119 / 166
Lab Y. Among these 636 servings, ten samples had a gluten content
of
P
20 ppm, not in compliance with the FDA’s gluten-free regula-
tion. Another 20 samples had gluten content between 5 and
20 ppm. Those 20 samples appeared on a first test to be in compli-
ance with the FDA’s gluten-free regulation. However, when ten
more aliquots (0.25 g) from each of those 20 ‘positive yet compli-
ant’ samples were analyzed, gluten contents of >20 ppm were
readily detected
( Table 1). Interestingly, four of the 20 samples
yielded both a BLQ (below the limit of quantification) result as well
as an ALQ (above the limit of quantification). This indicates the
possibility for a BLQ outcome to be obtained when ALQ level
sub-samples (of 0.25 g) also exist in the same ground serving
sample. Additionally, when looking at the average of the 11 gluten
values for each of those 20 samples, at least nine of them (and
potentially as many as 11 had we tested beyond a 160 ppm maxi-
mum) averaged >20 ppm, therefore ‘averaging’ non-compliant
relative to the FDA’s gluten-free regulation
( Table 1).
Gluten analysis of our in-market survey samples illustrates a
single gluten test of an oatmeal serving may be inadequate to accu-
rately reveal the gluten risk inherent in it (at least under current
best grinding practice). Kernel-based gluten contamination is the
prime suspect, since it increases the difficulty of a homogenous
grind due to gluten initially being centralized within a single con-
taminant kernel. This pill like form then needs to be adequately
ground and uniformly distributed for homogeneity to occur.
3.2. Gluten distribution in ground wheat-spiked oat groats
To test whether kernel-based contamination can lead to chronic
sample prep non-homogeneity, we prepared twelve pure oat groat
(50 g) samples each spiked with a wheat kernel (of approximately
0.027 g). Six of them were sent to each of two recognized laborato-
ries, Labs X and Y. The samples were ground, and the gluten
content of each sample was analyzed in triplicate (0.25 g per anal-
ysis). The remainders of the 12 ground samples were sent back to
the authors, and were then completely aliquot into 0.25 g portions
and 100% of them analyzed by the PepsiCo analytical team. Overall,
nearly 2300 analyses were conducted, varying from 184 to 196
results for each of the 12 spiked samples (
Supplemental material,
Table 1
). Based on 100% evaluation of these 12 samples, plots of
the individual distributions were created
( Fig. 1 ).
Fig. 1shows the distributions of 0.25-g test results are ‘skewed
right’, tending to follow log-normal distributions as determined via
chi-square goodness of fit tests. The log-normal approximations
are presented in
Fig. 1as dotted lines. This skewness suggests a
non-homogenous distribution of gluten particles despite the
spiked oat groat samples being ground under current ‘best’ prac-
tices by two well-established commercial analytical labs. The post
grinding particles produced by the contaminant kernels were
apparently not well dispersed throughout the ground samples
but rather tended to remain more highly concentrated in a small
subset of the possible 0.25 g test portions. We found the maximum
0.25-g test result obtained to vary from 12 to 57 times more than
the minimum for the sample. This difference in sample-to-sample
variation could be due to differences in gluten kernel hardness,
gluten content in the kernel itself, oat kernel hardness, ‘grind to
grind’ variability, and other influencers. Needless to say however,
with highly skewed distributions like this, the determination of
gluten content in oat groats via a single 0.25-g sample test
becomes error prone. This is because a few of the 0.25-g samples
possess large amounts of gluten while others have received just a
fraction of it. This leads to the potential for misdiagnosis (i.e., con-
cluding either a sample average is <20 ppm when it is not, or that
all possible test results are <20 ppm when they are not, depending
on one’s interpretation of the 20 ppm regulatory threshold).
This data set of spiked samples is relatively small (12 evalu-
ated), but it does allow for a rough estimation of probabilities of
obtaining a single reading possessing a value of <20 ppm given var-
ious true average gluten contents (in a 50-g sample). This means
we can roughly assess the potential for misdiagnosis, using the log-
normal distributional fits of test results
( Fig. 2 ). Doing so, based on
these outcomes, if one desires
P
95% confidence that a single com-
pliant reading (i.e., <20 ppm) does not come from a sample whose
average is actually >20 ppm, the true gluten sample average would
need to be >60 ppm, as
Fig. 2shows that a sample with an average
gluten content of 60 ppm has a probability of 0.05 (i.e., 5%) to yield
a <20 ppm gluten test result. In other words, for samples where the
true gluten average is high enough, i.e., >60 ppm, there is little risk
Table 1
Re-test of samples with >5 ppm and <20 ppm gluten content reveal non-homogeneous distributions of gluten post grinding.
Retests of in-market oatmeal finished goods
(For samples found positive for gluten on 1st ’0.25 g Test’ but compliant, i.e., >5 and <20 ppm)
Original 1st
0.25 g Test
Result
1st
Retest
Result
2nd
Retest
Result
3rd
Retest
Result
4th
Retest
Result
5th
Retest
Result
6th
Retest
Result
7th
Retest
Result
8th
Retest
Result
9th
Retest
Result
10th
Retest
Result
Resultant
Avg.
(n = 11)
Range of
Outcomes
1 6.5
BLQ BLQ
BLQ BLQ BLQ BLQ BLQ BLQ BLQ BLQ
2
7
2 6.6
14
10
13
7
8
8
39
BLQ 15
36
15
36
3 6.8
BLQ 6.5
BLQ BLQ BLQ BLQ BLQ BLQ BLQ 7.5
3
8
4 7.5
13
14
13
14
48
20
13
13
42
>160
>33
>153
5 7.7
34
16
14
6
20
8
BLQ 9
6.5
10
12
34
6 8.5
BLQ BLQ
BLQ 7.5
BLQ BLQ BLQ >160
13
BLQ
>18
>160
7 9.0
8.5
6.5
9.5
6
7
9
14
9
13
19
10
13
8 9.7
63
79
29
30
>160
31
30
70
65
>80
>59
>150
9 10.0
BLQ BLQ
BLQ BLQ BLQ BLQ BLQ BLQ BLQ BLQ
3
10
10 10.0
BLQ BLQ
BLQ BLQ BLQ BLQ BLQ BLQ BLQ BLQ
3
10
11 10.0
17
8
9.5
13
9.5
13
9
7
12
20
12
13
12 10.5
14
21
18
61
19
13
8.5
16
38
9.5
21
53
13 11.0
17
28
9.5
14
9.5
78
9.5
>80
24
12
>27
>71
14 12.5
15
8
BLQ 10
9
9
11
9
14
>160
>24
>160
15 13.0
BLQ BLQ
BLQ 33
13
6
107
146
BLQ 87
38
146
16 13.4
31
25
18
26
14
32
75
27
20
52
30
62
17 15.0
24
17
18
21
15
13
9
8
17
10
15
16
18 15.5
36
40
18
16
14
7.5
76
42
14
6.5
26
70
19 16.0
BLQ BLQ
6.5
BLQ BLQ >160
BLQ BLQ BLQ BLQ
>19
>160
20 18.0
54
26
24
118
>160
24
28
28
95
39
>56
>142
%Averaging
P
20 ppm 45–55%
172
R.D. Fritz et al. / Food Chemistry 216 (2017) 170–175