prior measures of neighborhood risk used in studies of health outcomes in youth with type 1 diabetes as it 1) used improved methods to derive the score, as other studies have used a composite summary score not calculated through factor analyses (16) 2) had lower likelihood of bias related to inclusion of indicators that might be confounded with glycemic control, such as the number of preventable hospitalizations in the neighborhood (17) and 3) included of a broader array of neighborhood characteristics than those in previous studies using factor analytic approaches to calculate the score (18). In addition, as noted earlier, none of the prior studies investigating the relationships between neighborhood adversity and glycemic control in youth with type 1 diabetes used census block group-level data to derive their scores.

In order to determine the NAI score for each participant, the MMQGIS add-on package was used within QGIS software (34) to geocode the participant’s address to latitude and longitude coordinates and to then link these to their respective census block group. In two cases, families resided within the same census block group. In order to avoid any confounding associated with potential lack of independence of the youth and family-level data and the neighborhood-level data, one of these families was randomly excluded, leaving a final N of 128.

Sociodemographic and Clinical Variables.

A self-report questionnaire was used to obtain information from the primary caregiver on demographic variables such as age, gender, annual family income and maternal education. The adolescent’s medical chart was reviewed to obtain information such as duration of diabetes and insulin delivery method.

Diabetes Family Conflict.

The revised Diabetes Family Conflict Scale (DFCS) (35) was used to evaluate family conflict associated with adolescent diabetes management. The DFCS is a 19-item questionnaire; conflict is rated on a 3-point scale (1 = never argue, 2 = sometimes argue, and 3 = always argue), with higher scores indicative of greater conflict. The parent-report version of the DFCS was used in the present study. Internal consistency of the measure was high (coefficient alpha=.90).

Glycemic control.

Hemoglobin A1c (HbA1c), a retrospective measure of average blood glucose during the past two to three months, was used to evaluate glycemic control. Values were obtained during study data collection visits using the Accubase test kit, which is FDA approved. The test used a capillary tube blood collection method instead of venipuncture and was therefore suitable for home-based data collection. High performance liquid chromatography (HPLC) was used to analyze the blood sample.

Analytic Plan

Missing data.

The NAI variable was missing five values which were imputed using the maximum likelihood expectancy maximization approach. The family income variable was missing four values, which were estimated with the sample median, as the income variable was strongly skewed. Using maximum likelihood approaches to missing data imputation can produce biased estimates when variables are skewed (36).

Data analyses.

Moderated multiple regression (MMR) analyses were used to test the main effects of NAI, DFCS and the hypothesis that neighborhood adversity moderated the relationship between family conflict and glycemic control. Because moderator effects/interactions are sensitive to distributional assumptions, scatter plots and bivariate regression were used to examine the regression of HbA1c on family conflict and neighborhood adversity. Locally Weighted Scatterplot Smoothing (LOWESS) regression was used to determine the form of the equations to be used in modeling the relationship between HbA1c and family conflict. Interaction terms were constructed as the cross products of main effect terms and entered into the equation after main effects and covariates. Continuous variables were mean-centered prior to forming cross product terms. Potential covariates in the models were determined empirically from a list of theoretically relevant candidates as follows: child age, child gender, duration of diabetes, insulin delivery method, caregiver educational attainment, family income, recruitment location. Covariates were first evaluated to determine if they were significantly related to HbA1c and either family conflict or neighborhood adversity at the trend level (p=.10). Three covariates that met this threshold were subsequently included in the regression models. These covariates were child age, family income and insulin delivery method.

Results

Family conflict (r = .242; P=.006) and neighborhood adversity (r=.318; P=.001) were both significantly related to HbA1c in bivariate analyses (Figure 1, panel A and B). The relationship between family conflict and neighborhood adversity (r=.041 P=.641) was nearly zero (Figure 1, panel C), indicating that these variables made independent contributions to the prediction of glycemic control.

Figure 1.

Scatter plots, regressions, and predicted values relevant to moderated multiple regression analysis. Regression of HbA_1c on Diabetes Family Conflict Scale (FC) and Neighborhood Adversity Index (NAI) (A, B). FC and NAI were not associated (panel C). Moderated multiple regression analysis (D-F): LOWESS regression of HbA1c on FC within high and low NAI groups (D). The LOWESS regressions showed some degree of nonlinearity around the midpoint of the Diabetes Family Conflict Scale. Moderated multiple regression tested the piecewise (E) and linear model (F). The tests of the interaction in both models was not significant.

