Forum for Sampling Practices

I am interested in the sampling plan philosophy people are using in the pharmaceutical industry for qualifications/validations.

I am aware of the work Dr. Taylor(author of Guide to Acceptance Sampling) has done on this subject. He states that ANSI/ASQ Z1.4 is intended for use on a series of lots and not for qualifications. He suggest an alternative approach for validation/qualifications (one-time) sampling plans. This is to select a plan based on the claim that can be made if you pass (associated with LTPD) and also to consider the chances of passing (AQL). He uses tables that are organized by LTPD.

Hi Ed,

Regardless of what method you use to arrive at sample size (n) and acceptance number (Ac), it is important to know what statement you can make as a result of passing the plan with data, or failing the plan. On the other hand, if you don't know the statements you can make as a result of using the plan, you are just following rules and don't know what you are doing.

Statements you can make:
Sampling plans carry out two statistical hypothesis tests simultaneously.

1. When you reject, you reject the hypothesis that the process/lot fraction defective (p') is equal to or less than than AQL. The level of significance of that decision is alpha -- the producer's risk.
2. When you accept, you reject the hypothesis that the process/lot fraction defective (p') is greater than or equal to LTPD. The level of significance of that decision is beta, the consumer's risk.

To explain exactly what these statements mean can become a mouthful for most people. The statements are best described graphically by referring to the OC Curve of the plan. The two statements above correspond to two points on the curve -- the producer's point and the consumer's point. I suspect that the FDA is more interested in the latter.

http://www.samplingplans.com/usingoccurves.htm

I find it useful to point to each of those points on an OC Curve as I explain what the two statements mean. I start the statement with the horizontal axis. Like: IF the true fraction defective is LTPD then the probability that the plan will accept is Beta. This means exactly the same thing as (2) above.

Know both AQL and RQL:
Standard Z1.4 contains sampling plans organized by AQL values, but various AQL plans in Z1.4 do not have any one specific producer's risk. In my opinion Z1.4 makes it too easy for a person to select a plan but not know what it does. Here is a critique of using the Z1.4 method to select a sample size and acceptance number.

http://www.samplingplans.com/aqlprimer.htm