For Restricted Mean Survival Time analysis, SAS documentation is incorrect. Suggest we add page for RMST analysis in SAS & R, to let people know about incorrect SAS guidance (and can then compare the correction to SAS).
Guidance here: https://support.sas.com/documentation/onlinedoc/stat/151/rmstreg.pdf
States in 2 places: page 8616 "The TAU= option in the PROC RMSTREG statement specifies the time limit that defines the RMST for this
analysis. If you omit this option, the largest event time from the input data is used. " and on page 8620: "TAU=value
specifies the upper time limit of the RMST. The default value is the largest event time, where value must be positive. "
However if you omit this option, the default uses the largest time (of patients with censoring or events). This results in incorrect analysis so Tau must be calculated prior to the procedure as the largest event time, then used as Tau (Do not leave to default).
Also very important to make sure you have the cnsr / event flag the right way around. if per the Guidance page 8615 you use: "Status, event indicator with the value 1 for death time and 0 for censored time." then make sure as per page page 8616 you code the model as: "model Time*Status(0)".
For Restricted Mean Survival Time analysis, SAS documentation is incorrect. Suggest we add page for RMST analysis in SAS & R, to let people know about incorrect SAS guidance (and can then compare the correction to SAS). Guidance here: https://support.sas.com/documentation/onlinedoc/stat/151/rmstreg.pdf States in 2 places: page 8616 "The TAU= option in the PROC RMSTREG statement specifies the time limit that defines the RMST for this analysis. If you omit this option, the largest event time from the input data is used. " and on page 8620: "TAU=value specifies the upper time limit of the RMST. The default value is the largest event time, where value must be positive. "
However if you omit this option, the default uses the largest time (of patients with censoring or events). This results in incorrect analysis so Tau must be calculated prior to the procedure as the largest event time, then used as Tau (Do not leave to default).
Also very important to make sure you have the cnsr / event flag the right way around. if per the Guidance page 8615 you use: "Status, event indicator with the value 1 for death time and 0 for censored time." then make sure as per page page 8616 you code the model as: "model Time*Status(0)".