Community credit scores and community socioeconomic deprivation in association with type 2 diabetes across an urban to rural spectrum in Pennsylvania: a case–control study

Introduction

Affecting 11% of the US adult population, diabetes is a significant public health challenge.1 Geographic disparities in type 2 diabetes (T2D) incidence and prevalence2–4 have motivated epidemiologic research regarding contexts that influence T2D risk. Identifying contextual risk factors can help target care and inform interventions aimed at distal built and social constructs to prevent T2D.

Such research commonly assesses community socioeconomic conditions due to their downstream influence on other contextual factors, including built and physical environments (eg, walkability and natural environments), social environments (eg, social cohesion, safety and healthcare access) and related health behaviours (eg, diet and physical activity).5 Community-level disadvantage is also associated with physiological stress, a risk factor for chronic disease such as T2D.6 Studies generally use deprivation indices that aggregate socioeconomic indicators to characterise community material deprivation.7 Numerous studies have reported associations between area-level deprivation and T2D prevalence, incidence and control,5 including our prior study of Pennsylvania communities.8

Other measures of socioeconomic conditions, such as community credit scores (CCS; individual credit scores averaged across a community), are less studied but may capture conditions uniquely related to health outcomes. As a community-level measure, CCS may represent communities’ access to financial resources and economic stability.9 10 For example, financial institutions use area-level credit scores to determine the interest rates, loans and other financial products available in certain areas. Businesses may use them to decide where to expand and which products to offer or market in specific areas. In these ways, CCS may influence communities’ physical, social and economic environments, which may in turn affect disease risk.11 CCS and community-level deprivation may therefore overlap in their contextual characterisations of communities, downstream impacts on other contextual factors and concomitant health outcomes; however, CCS may better represent communities’ economic stability and vitality and may have independent associations with health outcomes.

Few studies have examined CCS as a contextual factor nor its associations with health outcomes, including T2D. One exception, conducted in Philadelphia, found that CCS was associated with lower self-reported diabetes prevalence.9 No studies have evaluated CCS in association with health outcomes in non-urban US contexts, nor investigated whether CCS may capture novel, health-salient aspects of community socioeconomic conditions beyond traditional measures of community-level material deprivation. The primary aim of this study was to evaluate CCS in association with new onset of T2D across a geographically diverse region of Pennsylvania. We hypothesised that higher CCS is associated with lower T2D risk. Our secondary aim was to disentangle associations of CCS from community socioeconomic deprivation (CSD) in relation to T2D risk.

Methods

Results

The 15 888 cases of new onset T2D and 79 435 controls (65 084 unique individuals) were geocoded into 1070 communities (633 townships, 291 boroughs and 146 city census tracts). Characteristics of study individuals have been previously described.8 15 They had a mean age of 55 years and were predominately White and non-Hispanic, reflecting the study region’s racial and ethnic composition. Townships and boroughs had the least racial and ethnic diversity; approximately 98% of study individuals were White and 99% non-Hispanic in these community types. In city census tracts, 94.2% were White and 4.3% Black and 94.9% were non-Hispanic. T2D cases had higher historical use of Medical Assistance (6% vs 3% of controls), indicating lower household socioeconomic status. Cases were more likely than controls to live in a city census tract, less likely to live in a township and more likely to live in a community with lower CCS and higher CSD (table 1).

Distributions of community-level variables

There were few communities in the ‘fair’ or ‘low poor’ CCS categories, except within the highest CSD quartile in city census tracts (online supplemental figure 2). Spearman’s rho correlations between (continuous) CCS and CSD, calculated at the community-level, showed a strong negative correlation within city census tracts (−0.76) and moderate negative correlations in boroughs (−0.57) and townships (−0.43) (online supplemental figure 3). Three per cent of study communities, two-thirds of which were townships, had an unexpected discordance between CCS and CSD, including 12 communities with high CCS and CSD and 16 communities with low CCS and CSD (figure 1).

