Comparing methods to quantify experimental transmission of infectious agents.

Abstract

Transmission of an infectious agent can be quantified from experimental data using the transient-state (TS) algorithm. The TS algorithm is based on the stochastic SIR model and provides a time-dependent probability distribution over the number of infected individuals during an epidemic, with no need for the experiment to end in final-size (e.g., where no more infections can occur). Because of numerical limitations, the application of the TS algorithm is limited to populations with only a few individuals. We investigated the error of using the easily applicable, time-independent final-size (FS) algorithm knowing that the FS situation was not reached. We conclude that the methods based on the FS algorithm: (i) underestimate R(0), (ii) are liberal when testing H(0):R(0)1 against H(1):R(0)<1, (iii) are conservative when testing H(0):R(0)1 against H(1):R(0)>1, and (iv) are conservative when testing H(0):R(control)=R(treatment) against H(1):R(control)>R(treatment). Furthermore, a new method is presented to find a difference in transmission between two treatment groups (MaxDiff test). The MaxDiff test is compared to tests based on FS and TS algorithms. The TS test and the MaxDiff test were most powerful (approximately equally powerful) in finding a difference, whereas the FS test was less powerful (especially, when both R(control) and R(treatment) are >1).