RT Journal Article SR Electronic T1 In-Process Microbial Testing: Statistical Properties of a Rapid Alternative to Compendial Enumeration Methods JF PDA Journal of Pharmaceutical Science and Technology JO PDA J Pharm Sci Technol FD Parenteral Drug Association (PDA) SP 264 OP 269 DO 10.5731/pdajpst.2015.01019 VO 69 IS 2 A1 Emil M. Friedman A1 Mark Warner A1 Sam C. Shum A1 Fred Adair YR 2015 UL http://journal.pda.org/content/69/2/264.abstract AB In-process tests are used between manufacturing steps to avoid the cost of further processing material that is apt to fail its final tests. Rapid microbiological methods that return simple negative or positive results are attractive in this context because they are faster than the compendial methods used at product release. However, using a single such test will not reliably detect barely unacceptable material (sensitivity) without generating an undesirable number of false rejections (poor specificity). We quantify how to achieve a balance between the risks of false acceptance and false rejection by performing multiple rapid microbiological methods and applying an acceptance rule. We show how the end user can use a simple (and novel) graph to choose a sample size, the number of samples, and an acceptance rule that yield a good balance between the two risks while taking cost (number of tests) into account. LAY ABSTRACT: In-process tests are used between manufacturing steps to avoid the cost of further processing material that is apt to fail its final tests. Rapid microbiological methods that return simple negative or positive results are attractive in this context because they are faster than the compendial methods used at product release. However, using a single such test will not reliably detect barely unacceptable material (sensitivity) without generating an undesirable number of false rejections (poor specificity). We quantify how to achieve a balance between the risks of false acceptance and false rejection by performing multiple rapid microbiological methods and applying an acceptance rule. We show how the end user can use a simple (and novel) graph to choose a sample size, the number of samples, and an acceptance rule that yield a good balance between the two risks while taking cost (number of tests) into account.