Part of the microbiological sampling pillar of the TAAG Learning Center. If you are interested in sampling the plant environment, see environmental monitoring in food plants.
Table of contents
Why one sample is not enough
Microorganisms are not evenly distributed in a lot. They can concentrate in a few packages, in one zone of a sack or in clumps within a powder. That is why, when the decision is to accept or reject a lot of raw material or finished product, analyzing a single unit says very little. Codex Alimentarius sums it up in a warning worth keeping in mind: a well-designed sampling plan defines the probability of detecting microorganisms in a lot, but no plan can ensure the absence of a given organism.
This article deals with lot sampling. Sampling of surfaces and the plant environment is explained in the guides on surface sampling and environmental monitoring.
What a microbiological criterion is
According to the Codex principles for establishing and applying microbiological criteria (CXG 21-1997, revised in 2013), a criterion includes the microorganism of concern, the sampling plan with the number of units to take and the size of the analytical unit, the microbiological limits, the number of units that must meet them, the analytical methods and, depending on its purpose, an indication of the plan's statistical performance. A criterion without a sampling plan is just a number.
Codex distinguishes two types of plans. Variables plans use counts without grouping them and require knowing the distribution of microorganisms in the lot, so they are little used. Attributes plans classify each unit as conforming or nonconforming and are, in practice, the most widely used for accepting lots. Their only requirement is that units be taken with a probabilistic technique, such as simple or stratified random sampling, across the whole lot.
The four parameters: n, c, m and M
- n: number of sample units taken from the lot, independently and at random.
- c: maximum number of units that may give an unwanted result without the lot being rejected.
- m: microbiological limit that separates conforming units from nonconforming ones (in a two-class plan) or from marginally acceptable ones (in a three-class plan).
- M: in a three-class plan, the limit above which a unit is unacceptable.
Two-class plans
Each unit is conforming or nonconforming. It is the typical plan for pathogens, where presence or absence is tested in a defined amount and c is usually 0. For example, Regulation (EC) 2073/2005 requires for Salmonella in minced meat intended to be eaten raw n = 5 and c = 0, with absence in 25 g: a single positive in any of the five units is enough for the result to be unsatisfactory.
Three-class plans
They are used with counts. Units below m are conforming, those between m and M are marginally acceptable and those above M are unacceptable. In the same regulation, the process hygiene criterion for E. coli in minced meat is n = 5, c = 2, m = 50 CFU/g and M = 500 CFU/g. The result is satisfactory if all values are below m; acceptable if at most two of five fall between m and M and the rest are below m; and unsatisfactory if any value exceeds M or more than two fall between m and M.
How demanding a plan really is
A plan's performance is described by its operating characteristic curve (OC curve), which shows the probability of accepting a lot according to the proportion of defective units it contains. Codex asks that the plan report those statistical characteristics. A simple calculation for plans with c = 0 makes it evident:
- A lot in which 10% of the units are contaminated is accepted with a probability of 0.95, that is, about 59%, if 5 units are taken. With 30 units the probability drops to about 4%.
- If only 1% of the units are contaminated, a 5-unit plan accepts the lot about 95% of the time, and a 60-unit plan, about 55%.
The conclusion is uncomfortable but important: plans with few units detect heavily contaminated lots well and lots with low, scattered contamination very poorly. Increasing n makes the curve steeper and improves discrimination; lowering c reduces risk to the consumer but increases the probability of rejecting acceptable lots. The ICMSF's free sampling plan tool, a spreadsheet, estimates these probabilities for two- and three-class plans.
The ICMSF case scheme
The International Commission on Microbiological Specifications for Foods (ICMSF) proposed adjusting the stringency of the plan to the risk, with 15 cases that combine the severity of the hazard and what can happen to the food between sampling and consumption: conditions that reduce the hazard, leave it unchanged or may increase it. For serious hazards (cases 10 to 12) two-class plans with c = 0 use 5, 10 and 20 units; for severe hazards (cases 13 to 15), 15, 30 and 60 units.
A statistical analysis published by the ICMSF (Dahms, 2004), assuming a log-normal distribution with a standard deviation of 0.8 log, estimates that case 10 (n = 5, c = 0, absence in 25 g) rejects with 95% probability lots with a geometric mean of 1 CFU per 32 g, while case 15 (n = 60) reaches lots of 1 CFU per 526 g. The difference shows why foods for sensitive populations receive the strictest plans: Regulation 2073/2005 itself requires n = 30 for Salmonella in dried infant formula.
Sample unit, analytical unit and composite samples
The sample unit is what is taken from the lot; the analytical unit is the portion that is analyzed (for example, 25 g). The FDA's Bacteriological Analytical Manual (BAM) recommends obtaining at least 100 g per sample unit and classifies foods into three categories for Salmonella, with 60, 30 or 15 analytical units depending on consumer sensitivity and the existence of a subsequent lethal step.
To reduce the workload, the BAM allows combining fifteen 25 g units into a 375 g composite sample, unless the method indicates otherwise. Codex limits this practice: compositing of samples may be considered in presence or absence testing within a two-class plan, as long as it does not affect the result compared with analysis of individual units, but it is not appropriate for counts or for three-class plans, because it dilutes the concentration of each unit.
Practical rules that are often forgotten
- Do not repeat the analysis to "confirm" a lot: Codex states that, unless the plan provides for it, a lot should not be retested. Sampling until a negative is obtained defeats the statistical logic of the plan.
- The plan is a minimum when assessing a lot: Regulation 2073/2005 establishes that, if the aim is to assess the acceptability of a lot, its plans must be respected as a minimum. It only allows reducing the number of units when the operator demonstrates, with documented history, effective procedures based on HACCP.
- Moving windows to verify processes: when the aim is to verify a process and not to accept a lot, Codex describes the moving window approach, which always compares the last n results against the limits.
- Sampling beats method: a method with an excellent limit of detection does not compensate for a plan with few units in a lot with scattered contamination, as occurs in low-moisture foods.
Once taken, the units must reach the laboratory unchanged. Shipping conditions are detailed in the article on transport and storage of samples.
Conclusion
Attribute sampling plans turn a microbiological criterion into a decision with a known probability. Reading n, c, m and M correctly, and understanding the operating curve behind them, tells you which lots the plan detects and which it lets through. Adjusting stringency to risk, as the ICMSF proposes, respecting minimum plans and not repeating analyses to save a lot are the basis for sampling to do its job.
About TAAG
Discover how TAAG's molecular detection kits shorten turnaround time when a sampling plan requires analyzing many units per lot.
Frequently asked questions
What do n, c, m and M mean in a microbiological criterion?
n is the number of units taken from the lot; c, the maximum number of units with an unwanted result that is tolerated; m, the limit that separates conforming units; and M, in three-class plans, the limit above which a unit is unacceptable.
When is a two-class plan used and when a three-class plan?
Two-class plans are typically used for pathogens, with presence or absence testing and c = 0. Three-class plans are used with counts, for example of hygiene indicators, where a limited number of units between m and M is tolerated.
Does a negative n = 5, c = 0 plan guarantee the lot is free of the pathogen?
No. If 10% of the units in the lot were contaminated, that plan would accept it about 59% of the time. No sampling plan can ensure the absence of a microorganism; it only defines the probability of detecting it.
Can several units be mixed into a composite sample?
In presence or absence testing within a two-class plan, yes, as long as it does not affect the result compared with individual analysis. Codex states that it is not appropriate for counts or for three-class plans.