During PQ phase 3 data assessment to set up the 3 tier control system (Alert/Action/Specification), we calculated a high standard deviation (Sigma). Therefore, while setting up 3 tier control (Alert Level=Average + 2 Sigma; Action Level: Average + 3 sigma) we found the action level is too high (even exceeds the Specification).
Can anyone advise how we can handle such situation? Is there anyway we can handle the spike data that leads to a high value of Sigma?
You’re going to have to examine the data carefully and reduce sources that may influence a high sigma. This means mitigation (reduction) of the standard deviation. Microbial counts are not subject to normal gaussian distributions.
Did you test your samples in triplicate? By doing so you can determine the average (not the mean) and the %RSD (or SD). Thus, your sigma will be smaller and more representative of the sample population.
The spikes in data could be Outliers. Check if tgey are outliers using a Boxplot and, if they are, do a study on those specic data time, operator etc. If you conclude, those outliers are not possible, then you can remove them from the data set and then calculate the Sigma ( standard deviation). You will have a reduced standard deviation tgat reflects fairly the process SD.
Another method is to sub-group the data with a sub-group size of 3 or more to get pooled standard deviation which is likely to represent the process’ SD more correctly.