Dissertation
Identifying environmental risk factors and their use in predicting zoonotic diseases
University of Iowa
Doctor of Philosophy (PhD), University of Iowa
Autumn 2020
DOI: 10.17077/etd.005728
Abstract
Zoonotic diseases, diseases that can be transmitted between humans and animals which account for the majority of emerging infectious diseases, pose a consistent health burden on the human population. Transmission of zoonotic pathogens is driven by a complex interface between humans and non-human mammals via companionship, work, or environmental interactions. Two zoonotic disease that have a consistent public health threat in the United States are Influenza and Lyme disease. Influenza viruses, members of the Orthomyxoviridae family, are estimated to cause disease in 9.3 to 49 million Americans each year. Lyme disease, caused by Borrelia burgdorferi sensu lato, is the most widely reported vector-borne disease in the United States, and its incidence has been steadily increasing since 2000. Environmental factors are hypothesized to act as major contributors to transmission of these two zoonotic diseases.We conducted an epidemiological, longitudinal analysis using retrospective data to determine the association between flooding and influenza diagnoses in Iowa. Influenza diagnosis records were taken from a database of de-identified fully insured administrative claims data from a private health insurance company and influenza polymerase chain reaction tests from the Iowa State Hygienic Laboratory. Stream Height data was collected from United States Geology Survey (USGS) stream gauges across Iowa. Additional weather data was captured from meteorological aerodrome report (METAR) and surface synoptic observations (SYNOP) weather stations across Iowa. Lastly, population level risk factors were obtained from the 2010 United States Census and 2017 Agricultural census. Bayesian Poisson regression using conditional autoregression was used to determine the association between flooding and influenza risk. It was found that, on average, influenza diagnoses in Iowa increased by 1 % (95% CI: 1.00 – 1.02) with each day above flood stage. This relationship persisted after removing the 2009 – 2010 influenza pandemic year (Mean: 1.01 [95%CI: 1.00-1.02]). There was no association found between flooding and risk of Influenza-like illness (Mean: 1.00 [95%CI: 0.997-1.00]).
Next, we developed two-month influenza forecasting models using only environmental and selected influenza data. Influenza diagnosis records were captured from a database of de-identified fully-insured administrative claims data from a private health insurance company. Influenza strain data was captured influenza polymerase chain reaction test results from Iowa State Hygienic Lab. Weather data was taken from Global Historical Climatology Network (GHCN), METAR, and SYNOP stations and consisted of temperature, precipitations, snowfall, and humidity data. Lastly, flooding data from USGS stream gauges were used to implement flooding data into forecasting models. Ridge, lasso, elastic net and random forest regression models were utilized to fit two-month forecasting models for influenza diagnoses and burden categories. Random forest models had the best performance (RMSE = 11.04, MAE = 8.64, R2 = 0.71) for predicting influenza diagnoses. All models mimicked the timing of seasonal influenza but greatly underestimated the true influenza diagnoses rates. Similarly, random forest models had the best performance when predicting influenza burden categories (sensitivity = 0.40, specificity = 0.93), but still performed poorly when predicting high influenza burden months.
Lastly, we developed nowcasting models for regional Lyme disease rates using Google search history data. Lyme disease rates were calculated form Center for Disease Control and Prevention surveillance data. Google CorrelateTM was used to identify candidate search terms, and Google TrendsTM was used to collected monthly proportional hit data for each term. Hit data was collected per region. Elastic net regression models were fit and trained for five regions of the United States that incorporated all fifty states. It was found that elastic net regression models produced highly accurate predictions (RMSE: 0.921, MAE: 0.676 , and R2: 0.956) of Lyme disease rates in four of the five regions. Moreover, many of the search terms identified via Google CorrelateTM and the search terms that remained in final models were related to summer or summer-related actives.
In conclusion, the environment continually plays a key role in the epidemiology and transmission of zoonotic diseases. Flooding in Iowa has been shown to increase the risk of influenza diagnoses for affected populations. When incorporating flooding into influenza forecasting models, predictions can accurately mimic the timing of seasonal influenza but consistently underreport influenza diagnoses. Using Google search history from summer and summer related terms, highly accurate predictions of Lyme disease burden can be performed for four of five regions in the United States.
Details
- Title: Subtitle
- Identifying environmental risk factors and their use in predicting zoonotic diseases
- Creators
- Eric Kontowicz
- Contributors
- Christine A Petersen (Advisor)James Torner (Committee Member)Kelly K Baker (Committee Member)Margaret Carrel (Committee Member)Grant Brown (Committee Member)
- Resource Type
- Dissertation
- Degree Awarded
- Doctor of Philosophy (PhD), University of Iowa
- Degree in
- Epidemiology
- Date degree season
- Autumn 2020
- DOI
- 10.17077/etd.005728
- Publisher
- University of Iowa
- Number of pages
- xiv, 134 pages
- Copyright
- Copyright 2020 Eric Kontowicz
- Language
- English
- Description illustrations
- illustrations (some color), color maps
- Description bibliographic
- Includes bibliographical references (page 106-119).
- Public Abstract (ETD)
Zoonotic diseases, disease which can be transmitted between humans and animals, are major contributors to human disease burden each year. Zoonoses occur due to the complex relationship between the environment and the interconnected relationship between humans and non-human animals. Two major zoonotic disease in the United States are Influenza and Lyme disease.
Three studies were conducted to evaluate the role the environment plays in the transmission of influenza and Lyme disease. First, a population-level study was performed to evaluate the relationship between flooding and influenza in Iowa. Results found a consistent estimated 1% associated increase in influenza diagnoses per day above flood stage on average for Iowa. There was no association found between flooding and Influenza-like illness, a non-specific respiratory illness diagnosis. Next, a study was performed to determine the accuracy of influenza forecasts using only environmental data and limited influenza data. The influenza forecasts successfully mimicked the timing of seasonal influenza but poorly estimated influenza diagnoses and burden.
Lastly, nowcasting models were developed using Google search data for United States regional Lyme disease rates. Highly accurate predictions were made in four of the five regions based off the Lyme disease nowcasting models. It was identified that many of the key terms for accurate predictions were related to summer or summer actives and closely mimicked seasonal patterns of Lyme disease. These epidemiologic studies highlight the importance of the environment in population susceptibility to zoonotic diseases and the continual need for good disease monitoring methods.
- Academic Unit
- Epidemiology
- Record Identifier
- 9984035794702771
Metrics
28 File views/ downloads
101 Record Views