But fortunately the {EpiNow2} R package has a nice tool to take data on delayed outcomes and the delays from infection to that outcome, and reconstruct the number of new infections over time. https://epiforecasts.io/EpiNow2/dev/
Comments
Log in with your Bluesky account to leave a comment
One thing we've been wondering, @seabbs.bsky.social, is if EpiNow2 is used for the express purpose of recreating an infection timeseries based on case data, how long of a case timeseries can be inputted and the expectation be that the estimated infection timeseries is relatively stable?
In the example above, we show results from a Linton et al estimate of onset-to-death (mean 20 days), which gives infection peak around 18th March (figure below).
You'll notice that it isn't a 5 day difference in peak, because the 'convolution' is a delay and addition of all the events, rather than a simple shift in the overall curves...
Note this doesn't necessarily mean that measures introduced after the peak had no effect. At an epidemic peak the growth rate is zero: if no further decline had happened, the epidemic would have stayed at this level, in this case at 2500+ estimated fatal infections per day...
Yes, and perhaps more than this - the future can influence the present, if it is anticipated. Anticipating that a road will be closed due to snow can cause me to stay at home. Some of the measures introduced after the peak were widely anticipated - people adopted the restricted behaviour in advance.
And if social contacts had only reduced slightly further after the peak, we’d have seen a decline but a much more gradual one.
This shows the importance of thinking about assumptions in epidemic analysis - and value of packages like {EpiNow2} and {epiparameter} to compare different possibilities.
(The above is an expanded version of an earlier Bluesky discussion with @seabbs.bsky.social deep in a random conversation thread, but thought useful to outline in more detail in an easier-to-find new post.)
Comments
In the example above, we show results from a Linton et al estimate of onset-to-death (mean 20 days), which gives infection peak around 18th March (figure below).
This shows the importance of thinking about assumptions in epidemic analysis - and value of packages like {EpiNow2} and {epiparameter} to compare different possibilities.
(I ask because I read somewhere that using a distri curve has an impact)