Hello everyone,
I am using STATA 16.1 and I am having some trouble interpreting some of my mediation results from the med4way SSC command on survival data. I have a binary predictor and have 4 outcomes of interest: all-cause mortality, CVD mortality, cancer mortality and non-CVD/non-cancer mortality. I am investigating mediation by continuous vitamin D serum levels.
All-cause mortality:
I used the same code for the other outcomes with the appropriate stset.
CVD mortality:
Cancer mortality:
non-CVD/non-cancer mortality:
According to the med4way command I have calculated the mediation using the formula ((int med + pie)/TE)*100. Note that none of the mediation from interaction terms are not significant so I did not include them in the calculation. Therefore I used the formula (pie/TE)*100:
All-cause mortality: (-0.1179004/-0.1426803)*100= 82.6%
CVD mortality: (-0.1534318/-0.1783579)*100= 86.0%
Cancer mortality: (-0.0721859/-0.1530608) *100= 47.2%
Non-CVD/non-cancer mortality: (-0.1773257/-0.1044076)*100 = 169.8%
I am struggling a little bit with the non-CVD/non-cancer mediation result. I gather that it is possible for the total effect to be smaller than the indirect effect, but does it make sense that the mediation is over 100%? How may I interpret this?
Many thanks,
Andrew
I am using STATA 16.1 and I am having some trouble interpreting some of my mediation results from the med4way SSC command on survival data. I have a binary predictor and have 4 outcomes of interest: all-cause mortality, CVD mortality, cancer mortality and non-CVD/non-cancer mortality. I am investigating mediation by continuous vitamin D serum levels.
All-cause mortality:
Code:
stset survtimeyr, failure(died==1) xi: med4way i.sunbedusecat2 vitdserum i.agerec5 sex i.bmi_cat northco_s i.smokingstatus i.vigPA i.employed i.agecompletededu IMDadj depressionGP , a0(0) a1(1) m( 49.79) yreg(cox) mreg(linear)
I used the same code for the other outcomes with the appropriate stset.
CVD mortality:
Cancer mortality:
non-CVD/non-cancer mortality:
According to the med4way command I have calculated the mediation using the formula ((int med + pie)/TE)*100. Note that none of the mediation from interaction terms are not significant so I did not include them in the calculation. Therefore I used the formula (pie/TE)*100:
All-cause mortality: (-0.1179004/-0.1426803)*100= 82.6%
CVD mortality: (-0.1534318/-0.1783579)*100= 86.0%
Cancer mortality: (-0.0721859/-0.1530608) *100= 47.2%
Non-CVD/non-cancer mortality: (-0.1773257/-0.1044076)*100 = 169.8%
I am struggling a little bit with the non-CVD/non-cancer mediation result. I gather that it is possible for the total effect to be smaller than the indirect effect, but does it make sense that the mediation is over 100%? How may I interpret this?
Many thanks,
Andrew
