The 0,5 log₁₀ rule should never need to be applied to MPN results due to the inherent method variability.

If the PT scheme or the legislation on which it is based requires MPN values to be determined in duplicate, the results can be compared and if the tube combinations are credible, then the difference between the two results should not differ, in terms of decimal logarithm units, by more than 2,58 x  x 0,24 = 0,88 for the three-by-five tube method and 2,58 x  x 0,32 = 1,17  for the three-by-three tube method.

If two distributions with two replicates per distribution are compared, then the mean of the two should not differ, in terms of decimal logarithm units, by more than 2,58 × 0,24 = 0,62 for the three-by-five tube method and 2,58 × 0,32 = 0,83 for the three-by-three tube method.

8.3.8. Long-term performance assessment

8.3.8.1. General

Performance assessment in proficiency testing schemes is generally confined to assessment of results from single rounds, but there are instances where assessment in the longer term may be beneficial. Whilst this generally applies to external quality assessment schemes, some guidance is given for the sake of completeness.

Any method of assessing long-term performance shall ensure that laboratories undertaking enumeration examinations are not identified as “poor performers” by chance due to the number of organisms in the samples they receive.

 “Low” and “high” counts should be defined according to objective rules (e.g. Poisson model-based definition for low counts, percentiles or other methods for high counts), then used to determine those laboratories reporting such results more frequently over time than could be expected by chance.

...

...

...

8.3.8.2. Low count assessments

f chance is the only factor involved, the “tail-end” counts should be distributed at random. The results may be scrutinized to determine the scatter of tail-end results, between laboratories over a series of samples (e.g. using Cochran's Q-test).

If they are not distributed at random, a second stage analysis may be performed to determine the expected distribution of tail-end counts, amongst those laboratories reporting them, if they were simply due to natural variation between samples and not to laboratory performance effect. Then, contrasting those expected numbers of laboratories with the number actually observed highlights those laboratories that may have experienced problems. An example of performance assessment for samples containing low numbers is given in Table 2.

Table 2 - Observed and expected numbers of sets of results assuming random distribution of low results (from a distribution of low levels of Clostridium perfringens in drinking water samples, where variation in numbers between test items may necessarily exceed laboratory performance variation)

No. of “lows”

Observed

Expected

0

58

...

...

...

1

32

48,56

2

<