I've noticed for 499 client timeouts the upstream_connect_time nginx logged is sometimes an empty value, and sometimes a hyphen '-'.
I'm guessing this difference in how this field is logged means something - but I haven't seen this explained in the documentation. Would be interesting to know.
The upstream connection in this case is over https. Currently my best guess is perhaps this distinguishes between when a TCP connection was made (without completing the SSL handshake) vs no connection.