The point he is making is valid: these groups are not randomly selected and are not representative of one another. Hence, it is not that simple to extrapolate from one to the other.

Would free healthcare reduce deaths in uranium miners that are smokers? Probably some, but would it bring that to the same levels as non-smoking, non miners? Unlikely so. Hence it would be silly to pretend all the cancer deaths are due to lack of health care.

The smoking uranium miner example has a surprisingly realistic story, comes from real data; the dangers of radon in the basement were deduced from comparing cancer rates in smoking uranium miners with high radon in their basements, then they removed what they thought was the smoking and uranium contribution, then fitted to the general population.