I'm not sure, whether this article meets your problem, because you did not specify the disease(s) you want to investigate. But for a risk factor affecting various diseases this seems to represent the state of the art.

http://www.biomedcentral.com/1471-2288/11/48

By "risk-factor-specific incidence rates", do you mean population attributable risk (PAR)?

Thank you for your comments, and sorry for unclear question.

What I want to know is not relative risk or PAR per se. I wan to know the absolute incidence rate (unit : per 100,000 population) of a disease among a population with a specific risk factor.

For example, the absolute lung cancer incidence or mortality rate of the Japanese whole population is known by cancer registry or vital statistics, but the absolute lung cancer incidence/mortality rate of the Japanese smokers is unknown. What we know is relative risk of lung cancer for smokers vs. non-smokers. So my question is whether there is a way to allocate the absolute rate of overall population into smokers and non-smokers. I imagine I can do it by using the same concept as PAR.

Of course some cohort studies report absolute incidence/mortality rates according to smoking status, but these are values observed in a specific cohort. I want to know the values in the whole population.

Did it make sense?

We did just that in our paper on smoking in South Africa , see Lancet 2013. We followed Liu BQ on his paper on mortality from smoking in China.

Thanks Freddy! That's exactly what I was looking for.