Read on Twitter
Some #COVID19 rates: their meanings and implications (very roughly). Thread.
- recovery rate
- case fatality rate
- infection fatality rate
- mortality rate
- test positivity rate
- case detection rate
- recovery rate
- case fatality rate
- infection fatality rate
- mortality rate
- test positivity rate
- case detection rate
Recovery rate: % of recorded cases who have recovered. Starts very low at the beginning of a pandemic (no one recovered yet). Remains low during rapid growth. Rises as the pandemic slows and ends high. India at 38.73% is still quite early in this story... https://scroll.in/latest/962325/covid-19-states-announce-own-rules-during-lockdown-4-0-most-permit-public-transport
Case fatality rate: percentage of total recorded cases who have died. Treat with caution. If you're testing widely you lower the CFR - good. If you're under-reporting COVID deaths you also lower the CFR - hmmm. https://twitter.com/muradbanaji/status/1262719734522544131
Infection fatality rate (IFR): percentage of total cases who will die. This is the fatality rate we do really want to know. Should vary according to age structure of a country, levels of general health, access to healthcare, etc. Perhaps I'll do a thread on IFR estimates.
Mortality rate: that's the total number of people who die, say, per million population. At the end of an epidemic that's one number you want to look at to see how bad it's been. In the early stages of an epidemic: it's pretty meaningless.
Test positivity rate: % of total tests which are positive. Tells you if you're testing people with a high probability of having the infection. Used as a proxy for disease prevalence, but low numbers could just mean that you're testing the wrong people. https://twitter.com/muradbanaji/status/1256256032541601797
Case detection rate: the percentage of total infections being picked up in testing. We *do* want to know this, but it's hard to estimate - you need the true total infections i.e., "prevalence". Estimating prevalence requires either surveying (e.g., serosurveying), or modelling.