The LOWESS regression was performed in the total sample and within subgroups defined by high and low NAI scores. This analysis revealed a nonlinear trend in the regression of HbA1c on DFCS (see Figure 1 Panel D). When DFCS was < 37, there was a linear, positive relationships between HbA1c and DFCS at both low and high levels of NAI. When DFCS was >37; there was no relationship between HbA1c and DFCS at high levels of NAI, but a linear, negative relationship between HbA1c and DFCS at low levels of NAI. Because of this pattern in the data, one of the three MMR models testing the hypothesized interaction between NAI and DFCS used a piecewise linear regression model.

The results of the MMR analyses are shown in Table 2. Three different equations using different functional forms were tested in the analyses. Each model included NAI, DFCS, the NAI X DFCS interaction, and related covariates (child age, family income, and insulin delivery method). Model 1 tested the differences in the piecewise trends (see Table 2; and Figure 1, panel E). Model 2 tested the linear trend by group (Table 2; and Figure 1, panel F). NAI was dichotomized at the median and entered as group. Model 3 tested the linear by linear interaction (Table 2); NAI and DFCS were used as continuous variables in this model and the interaction term was evaluated by calculating the product of these variables (NAI x DFCS), centered at their means. Some evidence was found in Model 1 for a possible interaction effect based on the interaction contrast between the two piecewise segments when DFCS was greater than 37 (see Figure 1, panel E). However, the difference between these two piecewise contrasts was not significant (P = .277). Overall, in none of the three models was the test of the DFCS X NAI interaction found to be significant. The change in R² associated with adding the interaction to the model accounted for no more than 1.1% of the variance in any of the models tested.

Table 2.

Moderated multiple regression results testing the joint effect of family conflict and neighborhood adversity on metabolic control (HbA1c)

Model a	Test of interaction		Final model without interaction
	Change in R2	p-value	R2	adj R2
Model 1: Piecewise x Group b,c	.011	.416	.263	.226
Model 2: Linear x Group d,e	.003	.515	.240	.202
Model 3: Linear x Linear f	.004	.445	.248	.217

Note.

^aVariables in each model were Neighborhood Adversity Index (NAI), Diabetes Family Conflict Scale (DFCS), the NAI X DFCS interaction, and covariates (age, insulin regimen, and family income),

^bNeighborhood Adversity Index was dichotomized at the median and entered as Group,

^cTwo interaction contrasts are also tested with this model. The second piecewise contrast had a smaller P values (P = .26), reflecting the observed difference in HbA1c between the groups when family conflict scores were above 37; Above 37, the high group was flat and the low group trended down,

^dDiabetes Family Conflict Scale was centered prior to analysis,

^eWeighted and unweighted analyses were performed. There was very little difference in the two analysis and the unweighted results are shown here,

^fThe DFCS and NAI scales were centered prior to creating a product term (DFCS X NAI) to test the interaction; the full continuous scales for both measures were used in this model.

Based on the results of the MMR analyses, a final linear regression analysis was conducted excluding the NAI X DFCS interaction term and using NAI and DFCS as continuous variables. The regression coefficients of the model are shown in Table 3. This model accounted for 24.8 % of the variance in HbA1c. Child age (P=.002), DFCS (P=.034) and NAI (P=.013) each had significant effects on HbA1c. Based on the model, a one standard deviation increase in family conflict was associated with a 0.48% (.060 * 8.04) increase in HbA1c while a one standard deviation increase in NAI was associated with a 0.62% (.815 * .765) increase in HbA1c

Table 3.

