Thread by @RSSMedSection

If you are at #RSS2021Conf this morning, check out our invited session on Contagious disease #statistics and #misinterpretation of #COVID dynamics and data with @AdamJKucharski and @AyisSalma
11:40am, Exchange 10

https://twitter.com/RSSMedSection/status/1433715273262108673

First speaker is @AyisSalma, reflecting on the subtle but important difference of what we think and what data are.
#COVID19 statistics are easily misinterpreted

Statistical anomalies in calculations of #COVID19 infections and deaths meant it appeared that no-one ever recovered!
And these figures influenced government #publichealth decisions

The choice of denominator (population or subset of the population) is very important in #COVID19 research.
Confounding and collider bias impacting on estimates must also be considered

Can your #ArtificialIntelligence differentiate #cats 😸 from #COVID19?
Not always it seems!
deepai.org/publication/ca…

The second speaker is @AdamJKucharski on Making sense of #pandemic dynamics

Observed data are not very useful in epidemics. Data are incomplete, out of date, and often #biased.
Hindsight is a wonderful thing, but what can be done in real time?

Early methods included sequential #montecarlo modelling of the R statistic and infection rates.
But the estimate depends on the data source and must be interpreted in terms of data lag and under reporting

Scenarios can be modelled using a range of data sources and assumptions.
But it is it essential to #communicate the difference between #forecasts and #scenarios
❓What WILL happen vs what COULD happen?

Estimating R requires:
- Duration ⏳
- Opportunities 🤝
- Transition probability 🎲
- Susceptibility 🤒
But which data sources to use? And what really drives R?

Final reflections we can all agree with!
Modelling outbreaks and pandemics needs to be:
- Fast ⏩⏰
- Open 🧐
- Peer-reviewed 📖🤓
- Collaborative 🤝👫