ISPAM Stakeholder Panel Meeting Book 3-14-17

First page Table of contents Previous page 127 Next page Last page

Kernel-based gluten binary-like outcomes R. D. Fritz and Y. Chen 5

(Taylor, 1992). The kind of distribution uncovered in this survey (Fig. 1) complicates doing so however, as randomly selecting some servings for analysis may not adequately provide a representative inference on the rest. To investigate this, a sampling simulation was con- ducted where 10 000 samples of five, ten, twenty-five and fifty servings each were generated by randomly ‘se- lecting’ outcomes from the distribution of the 965 out- comes from the survey. Doing so, it was found that the probability of all servings selected being found compli- ant was 0.92, 0.84, 0.64 and 0.41 for samples of five, ten, twenty-five and fifty servings, respectively. So, with fifty servings evaluated for instance, about 40% of the time one will not get an indication of a compliance problem, getting all compliant outcomes. This probability increases when fewer servings are evaluated. So, with an underlying noncompliance rate of about one in fifty- seven servings, with ~ 1 in 161 being more than four times the regulatory maximum, sampling quantities in this range can fail to detect (with high confidence) inher- ent process and lot acceptance incapability. This is believed due to underlying statistical inferences that are being relied upon, which are undermined by the binary type distribution which kernel-based gluten contamina- tion has been shown to cause. Table 3 expands on this, showing the probability of selecting a contaminated serving in one to fifty tries for various rates of contamination present. This table is built on binomial distribution probabilities for pass/ fail type outcomes (i.e. attribute-based sampling) and further shows how noncompliance can go undetected for a time when modest sampling efforts are employed. When nonconformance rates are as high as one in ten servings, examination of five servings under this sce- nario provides less than a 50/50 chance of randomly selecting a serving that contains a gluten kernel.

Estimation of a wheat kernel’s contribution to gluten content in pure oats As mentioned, gluten-containing kernels of wheat, bar- ley and rye are the predominant source of gluten con- tamination in oats (Thompson, 2004; Hernando et al. , 2008; Thompson et al. , 2010; Koerner et al. , 2011). If not effectively mitigated, these kernel contaminants will survive the oatmeal production process intact (possibly being cut) and ultimately appear in a serving as indistinguishable flakes, consumed unknowingly by GF conscious consumers. Using wheat as an example, we have estimated the gluten contribution from a single kernel in a typical serving size of 40 g of otherwise pure oats (Table 2). This is based on ‘literature reported’ wheat protein content (2015 Crop Quality Report by US Wheat Associates, http://www.uswheat.org/cropQuality) and wheat gluten content (Shewry, 2009). We found that for the six predominant North American wheat vari- eties, a single wheat kernel will bring on average 65 – 129 mg kg 1 of gluten to 40 g of pure oats. This estimation suggests that gluten kernel contami- nants, including a cut or broken kernel, can lead to noncompliance at a serving size level, thereby posing a risk to GF oatmeal consumers. The binary-like circumstance of gluten outcomes cre- ates a sampling context similar to a pass/fail one. A serving fails when a gluten kernel or part of one exists in a serving, leading to noncompliance relative to glu- ten regulatory thresholds (e.g. > 20 ppm by FDA), and passes when one does not. ‘Attribute’-based sampling caters to binary type outcomes like this (Taylor, 1992). This type of sampling is in contrast to ‘variable sampling’, which assumes a few samples can provide information about the others around them. A key pre- requisite for variable sampling therefore is the ability to pick a sample that is ‘representative’ of the rest Sampling implications in assessment of kernel-based gluten contamination

Attribute-based sampling guidelines Sampling required to avoid this risk is also shown in Table 3. In the right most column are sample sizes

Table 2 Estimated gluten in 40 g of oats containing a kernel of North American varieties of wheat

Hard Red Winter

Hard Red Spring

Soft Red Winter

Soft White

Northern Durum

Desert Durum

Thousand kernel weight (g)*

29.1

30.4

32.6

35.3

39.2

Weight/kernel (g)

0.0291

0.0304

0.0326

0.0353

0.0392

0.048

% Protein*

12.7

14.1

10 80 65

10 80 71

13.5

13.4

% Gluten level in protein †

80 74

80 86

Gluten content in 40 g of oats containing 1 wheat kernel (ppm)

106

129

* Five year average values reported in 2015 Crop Quality Report by US Wheat Associates, http://www.uswheat.org/cropQuality. † Shewry (2009).

International Journal of Food Science and Technology 2016

Made with FlippingBook