Speaker
Description
<span class="s1">Generation intervals are distributions that describe the time between infection and onward transmission. They are a key epidemiological quantity because, together with the basic reproduction number R</span><span class="s2">0</span><span class="s1">, they determine the population-level growth rate of a pathogen r and its doubling time. Conversely, when fitting epidemiological models to data, assumed generation intervals, therefore, impact </span><span class="s3">R</span><span class="s4">0</span><span class="s1"> inference. This is well-known from studies that have used case data for </span><span class="s3">R</span><span class="s4">0</span><span class="s1"> inference, with many studies emphasizing the importance of choosing an accurate distribution for the generation interval, including both its mean and variation. In phylodynamic inference of </span><span class="s3">R</span><span class="s4">0</span><span class="s1">, however, components of the generation interval are often implicitly assumed. For instance, the birth-death skyline (BDSKY) model assumes exponentially distributed generation intervals due to its constant birth and death rates.<span class="Apple-converted-space"> </span></span>
<span class="s1">Here, we explore the impact of a misspecified generation interval on the estimation of </span><span class="s3">R</span><span class="s4">0</span><span class="s1"> in phylodynamic inference. Specifically, we assess the impact of generation interval misspecification when the ‘true’ interval is gamma-distributed with a coefficient of variation less than 1, consistent with many acutely-infecting viral pathogens including SARS-CoV-2. We find that misspecification of the generation interval by assuming an exponentially-distributed interval with the same mean as the true generation interval can result in </span><span class="s5">a </span><span class="s1">substantial underestimation of </span><span class="s3">R</span><span class="s4">0</span><span class="s1"> and high confidence in this underestimate. We then investigate quick-and-dirty approaches for adjusting the generation interval in phylodynamic analyses to successfully recover ‘true’ </span><span class="s3">R</span><span class="s4">0</span><span class="s1"> values, and highlight their shortcomings. Finally, we expand our phylodynamic analyses to SARS-CoV-2 Omicron sequences from South Africa and Botswana, providing real-world context to our findings. Together, this work highlights the importance of implicit generation interval assumptions in phylodynamic inference and points to the need for methodological development in this area to provide flexibility in </span><span class="s5">the </span><span class="s1">specification of generation intervals. </span>
