This thread by @AdamJKucharski is distorting the most fundamental aspects in the modelling of individual variation in susceptibility or exposure to infection introduced in medrxiv.org/content/10.110…
Here I address the main twists.
Here I address the main twists.
"If 'popular' people are mostly connected to other popular people, we'd expect faster outbreak that doesn't spread that widely. But if highly connected linked to less connected people, we'd expect slower epidemic that spreads further." Not true (pls see our Extended Data Fig 3)!
A common practice in network modelling is to fix the number of links and to compare the outcomes or rewiring those links in different ways. These comparisons are problematic because R0 changes with the rewiring causing effects such as those quoted above.
Instead, we fix R0 and find that mixing assortativity has little effect on the epidemic and associated herd immunity threshold (HIT).
"human interactions can be wonderfully dynamic over time... that person being immune would have made a big difference to outbreak dynamics at the time of the event, but possibly not at another point in time." This is orthogonal to our results!
These individual dynamics can be added whether individuals have equal contact propensity or not. Our estimates for variability in connectivity (meaning long-term contact propensity) apply whether rapid fluctuations are added to the model or not.
Note that rapid fluctuations alone do not affect the expected HIT as far as I reckon.
"As well as variation in contacts between individuals, variation in susceptibility can also influence outbreak size... Effect on dynamics will depend on why susceptibility is variable, and whether influences infection/disease." This is redundant to our approach!
Our approach is to quantify variation in susceptibility or exposure to infection holistically from the effect it has on the shape of epidemic curves. Admittedly we are not taking the reductionist approach to search for all the factors that contribute to that variation.
We infer overall variation in susceptibility to infection. Whether this is due to pre-existing immunity or something else is redundant to the approach, although such information could offer important means to test our inferences if colleagues engaged constructively.
Our models are constructed with inbuilt distributions characterised by a coefficient of variation (CV) - a parameter that can be estimated alongside several others by fitting the model to case data. Knowing CV is crucial for predicting HIT and forecasting the epidemic.
