[antasvara]

I think the standard calculation is per person-year, not per person. So a 22-year-old dying of a car crash shouldn't skew the statistics, because they only contribute 2 people-years to the 20-24 age group.

That being said, I can see a few plausible biases (though none of them explain the scale of the increase IMO):

1. CRC risk is correlated with some previously fatal, but now curable disease. The mechanism would be that your high-risk CRC patients would die due to yellow fever or something in 1970, meaning they don't have the chance to get CRC. The important thing is that it would have to "artificially" remove high-risk patients from the age group, but not low risk patients.

2. CRC risk is noticeably higher at 24 than it is at 20, and all-other-cause mortality is significantly lower today. That would lead to a higher proportion of 24-year-old "years" in the calculation.

3. People used to die of CRC before it was caught, which caused it to not be recorded as a cancer incidence.

1 seems unlikely, and even if true shouldn't make a big difference. 2 seems the most possible, but still unlikely to make a huge difference. I don't know enough about how they determine cause of death to know if 3 is a possible outcome.