How are type II boundaries updated with ERRLINE in PROC SEQTEST with BOUNDARYKEY=ALPHA?

[marghe]

Hello all,

I have a doubt about proc seqtest procedure. When the actual information level of the first analysis (and subsequent analyses) is higher than the one planned in proc seqdesign and the options ERRSPENDADJ=ERRLINE and BOUNDARYKEY=ALPHA are specified, type I error is preserved while type II errorr is updated and efficacy and futility boundaries are recalculated based on the boundaries calculated in proc seqdesign and provided in the BOUNDARY statement.

I'm trying to reproduce (manually) the calculations done by the SAS procedure, I get new type I error boundaries exactly like proc seqtest but I don't understand how to get the new type II error boundaries. I already had a look at the documentation and tried different methodologies including the one related to the ERRLINE described in SAS Help Center: Boundary Adjustments for Information Levels but I couldn't get the same values.

Is there anybody that can help me understanding how these boundaries are updated given the ones from proc seqdesign?

I can also add some codes with a simulated dataset but my issue is related to the mathematical computation rather than an issue with the procedure used with certain data.

[ballardw]

When you say something like " I couldn't get the same values" that means you need to provide data and code you used for calculation. Without knowing what you did we can't say why you get a different result. If you actually did this on paper by hand then that goes double.

If your result difference is "small" then I would look to possible rounding in calculations or a boundary value.

If your result is extremely different then likely there is a major difference in how you approached the calculations.

[marghe]

Thanks for your reply! Unfortunately, it isn't a matter of decimals, it's really about the formula used to compute beta boundaries. I don't know the function used to get beta boundaries from the boundaries calculated in the proc seqdesign.