stintreg with spline or fractional polynomial (or other ways to visualize non-linearity)

[Jeffrey Chan]

I am doing a study in which the outcome is type II censored. Interval-censored Weibull model is chosen as the primary analytic model, and I would like to visualize the association between the continuous independent variable (-inflam-) and the risk of the outcome both for reporting and for exploring non-linearity. I tried using fp for this, but it appears that stintreg is either not supported (it is indeed not included as a supported estcmd in the manual), or the predicted component is not something that I understand. I also tried using mkspline2 with xblc to plot RCS, but it keeps showing an error saying that the specified level of -inflam- was not observed (even when it showed up on the spreadsheet view??).

I also played a bit with partpred after mkspline but it again gave somewhat funny results. Interestingly, I tried to center the results at inflam==0 (so the HR can be interpreted as compared to when inflam==0) but there remained a CI at inflam==0, i.e. the CI bands did not converge at inflam==0. Not sure if there's something I'm missing or if stintreg is not supported.

I am out of ideas and would appreciate ideas / advice / suggestions. Thank you very much!!

[Rich Goldstein]

the help file for stintreg explicitly says that fp is allowed so I think you need to give us more information - there is good advice on asking questions in the FAQ for this forum; you might also want to look at

h makespline

for rcs code that might be easier

[Chen Samulsion]

Reminder: makespline is a Stata 18 command. Stata 18 Makespline vs mkspline