We will cover three issues in this Cleaning Memo. They are (a) how action and alert levels are defined, (b) how action and alert levels are used, and (c) how action and alert levels are established, all in the context of pharmaceutical process equipment cleaning validation.
The first issue is one of clarifying terminology (as should be obvious for regular readers of my writings, clarifying terminology is “near and dear” to my heart). Some companies refer to this as action/alert limits rather than action/alert levels (the term I use here and the term I try to use consistently). The reason I use “levels” is, first of all, it appears to be the preferred term in cleanroom environmental monitoring, although it is not used universally there. Secondly, and of equal importance, is that I prefer to distinguish a limit from a level. In cleaning validation, a limit typically refers to a residue value in a protocol that I don’t want to exceed. And if I exceed it, it represents a failure. Something went wrong either with the design of my cleaning process, with the design of my validation protocol,
with sampling, or with analysis (although other things could be added to that list).
An action/alert level is an indication of process control. While limits are typically values I don’t want to exceed, for action/alert levels I may have a range that I want the results to be within. That is, I ordinarily expect the value to be within in a typical range (high and low), and if it is outside that range, perhaps something is wrong or something has
changed in my cleaning process.
This really gets us into the second topic of how the action/alert levels are used. The main value of action/alerts levels is in routine monitoring of a cleaning process after the cleaning process is validated. The monitoring may be residue levels (for example, of highly hazardous actives or of microorganisms) or it may be a process parameter, such as temperature, pressure, cleaning agent concentration and time. Provided I select what to measure and establish a reasonable range of values, I may have some indication that my cleaning process is in a state of control or else I may have an indication that my cleaning process is out of control (or perhaps trending out of control). Now while I say the main purpose of alert and action levels is for routine monitoring, I will still do that routine monitoring for my validation protocol runs as part of the cleaning SOP; if that routine monitoring is done as part of the cleaning SOP, then in order to follow the SOP I must collect that routine monitoring data. In routine monitoring if I am outside the action levels (but still within any acceptance limit), that is a clear indication that an investigation is warranted; the result of that investigation may be correction and/or corrective actions. If I am outside the alert levels (but not outside the action levels), I will typically note it (for trending purposes) but may do little to investigate it further. If I am outside the alert level consistently, I may treat it as if were outside the action level.
Which bring us to the third issue of how to establish the action/alert levels. Since the levels are indications of process control, I would prefer to establish it on process data. For example, if I am collecting final rinse TOC data for routine monitoring purposes, I want to collect data for a large number (for right now, let’s say at least 20) of cleaning runs. I can then determine the mean TOC value and the standard deviation (SD) based on that large number of runs. Alert levels could be set at “mean ± 2 SD and then action levels set at “mean ± 3 SD”. I should clearly realize that it is statistically probable that in a small percentage of cases, I will be outside my range based on a mean ± 3 SD. This would be the ideal way to set action/alert levels, but clearly we don’t live in an ideal world. How do I set action/alert levels if I don’t have that large number of data points? So what are the alternatives?
Alternative #1: If there is a similar process where I can assume the data on my new process should be the very similar to the data for that other process, I can use the data from the other process to set action/alert levels. For example, if I have one biotech protein where I have established reliable action/alert levels for a TOC value in the final rinse, I might tentatively use that same data for a new protein.
Alternative #2: If my residue data from the design/development phase has consistently been <LOD or <LOQ (or if the data I actually collect is <LOD or <LOQ, I might want to set my alert/action levels based on a certain percentage of my acceptance limit. For example, if my acceptance limit were 25 mcg/swab, but my LOQ was 3 mcg/swab and
my LOD were only 1 mcg/swab, I would probably not want to spend a lot of time if one data point (one swab) were 5 mcg/swab. In that case I might set my alert level at 6.25 mcg/swab (25% of my acceptance limit) and my action level at 12.5 mcg/swab (50% of my acceptance limit). Clearly if one swab location was slightly higher than my LOQ, I probably shouldn’t be too concerned. On the other hand, if I sampled ten locations and five were in the neighbourhood of 8 mcg/swab, then clearly something has happened (something is different) and I might want to pay more attention to the situation.
For swab sampling, some might suggest that while I don’t have sufficient data to set alert/action levels before my protocol, if I collect 8 swab samples during each of three protocol runs, I would have a large number of data points and could establish levels based on the mean/SD criteria. The issue there is “is it really justified to say that I can treat the data statistically”. The swab locations are not the same population. Yes, they may be from the same equipment, but by definition if I have selected the worst-case locations for swabbing, I am not expecting all sampled locations to be equivalent. It is not enough to say they are all equivalent (the same population) because the data for each location was <LOD. If your data is all below LOD and if the LOD is significantly below your acceptance criterion, then you have a robust cleaning process. Good work! But setting limits based on those non-quantified results using the mean/SD criterion is a “pipe dream” (or just “busy work”). Treating such data statistically is like trying to average different swab location to show that you have passing results; it is not kosher.
Setting and using alert/action levels for routine monitoring in a cleaning process can be a valuable tool. It will not solve all your problems, but properly used they can help prevent minor problems from becoming serious problems.