Proc causalmed: Natural direct path exceeds 100%

[GLO1]

Hi,

I was doing a causal mediation analysis using proc CAUSALMED, but I have difficulties interpreting the results.

In an analysis we are using only one mediator; the natural direct decomposition shows a precentage of 118.71, the natural indirect path -18.71. In what way can I interpret these results?

Thank you in advance!

[MichaelL_SAS]

I assume that you are refereeing to the percentages reported in the “Percentage Decomposition of the Total Effect” table requested by the DECOMP option. Negative percentages for the natural direct effect (NDE) and natural indirect effect (NIE) can be reported when the NIE and NDE have opposite signs. Based on the total effect (TE) decomposition, TE = NDE + NIE, the combined percentages reported for the NDE and NIE should equal 100%. In your case it sounds like the NDE is larger than the total effect, so the NDE percentage, NDE/TE*100% results in a value greater than 100%, and therefore a negative NIE percentage is reported. In this type of situation these proportion measures for the NIE and NDE are less meaningful.

[GLO1]

@MichaelL_SAS  Thank you for your response, it is appreciated.

It is indeed true that NDE is bigger than TDE.

However, both NDE and NIE are significant. Theoretically, does this mean that the relationship between X (sex) en Y (dementia) is positively explained through the NDE, and negatively explained through NIE (mediator =macrovascular complications)? We already found that females had higher risk of dementia, and macrovascular complications also increase dementia. Am I right that in other words the NIE being negative would mean that less women have macrovascular disease, resulting in that the indirect natural effect is negative and does therefore decrease dementia risk for females through this path? Or does it just mean that it the higher risk of dementia is not explained by the mediator, because it only says something for males in this case.

Thank you

[MichaelL_SAS · Accepted solution]

Since I’m not familiar with the data for your example and the subject area I don’t want to comment specifically on it, but let me try a more general answer to your question. The natural indirect effect is interpreted as the effect the treatment has on the outcome through its effect on the mediator variable. So in the situation where the outcome is the risk of an event occurring, a negative NIE and positive NDE indicates (assuming all the necessary assumptions are satisfied) that the treatment’s effect on the mediator leads to a decreased risk of the outcome occurring, but the effect of treatment on the outcome through any remaining pathways (not through the mediator) increases the risk of the outcome occurring. One way this might occur is if the treatment has a positive effect on the outcome (directly), the mediator also has a positive effect on the outcome, but the treatment has a negative effect on the mediator leading to a negative NIE. Also, in some situations, it might be the case that both the NIE and NDE are significant, but the overall total effect is not. A formal definition of these effects in a counterfactual framework is provided in the “Conterfactual Framework for Defining Causal Mediation Effects” section of the PROC CAUSALMED documentation.