@sseb What you're trying to estimate doesn't really make sense unless you have an interaction between sex and BMI (without an interaction the HR is the same for both males and females). If you include an interaction I think it would be something like: estimate "&value. &year. sex level 1" bmis [-1, &ref.] [1, &value.] sex*bmis [-1, &ref. 1] [1, &value. 1]/ exp cl cov; estimate "&value. &year. sex level 2" bmis [-1, &ref.] [1, &value.] sex*bmis [-1, &ref. 2] [1, &value. 2]/ exp cl cov where sex is a class variable. This should be relatively easy to check since the interaction only matters for the non-reference level, so for the reference level of sex the two lines estimate "&value. &year. sex level 2 (ref)" bmis [-1, &ref.] [1, &value.] sex*bmis [-1, &ref. 2] [1, &value. 2]/ exp cl cov
estimate "&value. &year." bmis [-1, &ref.] [1, &value.] / exp cl cov should produce identical results, so you can just change which level is the reference and use the simpler code.
... View more