Matrix-Specific Method Validation of an Automated Most-Probable-Number System for Use in Measuring Bacteriological Quality of Grade "A" Milk Products.

The U.S. Food and Drug Administration (FDA) oversees a long-standing cooperative federal and state milk sanitation program that uses the grade "A" Pasteurized Milk Ordinance standards to maintain the safety of grade "A" milk sold in the United States. The Pasteurized Milk Ordinance requires that grade "A" milk samples be tested using validated total aerobic bacterial and coliform count methods. The objective of this project was to conduct an interlaboratory method validation study to compare performance of a film plate method with an automated most-probable-number method for total aerobic bacterial and coliform counts, using statistical approaches from international data standards. The matrix-specific validation study was administered concurrently with the FDA's annual milk proficiency test to compare method performance in five milk types. Eighteen analysts from nine laboratories analyzed test portions from 12 samples in triplicate. Statistics, including mean bias and matrix standard deviation, were calculated. Sample-specific bias of the alternative method for total aerobic count suggests that there are no large deviations within the population of samples considered. Based on analysis of 648 data points, mean bias of the alternative method across milk samples for total aerobic count was 0.013 log CFU/ml and the confidence interval for mean deviation was -0.066 to 0.009 log CFU/ml. These results indicate that the mean difference between the selected methods is small and not statistically significant. Matrix standard deviation was 0.077 log CFU/ml, showing that there is a low risk for large sample-specific bias based on milk matrix. Mean bias of the alternative method was -0.160 log CFU/ml for coliform count data. The 95% confidence interval was -0.210 to -0.100 log CFU/ml, indicating that mean deviation is significantly different from zero. The standard deviation of the sample-specific bias for coliform data was 0.033 log CFU/ml, indicating no significant effect of milk type.