Using universal kriging to improve neighborhood physical disorder measurement

Machine Learning Approaches for Measuring Neighborhood Environments in Epidemiologic Studies

Purpose of review

Innovations in information technology, initiatives by local governments to share administrative data, and growing inventories of data available from commercial data aggregators have immensely expanded the information available to describe neighborhood environments, supporting an approach to research we call Urban Health Informatics. This review evaluates the application of machine learning to this new wealth of data for studies of the effects of neighborhood environments on health.

Recent findings

Prominent machine learning applications in this field include automated image analysis of archived imagery such as Google Street View images, variable selection methods to identify neighborhood environment factors that predict health outcomes from large pools of exposure variables, and spatial interpolation methods to estimate neighborhood conditions across large geographic areas.

Summary

In each domain, we highlight successes and cautions in the application of machine learning, particularly highlighting legal issues in applying machine learning approaches to Google’s geo-spatial data.

Pathways between objective and perceived neighborhood factors among Black breast cancer survivors

BACKGROUND

Mounting evidence supports associations between objective neighborhood disorder, perceived neighborhood disorder, and health, yet alternative explanations involving socioeconomic and neighborhood social cohesion have been understudied. We tested pathways between objective and perceived neighborhood disorder, perceived neighborhood social cohesion, and socioeconomic factors within a longitudinal cohort.

METHODS

Demographic and socioeconomic information before diagnosis was obtained at interviews conducted approximately 10 months post-diagnosis from participants in the Women's Circle of Health Follow-up Study - a cohort of breast cancer survivors self-identifying as African American or Black women (n = 310). Neighborhood perceptions were obtained during follow-up interviews conducted approximately 24 months after diagnosis. Objective neighborhood disorder was from 9 items audited across 23,276 locations using Google Street View and scored to estimate disorder values at each participant's residential address at diagnosis. Census tract socioeconomic and demographic composition covariates were from the 2010 U.S. Census and American Community Survey. Pathways to perceived neighborhood disorder were built using structural equation modelling. Model fit was assessed from the comparative fit index and root mean square error approximation and associations were reported as standardized coefficients and 95% confidence intervals.

RESULTS

Higher perceived neighborhood disorder was associated with higher objective neighborhood disorder (β = 0.20, 95% CI: 0.06, 0.33), lower neighborhood social cohesion, and lower individual-level socioeconomic factors (final model root mean square error approximation 0.043 (90% CI: 0.013, 0.068)). Perceived neighborhood social cohesion was associated with individual-level socioeconomic factors and objective neighborhood disorder (β = - 0.11, 95% CI: - 0.24, 0.02).

CONCLUSION

Objective neighborhood disorder might be related to perceived disorder directly and indirectly through perceptions of neighborhood social cohesion.

Associations between neighborhood disinvestment and breast cancer outcomes within a populous state registry

BACKGROUND

Breast cancer (BrCa) outcomes vary by social environmental factors, but the role of built-environment factors is understudied. The authors investigated associations between environmental physical disorder-indicators of residential disrepair and disinvestment-and BrCa tumor prognostic factors (stage at diagnosis, tumor grade, triple-negative [negative for estrogen receptor, progesterone receptor, and HER2 receptor] BrCa) and survival within a large state cancer registry linkage.

METHODS

Data on sociodemographic, tumor, and vital status were derived from adult women who had invasive BrCa diagnosed from 2008 to 2017 ascertained from the New Jersey State Cancer Registry. Physical disorder was assessed through virtual neighborhood audits of 23,276 locations across New Jersey, and a personalized measure for the residential address of each woman with BrCa was estimated using universal kriging. Continuous covariates were z scored (mean ± standard deviation [SD], 0 ± 1) to reduce collinearity. Logistic regression models of tumor factors and accelerated failure time models of survival time to BrCa-specific death were built to investigate associations with physical disorder adjusted for covariates (with follow-up through 2019).

RESULTS

There were 3637 BrCa-specific deaths among 40,963 women with a median follow-up of 5.3 years. In adjusted models, a 1-SD increase in physical disorder was associated with higher odds of late-stage BrCa (odds ratio, 1.09; 95% confidence interval, 1.02-1.15). Physical disorder was not associated with tumor grade or triple-negative tumors. A 1-SD increase in physical disorder was associated with a 10.5% shorter survival time (95% confidence interval, 6.1%-14.6%) only among women who had early stage BrCa.

CONCLUSIONS

Physical disorder is associated with worse tumor prognostic factors and survival among women who have BrCa diagnosed at an early stage.

