# View Report Details

### Development of representative sampling plans for mycotoxins in food using distribution modelling

Project Code: C03055

11/11/2008

Cranfield University**Parsons, D**

**Executive Summary**

For consumer health to be effectively protected it is important that consumer exposure

to contaminants such as mycotoxins is minimised. The ability to effectively obtain a

representative sample for analysis, from a raw material or processed product, is

critical as part of a prevention strategy. It is widely acknowledged that designing

sampling plans for mycotoxins is particularly problematical because of the

heterogeneous distribution of these contaminants in bulk lots of different

commodities. Normal practice is to take several small quantities of the commodity,

known as incremental samples, from different locations. These are mixed together to

form the aggregate sample, from which a portion is extracted for analysis. Throughout

this report the process of collecting several incremental samples is referred to as

sampling, the set of discrete incremental samples is usually called the sample, and the

number of incremental samples is the sample size.

The aim of this project was to produce detailed and robust information on appropriate

sampling strategies for surveillance of mycotoxins in raw food commodities using a

statistical and modelling approach. The emphasis was on deoxynivalenol (DON) in

large lots of grain (in storage or bulk transport). Ochratoxin A (OTA) in the same

situations was also considered. The objectives were to:

1. Review the scientific literature and primary data associated with previous

sampling studies, directives and protocols.

2. Collate the available data sets for DON and OTA from those identified in the

literature review.

3. Analyse the spatial pattern of DON and OTA concentration for the data sets

collated, to assess if geostatistical analysis could be performed and to

determine if there is spatial structure in those data sets where geostatistical

analysis can be developed.

4. Develop a model that generates data sets of DON and OTA concentration with

desired spatial patterns and to compare different sampling strategies and

sample sizes.

5. Investigate the effect that sampling strategy and sample size have on deriving

the statistical properties of DON and OTA concentration in storage.

Geostatistics is a branch of statistics, developed in the earth sciences, which is

concerned with properties that vary in space in a way that is neither completely

random nor completely predictable. It assumes that measurements of a property, such

as the concentration of a chemical in soil, made close together are likely to be more

closely related (“spatially correlated”) than measurements made further apart. If so,

the property is said to have a spatial structure. It was proposed that geostatistics might

be appropriate for the analysis of data for mycotoxins, because the underlying fungal

infections could have a spatial structure.

Little information was found in the scientific literature regarding the origin or

scientific basis of general sampling plans, or on methodologies which could be used

to guide sampling of mycotoxins. Seven datasets containing information on DON or

OTA concentrations, or distribution of Fusarium were considered. Their spatial

patterns were analysed using geostatistical methods. Of the seven data sources, only

two sources of data for DON and one for Fusarium satisfied the requirements for

geostatistical analysis. With respect to stored grain contaminated with DON, there

was evidence for a spatial pattern with correlations between points up to 4 m apart.

Therefore, sampling strategies with points more widely spaced than this might not

adequately detect the DON concentration. The particular load of grain described was

originally chosen for study because a high level of contamination had been found, so

the results need to be treated with caution when attempting to generalise to other

situations. Results for the analysis of the spatial pattern of DON concentration and

fusarium in the field indicated that the spatial correlation was lost at distances around

50 m. The dataset for DON in stored grain also contained measurements of OTA; no

evidence of spatial structure was found for these.

The scarcity of data sources, supported by consideration of potential approaches, led

to the conclusion that a statistical simulation approach to distributions of mycotoxins

in stored grain was required. The aim was to simulate a wide range of plausible

distributions of DON and other mycotoxins in stored grain from a parsimonious set of

parameters characterising the distributions. Two versions of a model of mycotoxin

distributions and sampling based on generating data sets using geostatistics for DON

and classical statistics for OTA were developed. The interactive model allowed one

sampling at a time to be simulated and examined in detail. The batch model generated

data sets and samplings repeatedly to allow the accuracy (bias) and precision

(standard deviation) to be estimated. The model focused on the effect that the

sampling strategy and size have on determining the statistical properties of mycotoxin

concentration in a bulk container of a commodity such as wheat. Evidence of spatial

structure in the data indicated that both sampling strategy (location of sample points)

and sample size have an effect on the statistical properties of mycotoxin

concentration. The main conclusions are:

• For most sample sizes, a regular grid proved to be more consistent and accurate in

the estimation of the mean concentration of DON, which suggests that regular

sampling strategies should be preferred to random sampling, where possible.

• For both strategies, the accuracy of the estimation of the mean concentration

increased significantly up to sample sizes between 40 and 60 (depending on the

simulation). The effect of sample size was small when it exceeded 60 points,

which suggests that the maximum sample size required is of this order.

• These sample sizes are consistent with current recommendations for bulk cereals

(60 incremental samples for 10–20 t and 100 for 20–50 t).

• Similar conclusions apply to OTA, although the difference between regular and

random sampling was small and probably negligible for most sample sizes.

This study was the first application of geostatistical analysis to date on mycotoxin

contamination of bulk agricultural commodities. Several recommendations are made

for further research.

Some of the files on this site may be in a format that your computer can't read. However, you can download Readers and Viewers for the following document types below:

- PDF - Download Acrobat Reader
- DOC - Download MS Word Viewer
- XLS - Download MS Excel Viewer
- PPT - Download MS Powerpoint Viewer