Environmental Health Nexus: Designing Predictive Models for Improving Public Health Interventions

Thumbnail Image

Persistent link to this item

View Statistics

Journal Title

Journal ISSN

Volume Title


Environmental Health Nexus: Designing Predictive Models for Improving Public Health Interventions

Published Date




Thesis or Dissertation


The environment embodies all surroundings of humans, including natural (e.g., climate, rivers, and animals) and built (e.g., roads and buildings) components. The environment is closely related to population health both directly and indirectly. Ambient temperature exposure and air pollution, for example, can directly affect population health through its direct impacts on human cardiovascular and respiratory functions. Rainfall, on the other hand, can indirectly affect population health through its impacts on disease-transmitting vectors, such as mosquitoes. The U.S. Global Change Research Group and the Intergovernmental Panel on Climate Change both highlight the importance of the environment on population health. Environmental health is a challenging research topic for a variety of reasons. First, it is difficult to select the appropriate environmental indicators. Thanks to technological advancements in instrument precision, remote sensing, and many other fields, there is now an unprecedented amount of environmental information available to researchers. Although data availability issues still exist, the bigger question now is how to select information that is most relevant and appropriate to answer research questions. For example, when studying the epidemiological link between ambient temperature and population health, the most fundamental task is to select the appropriate indicator for ambient temperature. Because there are over 60 potential indicators that are all designed to approximate temperature perceived by the human body, this task can be a challenge. Second, along with the wide range of indicators comes a large volume of environmental data that is now available. Some ambient environmental indicators, such as air temperature, are available on a three-hour basis globally with high resolution since the 1980s. Technologies such as geographic information systems (GIS) have empowered public health to access this information. However, extracting this information for public health purpose is not always easy and may involve specialized technical expertise. Furthermore, incorporating this high-granularity data with traditionally scarcer public health data also entails technical difficulties. Last but not least, from a computational standpoint, it is challenging to work with high-dimensional data, especially given different research objectives. Environmental health issues do not usually deal with only a pair of exposure and response factors because no environmental factor exists independently. When studying dengue fever, for example, the link between temperature and disease occurrences is not two-dimensional because climate (e.g., rainfall), environmental (e.g., river network, non-human primates), and societal factors (e.g., human mobility network) are also involved. Reducing high-dimensional data to the essentials in order to meet research objectives is easier said than done. It involves sophisticated quantitative methods such as complexity science. It also largely depends on the specific research questions, e.g., if the model used to study the environmental health issue is for risk assessment, risk comparison, or disease forecast. Despite the technical challenges, environmental information has tremendous potential in terms of ecosystem service for population health research. Existing research has already generated many valuable outcomes with great real-life implications. However, the uptake rates of such knowledge among public health policy- and decision-makers remain low. An important underlying reason is that current knowledge contains little necessary details needed to influence policies and decisions. Moreover, policy- and decision-makers often lack the technical expertise to translate the results from research articles into valuable information to their specific context. The relationship between research and policy is predominantly driven by the research, i.e., the supply-side of epidemiological knowledge. Such supply-driven model has already been proven to be suboptimal in terms of maximizing the impact of research. Targeting the challenges discussed above, this dissertation focuses on designing quantitative predictive models for improving environmental health policy and decisions. More specifically, it generates evidence-based science to improve policies and decisions with respect to risk communication, impact assessment, and intervention planning. Although the end-users of specific studies included are environmental health managers and practitioners, the knowledge generated is also valuable to environmental health and methods researchers. Within the overarching theme, two projects were completed over the course of this dissertation. The first project used environmental information to forecast infectious disease outbreaks. Infectious diseases that rely on vector-borne, water-borne, air-borne, and zoonotic transmissions are all considered environmentally sensitive infectious diseases. Two studies were completed for influenza outbreaks in the U.S. and dengue fever outbreaks in San Juan, Puerto Rico and Iquitos, Peru. The research objective was to design statistical models that maximize forecast accuracy in terms of future outbreak timing and magnitude. Meteorological factors such as temperature, humidity, and precipitation were considered. The end-users in these projects were the U.S. Centers for Disease Control and Prevention, the National Oceanic and Atmospheric Administration, and local public health agencies. The end-goal was to reduce disease burden through preventative intervention planning. The forecasting methods used in these disease forecast models were uniquely designed for environmentally sensitive infectious diseases. Based on the nature of the transmission mechanisms involved, the models considered substantial temporal delays between the environmental exposure and population health responses. Traditionally, researchers have relied on measurements such as auto-correlation and partial auto-correlation coefficients to assess these temporal delays. However, these coefficients are constrained by linear assumptions. In this project, mutual information (a concept in information theory) was adopted as an alternative measure that quantifies the delayed relationship between environmental exposure and health response. The second component was to design evidence-based and policy-oriented models for managing population health risks associated with ambient temperature exposure. This component was a collaborative effort with the Minnesota Department of Health and the U.S. Centers for Disease Control and Prevention. The study site is the Minneapolis-St. Paul Twin Cities Metropolitan Area. The environmental indicator to measure ambient temperature exposure was selected using a data-driven approach. The risk assessment models aim at improving the quality of public health policy and decision-making. This project expands on the existing risk assessment methods by developing various modifications and extensions to meet the needs of risk communication, impact assessment, and intervention planning. In this dissertation, three studies from this second project (ambient temperature) are presented. The work described above has important epidemiological, methodological, and policy implications. It also contributes to a bigger picture, which is to design decision support tools for environmental health management. An ideal decision support tool should combine general and universal patterns in epidemiology with the local public health context to optimize policies and decisions under uncertain scenarios. This type of tool has been developed for ecosystem management (e.g., wetland restoration) and for chronic care. However, for environmental health issues, this type of tool does not yet exist. Currently, environmental health management still largely relies on past experiences of policy and decision makers. Essential knowledge needed for creating such decision support tools is not yet fully available. This dissertation provides some of the missing answers, with the ultimate goal of rationalizing and optimizing public health policies and decisions regarding environmental health intervention.


University of Minnesota Ph.D. dissertation. May 2018. Major: Environmental Health. Advisor: Matteo Convertino. 1 computer file (PDF); xx, 197 pages.

Related to




Series/Report Number

Funding information

Isbn identifier

Doi identifier

Previously Published Citation

Suggested citation

Liu, Yang. (2018). Environmental Health Nexus: Designing Predictive Models for Improving Public Health Interventions. Retrieved from the University Digital Conservancy, https://hdl.handle.net/11299/199044.

Content distributed via the University Digital Conservancy may be subject to additional license and use restrictions applied by the depositor. By using these files, users agree to the Terms of Use. Materials in the UDC may contain content that is disturbing and/or harmful. For more information, please see our statement on harmful content in digital repositories.