Assessing Google Street View Image Availability in Latin American Cities

Virtual audits using Google Street View are an increasingly popular method of assessing neighborhood environments for health and urban planning research. However, the validity of these studies may be threatened by issues of image availability, image age, and variance of image age, particularly in the Global South. This study identifies patterns of Street View image availability, image age, and image age variance across cities in Latin America and assesses relationships between these measures and measures of resident socioeconomic conditions. Image availability was assessed at 530,308 near-road points within the boundaries of 371 Latin American cities described by the SALURBAL (Salud Urbana en America Latina) project. At the subcity level, mixed-effect linear and logistic models were used to assess relationships between measures of socioeconomic conditions and image availability, average image age, and the standard deviation of image age. Street View imagery was available at 239,394 points (45.1%) of the total sampled, and rates of image availability varied widely between cities and countries. Subcity units with higher scores on measures of socioeconomic conditions had higher rates of image availability (OR = 1.11 per point increase of combined index, p < 0.001) and the imagery was newer on average (- 1.15 months per point increase of combined index, p < 0.001), but image capture date within these areas varied more (0.59-month increase in standard deviation of image age per point increase of combined index, p < 0.001). All three assessed threats to the validity of Street View virtual audit studies spatially covary with measures of socioeconomic conditions in Latin American cities. Researchers should be attentive to these issues when using Street View imagery.

Drop-And-Spin Virtual Neighborhood Auditing: Assessing Built Environment for Linkage to Health Studies

INTRODUCTION

Various built environment factors might influence certain health behaviors and outcomes. Reliable, resource-efficient methods that are feasible for assessing built environment characteristics across large geographies are needed for larger, more robust studies. This paper reports the item response prevalence, reliability, and rating time of a new virtual neighborhood audit protocol, drop-and-spin auditing, developed for assessment of walkability and physical disorder characteristics across large geographic areas.

METHODS

Drop-and-spin auditing, a method where a Google Street View scene was rated by spinning 360° around a point location, was developed using a modified version of the virtual audit tool Computer Assisted Neighborhood Visual Assessment System. Approximately 8,000 locations within Essex County, New Jersey were assessed by 11 trained auditors. Using a standardized protocol, 32 built environment items per a location within Google Street View were audited. Test-retest and inter-rater κ statistics were from a 5% subsample of locations. Data were collected in 2017-2018 and analyzed in 2018.

RESULTS

Roughly 70% of Google Street View scenes had sidewalks. Among those, two thirds were in good condition. At least 5 obvious items of garbage or litter were present in 41% of Google Street View scenes. Maximum test-retest reliability indicated substantial agreement (κ ≥0.61) for all items. Inter-rater reliability of each item, generally, was lower than test-retest reliability. The median time to rate each item was 7.3 seconds.

CONCLUSIONS

Compared with segment-based protocols, drop-and-spin virtual neighborhood auditing is quicker and similarly reliable for assessing built environment characteristics. Assessment of large geographies may be more feasible using drop-and-spin virtual auditing.

The linkage between the perception of neighbourhood and physical activity in Guangzhou, China: using street view imagery with deep learning techniques

BACKGROUND

Neighbourhood environment characteristics have been found to be associated with residents' willingness to conduct physical activity (PA). Traditional methods to assess perceived neighbourhood environment characteristics are often subjective, costly, and time-consuming, and can be applied only on a small scale. Recent developments in deep learning algorithms and the recent availability of street view images enable researchers to assess multiple aspects of neighbourhood environment perceptions more efficiently on a large scale. This study aims to examine the relationship between each of six neighbourhood environment perceptual indicators-namely, wealthy, safe, lively, depressing, boring and beautiful-and residents' time spent on PA in Guangzhou, China.

METHODS

A human-machine adversarial scoring system was developed to predict perceptions of neighbourhood environments based on Tencent Street View imagery and deep learning techniques. Image segmentation was conducted using a fully convolutional neural network (FCN-8s) and annotated ADE20k data. A human-machine adversarial scoring system was constructed based on a random forest model and image ratings by 30 volunteers. Multilevel linear regressions were used to examine the association between each of the six indicators and time spent on PA among 808 residents living in 35 neighbourhoods.

RESULTS

Total PA time was positively associated with the scores for "safe" [Coef. = 1.495, SE = 0.558], "lively" [1.635, 0.789] and "beautiful" [1.009, 0.404]. It was negatively associated with the scores for "depressing" [- 1.232, 0.588] and "boring" [- 1.227, 0.603]. No significant linkage was found between total PA time and the "wealthy" score. PA was further categorised into three intensity levels. More neighbourhood perceptual indicators were associated with higher intensity PA. The scores for "safe" and "depressing" were significantly related to all three intensity levels of PA.

CONCLUSIONS

People living in perceived safe, lively and beautiful neighbourhoods were more likely to engage in PA, and people living in perceived boring and depressing neighbourhoods were less likely to engage in PA. Additionally, the relationship between neighbourhood perception and PA varies across different PA intensity levels. A combination of Tencent Street View imagery and deep learning techniques provides an accurate tool to automatically assess neighbourhood environment exposure for Chinese large cities.