The project consists of four parts: (1) database development, (2) modeling and analytics, (3) visualization and web development, and (4) community engagement and application. The first two parts are associated with the actual building, training, and testing of machine learning models. The targets are the various health outcomes, namely the prevalence of common non-communicable chronic diseases such as coronary heart disease, cancer, diabetes, poor mental health, obesity, and stroke. The actual health outcomes used in training and testing the models are accessed from the CDC’s 500 Cities Project. The features are created based on three data sources, namely the CDC’s Social Vulnerability Index (SVI) dataset, the EPA’s Smart Location Database (SLD), and the 311 service request datasets accessed from each municipality. Sixty features (i.e., predictor variables) are considered, which characterize the social environment, the physical environment, and the aspects and degrees of neighborhood disorder. A variety of machine learning algorithms are applied and compared, including Ridge Regression, Lasso Regression, Elastic Net, Support Vector Machine, Decision Tree, Random Forest, Extra Trees, and Gradient Boosting. To improve the model performance, the model hyperparameters are fine-tuned using 10-fold cross-validation. Different sets of features are also experimented with.
It is shown that the tract-level prevalence for the common non-communicable chronic diseases can be reasonably well predicted based on the publicly available datasets. Furthermore, two major findings have been yielded from this study: (1) the sociodemographic and socioeconomic variables are the strongest predictors for tract-level health outcomes; (2) the historical records of 311 service requests can be a useful complementary data source because the information distilled from the 311 data often helps improve the models’ performance.
The datasets and the predictive models are published online. Users can play with the models interactively by using the web tools we developed. The web tools can help the public and city officials evaluate future scenarios and understand how changes in the neighborhood conditions can lead to changes in the health outcomes.
Data Sources
The census tract-level health data are drawn from the 500 Cities Project dataset. (https://chronicdata.cdc.gov/browse?category=500+Cities)
The built environment variables are calculated based on EPA’s Smart Location Database (SLD). (https://www.epa.gov/smartgrowth/smart-location-mapping#SLD)
The socioeconomic and sociodemographic variables are extracted from CDC’s Social Vulnerability Index (SVI) dataset. (https://svi.cdc.gov/data-and-tools-download.html)
The 311 data are accessed from the open data portal of each municipality:
Austin: https://data.austintexas.gov/Utilities-and-City-Services/Austin-311-Public-Data/xwdj-i9he
Baltimore: https://data.baltimorecity.gov/City-Services/311-Customer-Service-Requests/9agw-sxsr
Boston: https://data.boston.gov/dataset/311-service-requests
Dallas: https://www.dallasopendata.com/City-Services/311-Service-Requests-October-1-2016-to-September-3/shgm-yzbp
https://www.dallasopendata.com/City-Services/311-Service-Requests-October-1-2018-to-Present-/m36q-vtbr
Washington, D.C.: https://opendata.dc.gov/datasets/311-city-service-requests-in-2019
Houston: http://www.houstontx.gov/311/
Los Angeles: https://data.lacity.org/browse?q=311&sortBy=relevance&page=2
New York City: https://nycopendata.socrata.com/Social-Services/311-Service-Requests-from-2010-to-Present/erm2-nwe9
San Antonio: https://data.sanantonio.gov/dataset/service-calls
San Francisco: https://data.sfgov.org/City-Infrastructure/311-Cases/vw6y-z8j6/data
Abbreviations and Descriptions of Variables
|
Variable |
Abbreviation |
Data Source |
|
Outcome variable
|
|
|
|
Arthritis among adults aged ≥ 18 years (%) |
ARTHRITIS |
CDC’s 500 Cities Project |
|
High blood pressure among adults aged ≥ 18 years (%) |
BPHIGH |
|
|
Cancer (excluding skin cancer) among adults aged ≥ 18 years (%) |
CANCER |
|
|
Current asthma prevalence among adults aged ≥ 18 years (%) |
CASTHMA |
|
|
Coronary heart disease among adults aged ≥ 18 years (%) |
CHD |
|
|
Chronic obstructive pulmonary disease among adults aged ≥ 18 years (%) |
COPD |
|
|
Diagnosed diabetes among adults aged ≥ 18 years (%) |
DIABETES |
|
|
High cholesterol among adults aged ≥ 18 years who have been screened in the past 5 years (%) |
HIGHCHOL |
|
|
Chronic kidney disease among adults aged ≥ 18 years (%) |
KIDNEY |
|
|
Mental health not good for ≥ 14 days among adults aged ≥ 18 years (%) |
MHLTH |
|
|
Physical health not good for ≥ 14 days among adults aged ≥ 18 years (%) |
PHLTH |
|
|
Stroke among adults aged ≥ 18 years (%) |
