sure, the overall sample size of cyclists is large (10^3+) but the number of events leading to mean conclusion of the YouTube video is small (38 cyclists vs 3620 non-active). Hence the very large error bars.

CVD and cancer mortality are very low in terms of absolute events (11 and 15 events vs. 800 and 1000+) hence the huge error bars.

It seems well conducted but I'd still argue small sample size of a small population.

Glaswegians are made of different stuff than the rest of us , so I wouldn't relate this to the North American population.