Statistical Techniques in Research - Application of 820.250

[dizzyjock]

Hi all

I work for a medical software firm, making Class 2 Picture Archiving devices.

Our R&D group do the fundamental work on developing algorithms etc, before handing over code and some documentation to Engineering to incorporate into the product. Reading 820.250, has anyone had any experience of that particular requirement being applied to the handover from R & D to Engineering ?
My rationale is this product characteristic (which is critical to the product) requires verification.
So, do we need a procedure to identify these statistical techniques to be used before it's handed over ?
Or, can we document the statistical rationale used in the science documentation and science test results on a case by case basis (given how varied the work is, this would be easiest).

Comments much appreciated

 

[Marc]

I'm not familiar with 820.250. You're providing software and/or firmware?

 

[dizzyjock]

Software only. I searched for some reference material on Subpart O : 820.250 Statistical Techniques, but it's pretty thin on the ground, and now mainly references CAPA.

 

[Al Rosen]

Hi all

I work for a medical software firm, making Class 2 Picture Archiving devices.

Our R&D group do the fundamental work on developing algorithms etc, before handing over code and some documentation to Engineering to incorporate into the product. Reading 820.250, has anyone had any experience of that particular requirement being applied to the handover from R & D to Engineering ?
My rationale is this product characteristic (which is critical to the product) requires verification.
So, do we need a procedure to identify these statistical techniques to be used before it's handed over ?
Or, can we document the statistical rationale used in the science documentation and science test results on a case by case basis (given how varied the work is, this would be easiest).

Comments much appreciated

The verification process is performed during the development of the device prior to the Design Transfer to production. Validation is performed on production product not prototypes, i.e after transfer to production and on production devices.

820.250 addresses statistical techniques regarding verification of processes and product characteristics. It requires that you identify where you will need to use sampling, rather than 100% verification and that when you use these techniques, they are determined on statistical principles. If you chose a sampling plan, you should be able to justify it and show that it is based on a recognized standard like ANSI/ASQ Z1.4.

Look at the FDA Guidance Document
General Principles of Software Validation; Final Guidance for Industry and FDA Staff.

 

[Weiner Dog]

There is a difference between device product software v hardware validation/verification testing. Additionally, design verification/validation is not the same as process validation/ acceptance testing. (During the design process, you are probably conducting process validation studies as well.)

During the design software validation process, software has to be tested once (i.e. because you are validating the coded program to see if it meets your predetermined specifications); whereas for process validation, you have to determine statistically how many units to test (i.e because you are determining whether the manufacturing process is consistent over-time to manufacture the units that meet predetermined specs- not only for hardware but software too). For parts of the design validation process (such as during actual or simulated use tests), if you state that you are going to test "X" units, then you have to prove statistically why you chose these samples.