If you use the CURVE= option and the GROUPSURVIVAL= option in the TWOSAMPLESURVIVAL statement in PROC POWER to specify a piecewise linear survival curve, then the survival rates must be strictly decreasing. However, PROC POWER fails to detect an invalid curve that is flat between any time intervals, and incorrect results are produced.

To circumvent the problem, modify one or more of the survival rates so that the curve is strictly decreasing.

For example, suppose the following control curve is specified:

   CURVE("Control") = (2 4 6 8 12 24):(1.00 0.92 0.69 0.62 0.46 0.18)

Since the survival rate in all survival curves is assumed to be 1 at time 0, this curve has a flat portion because the survival rate is also 1 in the first (time, survival) pair. To correct the problem in this case, replace the survival rate of 1 with a smaller value that is close to 1, such as 0.999.

Operating System and Release Information