Different results in a interaction model and effect measure modification model

[TowfiqulAlam]

Hello, I am trying to see if the association between education level and ideal cardiovascular health is different for Asian who stayed less than 10 years in the US vs. more than 10 years in the US. 

For the interaction effect I have constituted an interaction term (education level*length of stay)

For the effect measure modification, I have created flag variable where I have separated them into two strata, less than 10 years and more than 10 years.

I have separately ran the interaction effect model and the effect measure modification model. I am getting slightly different result for these two models. I thought the result would be the same as interaction and effect measure modification supposed to be the same. Can anyone help me clarifying the cause of this different result?

Interaction model:

PROC SURVEYLOGISTIC DATA=test8 NOMCAR;
STRATA SDMVSTRA;
CLUSTER SDMVPSU;
WEIGHT WTMEC6YR;
DOMAIN FLAG2; *FLAG2 is the Asian Americans population
CLASS Educ_Lvl (param=ref ref='4') GE10Yrs;*Compare <HS, HS, and some college to College graduate or more; *GE10Yrs is years in US (<10 vs 10+);
MODEL Metrics5cat (Event="1")= Educ_Lvl GE10Yrs Educ_Lvl*GE10Yrs/EXPB; *constituing an interaction term (education level*length of stay in US),*event 1:CV health:ideal;
Title "Overall type 3 effect/joint effect of education and length of stay on ideal CV health for Asians Americans";
RUN;

Result for interaction model:

Effect measure modification model:

PROC SURVEYLOGISTIC DATA=test8 NOMCAR;
STRATA SDMVSTRA;
CLUSTER SDMVPSU;
WEIGHT WTMEC6YR;
DOMAIN FLAG3; *Created 2 groups of Asian Americans stratified by years in US (<10 vs 10+); *FLAG3=2 is stay in US >10 yrs;
CLASS Educ_Lvl (param=ref ref='4');*Compare <HS, HS, and some college to College graduate or more;
MODEL Metrics5cat (Event="1")= Educ_Lvl; *Model 1-unadjusted; *event 1:CV health:ideal;
Title "Unadjusted effect measure modification of ideal CV health for Asians Americans";
RUN;

Result for the effect measure modification: