I am interested in running lasso inference models (starting with dsregress), but I have non-trivial amounts of missing data in my sample. When estimating a model using OLS, I first dropped observations that had missing data for my variables of interest, then dealt with missing data for covariates in the following way: replace missing values with 0 and include an indicator variable in the regression for whether the value was originally missing, as shown below:

Code:
foreach var of varlist hhincQ par_edcat hsgpa_orth sat_orth badbehavior_orth int_restless_orth earlysex_orth ///
    enrich_index contact_biomom contact_biodad notinmsa overweight obese religAnCont {
    gen `var'_d=`var'==.
    replace `var'=0 if `var'_d==1
}
What I would like to do with Lasso is force it to pick groups of two variables together - the original variable and its missing indicator. Is there a way to do this? If not, how might one suggest dealing with missing data for a large number of covariates and relatively small sample in lasso?

Thanks very much!