Why might our estimates of epidemiological delays be biased?
We are trying to estimate an incubation period. For person A we know exposure happened on day 1 and symptom onset on day 3.
We are trying to estimate an incubation period. For person A we know exposure happened on day 1 and symptom onset on day 3.
We are trying to estimate an incubation period. For person A we know exposure happened on day 1 and symptom onset on day 3.
We are trying to estimate an incubation period. For person A we know exposure happened on day 1 and symptom onset on day 3.
The true incubation period of A could be anywhere between 1 and 3 days (but not all equally likely).
We are trying to estimate an incubation period. Each arrow represents one person with an associated pair of events (infection and symptom onset).
We are trying to estimate an incubation period. Each arrow represents one person with an associated pair of events (infection and symptom onset).
We are trying to estimate an incubation period. Each arrow represents one person with an associated pair of events (infection and symptom onset).
We are trying to estimate an incubation period. Each arrow represents one person with an associated pair of events (infection and symptom onset)
On the day of analysis we have not observed some onsets yet. The delay from infection to onset for these delays tended to be longer. This is made worse during periods of exponential growth.
We are trying to estimate an incubation period. Each arrow represents one person with an associated pair of events (infection and symptom onset)
We need to account for infections with longer delays that we haven’t yet observed. We can best do this by using a lognormal distribution for the delays that we do have data on.
Introduction to biases in epidemiological delays