I found one paper for your reference. Hope it helps for your study.

Thank Zhijie. I calculated the formula. But, the result for SE is not rational. 

You estimate the difference between regression coefficients for the two items (the one to be replaced and the one replacing) when both items are included as continuous terms in a single model. Separately you compute the 95% CIs using a variance–covariance matrix for the two items.

I have published an analysis using this method (based on Walter Willets) in the attached link and can help you with some Stata code if it would be helpful,

Thank you very much Laura. It can be very important help to me if you can send the stata command. Can you share the command with me? 

Hope this helps.

regress item_to_be_substituted substitution_item `model'

nlcom exp (_b[substitution_item] -_b[item_to_be_substituted]

Regress can be replaced with any other command e.g. stcox etc. just use whatever your main analysis uses.

If you need further help, please feel free to contact me at my email: [email protected]

We want to do a substitution analysis to estimate substitution effects of healthy foods whole grains for unhealthy refined grains Can you help us for calculation, please? I want a sas code to compute the substitution model.

Thanks

Regards

Lakshmi priya

We also want to do a substitution analysis to estimate substitution effects of less healthy food to replace fish consumption using Cox regression model. But the difference of two variables (e.g. fish and meat consumption) was very small (e.g. 0.03) and SE was very large (e.g. 47), which are not normal compared with previous articles (e.g. HR=0.98). May I ask for the detail information on calculation of these two parameters? Is the formula provided by Walter Willets also useble in Cox regression model?