1 of 55

The life cycle of a modeling project:��Investigating the potential impact of structured human movement on arbovirus transmission�

Bobby Reiner

University of Washington

MMED 2024

000

2 of 55

Goals

Chart the progress of a research project

Conception
Hypothesis formulation
Study design
Data collected
Hypothesis tested
Paper published

Develop a dynamic model that uses the data to develop new ideas from the study

To stress the central importance of data – and the power of dynamic modeling

3 of 55

Public Health, Epidemiology, & Models (Day 1)

Simple Models (Day 1)

Foundations of Dynamic Modeling (Day 1)

(Hidden) Assumptions of Simple ODE’s (Day 2)

Breaking Assumptions!

Consequences of Heterogeneity (Day 6)

Introduction stochastic simulation models (Day 3)

Heterogeneity tutorial�(Day 6)

Introduction to Infectious Disease Data (Day 1)

Thinking about Data�(Day 2)

Data management and cleaning (Day 9)

Creating a Model World�(Day 4)

Study design and analysis in epidemiology (Day 3)

Introduction to Statistical Philosophy (Day 4)

Variability, Sampling Distributions, & Simulation (Day 10)

HIV in Harare tutorial�(Day 3)

Integration!

Introduction to Likelihood (Day 4)

Fitting Dynamic Models I – III (Day 5, 8, & 9)

Modeling for Policy (Day 11)

Model Assessment (Day 10)

MCMC Lab (Day 9)

MLE Fitting SIR model to prevalence data (Day 5)

Likelihood Lab (Day 4)

4 of 55

The first step

The world around us is complicated
How can we distill our observations into a testable hypothesis?

DATA!

Many ways to look at data…

Look for patterns in the numbers
Plot graphs
Make tables
Statistical analysis

Preliminary analysis may be sufficient to:

Decide if the hypothesis is correct
Publish the data

But … sometimes this process does not give the whole story and may fail to capture system dynamics

5 of 55

Questions to consider

What kinds of data are useful for modeling?
How important is the (known) biology of the system?
Which nuances can be safely ignored? What is vital?
When is it OK to take data or parameter estimates from other studies and use them in my model?
Remember, science is a team sport!

6 of 55

Why model at all?

We have a problem we can’t solve any other way
We can learn something from the modeling that we could not otherwise learn
To learn what we are missing & what further work we need to do
But … [sometimes] there are more interesting things in data than anticipated from the original experimental design and hypothesis

7 of 55

As is often the case, our disease modeling story begins at the end of a different, successful modeling project…

but first, some background

8 of 55

Dengue

Dengue (DENV) is a mosquito-borne viral infection caused by any of four related, but antigenically distinct, virus serotypes (DENV-1, -2, -3, and -4)

Approximately half of the world’s population is at risk from DENV, with approximately 50-100 million reported cases of dengue fever and more than 20,000 deaths from dengue hemorrhagic fever

The primary vector of dengue, Aedes aegypti, feeds throughout the day, is well adapted to urban and peri-urban environments and has a short dispersal range (<100 m)

9 of 55

Dengue in Iquitos

Iquitos, Peru

Iquitos has a population of 400,000. The city, located at the beginning of the Amazon river, is relatively isolated from other large cities.

In Iquitos, where within-city movement is extremely easy using motorized rickshaws (motocarros), transmission patterns do not appear extremely local and focal.

Human movement was hypothesized as the driver of spatial patterns of transmission, and a contact-tracing trial was initiated

10 of 55

Dengue in Iquitos

Iquitos, Peru

Iquitos has a population of 400,000. The city, located at the beginning of the Amazon river, is relatively isolated from other large cities.

In Iquitos, where within-city movement is extremely easy using motorized rickshaws (motocarros), transmission patterns do not appear extremely local and focal.

Human movement was hypothesized as the driver of spatial patterns of transmission, and a contact-tracing trial was initiated

Stoddard et al, PNAS (2013)

11 of 55

Dengue in Iquitos

2009 Cluster Trials

From two neighborhoods (Tupac and Maynas), active case detection was initiated to find actively infectious individuals.

Once an individual (index) was identified, they were asked what homes they had visited and every member of those homes was tested for DENV

For each DENV+ cluster initiated, a DENV- cluster was also created.

Stoddard et al, PNAS (2013)

12 of 55

Dengue in Iquitos

DENV+ Cluster

DENV- Cluster

Stoddard et al, PNAS (2013)

13 of 55

Dengue in Iquitos

DENV+ Cluster

Stoddard et al, PNAS (2013)

DENV- Cluster

Results

There was a striking, significant difference in risk (yellow+red+blue segments).

The increased risk was independent of distance from index home.

Conclusion

“Fine-scale human movements… underlie patterns of infection and result in pronounced temporal and spatial heterogeneity in dengue incidence”

14 of 55

Dengue in Iquitos

Stoddard et al, PNAS (2013)

* This includes the index’s home, but not themselves

15 of 55

Fundamental dengue epi

Primary infection

Host

Vector

16 of 55

Fundamental dengue epi

Primary infection

Host

Vector

17 of 55

Fundamental dengue epi

Primary infection

Secondary infection

Host

Vector

!!!

