// stpm3 releases to SSC

2023-05-29 Version 1.00 First SSC release
2023-05-29 Version 1.01 Changes so now works in Stata 16
2023-08-16 Version 1.02 Various
                          -- Fixed bug in centiles option
                          -- toffset option added
                          -- Fixed using rmst without ci option
2023-09-01 Version 1.03 Various
                          -- Fixed cumhazard option for scale(hazard) models
                          -- Fixed bug with dftvc(1) when using ode
                          -- CIFs in competing risks models would error out
                             when tvcs differed between models.
                          -- fixed dftvc(1) whne using ode
                          -- fixed using rmst without ci option
2023-08-24 Version 1.04  Various
                          -- dftvc(1)  could produce incorrect predictions with ODEs
                          -- log hazard model predictions with untransformed time when using ode
                             Now works
							 
2023-10-25 Version 1.05  Various
                         -- failure and survival option fixed for competing risks
                            models (broken when added toffset() option)
                         -- fixed a potential error when using merge option 
                            with generated timevar.
                         -- including interactions with no main effect could
                            results in eroneous error when making predictions.
						 
2024-01-25 Version 1.06 Various
                         -- level option did not work with non-integer values
                         -- stpm3km - could not combine factor with frame option
                         -- added ignore suboption to at() options, which gives a
                            prediction even if listed variable not in model.
                         -- some predictions of relative survival model when
                            incorporating expected rates would give error when not
                            using ci option. Now fixed.

2024-02-21 Version 1.07 --  Changed the way expected rmst is calculated. Previous
                            used ODEs now use Gauss-Legendre quadrature. Note 
                            this is just expected survival, ODEs still used for anything
                            based on model parameters.

2024-05-07 Version 1.08 Various
                         -- lnhazard mode can now be fitted in Python to improve speed
                         -- better errors if no observations or no events.
                         -- fixed error with many (6+) extended functions
                            and no interactions.
                         -- tanhsinh quadrature is now the default quadrature   
                            method for scale(lnhazard) models when the spline
                            function is a function of log time. Otherwise
                            Gauss-Legendre quadrature is used.
                         -- Quadrature for log-hazard models is more accurate
                            due to using three-part quadrature.

2024-05-08 Version 1.09  -- fixed prediction error due to change of naming of function

2024-05-30 Version 1.10  -- Competing risks model predictions would lead to error
                            if first model had tvc() option, but subsequent models
                            did not. Now fixed.
                         -- previous release broke crude probabilities due 
                            to missing file in mata library
                            
2024-06-24 Version 1.11  -- Now can predict centiles for lnhazard models 
                            Tided notes for timevar, for using when mergimg frames