Request: Please can SAS include the option to calculate test statistics for the unadjusted and stratified MN method.
Background: Miettinen-Nurminen (MN) methods are increasingly being requested by regulatory agencies. This is particularly relevant for non-inferiority trials where it's appropriate for the variance to be estimated under the null hypothesis.
Within the PROC FREQ procedure, it's currently possible to calculate the treatment differences (and CIs) for both the unadjusted MN and the stratified MN methods as well as other MN tests such as the Summary Score method (also referred to as Agresti Score). However, whilst for the Summary Score method SAS allows the option to calculate the test statistics this is not yet available for the unadjusted MN method or the stratified MN method.
These methods are important as they estimate variance under the null hypothesis, specifically the test statistics are of particular useful for:
- testing superiority and presenting a p-value
- studies with interim analyses, particularly when multiplicity is controlled via a group sequential design
i.e. with alpha boundaries less than the typical 1-sided 2.5%/2-sided 5% alpha
Example SAS code (for reference):
Note: data not supplied due to proprietary nature, however code shared to illustrate current options/limitations.
- The unadjusted MN treatment difference can be calculated using:
proc freq data=DATA;
tables TRT*AVAL / riskdiff(CL=(MN) noninferiority margin=0.1);
run;
Note: whilst the (unadjusted) Wald test statistics are calculated, it would be helpful if the unadjusted MN test statistics could also be provided. - The stratified MN Score treatment difference can be calculated using:
proc freq data=DATA;
tables STRATA*TRT*AVAL / riskdiff(CL=(MN) noninferiority margin=0.1) COMMONRISKDIFF(CL=(MN score) test=(score));
run;
Note: whilst the (adjusted) Summary Score (also referred to as Agresti Score) test statistics are calculated, it would be helpful if the stratified MN test statistics could also be provided. Specifically, if MN could also be an option within COMMONRISKDIFF(TEST=( ))
