Methods to Improve Existing Heat Wave Surveillance Systems

Elevated and prolonged exposure to extreme heat is an important cause of excess summertime mortality and morbidity. To protect people from health threats, some governments are currently operating syndromic surveillance systems. However, a lack of resources to support time- and labor- intensive diagnostic and reporting processes make it difficult establishing region-specific surveillance systems. Big data created by social media and web search may improve upon the current syndromic surveillance systems by directly capturing people’s individual and subjective thoughts and feelings during heat waves. The primary objectives of the dissertation are to improve existing heat wave and health surveillance systems by testing current heat exposure metrics, checking system improvements with social media/web search data, and studying differential vulnerability to extreme heat exposure. In order to conduct the research, this dissertation employed two popular statistical techniques: time series and case-crossover analysis. Chapter 2 examines the relationship between the count of heat-related tweets and heat exposure. For this, I collected Twitter data focusing on six different heat-related themes (air conditioning, cooling center, dehydration, electrical outage, energy assistance, and heat) for 182 days from May 7 to November 3, 2014. First, exploratory linear regression associated outdoor heat exposure to the theme-specific tweet counts for five study cities (Los Angeles, New York, Chicago, Houston, and Atlanta). Next, autoregressive integrated moving average (ARIMA) time series models formally associated heat exposure to the combined count of heat and air conditioning tweets while controlling for temporal autocorrelation. Finally, I examined the spatial and temporal distribution of energy assistance and cooling center tweets. The result indicates that the number of tweets in most themes exhibited a significant positive relationship with maximum temperature. The ARIMA model results suggest that each city shows a slightly different relationship between heat exposure and the tweet count. A one-degree change in the temperature correspondingly increased the Box-Cox transformed tweets by 0.09 for Atlanta, 0.07 for Los Angeles, and 0.01 for New York City. The energy assistance and cooling center theme tweets suggest that only a few municipalities used Twitter for public service announcements. The timing of the energy assistance tweets also indicates that most jurisdictions provide heating instead of cooling energy assistance. Chapter 3 aims to investigate the relationship between heat-related web searches, social media messages, and heat-related health outcomes. I collected Twitter messages that mentioned “air conditioning (AC)” and “heat” and Google search data that included weather, medical, recreational, and adaptation information from May 7 to November 3, 2014, focusing on the state of Florida, U.S. I separately associated web data against two different sources of health outcomes (emergency department (ED) and hospital admissions) and five disease categories (cardiovascular disease, dehydration, heat-related illness, renal disease, and respiratory disease). Seasonal and subseasonal temporal cycles were controlled using autoregressive moving average-generalized autoregressive conditional heteroscedasticity (ARMA-GARCH) and generalized linear model (GLM). The results show that the number of heat-related illness and dehydration cases exhibited a significant positive relationship with web data. Specifically, heat-related illness cases showed positive associations with messages (heat, AC) and web searches (drink, heat stroke, park, swim, and tired). In addition, terms such as park, pool, swim, and water tended to show a consistent positive relationship with dehydration cases. However, I found inconsistent relationships between renal illness and web data. Web data also did not improve the models for cardiovascular and respiratory illness cases. These findings suggest web data created by social medias and search engines could improve the current syndromic surveillance systems. In particular, heat-related illness and dehydration cases were positively related with web data. This study also shows that activity patterns for reducing heat stress are associated with several health outcomes. Chapter 2 and chapter 3 suggest that web data could benefit both regions without the systems and persistently hot and humid climates where excess heat early warning systems may be less effective. Chapter 4 investigates whether there is a difference between five different types of heat sensitive health outcomes (cardiovascular disease, dehydration, heat-related illness, renal disease, and respiratory disease) between undocumented immigrants and US citizens. This study also examines if the impact of heat exposure on health by citizenship status is further modified by sex, age, or race/ethnicity. I conducted a case-crossover analysis to assess different heat-related health impact by citizenships, focusing on the warm season (May through September) from 2008 to 2012 in Florida. I reported separate case-crossover models for each health outcome and type of healthcare visit (emergency department, hospitalization). I stratified the data by immigration status and then added interaction terms to understand the impact of sex, age, or race/ethnicity. For both groups, higher temperature raised the risk of all heat-related health outcomes and healthcare visits. This analysis suggest undocumented people (ED: 1.127, 95 % CI: 1.056 ~ 1.204; hospitalization: 1.061, 95 % CI: 1.046 ~ 1.076) have moderately higher renal disease ORs than US citizens (ED: 1.069, 95 % CI: 1.059 ~ 1.078; hospitalization: 1.051, 95 % CI: 1.049 ~ 1.053). In addition, male US citizens had significantly higher ORs than female citizens for both ED (male: 1.080, 95 % CI: 1.076 ~ 1.085; female: 1.060, 95 % CI: 1.056 ~ 1.064) and hospitalization (male: 1.063, 95 % CI: 1.060 ~ 1.066; female: 1.054, 95 % CI: 1.052 ~ 1.057). This study documents some heat and health inequalities between US citizens and undocumented immigrants.