Hi Stata users

I've asked beforre, but realize that my post may have been unclear, so hopefully I can make it clearer here:

I'm running Poisson regression to model diabetes incidence rate over four different time periods, using margins to calculate the incidence.
I'm interested in the incidence rates for different demographic groups. But I'm uncertain about two issues:
1. If I want to calculate the rates for each year for different ethnicities: In the Margins, should I use the option 'over' or '#' (see examples below).
2. If I want to compare the incidence rate for e.g. African Americans (coded as two in the ethnicity variable) at two different time points, I have tried to use the 'nlcom' command, but I cannot make it work.

Code:
//Survey weight adjusted Poisson -
svy: poisson _d i.year i.ethnicity i.sex age_split year_split, offset(logtimerisk)

//Predicted margins to give the incidence rates for each ethnicity - with the 'over' option.
margins, over(year ethnicity) expression(predict(n)/FUT_split)

//Predicted margins to give the incidence rates for each ethnicity - with the 'over' option.
margins year#ethnicity, expression(predict(n)/FUT_split)

// CALCULATE INCIDENCE RATE RATIO - for the same ethnicity in two different calendar years:
nlcom exp(_b[2017.year#2.ethnicity] - _b[2005.year#2.ethnicity])

// here I get the error:
[2017.year#2.ethnicity] not found
r(111);
Kind regards
Jannie