I am analysing data from an open cohort and because there is a selection bias associated with the exposure, I calculated the propensity score (PS) and used it as covariate in my regression model (PS covariate adjustment). I am aware that many prefer other PS methods such as matching and weighting but I wanted to keep it simple for now. Also, I am also comparing results without PS adjustment. Ideally, I would like to conduct additional analyses on a sub-set of the study population (around 40% of the sample). However, the PS was generated considering the entire population, therefore, I am questioning the validity of this approach. Certainly I am comparing results with those without PS adjustment to avoid model misspecification. Also, because these are only explorative analyses I was wondering whether there is a rationale to go ahead with sub-analyses using the PS calculated considering the whole study population.
The relevant question is whether in the subsample the covariates are independent of treatment conditional on the propensity score. If so, then by using the same logic as in the whole sample, you can validly perform subgroup analyses that have the same causal interpretation as that of the analysis with the full sample. Green & Stuart (2014) (DOI:10.1037/a0036515) describe the various alternatives for subgroup analysis after propensity score adjustment.
One of the problems with using the propensity score in a regression model is that balance assessment is not straightforward. That is one of the advantages of propensity score matching or weighting.