Validation Status
The %JM macro has been extensively tested but not followed a formal process of documented software validation [General Principles of Software Validation; Final Guidance for Industry and FDA]. The use of fully validated software may required in fields such as analysis submitted to regulatory authorities (e.g. for the registration of new medicines).
This limitation may be addressed with one of the following strategies:
a) Use %JM macro, e.g. for sensitivity analysis, and add a warning note (e.g. in the statistical analysis plan): “The %JM macro is academic non-validated software. No other fully validated software (e.g. a supported SAS procedure) is available yet for this family of models.”
b) Use the %JM macro as an one-off (single use) program and the JM R package as its corresponding QC program (or vice versa) to verify that both programs produce the same output. Since both the %JM macro and the JM R package were independently programmed (different programmers in a separate environment using distinct approximation and optimization routines), matching results provide strong evidence of a validated status for the output.
c) Program the model yourself for a specific model/data (single use program) and use the %JM macro as a QC program (or vice versa).
This limitation may be addressed with one of the following strategies:
a) Use %JM macro, e.g. for sensitivity analysis, and add a warning note (e.g. in the statistical analysis plan): “The %JM macro is academic non-validated software. No other fully validated software (e.g. a supported SAS procedure) is available yet for this family of models.”
b) Use the %JM macro as an one-off (single use) program and the JM R package as its corresponding QC program (or vice versa) to verify that both programs produce the same output. Since both the %JM macro and the JM R package were independently programmed (different programmers in a separate environment using distinct approximation and optimization routines), matching results provide strong evidence of a validated status for the output.
c) Program the model yourself for a specific model/data (single use program) and use the %JM macro as a QC program (or vice versa).