PROC CAUSALMED for mediation analysis

Analu — Wed, 13 Mar 2024 15:43:48 GMT

Dear SAS users,

I am doing a mediation analysis using cross-sectional data (n=7,173):

X= family tension (binary)
M= sleep disturbance (continuous)
Y= Depression (binary and rare outcome in my population)

There is no interaction between X (family tension) and M (sleep disturbance)

The equations are the following:

Logit {P(depression=1|a,m,c)}=θ0+ θ1family_tension+ θ2sleep_disturbance+θ3family_tension*sleep_distrubance+θ’4covariates

E[sleep_disturbance|family_tension,covariates]=β0+ β1dispute + β’2covariates

When I run my mediation analysis using @PROCCAUSALMED, I obtain the following results:

Total effect (OR= 4.05, 95%CI, 3.06-5.05)
Natural direct effect (OR=2.39, 95%CI, 1.88-2.90)
Natural indirect effect (OR=1.70, 95%CI, 1.48-1.91).

If I run a logistic regression between X and Y (without the mediator and including covariates), I would expect to get the OR=4.05 as in the CAUSALMED result of the total effect, however I get a OR= 3.06, 95%CI, 2.49-3.76).

Does anyone know why I am not getting the same OR? I would expect to also get OR=4.05 from a logistic regression (x->y without the mediator), because I had understood the total effect in the causal mediation would be equivalent to the reduced model using traditional mediation (x->y without the mediator).

Perhaps I am mixing concepts, your help is very wellcome. Thank you so much in advance,

Ana

topic PROC CAUSALMED for mediation analysis in SAS Programming

PROC CAUSALMED for mediation analysis