Part of the fundamentals and standardization pillar of the TAAG Learning Center. It goes deeper into what we saw on how CFU are interpreted and on the microbiological enrichment step.
Table of contents
A microbiological result always comes with a margin
A report says "Salmonella: not detected in 25 g" or "aerobic mesophiles: 1.2 × 105 CFU/g". Both results look definitive, but neither entirely is. The first depends on the probability that the method detects the few cells that might be present in the portion analyzed. The second is an estimate with a dispersion that, in microbiology, is much greater than in a chemical analysis. Understanding those margins is what allows good decisions when a result falls close to a limit.
Why microbiology is different
Bacteria are not distributed in a food like a dissolved substance. They are discrete particles, often clustered, and when they are at low concentration their presence in a test portion is a matter of chance. If a lot has on average one Salmonella cell per 25 g, some 25 g portions will have none, others will have one and others two or more. That random distribution follows a Poisson distribution, and sets a theoretical limit that no method can exceed.
The theoretical ceiling of any method
With a Poisson distribution, the probability that a portion contains at least one cell is 1 − e−λ, where λ is the mean number of cells per portion. With an average of 1 CFU per portion, that probability is around 63%: even a perfect method would fail in more than a third of cases, simply because the cell is not in the portion. To reach 50% detection an average of ln 2, about 0.69 CFU per portion, is enough; to reach 95%, about 3 CFU per portion are needed.
Qualitative methods: from LOD to LOD50
In a detection method (presence or absence), the limit of detection is not expressed as a fixed number, but as a probability of detection (POD) curve as a function of contamination. Wilrich and Wilrich described in 2009, in the Journal of AOAC International, a statistical model to estimate that curve and the LODp, the contamination that the method detects with a probability p. The most widely used value is the LOD50: the concentration at which the method gives a positive in half of the portions analyzed.
The LOD50 makes it possible to compare methods with each other and with the theoretical ceiling. ISO 16140-2 uses it to calculate the relative level of detection (RLOD) between an alternative method and the reference one, and ISO 16140-3 asks each laboratory to estimate its own eLOD50 when verifying a method and check that it does not exceed four times that of the validation. The procedure is explained in the article on validation and verification of methods.
What "absence in 25 g" means
A negative result means the method did not detect the microorganism in that portion, not that the lot is free of it. That is why absence criteria are applied over several sample units, and why the enrichment step matters so much: it is what allows a single cell, often damaged, to multiply to detectable levels. How many units to analyze and what probability of detecting a contaminated lot each plan offers is covered in the article on attribute sampling plans.
Quantitative methods: the limit of quantification
In a plate count, the lower limit is set by the lowest dilution plated. If 1 mL of the 10−1 dilution is plated and no colony grows, the result is not "0 CFU/g", but "less than 10 CFU/g" (in the ISO 19036 guide's example, with two dilutions without colonies, the calculated limit is 9.1 CFU/g). Reporting "0" is one of the interpretation errors we already discussed in the article on CFU.
Measurement uncertainty in counts: ISO 19036:2019
ISO/IEC 17025 requires accredited laboratories to evaluate the measurement uncertainty of their tests. For food microbiology counts, the reference is ISO 19036:2019, published in October 2019 as a revision of the 2006 technical specification. The ISO working group's presentation recalls that quantitative results in microbiology are highly variable, on the order of 0.1 to 1 log10.
The three components
- Technical uncertainty: the variability of the analysis steps (weighing, dilutions, plating, incubation, counting, confirmation, media lots, analysts). It is preferably estimated as the intralaboratory reproducibility standard deviation (sIR), analyzing at least 10 samples with two results each under conditions that vary as much as in routine work. It is usually the largest component.
- Matrix uncertainty: the extent to which the test portion does not represent the laboratory sample. For liquids or well-homogenized samples a fixed value of 0.1 log10 can be used; in solid or multi-component foods it can be much greater.
- Distributional uncertainty: the random Poisson variation. It is calculated from the total number of colonies counted (0.4343/√ΣC in log10) and weighs most when few colonies are counted.
Expanded uncertainty and how it is reported
The components are combined as the square root of the sum of their squares, and the result is multiplied by a coverage factor k = 2, which corresponds approximately to a 95% confidence level. The standard also allows a simplified option, using only reproducibility, warning that it may underestimate uncertainty in solid foods or at low levels.
The ISO guide's example shows it well: a count of 1.00 × 105 CFU/g, with an expanded uncertainty of 0.31 log10, is reported as 5.00 ± 0.31 log10 CFU/g. In natural values, the interval runs from 4.90 × 104 to 2.04 × 105 CFU/g: more than double from one end to the other.
What measurement uncertainty does not include
ISO 19036 leaves out two sources of error. The first is sampling uncertainty, the error of drawing the laboratory sample from the lot, which can be large but is not attributed to the laboratory. The second is bias: in food microbiology there is no "true" value independent of the method against which to compare a routine result.
From number to decision
Imagine a count of 1.2 × 102 CFU/g against a limit of 100 CFU/g. With a typical expanded uncertainty of several tenths of a log, the true value could be below the limit. Does the product comply or not? The answer depends on the decision rule. ISO/IEC 17025 asks that, when the laboratory declares conformity, it document the rule it uses and the risk of accepting or rejecting by mistake. A common criterion considers a result nonconforming only if it exceeds the limit by more than the expanded uncertainty; another, more conservative one, considers conforming only what falls below the limit minus that uncertainty. Choosing between them is a risk management decision, not a laboratory detail.
Common mistakes
- Reading "not detected" as a guarantee of absence in the whole lot.
- Comparing methods only by their turnaround time, without checking their LOD50 in the matrix of interest.
- Reporting "0 CFU/g" instead of "< limit of quantification".
- Treating differences of 0.2 or 0.3 log between two counts as a real process change.
- Requesting a statement of conformity without first agreeing on the decision rule.
Conclusion
In microbiology, a result never comes alone: it carries a probability of detection if qualitative and an uncertainty if quantitative. The LOD50 makes it possible to compare methods and understand what a negative really means; ISO 19036 makes it possible to put an interval around each count. Using those margins, together with an agreed decision rule, is what avoids releasing or rejecting a lot over a difference that the method itself cannot distinguish.
About TAAG
Discover the performance data of TAAG's molecular detection kits, based on AiGOR™ RT-qPCR technology, and how they are documented in each validation.
Frequently asked questions
What is the LOD50 of a microbiological method?
It is the concentration of the microorganism at which a detection method gives a positive in 50% of the portions analyzed. It is used because detection in microbiology is probabilistic: even a perfect method reaches only 50% detection with about 0.69 CFU on average per portion, because of the random distribution of cells.
Why does a "not detected" result not mean zero?
Because the method analyzes only a portion of the sample and has a limit of detection. If the microorganism is at low concentration or unevenly distributed, it may not be in the portion analyzed or may not reach detectable levels.
Which standard is used to estimate the uncertainty of a microbiological count?
ISO 19036:2019. It combines technical, matrix and distributional uncertainty, and expresses the result as expanded uncertainty with a coverage factor k = 2, approximately 95% confidence.
Does measurement uncertainty include the sampling error of the lot?
No. ISO 19036 expressly excludes sampling uncertainty, that is, the error of drawing the sample from the lot. That risk is managed through the design of the sampling plan.