18 of 55

Observation

19 of 55

Observation

~40% of the locations have more than 1 concurrently infectious individual

20 of 55

Observation

~40% of the locations have more than 1 concurrently infectious individual

21 of 55

Observation

~40% of the locations have more than 1 concurrently infectious individual

~15% of the locations have more than 1 concurrently infectious individual

22 of 55

Development of modeling study concept

What is your question?
Why is it interesting?
Who might be interested?
Can it be narrowed down to a question about specific quantitative relationships?

23 of 55

Hypotheses

During latent period

H₀: No secondary structure within contact networks

24 of 55

Hypotheses

During latent period

During infectious period

H₀: No secondary structure within contact networks

25 of 55

Hypotheses

During latent period

H_a: Contact networks contain overlapping movement

26 of 55

Hypotheses

During latent period

During infectious period

H_a: Contact networks contain overlapping movement

27 of 55

Hypotheses

H_a: Contact networks contain overlapping movement

28 of 55

Hypotheses

H_a: Contact networks contain overlapping movement

29 of 55

Review of Literature & Available Data

Who has tried to answer this before and how did they do it?

Empirical studies
Modeling studies

Are there short-comings in these studies?
Find useful parameter estimates or data sets

30 of 55

Construction of modeling framework

Options?

Heterogeneous human movement / household size
Heterogeneous mosquito populations / movement
Mosquito satiation?

Describe approach

Describe simulation study

What did we learn?

31 of 55

Model Formulation

Overlapping movement ABM

To test this hypothesis, we built an agent based model of a neighborhood

Followed an SEIR-SEI model framework

32 of 55

Model Formulation

Overlapping movement ABM

To test this hypothesis, we built an agent based model of a neighborhood

Followed an SEIR-SEI model framework

20x25 houses on a grid

Number of residents per house based on Iquitos

Each individual had a “activity space” of 6 homes

Each house was a member of a “community” of 6 homes.

33 of 55

Model Formulation

Quantifying “socially structured movement”

Recall our hypothesis is that the “strong” within-community movement leads to shared risk amongst multiple houses within an individuals activity space

This requires us to define “strong”… Moreover this requires us to have a quantitative metric of “strength” of within-community movement that we can vary within our modeling framework.

34 of 55

Model Formulation

Overlapping movement ABM

35 of 55

Overlapping movement ABM

36 of 55

Overlapping movement ABM

37 of 55

Overlapping movement ABM

38 of 55

Model Formulation

Overlapping movement ABM

39 of 55

Overlapping movement ABM

40 of 55

Overlapping movement ABM

41 of 55

Problems / thoughts on the modeling

Limitations

Simplistic understanding of movement
Ignores numerous aspects of dengue transmission

Future work beyond the scope of this analysis

The work implies the importance of the overlap of the movement of multiple members within a household / community. This is directly measureable….

42 of 55

Writing the model and producing output

What are the graphical outputs that will display the answer(s) to my question?
Coding and debugging (and commenting)
Version control
Checking code to verify methods
Write your methods at this stage!

43 of 55

Model results

44 of 55

Model results

Overlapping movement ABM

RCR et al, Epidemics (2014)

45 of 55

Model Validation & Robustness

Sensitivity analyses

What (wrong) assumptions may be most important?

Model validation

How can we know our code is doing what we think it is?

Comparison to alternative models

Are there other formulations we can try out to compare results?

46 of 55

Write-up of results, discussion, abstract, and intro

State assumptions clearly
Try to wear your reviewer hat

But not too much!!!

47 of 55

Choose your journal

Aim high (but realistic)

How many journals are you willing to try?

Sometimes you just need to get it out so you can move on

48 of 55

Choose your journal

Where are your citations coming from?
Journal scope / statement
Other articles in that journal
Audience
How technical will your article be?
Text, figure, table limits

49 of 55

Reviewer feedback

A reviewer with a keen eye asked the following question

Do the strong constraints you enforce to match the observed patterns change the population dynamics?

In other words: are there obvious/important features of your model that don’t match reality that you ignore in your effort to match one specific feature?

50 of 55

Reviewer’s experiment

RCR et al, Epidemics (2014)

51 of 55

Reviewer’s experiment

Overlapping movement ABM

RCR et al, Epidemics (2014)

No mosquito movement

52 of 55

Reviewer’s experiment

Overlapping movement ABM

RCR et al, Epidemics (2014)

Any mosquito movement

53 of 55

Dengue in Iquitos

Conclusions

Strong social groups are consistent with observed spatio-temporal patterns of transmission

This secondary structure is a “hidden heterogeneity”

54 of 55

Conclusion

Work with the best data you can find
Look at and play with the data
Start with simple numerical or graphical summaries, then statistical models
Go to dynamic models only where these will tell you things you cannot find out using simpler approaches
Let the data drive the model not the other way around
Look for simple solutions: only add complexity when essential
Think
Be patient and try not to take reviewer feedback too personally
Enjoy!

55 of 55

000

This presentation is made available through a Creative Commons Attribution-Noncommercial license. Details of the license and permitted uses are available at� http://creativecommons.org/licenses/by/3.0/

Attribution:

Clinic on the Dynamical Approaches to Infectious Disease Data

Source URL:

For further information please contact figshare@ici3d.org.