Show simple item record

dc.contributor.advisorCastro, Marcia
dc.contributor.authorJay, Jonathan
dc.date.accessioned2018-12-20T13:45:20Z
dash.embargo.terms2019-05-01
dc.date.created2018-05
dc.date.issued2018-04-20
dc.date.submitted2018
dc.identifier.urihttp://nrs.harvard.edu/urn-3:HUL.InstRepos:37945619*
dc.description.abstractCities are vital units of analysis for public health problems, both in the United States and globally. City-dwellers, who account for more than half the world’s population, face a characteristic set of health risks. Many of these risks are closely related to housing, transportation, sanitation and other social determinants of health for which local governments are responsible. For better or worse, therefore, local government performance produces health outcomes. This DELTA doctoral thesis examines three novel methods intended to improve local government service delivery and prevent adverse health outcomes. Each uses machine learning algorithms and publicly available datasets to generate actionable predictions. (1) In Portland, Oregon, we used a random forest model and city property records to rank over 200,000 properties according to fire risk. In a statistically simulated field trial, we found that following these rankings could improve the efficiency of fire safety inspections by 1.5 times and home safety visits by up to 9.0 times. (2) In Fortaleza, Brazil, we combined convolutional neural networks and ridge regression to predict dengue case incidence based on block-level satellite imagery. In five-fold cross-validation, our model explained 73% of variance in log-normalized case counts. (3) Using Google Street view images and city records from Detroit, Michigan, I trained a convolutional neural network to identify physically distressed properties. In a balanced testing set of 400 images, the model correctly classified 87% of images as either distressed or normal condition, with an area under the curve statistic of 0.94. Each of these methods is novel, meets appropriate standards for accuracy, and aligns with accepted preventive interventions for important public health problems. Nonetheless, it does not automatically follow that the proposed public health benefits will outweigh potential societal costs. The final section of this document reviews additional factors, such as equity, ethics and staff motivation, that local governments must consider prior to adopting predictive strategies.
dc.format.mimetypeapplication/pdf
dc.language.isoen
dash.licenseLAA
dc.subjectHealth Sciences, Public Health
dc.titlePublic Health in the Predictable City
dc.typeThesis or Dissertation
dash.depositing.authorJay, Jonathan
dash.embargo.until2019-05-01
dc.date.available2018-12-20T13:45:20Z
thesis.degree.date2018
thesis.degree.grantorHarvard T.H. Chan School of Public Health
thesis.degree.levelDoctoral
thesis.degree.nameDoctor of Public Health (DrPH)
dc.contributor.committeeMemberHemenway, David
dc.contributor.committeeMemberLeary, Kimberlyn
dc.type.materialtext
thesis.degree.departmentPublic Health
dash.identifier.vireohttp://etds.lib.harvard.edu/hsph/admin/view/238
dc.description.keywordsUrban health; machine learning; fire prevention; dengue; housing; neighborhood effects
dc.identifier.orcid0000-0002-7543-4247
dash.author.emailjonjay04@gmail.com


Files in this item

Thumbnail

This item appears in the following Collection(s)

Show simple item record