Request reuse permission

Figure 1

Map of study communities in Pennsylvania, USA, created using ArcGIS V.10.4 (ESRI, Redlands, California, USA). Red and blue cross-hatching shows communities with discordant measures of community socioeconomic deprivation (CSD) and community credit scores (CCS) (ie, communities with high deprivation and high CCS and vice versa).

Spearman’s rho correlations between CCS and the proportion of the population in each community who had ever received Medical Assistance showed moderately weak negative correlations (city census tracts: −0.32; boroughs: −0.23 and townships: −0.24). Because correlations appeared to be influenced by communities containing only a few study individuals, we recalculated correlations for communities with at least five study individuals (n=750 communities). Correlations strengthened (city census tracts: −0.64; boroughs: −0.45 and townships: −0.33) (online supplemental figure 4).

Discussion

This study is among the first to evaluate area-level credit scores as a contextual factor in relation to an objective health outcome and across a heterogeneous geography representing the urban–rural spectrum. As hypothesised, higher CCS was associated with lower T2D risk, though the strength and pattern of associations differed by community type. Furthermore, CCS and CSD appeared to be independently associated with T2D in a limited set of communities, providing evidence of their unique characterisations of community socioeconomic context. However, questions remain as to how the mechanisms by which CCS influences T2D differ from that of area-level material deprivation and whether CCS reflects contextual or compositional (ie, individual-level) effects.

Non-stratified model results revealed a trend of lower T2D risk with increasing CCS. Community-stratified models revealed idiosyncrasies in this relationship. First, consistent associations only for the highest CCS category across community types indicated that our findings were limited to the high end of the CCS distribution. Second, the strongest associations appeared in city census tracts. This is partially explained by a more even distribution of city census tracts across CCS categories, whereas few boroughs and townships were in the lowest CCS category, but could also reflect a stronger link between CCS and downstream environmental or behavioural factors that influence T2D onset. This could occur if populations with greater homogeneity in credit scores were segregated into distinct communities—a situation more likely in metropolitan areas than large rural townships. Unfortunately, we could not explore CCS variance with the available data. Third, individuals in townships with ‘low fair’ CCS had lower odds of T2D than those in ‘high fair’ townships, which was an unexpected result.

Given higher proportions of racially and ethnically minoritised groups in city census tracts, the stronger associations of CCS and T2D in this community type could potentially reflect a differential impact of CCS among these groups. It is plausible that due to residential segregation and discrimination, minoritised groups may experience the community socioeconomic context differently than majority groups, with concomitant impacts on health status, including T2D. To explore the influence of race and ethnicity on associations of CCS and T2D, we examined whether these factors modified CCS–T2D associations in city census tracts but found no supporting evidence. However, this exploratory analysis was limited by small sample sizes across the distribution of CCS categories and by the fact that—based on 2010 Census data—only half of city census tracts in our geography comprised<80% White individuals (a conservative definition of racially diverse communities21).

Credit score data are proprietary and thus less accessible than deprivation measures, which are commonly created with publicly available data. To justify investment in credit score data, CCS should uniquely characterise community context beyond area-level material deprivation, which is ubiquitously studied and linked to numerous health outcomes,22 including T2D.5 8 Through stratified models, we examined independent associations of CCS and CSD. A strong negative correlation between CCS and CSD in city census tracts and few low CCS/low CSD communities across community types constrained this analysis and demonstrated the inter-relatedness of CCS and CSD. However, CCS remained protective of T2D risk among the highest CCS communities, regardless of CSD and among high CCS communities, CSD remained associated with T2D. Furthermore, a small proportion of communities—particularly townships—had unexpectedly discordant measures for CCS and CSD (ie, both high or both low). Taken together, these findings suggest that CSD and CCS measure different community features.

