Mapping tick dynamics and tick bite risk using data-driven approaches and volunteered observations

Slides used during my presentation at the University of Idaho's TickBase Project 1st Annual Meeting, held 23-24 March 2021. This presentation contains a quick tour through my PhD research, focusing more on the scientific communication and the storytelling rather than in the technical details.

Programme here: https://tickbase.net/wp-content/uploads/2021/03/TickBase-Annual-Meeting_Schedule-Public.pdf

Transcript

1. Mapping tick dynamics and tick bite risk using data-driven approaches and volunteered observations Dr. Irene Garcia-Marti Prof. Dr. Raul Zurita-Milla In collaboration with: Dr. A Swart MG Harms, MSc Dr. AJH van Vliet Dr. CC van den Wijngaard
2. Scope
3. the Netherlands 101 • BBOX: 350 x 300 km, 45km2 of land • Elevation: -7m – 322m • Population: ~ 17M • ~85% urbanization, little amount of forests • Population is mainly concentrated in Randstadt (8M) • Enthusiasm about outdoor leisure and nature • People are more exposed to ticks
4. An increasing problem • Current situation: • 1 million tick bites per year • 2% contract disease • 27,000 cases per year • Cost of treatment: • 5700 euros per patient (~6800 USD) • Yearly lump sum: ~20M euros (~24M USD) • Disease burden: 10.55 DALYs • Source: • The cost of Lyme borreliosis (van den Wijngaard et al., 2017) • Socio-economic changes: • Urban demographics • Increased healthier lifestyles Source: https://www.rivm.nl/ziekte-van-lyme Source: Dutch Institute forPublicHealth and the Environment(RIVM)
5. Conceptualizing the problem Risk Probability that an individual will become ill or die within a stated period of time [. . . ] Hazard Inherent capability of an agent [. . . ] to have an adversely health effect Exposure Proximity and/or contact with a source of a disease agent in such a manner that effective transmission of the agent or harmful effects of the agent may occur Vulnerability The conditions determined by physical, social, economic and environmental factors [...] increasing the susceptibility to the impacts of hazards Sources: A dictionary of Epidemiology (Last, 2001) Ecology and prevention of Lyme borreliosis (Braks et al., 2016) UN Disaster Risk Reduction An approach from Risk Assessment: R = H × E × V
6. Grounding the problem Risk risk of tick bite Hazard tick dynamics Exposure human exposure Vulnerability unknown, thus constant An approach from Risk Assessment: R = H × E × V
7. Data collections
8. Citizen monitoring networks • Tick bites risk and tick dynamics are very local phenomena: • Unseen by: sensor networks, satellites, simulations • Certainly you can proxy! • Importance of grass-root campaigns: citizen science • Enabling citizens with digital devices and technologies to report conditions on Earth surface (Goodchild, 2007) 2006 - 2012 2012 - present
9. Citizens collecting data about ticks Source: RIVM Roughly 27,000 cases of Lyme borreliosis per year in the Netherlands
10. Tick dynamics • 15 volunteers, 9 years = ~3,000 samples • Counted: larvae, nymphs, adults • Sampling strategy: blanket dragging • Sites in different landscapes (forest, dunes) • Two transects/site, manually inspected • Gaps based in availability/meteorology • Read more on (Gassner, 2011)
11. Tick dynamics
12. Tick dynamics
13. Tick bites • 11 years = ~ 47,000 samples • Content: tick bite report, containing location • Combination of two citizen science networks: • Natuurkalender (2006 – 2012) • Tekenradar (2012 – present)
14. Tick bites
15. Tick bites Why a machine learning algorithm should work, out-of-the-box, in such 'difficult' conditions?
16. Modelling
17. What are we modelling? Risk risk of tick bite Hazard tick dynamics Exposure human exposure Vulnerability unknown, thus constant An approach from Risk Assessment: R = H × E × V
18. Modelling tick data Obtaining indicators of tick dynamics and tick bite riskto assist at creating tick mitigation and awareness campaigns
19. Tick dynamics
20. Tick dynamics • Unclear from literature: • Broad view on the most important variables driving tick dynamics • Most suitable temporal units to model tick dynamics • Our approach: • Building an array of explanatory variables • Tick habitat: composition, mast years • Weather: T, P, EV, RH, SD, VP (at 11 time-scales) • Vegetation: NDVI, EVI, NDWI • Land use: land cover type • Applying a data-driven method: Random Forest • With a tweak: time-awareness • Tick counts transformed into monthly Z-scores • Constrained and normalize range of tick counts Can we build a data-driven model predicting daily tick activity over the country?
21. Tick dynamics Model selection:
23. R2 ≥ 0.7 R2 < 0.7
24. Tick dynamics
25. Human exposure tick dynamics ≠ tick bite risk TICK BITES PER 1KM GRID CELL (2006-2016) R = H × E E = R ÷ H
26. Human exposure Recreational patterns along forest edges Not all forests are heavilyvisited
27. Tick bite risk • From the reported tick bites, we had: • R = number of tick bites per pixel (descriptive model) • Could we make a ... ? • R = F(covariates) (predictive model) • Covariates: H × E • What are measures of human exposure? • Based on (Zeimes, 2014): • Framework based in measures of attractiveness and accessibility as proxies of human exposure • Including forests, wildlife, and waterbody characteristics as proxies of tick hazard R = H × E
28. Tick bite risk 1 – 19 : human exposure 20 – 21: tick hazard
29. Tick bite risk • Modelling challenges: • Canonical ML algorithms might not handle well disproportioned data (Krawczyk, 2016) • Implies that a data-driven method produces results biased towards the majority class • How to learn from a dataset with heavy disproportions? • Zero-inflation • Overdispersion
30. Tick bite risk A Random Forest capable of learning from zero-inflated and overdispersed data
31. Tick bite risk
32. Tick bite risk
33. Wrapping up
34. Tick Hazard Human exposure Tick bite risk
35. Publications
36. Conclusions •Illustrating relationship between R, H, E •Atmospheric water levels drive tick dynamics •Tick bite risk is influenced by human exposure •Principles can be applicable to other tick-borne diseases Scientific •Modify method to handle time-aware observations •Modify method to handle skewed and zero-inflated distributions •Devised method to estimate human exposure to ticks without ICT data •Guidelines to explore patterns of risk Methodological •Monitoring elusive phenomena at unprecedented levels of detail •Observations can have sufficient quality to support scientific research •Sharing creative approaches at validating citizen science observations Citizen Science
37. Additional ideas ACQUIRING ICT DATA FOR BETTER HUMAN EXPOSURE METRICS ADDING DYNAMIC TIME- SERIES OF WILDLIFE SUSTAINED ENGAGEMENT WITH THE COMMUNITY INVESTIGATING OTHER WATER-BASED FEATURES
38. Thanks! Questions? email: garciamarti@knmi.nl web: irenegarciamarti.com PhD data deposit: https://doi.org/10.17026/dans-zre-tggd