Thread by @AdamJKucharski

Lots of people are now getting interested in infectious disease epidemiology (which is great), but I wanted to do a quick thread on some common mistakes/misunderstandings to watch out for if you're new to these sorts of datasets/questions... 1/

Epi often involves calculating proportions. But make sure the values on the top and bottom of the fraction are being compared in a sensible way. A good example is fatality risk - often people aren't comparing what they think they're comparing...

https://twitter.com/AdamJKucharski/status/1229708001243795458?s=20

2/

The most reliable studies often follow people over time (i.e. prospective), but in real-time we may only have backward-looking studies (i.e. retrospective). Look at differences in this contact-based prospective data and symptom-based retrospective data medrxiv.org/content/10.110… 3/

Watch out for how data are grouped. There are lots of hospitalisations in 20-44 group here, but this spans a 25 year group: the older groups only span 10 year periods. Per capita, risk of hospitalisation is lower in 20-44 group than some older groups cdc.gov/mmwr/volumes/6… 4/

Fitting curves to cumulative data can be misleading. The below two plots show the same data, but the cumulative data is smoother and gives a good exponential fit: the raw data tells a different story. thelancet.com/journals/lance…
More on model fitting here: royalsocietypublishing.org/doi/full/10.10… 5/

Many of the COVID-19 studies coming out were doing in 'under control' settings (i.e. reproduction number <1), so the risks calculated may not be reflective of the actual risk in situations where the outbreak is growing, e.g.

https://twitter.com/AdamJKucharski/status/1235124770896826369?s=20

6/

It's tempting to look at genetic data and get carried away with a 'the virus is mutating!' narrative, but all viruses mutate over time - and generally this doesn't affect outbreak in any meaningful way, e.g.
nature.com/articles/s4156…
virology.ws/2016/04/14/zik… 7/

A good rule of thumb in outbreak analysis is that if something looks interesting/unusual, the most likely explanation is that it's a quirk of how you've interpreted the data. I've outlined some common mistakes above, but some more tips here:
bmj.com/about-bmj/reso… 8/8