I gave a talk on writing academic scientific papers, and writing more generally
github.com/ChrisHIV/tea...

Congrats to @jdrakephd.bsky.social on the important achievement of winning our* infectious disease pub quiz. Honorable mentions to Corin Yeats, Luca Ferretti, @christophraser.bsky.social and @aliciagill.bsky.social
* the Pandemic Sciences Institute; Data, Epidemiology and Analytics section

It's a rare day when Adam Kucharski and my daily chocolate are equally informative

A quick worked example of plotting survival curves on simulated data, comparing the Kaplan-Meier estimator with the Cox proportional hazards model:
htmlpreview.github.io?https://gith...
I'm very far from an expert in survival analysis - let me know if I said anything silly 😬

New in Science: app-based contact tracing for covid, as well as reducing transmission, generated data for epidemic monitoring & analysis in real-time with unprecedented resolution. E.g. below: sharp spikes in transmission in England during last Euros, 2021
Read more at 045.medsci.ox.ac.uk/monitoring

Inference with two correlated variables either or both of which may be censored: worked Bayesian example with Stan code htmlpreview.github.io?https://gith...

🙏

Festive math puzzle

5) This is encouraging validation of the risk model used in the NHS COVID-19 App, and for the prospects of digital contact tracing and precision epidemiology more generally. 👇

We tested whether the app could have done better by using machine learning methods to estimate risk. We found that simple measures, such as the exposure duration, performed almost as well.

Once we know risk factors, how helpful that is depends on their distribution in the population. We show that here, a-c, for all 7 million contacts (blue) and for those testing positive with the background subtracted (‘transmissions’, red). Transmissions are shifted to higher risk.

Shown here is the estimated probability of getting infected (and later reporting positive) by one specific case during a 30-minute exposure to them, as a function of the app-measured risk score. When the app thinks there’s twice as much risk of transmission, there really is!

We estimated and removed the background risk of getting infected, and grouped the contacts by both exposure duration and app-estimated risk score per unit time (derived from proximity). The fraction of contacts testing positive increased with both factors separately:

New in Nature: we used digital measurements for 7 million people exposed to confirmed COVID-19 cases to determine risks for virus transmission, captured well by the NHS COVID-19 App. We found that number of hours of exposure is a major predictor, along with proximity.
tinyurl.com/25tfmzee