As a brief review: I picked a location from one of each of these six clusters, as well as Bajonapo in eastern Ecuador and a Karen village in Myanmar. I pulled climate from each location from COMPS.
For Africa I pulled the relative vector abundance of funestus, gambiae, and arabiensis from Catherine Moyes’ raster. For Bajonapo I assumed all darlingi, while Karen has a 60/40 maculatus/minimus mix. Key vector species params are:
Name
Adult Life Expectancy(Days)
Anthropophily(%)
Endophily(%)
arabiensis
20
75
50
darlingi
20
50
60
funestus
20
65
85
gambiae
20
85
85
maculatus
20
50
1
minimus
20
50
60
All villages were run with the same age distribution, an initial population of 2000, and a ten-year burnin with the age distributions already at equilibrium.
I tested ITNs at coverages from 0 to 80, distributed on Jan 1 every year for three years with a 180-day discarding halflife, ten runs for each site-coverage-initial_prevalence. All other ITN parameters were kept at their defaults, and were the same for each site(90% initial blocking with 2-year exponential decay, 60% initial killing with 4-year exponential decay).
Results
Preliminary results for all eight sites are below. The x- and y- axes are the annual average PFPr 2-10 for the final year of the burnin and three-year intervention sims, respectively:
In general we see the sort of curve we want, with behaviors we’d expect(like more residual transmission among the less anthro- and endo- philic vectors in Ecuador and Myanmar), but it’s surprising that ITNs are so effective at high coverages, even for quite high initial prevalences.
To test more realistic scenarios, I tried two things:
Reducing the number of ITN distributions to only one or two(still on Jan 1st of the 1st and 2nd years, respectively).
Simulating importation by adding a“recurring outbreak” that randomly infects some proportion of the population. I dropped these once a month, and tested different outbreak proportions: 0.1%, 0.5%, or 1% of the population infected each month.
Here’s what those results look like, just for Moine:
The bottom right plot is exactly the same as in the initial plot for Moine. We note two different behaviors:
Without regular net distributions, prevalence snaps back to pre-intervention values in medium-to-high transmission settings by year three;
Adding importation(but keeping annual net distributions) keeps the shape of the curve, but lifts it such that we always have some nonzero residual transmission.
Next Steps
I’d like to test one other behavior: correlated intervention distribution, such that there is persistently some portion of the population that never receives nets. From there, we can make final decisions about what parameters we want to pick for our next“big run”
It would be useful to do a similar analysis to this for IRS and ACTs to make sure we like what we’re seeing with those intervention behaviors in isolation before we start running interactions.
Finally, you may have wondered why some of the sites in the initial plot looked noisier than others. It’s because my initial guess at what larval habitat range it was worth sweeping over to capture a good set of initial prevalences was a bit, uh, patchy:
I got good coverage of initial prevalences for Moine, but spend too much time at the high end of larval habitat in general and miss out in particular on a lot of Karen’s lower-prevalence parameter space. I just finished a burnin run that does a better job of capturing the parameter range we want, will run simulations from these in the future:
Methods
Results
Next Steps
Update 1: IRS and ACT Results