1 of 29

Compounding impacts of climate and land use change on human schistosomiasis in Brazil

Alyson L. Singleton, MA

De Leo and Mordecai Labs, Stanford University

asinglet@stanford.edu

2 of 29

Schistosomiasis

  • Debilitating parasitic disease

asinglet@stanford.edu

3 of 29

Schistosomiasis

  • Debilitating parasitic disease
  • Current burden of disease
    • Worldwide: 200 million infected
    • Brazil: 2-6 million infected

asinglet@stanford.edu

4 of 29

Schistosomiasis

  • Debilitating parasitic disease
  • Current burden of disease
    • Worldwide: 200 million infected
    • Brazil: 2-6 million infected
  • Schistosoma parasite depends on Biomphalaria snails and human beings to complete life cycle

asinglet@stanford.edu

5 of 29

Schistosomiasis

  • Debilitating parasitic disease
  • Current burden of disease
    • Worldwide: 200 million infected
    • Brazil: 2-6 million infected
  • Schistosoma parasite depends on Biomphalaria snails and human beings to complete life cycle
  • Predominantly impacts communities that rely on open water resources for daily life

asinglet@stanford.edu

6 of 29

Project Goals

asinglet@stanford.edu

7 of 29

​

Create high-resolution maps for current risk of schistosomiasis transmission in Brazil

​

Project Goals

asinglet@stanford.edu

8 of 29

​

Create high-resolution maps for current risk of schistosomiasis transmission in Brazil

​

Capture schistosomiasis sensitivity to environmental conditions and human impacts

Project Goals

asinglet@stanford.edu

9 of 29

​

Create high-resolution maps for current risk of schistosomiasis transmission in Brazil

​

Capture schistosomiasis sensitivity to environmental conditions and human impacts

            • Temperature (snail thermal dependencies)
            • Land use change (agriculture, urbanization)
            • Human demography change (migration, water needs)

Project Goals

asinglet@stanford.edu

10 of 29

Species Distribution Modeling (SDM)

asinglet@stanford.edu

11 of 29

Species Distribution Modeling (SDM)

  • Human health related use case:

Diseases that require non-human species for transmission

​

​

asinglet@stanford.edu

12 of 29

Species Distribution Modeling (SDM)

  • Human health related use case:

Diseases that require non-human species for transmission

​

​

asinglet@stanford.edu

Environmental Covariate Data

Species

Occurrence Data

+

Input

13 of 29

Species Distribution Modeling (SDM)

  • Human health related use case:

Diseases that require non-human species for transmission

​

​

asinglet@stanford.edu

Species

Occurrence Data

+

Input

Environmental Covariate Data

Statistical Learning Models

🡪

14 of 29

Species Distribution Modeling (SDM)

  • Human health related use case:

Diseases that require non-human species for transmission

​

​

asinglet@stanford.edu

Species

Occurrence Data

🡪

Statistical Learning Models

🡪

Species Ecologic Suitability Probabilities

+

Input

Output

Environmental Covariate Data

15 of 29

Species Distribution Modeling (SDM)

  • Human health related use case:

Diseases that require non-human species for transmission

​

​

asinglet@stanford.edu

Species

Occurrence Data

🡪

Statistical Learning Models

🡪

Species Ecologic Suitability Probabilities

+

Input

Output

  • SDMs allow for non-linearities and variable interaction across heterogenous geographies

​

Environmental Covariate Data

16 of 29

Preliminary Results

asinglet@stanford.edu

17 of 29

Preliminary Results

asinglet@stanford.edu

18 of 29

Preliminary Results

asinglet@stanford.edu

Random Forest

  • Merges multiple decision trees

​

19 of 29

Preliminary Results

asinglet@stanford.edu

Random Forest

  • Merges multiple decision trees

​

Background Sampling

  • Use samples of other snail species to control for sampling bias

​

20 of 29

Preliminary Results

asinglet@stanford.edu

Background Sampling

  • Use samples of other snail species to control for sampling bias

​

Testing and validation

​

Random Forest

  • Merges multiple decision trees

​

21 of 29

Preliminary Results

asinglet@stanford.edu

Background Sampling

  • Use samples of other snail species to control for sampling bias

​

Testing and validation

​

Random Forest

  • Merges multiple decision trees

​

22 of 29

Preliminary Results

asinglet@stanford.edu

Background Sampling

  • Use samples of other snail species to control for sampling bias

​

Testing and validation

Out of sample AUC

  • Mean: 0.667
  • 95% CI: [0.616, 0.727]

Random Forest

  • Merges multiple decision trees

​

23 of 29

Future directions and applications

asinglet@stanford.edu

24 of 29

Future directions and applications

Short term: Investigate best SDM methodological choices

asinglet@stanford.edu

25 of 29

Future directions and applications

Short term: Investigate best SDM methodological choices

    • Model structure: random forest, maximum entropy, extreme gradient boosting
    • Geographic extent: municipality, regional, national
    • Background sampling: sampling bias, no true absences

​

asinglet@stanford.edu

26 of 29

Future directions and applications

Short term: Investigate best SDM methodological choices

    • Model structure: random forest, maximum entropy, extreme gradient boosting
    • Geographic extent: municipality, regional, national
    • Background sampling: sampling bias, no true absences

Long term: Projections under global change scenarios

    • Ecological: Temperature and precipitation change
    • Social: Dams, water access, demography change

​

asinglet@stanford.edu

27 of 29

Future directions and applications

Short term: Investigate best SDM methodological choices

    • Model structure: random forest, maximum entropy, extreme gradient boosting
    • Geographic extent: municipality, regional, national
    • Background sampling: sampling bias, no true absences

Long term: Projections under global change scenarios

    • Ecological: Temperature and precipitation change
    • Social: Dams, water access, demography change

​

Predictions will inform future schistosomiasis prevention and mitigation efforts

​

asinglet@stanford.edu

28 of 29

Future directions and applications

Short term: Investigate best SDM methodological choices

    • Model structure: random forest, maximum entropy, extreme gradient boosting
    • Geographic extent: municipality, regional, national
    • Background sampling: sampling bias, no true absences

Long term: Projections under global change scenarios

    • Ecological: Temperature and precipitation change
    • Social: Dams, water access, demography change

​

Predictions will inform future schistosomiasis prevention and mitigation efforts

Species distribution modeling can be a useful global health tool

​

asinglet@stanford.edu

29 of 29

Jürg Utzinger

Anna-Sofie Stensgaard

Guojin Yang

Fiona Allan

Julia Jones

  • Four countries across four continents
  • >15 institutions
  • Climatology, hydrology, ecology, tropical diseases, parasitology, public health, aquaculture, disease dynamics

​

Collaboration and our interdisciplinary team

Erin Mordecai

Caroline Glidden

Tejas Athni

Aly Singleton

Devin Kirk