Comparison of Different Calculation Approaches for Defining Microbiological Control Levels Based on Historical Data

Area of Science:

• Microbiology
• Pharmaceutical Manufacturing
• Statistical Modeling

Background:

• Microorganism counts are monitored in pharmaceutical manufacturing to ensure product quality.
• Setting microbiological control levels is crucial for detecting and investigating high microbial counts.
• Accurate control levels should be based on historical data for reliable future predictions.

Purpose of the Study:

• To compare different statistical approaches for calculating microbiological control levels.
• To evaluate the predictive power of various methods using real microbiological data and simulations.
• To identify the most suitable method for setting reliable microbiological control levels.

Main Methods:

• Simulated experiments using real microbiological data.
• Comparison of normal, Poisson, non-parametric Excel percentile, negative binomial, zero-inflated negative binomial, and gamma distributions.
• Evaluation of predictive power for future values and robustness towards outliers.

Main Results:

• Normal and Poisson distributions were inappropriate for setting microbiological control levels.
• The non-parametric Excel percentile showed good predictive power but was sensitive to outliers.
• Parametric models, particularly the gamma distribution, demonstrated strong predictive power for high percentiles (≥98%).

Conclusions:

• The gamma distribution is a viable and potentially superior option for calculating microbiological control levels, especially at high percentiles.
• It offers better fit to the upper end of data distributions compared to other models.
• The gamma distribution is easily calculable in standard software, making it a practical alternative.