STROKE |
|
|
All teeth lost among adults aged ≥ 65 years (%) |
TEETHLOST |
|
|
Binge drinking prevalence among adults aged ≥ 18 years (%) |
BINGE |
|
|
Current smoking among adults aged ≥ 18 years (%) |
CSMOKING |
|
|
No leisure-time physical activity among adults aged ≥ 18 years |
LPA |
|
|
Obesity among adults aged ≥ 18 years |
OBESITY |
|
|
Sleeping less than 7 hours among adults aged ≥ 18 years |
SLEEP |
|
|
Note: The column names for the predicted health outcome values are made simply by prefixing a lowercase ‘p’ before the variable names shown above. For example, ‘ARTHRITIS’ becomes ‘pARTHRITIS’. |
||
|
Predictor variable
|
|
|
|
Percentage of persons below poverty |
P_POV |
CDC’s SVI data |
|
Percentage of civilian (age 16+) unemployed estimate |
P_UNEMP |
|
|
Per capita income (US$) |
PCI |
|
|
Percentage of persons with no high school diploma (age 25+) |
P_NOHSDP |
|
|
Percentage of persons aged 65 and older |
P_AGE65P |
|
|
Percentage of persons aged 17 and younger |
P_AGE17M |
|
|
Percentage of civilian noninstitutionalized population with a disability |
P_DISABL |
|
|
Percentage of single parent households with children under 18 |
P_SNGPNT |
|
|
Percentage minority (all persons except white, non-Hispanic) |
P_MINRTY |
|
|
Percentage of persons (age 5+) who speak English “less than well” |
P_LIMENG |
|
|
Percentage of housing in structures with 10 or more units |
P_MUNIT |
|
|
Percentage of mobile homes |
P_MOBILE |
|
|
Percentage of occupied housing units with more people than rooms |
P_CROWD |
|
|
Percentage of households with no vehicle available |
P_NOVEH |
|
|
Percentage of persons in institutionalized group quarters |
P_GROUPQ |
|
|
Percentage uninsured in the total civilian noninstitutionalized population |
P_UNINSUR |
|
|
Percent of population that is working aged |
P_WRKAGE |
EPA’s Smart Location Database |
|
Percent of one-car households |
P_AO1 |
|
|
Percent of two-plus-car households |
P_AO2P |
|
|
Percentage of low-wage workers (earning $1250/month or less) among total workers (home location) |
P_LOWWAGEr |
|
|
Percentage of low-wage workers (earning $1250/month or less) among total workers (work location) |
P_LOWWAGEe |
|
|
Gross residential density (HU/acre) on unprotected land |
D_HH |
|
|
Gross population density (people/acre) on unprotected land |
D_POP |
|
|
Gross employment density (jobs/acre) on unprotected land |
D_EMP |
|
|
Gross activity density (employment + HUs) on unprotected land |
D_HUEMP |
|
|
Gross retail (5-tier) employment density (jobs/acre) on unprotected land |
D_EMP_RET |
|
|
Gross office (5-tier) employment density (jobs/acre) on unprotected land |
D_EMP_OFF |
|
|
Gross industrial (5-tier) employment density (jobs/acre) on unprotected land |
D_EMP_IND |
|
|
Gross service (5-tier) employment density (jobs/acre) on unprotected land |
D_EMP_SVC |
|
|
Gross entertainment (5-tier) employment density (jobs/acre) on unprotected land |
D_EMP_ENT |
|
|
Jobs per household |
JOBSPERHH |
|
|
5-tier employment entropy (denominator set to observed employment types in the census tract) |
EMPMIX |
|
|
Employment and household entropy |
EMPHHMIX |
|
|
Employment and household entropy (based on vehicle trip production and trip attractions including all 5 employment categories) |
TRIPMIX |
|
|
Trip productions and trip attractions equilibrium index |
TRIPEQ |
|
|
Household workers per job, by census tract |
WRKSPERJOB |
|
|
Household workers per job equilibrium index |
HHWRKJOBEQ |
|
|
Total road network density |
D_RD |
|
|
Network density in terms of facility miles of auto-oriented links per square mile |
D_RD_AO |
|
|
Network density in terms of facility miles of multi-modal links per square mile |
D_RD_MM |
|
|
Network density in terms of facility miles of pedestrian-oriented links per square mile |
D_RD_PO |
|
|
Street intersection density (auto-oriented intersections eliminated) |
D_X_EXCLAO |
|
|
Intersection density in terms of auto-oriented intersections per square mile |
D_X_AO |
|
|
Intersection density in terms of multi-modal intersections having three legs per square mile |
D_X_MM3 |
|
|
Intersection density in terms of multi-modal intersections having four or more legs per square mile |
D_X_MM4 |
|
|
Intersection density in terms of pedestrian-oriented intersections having three legs per square mile |
D_X_PO3 |
|
|
Intersection density in terms of pedestrian-oriented intersections having four or more legs per square mile |
D_X_PO4 |
|
|
Proportion of census tract employment within ¼ mile of fixed-guideway transit stop |
P_EMP025 |
|
|
Proportion of census tract employment within ½ mile of fixed-guideway transit stop |
P_EMP050 |
|
|
Aggregate frequency of transit service per square mile |
D_TRANSIT |
|
