MSYNC 2021-02-03: Heterogeneous biting

Background

  • Variance in biting risk is an important epidemiological parameter.
  • A recent paper (Cooper et al 2019) suggested that biting risk follows a roughly Pareto distribution, with 20% of households receiving 80% of the bites.
  • Current heterogeneous biting scheme that most people use is an exponential distribution with scale factor = 1 
  • Bottom line: new distribution better fits data for low and moderate transmission settings, and this distribution is more resistant to elimination.

Methods

  • Took raw data of mosquito counts from 3 sites in Uganda from Cooper et al. 2019.  This contains number of mosquitos collected at each household in the study, for each month.
  • Generated model data for different heterogeneous biting distributions to find best fit.


Fitting raw data

  • First, took each month of data and normalized household-level counts by the monthly average.  This allows us to focus on the monthly variance of risk, ignoring seasonal differences
  • Then, compute normalized, Euclidean distance between observed distribution and model distribution



  • Log-normal is provide better fit for lower transmission area (Jinja, Kanungu)
  • Exponential is better fit for very high transmission area (Tororo)




How different is this new distribution?


How does this show up in the simulations?

(constant transmission sims, for clarity)

  • Uniform biting tends to have higher prevalence than heterogeneous biting
  • Uniform biting explores the antigen space faster because people are less likely to be capped out on their allowed infections.
  • The more heterogeneous the biting, the lower the prevalence in general

HOWEVER, the more heterogeneous biting is more resistant to elimination:

Takeaways

  • Uniform biting risk is probably never what you want, except for debugging.  
  • In other words, make sure Enable_Demographics_Risk is set to 1, and, that you are setting a distribution of risk in the demographics file.
  • In low and moderate settings, a log-normal distribution of risk with sigma ~1.6 is better fit to data than exponential distribution.  
  • The new distribution has a more extended tail of high-risk individuals
  • These high-risk individuals make the overall population more resistant to elimination