How much could potential biases in early data affect our current understanding of COVID-19? This question crops up a lot, so let's look at three crucial aspects of that early data: transmission dynamics, impact of control measures and disease severity... 1/
In the early stages of the outbreak in China, estimates for the reproduction number were in the 1.5–4 range (e.g. eurosurveillance.org/content/10.280…). Subsequent analysis of data from multiple countries has estimated similar transmissibility elsewhere cmmid.github.io/topics/covid19… 2/
When we analysed changes in transmission in Wuhan in late Jan, we used mostly international data (exported cases, evacuation flights). Even without China case reports from that period, we concluded there had been a substantial decline in transmission. thelancet.com/journals/lanin… 3/
Finally, fatality risk: estimates based on China data put the fatality risk at ~1.4% for symptomatic cases (thelancet.com/journals/lanin…). This is consistent with other estimates from Korea (medrxiv.org/content/10.110…) and using Diamond Princess data (eurosurveillance.org/content/10.280…). 4/
In any outbreak - whether flu, Ebola, Zika, or COVID - we must start with the assumption we're only seeing a fraction of the picture. Given the potential biases in data, it's therefore important to find conclusions that can hold up despite these biases. 5/5
