New package on SSC - stipw: Inverse probability weighted parametric survival models with variance obtained via M-estimation

[Micki Hill]

stipw is now available on the SSC archive, as presented at the 2021 UK Stata conference.

stipw performs an inverse probability weighted analysis on survival data using parametric survival models (modelled using streg or stpm2) for the outcome model. Importantly, the variance of the parameters in the outcome model are obtained via M-estimation and therefore include the associated uncertainty in the weight estimation. Both stabilised and unstabilised weights are supported. Standard post-estimation that can be used following streg and stpm2 can be used following stipw. A manuscript of the methods of this work is in draft.

Authors: Micki Hill, Paul C Lambert, Michael J Crowther

Thank you to Kit Baum for uploading stipw to the SSC archive.

[Semaw Abera]

Dear Dr. Hill,
Thank you for this contribution.
Two quick questions:
I have successfully performed 20 multiple imputations on my dataset and run “stipw” under the Royston-Parmar (RP) parametric survival model. My questions are:
1) The stabilized weights (_stipw_weight) are assigned only to the non-missing observation. Will the stabilized weights (_stipw_weight) automatically assigned to all observations after the values are imputed following the 20 multiple imputations? If not, how can I include the “stipw” generated stabilized weights ((_stipw_weight) to all the observations for the multiply imputed observations?
2) How can I generate the weighted Kaplan Meier and restricted mean survival time (RMST) following the suggested predictions for the multiply imputed dataset. When I run the available predictions (example, "mi predict surv, survival ci") , STATA says “requested action not valid after most recent estimation command (requested action not valid after most recent estimation command (stata error= r(321)). Any help?

Thank you in advance for your timely response.