Limits of alert and action for the environmental microbiological control – practical solution

INTRODUCTION

The environmental microbiological control is indispensable in the pharmaceutical industry to assure the quality of the pharmaceutical products.

The environmental control not only is required for the sterile product elaboration, but also for nonsterile products, where the microbiological contamination can seriously affect the quality of pharmaceutical products. Some microorganisms can be pathogenic, whereas in certain cases the contamination can produce a diminution of the title of the API (Active Pharmaceutical Ingredients) or a deterioration in the physical characteristics of the product (1).

In order to carry out the environmental control, appropriate alert and action limits must settle down. These limits must be determined having in consideration that the primary target is to detect conditions that move away of the habitual behavior of the productive atmosphere (2).

A concept mistaken in the pharmaceutical industry has been to apply bacteriological norms established and to use them to elaborate specifications of product liberation (3). The norms usually include recommendable limits for the microbiological control of the areas in which are elaborated sterile products, but they clarify that these values do not represent specifications.

Therefore, to determine the values of alert and action from the historical behavior is useful as much for areas where nonsterile products are elaborated, where the norms do not establish reference limits, how for those used in the sterile product elaboration, in which the established limits can be smaller to the recommended ones by the effective dispositions, allowing a greater microbiological control of the atmosphere.

Although at the moment there is a general consensus on the necessity to establish the limits from the historical values, a general criterion does not exist to carry out this task.

The intention of this article is to present a suitable tool for the determination of the values of alert and action from the historical behavior, using statistical concepts.

USE OF THE POISSON DISTRIBUTION FOR THE DETERMINATION OF THE LIMITS

In one first instance advanced in the search of a method to determine the action and alert limits, supposing that the values obtained in the microbiological control came near to a Poisson distribution.

The election of this distribution was made because one of the common applications of the same one is the prediction of the number of events by unit of time or unit of area.

In order to evaluate if the data collected in the microbiological control behave following a distribution determined, we must raise a test of adjustment kindness (4) that allows us to measure the discrepancy between the observed results and the waited for results, and a hypothesis test that allows us to reach a conclusion.

We took the following point from sampling:

· Point of Sampling: Ag – Packaged of nonsterile Liquids in bottles (Figure 1)

· Method: Air sampling by Sedimentation

· Time of sampling: 1 hour

· Count of: Bacteria

· Collected data:

We used a test of hypothesis with the following null Hypothesis (Ho):

Ho: The data adjust to a Poisson distribution

For each point of sampling, we classify the data observed in k ranks or classes; the number of observations (o) in each rank is entered and the amount of observations calculates that would be hoped to obtain (e) in each rank with the selected distribution.

Next, we applied the Ji-square distribution , in order to measure the discrepancy between observed and the awaited value:

Ji-square test:

• Ek = awaited value

• Ok = observed value

• n = number of classes

The calculated value is compared with the reduced which is tab, for a level of significance (a) =0,05 and 3 degrees of freedom (n):

reduced 7,81

The calculated is greater to the reduced then the null Hypothesis is not admitted, therefore the data do not follow a Poisson distribution.

This discrepancy between the observed data and the theoreticians (taking into account a Poisson distribution) can be observed with greater clarity in Figure 2.

Similar results were founded in the points of sampling, concluding that, in most of the cases, the use of the Poisson distribution does not turn out suitable to describe the behavior of the historical data.