Advances in the estimation of the reproduction number from compartmental models using contemporary Monte Carlo methods
View/ Open
Date
2024-03-08Author
Andrade, Jair
Metadata
Show full item recordUsage
This item's downloads: 13 (view details)
Abstract
The reproduction number represents the average number of secondary cases
generated by a primary case. If the population is completely susceptible, it is
referred to as the basic reproduction number (<0) and theoretically determines
whether the pathogen can invade the population. Moreover, its magnitude
is proportional to the effort needed to control the disease. Conversely, if
the infection is spreading, it is referred to as the effective reproduction
number (<t). It serves as an indicator of how extrinsic and intrinsic factors
have affected transmission at any given time. Both <0 and <t can be
estimated from compartmental models fitted to time series data. However,
these estimates are sensitive to both model assumptions and calibration
methods. Here, we show that by adhering to a rigorous inference workflow
and utilising state-of-the-art algorithms and visualisation tools, one can obtain
robust estimates. Using Hamiltonian Monte Carlo in a Bayesian approach,
we found a linear relationship between the mean generation time and <0.
This discovery allowed us to formulate a parameterisation that produces
accurate <0 estimates regardless of the distribution of the epidemiological
delays. On the other hand, we demonstrated, through a complementary
workflow that spanned three Data Generating Processes (semi-deterministic
and deterministic) and both schools of thought for statistical inference, that
incorporating mobility data into the workflow can reduce the uncertainty in
<t estimates. Nevertheless, this incorporation requires caution, given that
mobility data can only be a proxy measurement of the transmission rate.
Our results emphasise the importance of envisioning model calibration as
a learning process that confronts embedded assumptions. We anticipate
these findings will serve as a reference point for modellers that fit SIR-like
structures to time-series data. These guidelines include which information
to prioritise, how to approach the inference procedure, and how to interpret
calibration results.