Issue
How should environmental monitoring data be trended to assure microbiological control of aseptic processing areas?
Introduction
This guidance establishes the need for trending of environmental monitoring data and gives recommendations on aspects of trending such as categorization of data, frequency of trending, trend definition, and content of trend reports.
Recommendations & Rationale for Recommendations
It is a regulatory expectation 1,2that producers of aseptically produced drug product have an understanding of how the environment effects their products. From a microbiological perspective, this means review of environmental monitoring (EM) results from each batch supplemented with periodic review of EM data over extended periods of time.
The goal of this second review is to take a broad look at the data and identify notable changes that may be masked during batch review of the data. Detecting these changes will allow corrective actions to be initiated, solving the problem before it gets out of control.
Why trend EM data?
Continued microbiological control over aseptic processing areas requires information over the short and long term. Long-term information must be of sufficient duration (i.e. monthly, quarterly, yearly) to proactively evaluate changes in the frequency of contamination as well as changes to the microbial populations. Notable changes in either parameter may trigger an investigation into the probable cause and potential corrective action.
Further, the availability of data over time can assist with investigations into batch related microbial excursions (e.g. exceeding action or alert limits).
In addition to the scientific arguments for an EM trending program, there are regulatory reasons as well. The FDA’s draft guidance on aseptic processing establishes clear expectations for EM data to be reviewed as a batch release criterion, as well as over time to assess continued microbiological health of the manufacturing environment.
How should the data be organized?
EM data in isolation provides a snapshot of microbial control at a discrete time point, but does not provide information on adverse trends that may be developing. It is only when it is put in a framework of whom, where, and appropriate timeframe does is become useful for trending purposes. Therefore, it is vitally important to determine the type of trending information needed prior to monitoring. For example, if it is important to summarize the ongoing microbiological status of a Plant’s Grade A areas, data segregated solely by time and room number will not be sufficient. The key is to design data classifications prior to needing the trend reports so that the reports can be assembled easily and provide the required information in as short a time as possible.
Suggested classifications of EM data include:
- Location (including air classification)
- Date
- Shift
- Lot
- Room
- Operator
- Process
- Isolate Recovered
How often should trend data be reviewed?
Trend data should be reviewed at sufficiently brief intervals to allow timely response to contamination problems before serious threats are realized. However, trending should also be of sufficient length to analyze seasonal effects. These opposing goals necessitate multiple frequencies of data review and report generation. In addition to the batch release EM results, reports on a monthly or even weekly basis are recommended for rapid response to potential problems.
Quarterly or yearly reports should be sufficient to recognize seasonal variations in microbial control. It is also important to remember that the criticality of the area will have an impact on how frequently trend information is compiled. For example, it may be sensible to examine Grade A and B data for both short and long term trends while analysing Grade C and D data only for long term trends.
How should the data be analysed and what constitutes a trend that requires investigation?
The goal of trending is to identify gradually deteriorating environmental conditions that may eventually lead to product contamination. A thorough knowledge of the environment and the processes that take place within that environment can accomplish this goal.
When EM data is analysed with appropriate statistical and qualitative techniques, knowledge of the environment becomes more significant and decisions based on this knowledge easier to defend. The easiest form of trending is to consider a certain number of alert level excursions equivalent to an action level excursion. This type of trending is good for quickly spotting an adverse situation over a relatively short timeframe (e.g. weekly trending). Commonly, three alert level excursions in ten consecutive samples are sufficient to initiate an investigation. However, the nature of the operations conducted at the particular monitoring point should be considered when deciding on a number.
This approach can be applied to a single sampling point, or more conservatively, to an entire air classification within a production suite. While alert level excursions will show adverse trends in the short term, more advanced statistical treatment is required for longer-term data analysis. When deciding on a statistical tool for use with EM trending, it is important to keep two things in mind. First, make sure the statistical tools are appropriate for their intended use.
For example, it is inappropriate to use statistical approaches that rely on normally distributed data to analyse for trends in Grades A and B count data, where the majority of counts are zeros. Second, after a statistical tool is chosen, establish trend targets (i.e. control levels) by analysing at least a year of historical data. This will ensure a more accurate target that is free of short-term bias.
Percentage Nonconforming.
It is well known that EM count data can be widely variable and statistical treatments that rely on counts are unreliable at best. Therefore, one method for identifying long-term trends is to examine the proportion (i.e. percentage) of nonconforming EM results. This method merely puts a statistical foundation under what many plants are already doing to identify trends and is analogous to using p charts for controlling processes that yield attribute data3.
One benefit of using this method is that the definition of nonconforming results is flexible. They can be alert level excursions, identification of certain genera or groups (e.g. Bacillus, Gram negatives), or both. In other words, both qualitative and quantitative trends can be analysed with this tool.
Another significant benefit is that determining control levels and generating p-charts are relatively straightforward and require no special statistical analysis software.
Assuming the process is constant, the underlying distribution of percent nonconformities will be the binomial distribution. Using this technique, historical data is analyzed for the percentage of EM samples that produced nonconformities, or p.
Then, an upper control level can be established based on the chosen standard deviation (e.g. +2σ or +3_). Control levels are set with the following formula:
ppp )1( − /+σ)
n
where,
p = proportion nonconforming (from historical data)
σ = standard deviation
n= sample size
Since sample sizes are rarely identical from one time period to the next, the upper control level can be established once using an average sample size or with each trend analysis using the sample size for the time period in question. Once the upper control level is established, it can be compared to the percentage of EM data points that do not conform to the chosen attribute.
Further, it is possible to construct a control chart consisting of the predetermined percent non-conforming, upper control levels, and each data point representing the percentage non-conforming from a particular time point. This will provide a visual representation of how each time interval compares with each other in terms of percentage of EM data non-conforming. It will also illustrate how percentage nonconformities perform relative to the established control levels.
Alert Level Time Interval.
Another method of long-term EM data trending is to determine the expected time interval between alert level excursions. Time interval distributions are most often positively skewed and for this reason are particularly well suited to the log-normal or Weibull distributions. As above, fitting EM data to these distributions will allow setting of the expected time interval between alert level excursions. Two alert level excursions within this interval would then be an indicator of a potential adverse trend.
While this method can provide valuable insight into EM trends, it may not be as practical as the percentage non-conforming approach described above.
Skewed distributions are relatively complicated and may require dedicated statistical software to analyse. Further, tracking of time intervals between alert level excursions is not usually performed, so reprocessing of the raw data to generate this new parameter will most likely be necessary.
It is recommended that environmental monitoring programs include procedures for trending data in the aseptic manufacturing practice.
It also recommended to
- Trending as tabulation or graphical plot of actual data used to determine ongoing performance.
- Assigning trend review responsibility
- Suggesting tabulation tools
- Defining the purpose for trending
- Delineating responses to adverse trends