CCS and CSD thus seem to each capture health-salient features that do not fully overlap. Considering the observed CCS–T2D associations primarily showed a protective effect of ‘good’ CCS, we conjecture CCS may be a stronger measure of community economic stability and vitality. Area-level credit scores are used by financial institutions to determine the financial resources (eg, loans and interest rates) available to an area10 and by businesses to assess the financial potential of markets. Access to favourable loans is intrinsically tied to housing quality and the availability and quality of amenities in a community, characteristics that persist over time, as demonstrated by studies of historic redlining.23 CCS may therefore influence the physical environment (eg, placement of food, physical activity, tobacco and alcohol outlets; public amenities such as green space, schools and libraries) and social environment (eg, neighbourhood cohesion and safety) in ways that are similar to, but not fully overlapping with CSD. We theorise that CCS may better measure health-promoting features of communities, while CSD may better capture community-level disadvantage that adversely affects health.

Our findings revealed variation in CCS–T2D associations by CSD strata, with inconsistent patterns across community type. Associations attenuated in the most deprived city census tracts and boroughs, suggesting that the positive impact of CCS on T2D risk was offset, to some degree, by higher deprivation. The opposite was true in townships: CCS–T2D associations strengthened in townships with high CSD. One potential explanation for this inconsistency pertains to the measurement of CSD, with quartile cut-offs determined within community type rather than across all communities. For this reason, the most deprived townships did not have as extreme ranges of CSD as did city census tracts, for example, and so the degree of deprivation in townships may not have been sufficient to offset the positive impact of CCS. However, CIs for CCS–T2D associations encompassed the point estimates across strata, so stratum-specific differences may not be highly meaningful.

CCS–T2D associations may arise under two conditions. As a contextual measure, higher CCS would impact all individuals in a particular community, leading to lower population-level T2D risk,24 as described above. Alternatively, as a compositional measure, the observed CCS–T2D associations could be due to individual-level effects of credit scores on T2D, as demonstrated in prior studies of chronic disease outcomes.25 26 Such associations could arise due to a link between human capital factors (such as educational attainment, cognitive ability and self-control) with T2D, mediated through health behaviours.26 Associations of individual credit scores and T2D could similarly occur if high credit scores reflect higher individual socioeconomic position, which is consistently associated with lower T2D risk.11 However, we observed weak to moderate correlations between CCS and the proportion of individuals in each community with a history of using Medical Assistance, a proxy for household socioeconomic status. Similarly, a study in Philadelphia observed moderate correlations (from −0.78 to 0.49) between neighborhood-level credit scores and neighborhood-level measures of individual socioeconomic position such as income, educational attainment and median housing value,10 and nationally, individual credit scores were only moderately correlated with household income, with wide variation in credit score distributions within income groups.27 Deciphering whether CCS represents a contextual or compositional measure is further complicated by evidence suggesting the local credit economy underpins individual credit scores, which in turn may influence an area’s overall creditworthiness.9 For example, research found that residents of areas recovering from a local economic downturn received lower credit ratings than individuals who had the same credit history but lived in more economically robust areas.28 Furthermore, individual credit scores have been shown to impact residential location decisions and thus the community characteristics to which individuals are exposed, such as access to quality education, proximity to amenities, crime rates and air quality.29

This study had limitations. CCS had a limited range in the study geography; it was skewed towards higher categories, with few communities scoring ‘poor’ but also no communities scoring above ‘good’. This particularly limited our ability to assess associations in boroughs. Additionally, although the Geisinger primary care population is representative of the region,14 the region itself (and therefore our study population) lacks racial and ethnic diversity, particularly in boroughs and townships. For this reason, we only controlled for race and ethnicity at the individual level and not for community-level racial and ethnic composition. The generalisability of our findings to more diverse populations remains unknown, particularly considering that Black and Hispanic individuals are less likely to have a credit score on record30 and that credit score models penalise borrowers for using some credit types that are disproportionately used by minoritised and economically disadvantaged individuals.31

Conclusion

There is great interest in understanding community contexts that confer risk for T2D to inform policy and programmatic efforts. We observed associations of high CCS with lower T2D risk, and in a subset of mainly non-urban communities, associations were independent of CSD. Thus, in some communities, CCS may capture novel, health-promoting aspects of community socioeconomic conditions. However, future research on CCS must weigh the limitations of this data—namely, its propriety nature—and continue to explore mechanisms by which CCS influences health.