Community Seminar 2023-07-05 - James Hay - The most important operator in infectious disease modelling, and how to use it with biomarker data

Community seminar tomorrow at 2 pm UK time.

Please use this thread for asynchronous question-asking!

Here are the slides (slightly redacted): epinowcasts2023_redacted.pdf - Google Drive

1 Like

Amazing, thanks James. That was a great talk. Everyone else I will share the recording (for real this time) on youtube shortly along with previous talks (@jameshay give me a ping about this).

Pulled out the references for ease of finding:



Everything is a convolution

Ct values

Thanks for the great talk James! Your work on Ct values during the pandemic inspired me to think about how one could use a Ct value or continuous quantitative measure of viral load to get a probabilistic estimate of the age of infection for a single individual, if you had sufficient information on viral kinetics. At the time, this seemed relevant for contact tracing (e.g. to guide backwards contact-tracing by identifying the source of an infection), and also potentially useful for guiding individual decisions regarding treatments (e.g. when it is better to take an anti-viral vs a cortico-steroid). I’m wondering if you could speak a bit to that direction of utilizing Ct values and whether you see it as one worth pursuing in certain contexts (e.g. high density settings or in hospitalized patients to inform treatment decision-making)? For fun/out of anger my old colleague and I did a quick analysis similar to this after Dr. Fauci accidentally revealed Trump’s Ct value as he was recovering from COVID-19 in 2020. We also worked with the contact-tracing team at the university I was at, and they were interested in using this method to guide contact-tracing efforts to better identify super-spreading events.

1 Like

This is a cool analysis! I don’t see any reason why we shouldn’t generate these uncertain estimates, it’s just a question of what they should be used for. Some of the clinicians we collaborate with were definitely using Ct as a proxy for infection recency, and they would use them in a qualitative way to triage, combined with other contextual information. Others would not trust the single measurement as we are familiar with hearing. So the analysis you proposed is already being used informally.

I agree, this is all just about back-calculating time-of-infection, and clearly any biomarker measurements which inform that could be used, provided a reliable model is available and we accurately capture the huge uncertainty (as you demonstrate from a single measurement). And ofc, if we had measurements over time, or measurements of multiple biomarkers from a single time point then we could probably get some quite precise estimates on TSI. This paper from Jamie Lloyd-Smith’s lab, Prager et al. Linking longitudinal and cross-sectional biomarker data to understand host-pathogen dynamics: Leptospira in California sea lions (Zalophus californianus) as a case study - PMC looks at this idea of combining biomarkers. There is also this great paper: Estimating Time of Infection Using Prior Serological and Individual Information Can Greatly Improve Incidence Estimation of Human and Wildlife Infections.

I think there is a really interesting idea there on epidemiological data synthesis – the whole “who infected who” question. People use sequence data to estimate transmission trees; people use linelist data from households to estimate latent infection times/pairs; and IIRC there is work using joint likelihoods to combine these data. I don’t see why one couldn’t also include a likelihood for time-of-infection from a single Ct value to help pinpoint when an infection was likely to have started. Particularly because if you’ve got a sequence, you probably already have a Ct value for free.


Looking forward to viewing the recording.

Re this Ct discussion, agree plenty to work on. Some tough-but-interesting questions about how single measures might be used, both in terms of getting a reference curve, but also how to deal with more banal ones like “are we in the waxing or waning stage of the infection?”.

Strikes me as equally interesting to think about more-than-one testing and then getting obsessively fiddly about e.g. the incremental informational contribution of each additional draw. Or whether we can get Ct-like surrogates from RDTs (e.g. line intensity).

1 Like