• Author / Uploaded
• Phạm Quang Huy

Citation preview

Downloaded from journal.pda.org on November 16, 2016

COMMENTARY

Setting Alert/Action Limits for Environmental Monitoring Programs JAMES D. WILSON Pharmaceutical Systems, Incorporated, Mundelein, Illinois

The principle purposes of an environmental monitoring program are to provide an assessment of the general microbiological cleanliness of an operation and to assess the state of environmental systems control. Being cognizant of the numbers and kinds of microbes present, a microbiologist can assist in designing an effective control system(s); however, no environmental monitoring program, regardless of its design, can provide the level of confidence desired without subsystem management which is accomplished by adherence to current good manufacturing practices, facility design control, effective supervision, sound corrective action steps, and proper employee training. An environmental program is comprehensive. One important component of an environmental surveillance system is the establishment of alert and action levels. It is dangerous to consider alert/action limits independently of other components of the system. The setting of absolute numerical limits for microbiological monitoring, from either published reports or guidelines, should be approached with caution due to the diversity of operations and inherent variability of sampling/testing methodologies, notwithstanding statistical implications. There are several methods in use for establishment of alert/action levels. There is no clear consensus among pharmaceutical/device manufacturers on a preferred method. Challenges to microbiologists apparently occur when wide ranges in results are observed or when limited data are available. The challenge for the microbiologist is exacerbated by the adoption of arbitrary limits imposed by some Food and Drug Administration (FDA) investigators. If historically based alert/action levels are desired, acceptable microbial levels must be established based on sound statistical surveys of historical data that have been collected on a consistent basis using identical methods. Historically, environmental data have been treated as means ± 2 or 3 SD, cumulative frequencies, log transformations, etc., but these methods can ultimately lead to levels that may not represent the actual process capability and/or an understanding of the same. Setting alert/action levels using standard statistical tools such as means ± 2 or 3 SD, while providing a method of identifying out-of-limit trends, can be misleading because (1) these methods are based on a normal distribution and (2) these methods are based on two-tail probability. In practice, environmental data are usually not normally distributed; environmental data histograms generally resemble poisson

Received January 25, 1997. Accepted for publication April 10, 1997. Correspondence address: Pharmaceutical Systems, Inc., 102 Terrace Dr., Mundelein, IL 60060-3826. Vol. 51, No. 4 / July-August 1997

or negative exponential distributions. Regardless of the method employed, however, a special concern is how one deals with outliers and clusters of unusual results in the data. Including data taken from a period of unusually high counts, where the process was out of control, will lead to inappropriately high alert/action limits. Such clusters of obviously out of control data should be excluded before alert/action limits are calculated. For example, suppose a plot of the data versus date showed that counts from July and August were about twice as high as typical. An investigation showed that a new technician had been using poor technique and was retrained at the end of August. If this out of control period is left in, the alert/action limits will be calculated inappropriately high. The out of control period is not part of the underlying process capability and should not be used in calculating limits. Several ways to identify unusually high clusters of data include (1) plots of the data over time; (2) histograms; and (3) statistical process control charts. The data should always be plotted versus date for separate physical locations to help identify unusual data points. Once the data have been cleaned up by removing obvious out of control points, the underlying process capability can be determined. Several methods may then be used to determine appropriate alert/action limits on the cleaned up data: 1. Direct calculation of percentiles If there are enough data points (perhaps 100 or more), the action level may be set to equal the 99th percentile of the data; the alert level may be set to equal the 95th percentile. This method is probably the most desirable because it does not make any assumption about the shape of the distribution of the data. 2. Poisson Counts often follow the poisson distribution. If the data appear to be Poisson distributed, the estimated 99th and 95th percentiles may be calculated with tables of the Poisson distribution or available software. 3. Negative Exponential Distribution Although microbial counts are not continuous data, the exponential distribution may provide a reasonable approximation. The estimated 99th and 95th percentiles may be calculated by multiplying the mean by 4.6 (for the 99th percentile) or 3.0 (for the 95th percentile). 4. Other Distributions If the data appear to be approximated by the normal distribution, which does not often occur with microbial data, the upper one-sided 99th and 95th percentiles 161

Downloaded from journal.pda.org on November 16, 2016

may be approximated by the mean ± 2.326 SD (99th) and by the mean ± 1.645 SD (95th). Sometimes taking a transformation of the data, such as square root or logarithm, will normalize the data. Other distributions. such as the Weibull, may also be used. It is advisable to compare the alert/action limits obtained with a histogram of the data. If something is dramatically wrong with the method used, it will be obvious when locating where the alert/action limits are on the histogram.

162

This deficiency can happen when the method and distribution used does not match the actual distribution of the data. This approach provides a reference point for adopting official or corporate alert/action levels by using a sound scientific rationale. Acknowledgment The valuable assistance of Mark Varney is gratefully acknowledged.

PDA Journal of Pharmaceutical Science & Technology