Before I go to my questions, I'll give a short background of my research proposal. Most longitudinal BMI-mortality observational studies use Cox PH to estimate the HRs for any given BMI category (i.e., underweight, normal, overweight, obese), controlling for co-existing illness (e.g., diabetes). However, due to the possibility of reverse causality between disease and BMI (to which the so-called 'obesity paradox' and the J-shaped relationship between BMI and mortality are attributed), these studies exclude the first 5 years of their data to remove individuals who are expected to die because of their underlying disease, and not because of their current BMI per se.
In my research, I proposed to use a cross-lagged panel model approach to address the issue of reverse causality, including smoking, age and sex as confounders... and thereafter, model survival. I wanted to see if the mortality HRs for each BMI category would differ using this approach. The cross-lagged panel model which I constructed using the model builder function of Stata, likewise incorporating survival into it, is as follows... based on my understanding:
Array
In this model, the disease/comorbidity in question is diabetes. Diabetes (categorical variable, y/n), BMI (ordinal variable; underweight, normal, overweight, obese 1, obese 2, obese 3), current smoking (categorical, y/n), age (continuous) and sex (m/f) were all recorded at 3 different time points. All variables in the model are observed data, no latent variables. Note the arrows added also accounted for autoregression.
1. With this way of constructing the model, is my understanding correct if I say that in order to estimate the HRs in this case, all direct and indirect effects (coefficients) from the predictor (e.g., diabetes1) to the outcome (timedth = time to death) be taken added then converted to HR?
2. When performing survival analysis under GSEM, should the format of the data be the same as when you would do the usual survival analysis in Stata?
3. Does this take censored data into consideration automatically, or should I create a separate dummy variable for when the data becomes censored?
Thank you so very much.
0 Response to Performing survival analysis under GSEM
Post a Comment