1/14 Dear #epitwitter,
Our first peer reviewed #COVID modelling paper has just been made available online by the grand classic Journal of Theoretical Biology 🙏authors.elsevier.com/sd/article/S00…
Others will follow shortly I believe.
2/14 Joined this platform in June 2019 to say some forms of individual variation impact epidemic dynamics hugely. Not all variation matters and explanation isn’t "nonlinearities this and that". Key process can be described linearly, intuitively, and quantifiably by inference.
3/14 When covid-19 emerged I couldn’t but apply concepts and new inference procedures to the pandemic. Also couldn’t but communicate results through this and other platforms.
This post summarises concepts, results, how they were received, and how mission has been accomplished.
4/14 When a population consisting of individuals with different levels of susceptibility or exposure to infection is invaded by a pathogen, those with higher susceptibility or exposure tend to be infected first depleting susceptibles faster than a homogeneous model would assume.
5/14 This results in lower herd immunity thresholds and smaller epidemics than an insufficiently heterogeneous model might have predicted:
6/14 Our first preprint saying this was submitted to medRxiv on the 27 April 2020 (memorable day for me as it was the first birthday of my first grandson, and it would have been the 81st birthday of my father).
7/14 Your reaction, dear #epitwitter, was beyond comprehension. You hated this work (which I loved) for some reason. You resorted to most imaginative arguments and means to discredit the work and hurt careers/livelihoods of those involved. Here main arguments and our responses:
8/14 1- Not new theory.
Re: Right (see Introduction to our JTB paper) but infectious disease epidemiology (IDE) researchers don’t make full use it. They prefer to model a few specific traits (such as age and contact degree) and ignore unmeasured variation, inflating epidemics.
9/14 2- Behaviours are complex, contact networks rewire, and what not.
Re: Yes, but these micro processes are implicit in our macro inferences. We infer variation from its effects on epidemic curves. Variation cancelled by rewiring or whatever is excluded by inference approach.
10/14 3- Gamma distributions not good representations of individual variation in exposure.
Re: They seem to be. See Section 5.4.1 of JTB paper where we review contact surveys and fit gamma distributions to published data. In any case conclusions not limited to gamma distributions
11/14 4- Non-pharmaceutical interventions (NPIs) are under-represented.
Re: No, they are not. Our inferences result in similar NPI efficacy (but not effectiveness; see below) to those reported by other groups.
12/14 I don’t recall further criticisms and those I recall seem straightforward to address on retrospect. So, mission to establish and communicate what individual variation in susceptibility and exposure does, and how it can be estimated and used has been accomplished.
13/14 Sad, however, that our models were not used to inform policy. Sad because although we estimate similar NPI efficacy to other research groups, the impact of such measures on pandemic is much lower according to our models because unmitigated epidemics would have been smaller.
14/14 To conclude, we might have controlled the pandemic almost as effectively with less restrictive measures.
PS1: With this, I now unblock and unmute those accounts that I had to shield myself against to protect wellbeing of myself and my family, my work, and our team.
PS2: Sorry the JTB paper is not accessible without subscription. Another nefarious effect of the campaign to suppress out work is that we could not secure any funding (not even dream to try) so could not pay open access. Here is the equivalent on medRxiv: medrxiv.org/content/10.110…
• • •
Missing some Tweet in this thread? You can try to
force a refresh
Almost 2 years since our low-HIT (herd immunity threshold) preprint was released the paper has been peer reviewed and accepted for publication in a scientific journal (specific details soon).
Following the initial preprint submitted to medRxiv in April 2020 the theory that individual variation in susceptibility and exposure (frailty variation) to infection lowers HIT and epidemic final size was featured in many news and science outlets, eg:
Two years ago had privilege to be: offered/accepting Global Talent Research Professorship at Strathclyde University @StrathMathStat; awarded Habilitation from Porto University. Both in recognition for research/teaching variation/selection in epidemiology/ecology/evolution.
Then pandemic emerged and took all time/attention. A lot has been sacrificed but I believe for good causes: research-wise immediately; policy-wise may take longer.
Then decided to write thread about what's killing me. Not depressing (on contrary). It's my duty to make the world understand this +ve thing before I die [not that I think I'll die soon; that was just in dream]
PLEASE READ IF YOU CAN!
For more than 10 years I've been researching with collaborators (including @mlipsitch@GrahamMedley) why epidemic models tend to exaggerate epidemic sizes and overestimate intervention impacts (particularly vaccines but also NPIs): journals.plos.org/plospathogens/…
1/ Many times I have been asked why communication around herd immunity threshold (HIT) was so confusing in this pandemic. I have even been asked whether experts really understand it. Here is my answer:
2/ The concept is well understood among mathematical epidemiologists. In my view what went terribly wrong was the politicised way in which the HIT was used in this pandemic.
3/ The HIT is the percentage of the population that needs to be immune (prior to vaccination immunity was a natural outcome of recovery from infection) before the epidemic peaks and subsides.
Last year, prominent modelling groups dismissed our Covid work (known for incorporating individual variation in susceptibility and exposure, and estimating low herd immunity thresholds) by claiming our stylised contact-reduction profile (Rc) wasn't close to government NPIs..
England and Scotland (with our latest stylised Rc):
This week, I had a few moments to spare and implemented the same model and fittings with the stringency index that tracks government response. The results are almost identical...