Coefficients in final multiple regression model of HbA1c on family conflict, neighborhood adversity, and selected covariates

Predictor	B	β	95% CI	p-value
Diabetes Family Conflict (DFCS)	.060	.174	.004, .115	.034
Neighborhood Adversity Index (NAI)	.815	.226	.177, 1.453	.013
Child Age	.419	.252	.156, .681	.002
Family Income	−.067	−.063	−.256, .122	.482
Insulin Regimen	−1.001	−.154	−2.076, .073	.068

Note. Insulin regimen was coded as other regimen (1) vs pump (2)

Discussion

The purpose of the present study was to evaluate the associations between family conflict, neighborhood adversity and glycemic control in a sample of young Black adolescents with type 1 diabetes residing in two major metropolitan areas in the US and to determine if neighborhood adversity moderated the relationship between family conflict and glycemic control. The majority of youth participating in the study resided in families living at or around the U.S. federal poverty line. Findings from the present study showed that diabetes-related family conflict and neighborhood adversity both had significant, independent effects on glycemic control even after controlling for confounders such as child age, family-level socioeconomic variables (family income) and type of insulin management regimen. This finding is noteworthy in demonstrating that variability in degree of neighborhood adversity predicts diabetes-related health outcomes in Black children with type 1 diabetes, even within a sample characterized by significant socioeconomic disadvantage. In particular, in the present study, a one standard deviation increase in neighborhood adversity was associated with 0.6% increase in HbA1c, which is clinically meaningful (37). It has previously been suggested that the socio-economic disadvantage and disorder created by poor neighborhood quality may cause suboptimal health for residents through psychological stress processes (38, 39). Prior work by our group also lends support to the possibility that socio-economic disadvantage in minority youth with type 1 diabetes may led to suboptimal glycemic control through elevated diabetes-related stress and dysregulated cortisol (40). This study found that self-reported socio-economic disadvantage was associated with poorer glycemic control via indirect pathways linking disadvantage to higher perceived diabetes stress, higher perceived diabetes stress to dysregulated cortisol, and, in turn, dysregulated cortisol to elevated HbA1c. More studies to understand how neighborhood risk factors affect health outcomes in youth with type 1 diabetes, and testing the possibility that stress-related pathways may be implicated, are warranted.

In the present study, family conflict and neighborhood adversity were not significantly associated. A possible explanation is that we used a measure of diabetes-specific family conflict rather than general family conflict. On the other hand, prior studies testing models that included both family and neighborhood risk factors have shown mixed results in this regard, with some demonstrating significant but low order associations between family and neighborhood-level risk factors (25, 41) and some demonstrating no relationship (21).

Findings from the present study did not show that neighborhood adversity played a significant role in moderating the relationship between family conflict and glycemic control. The relatively small number of families reporting high levels of family conflict may have affected our ability to detect such effects. In addition, we chose to focus on family conflict as our measure of family relationships. Other aspects of family relationships that have been shown to be related to glycemic control among youth with type 1 diabetes, such as family warmth, support and parental involvement (42–44), were not measured in the present study. Such variables could be explored in future studies for their potential to interact with neighborhood risk factors to predict diabetes health outcomes. Likewise, culturally relevant aspects of family relationships that have previously been demonstrated to interact with neighborhood factors to protect against suboptimal health outcomes in studies of Black youth, such as kinship support (45), were not included in the present study.

Study limitations include the cross-sectional nature of the analyses and the urban nature of the sample, which limits the generalizability of findings to Black youth living outside of urban areas. In addition, the sample consisted of youth and families who agreed to participate in a clinical trial and therefore may differ from the broader population of minority youth or from those who declined to participate in the trial. Reports of family conflict were obtained from parents, but not from the youth; adolescent and parent perspectives on conflict related to diabetes management may have differed which might also have affected the findings (46). Additional studies with longitudinal data are needed to establish a causal relationship between neighborhood adversity and glycemic control and to evaluate the relative importance of family versus neighborhood risk factors for predicting diabetes health outcomes over time.

In summary, young Black adolescents with type 1 diabetes are at increased risk of suboptimal glycemic outcomes and diabetes complications. Findings from the current study highlight the importance for diabetes care providers caring for youth with type 1 diabetes to screen not only for individual and family risk factors, but also for broader social determinants of health such as neighborhood adversity that may impact ability to maintain optimal glycemic control. Calls for hospitals and practices to conduct routine screening of factors such as housing instability and inadequate transportation, which disproportionately affect youth living in risky neighborhoods, continue to increase (47); brief, self-report screening tools for social determinants of health that include assessments of neighborhood risk factors are now becoming more widely available (48). Such screening could allow providers to identify which families are most in need of additional resources and may lead to increased referrals to sources of support, which in turn may help to optimize diabetes health outcomes in Black youth.