Jump to content

  • Quick Navigation
Photo

Determining the representative testing sample size for a batch

Share this

  • You cannot start a new topic
  • Please log in to reply
12 replies to this topic

Inverse

    Grade - MIFSQN

  • IFSQN Member
  • 53 posts
  • 4 thanks
2
Neutral

  • United Kingdom
    United Kingdom

Posted 29 September 2016 - 12:18 PM

Hi all,

 

I am not sure if there was such topic but I couldn't find it and really hope that someone will be able to help me out!

 

I am working with dried flaked herbs and I am in the middle of building our testing regime. Every batch (approx 200kg) is tested for microbes and allergens (our customer's requirement) but only one sample (100g) is sent out. Do you think I need to send more samples in order to assure product is free from allergens and contaminants or one sample (picked from different barrels of one batch) is good enough? I understand that 100g doesn't represent 200kg, so I was wondering if anyone works with similar products and could help me out? 

 

Thank you!

 



Gerard H.

    Grade - SIFSQN

  • IFSQN Senior
  • 411 posts
  • 131 thanks
44
Excellent

  • France
    France
  • Gender:Male

Posted 29 September 2016 - 12:59 PM

Dear Mr., Mrs.,

 

That is an interesting topic. You can consider the following (there probably will be more people reacting to your question):

 

  • Take a sample at the begin, at the middle and at the end of the batch. (3 samples). For some test plans, 5 samples are required, but if that's not the case, I would start with 3.
  • 100 g seems a lot for the analyses you have described (but of course it depends on what you are testing). It's just a remark, because it will be pity when 75 g of each sample is thrown away.

I hope to have informed you sufficiently.

 

Kind regards,

 

Gerard Heerkens



Inverse

    Grade - MIFSQN

  • IFSQN Member
  • 53 posts
  • 4 thanks
2
Neutral

  • United Kingdom
    United Kingdom

Posted 29 September 2016 - 01:26 PM

Gerard,

 

Thank you for the quick response! Our batches are split into barrels (each hold at least 35kg). Do you think it's representative enough to test every other barrel or should we test each? For micro we send only 20g of the sample (one per barrel), whereas allergens require more (and they are expensive), so we test only once per batch. One of our product may contain traces of crustacean and molluscs, so we need to test for allergens but I'm not sure how often.

 

On top of that, our customers require allergen-free product, so how many clear test results should we receive to prove that implemented procedures work?

 

Thank you,

Inga



Charles.C

    Grade - FIFSQN

  • IFSQN Moderator
  • 20,542 posts
  • 5662 thanks
1,544
Excellent

  • Earth
    Earth
  • Gender:Male
  • Interests:SF
    TV
    Movies

Posted 29 September 2016 - 02:07 PM

Hi all,

 

I am not sure if there was such topic but I couldn't find it and really hope that someone will be able to help me out!

 

I am working with dried flaked herbs and I am in the middle of building our testing regime. Every batch (approx 200kg) is tested for microbes and allergens (our customer's requirement) but only one sample (100g) is sent out. Do you think I need to send more samples in order to assure product is free from allergens and contaminants or one sample (picked from different barrels of one batch) is good enough? I understand that 100g doesn't represent 200kg, so I was wondering if anyone works with similar products and could help me out? 

 

Thank you!

Hi Inverse,

 

Yr question is a classic sampling query. It is a similar one to people who wish to be confident that their collection (lot) of packed cartons does not contain Salmonella.

 

The short answer to yr question is that the number of samples taken from a lot, not so much the size of the lot, dictates the degree of confidence with which one can assert that the percentage of units in the batch exhibiting a target characteristic X is less than Y%. Generally the size of the lot is not so important if much larger than the sample size (eg > approx. 10x from memory).

 

For example if yr lot contained 10,000 boxes of 25g each and you took 10 random boxes to test for X, a negative result for all samples would mean that you can assert with a 95% confidence that in the whole lot, the percentage Y of boxes containing X is less than 30% !. (100 negative samples would give a Y of 3%.)

 

The above assumes a uniform distribution of X throughout the lot which is unlikely in practice so that the above comments may be somewhat optimistic.

 

To put it another way, if you believe that the lot contains 4% defective units, how many samples do you have to take to have a 95% probability of finding one defective unit ? The answer is 75 !

 

As you can see sampling to detect low levels of a defect with a decent probability of success requires many, many samples. Conversely a satisfactory (negative) result for a small number of samples does not really prove very much.

 

Basically yr OP needs a decision as to how precise/confident you wish to be that the lot is "free" of allergens.


Kind Regards,

 

Charles.C


Thanked by 1 Member:

Gerard H.

    Grade - SIFSQN

  • IFSQN Senior
  • 411 posts
  • 131 thanks
44
Excellent

  • France
    France
  • Gender:Male

Posted 30 September 2016 - 08:22 AM

Dear Inga,

 

Referring to your second question, I like to give some information about the First, Middle and End sampling of a batch testing. I just like to add, that I don't know your factory, so it is important to translate and adapt it to your situation and customer requirements.

 

If you like to see whether the batch is allergen free, free of traces. Where does the risk of contamination occurs? From a distant perspective I would say, that the allergen could be present on the line --> so First product to analyse. Or there may be contamination during the day. So Middle and End product.

 

  • If the 3 samples are allergen free, it says something that makes sense about your batch.
  • If 1 of the 3 is contaminated, you will have much information, however work to do (Resampling, more sampling and Root cause analysis). You have to consider the interesting statistical analysis of Mr. Charles C.

So there are many perspectives to look to your question. I hope that these comments are helpful for you to get toward a satisfying Sampling plan.

 

Kind regards,

 

Gerard Heerkens



redfox

    Grade - SIFSQN

  • IFSQN Senior
  • 481 posts
  • 163 thanks
24
Excellent

  • Philippines
    Philippines

Posted 03 October 2016 - 03:36 AM

Hello Inverse,

 

Please see attachment. It may help. We are using this now.

 

regards,

redfox

 

Attached File  Sample Size_ifsqn1.pdf   314.78KB   290 downloads

]Attached File  The Special Inspection Levels in the AQL Tables_ifsqn3.docx   89.63KB   216 downloads

Attached File  Sample Size Chart_ ifsqn2.pdf   66.55KB   246 downloads



Thanked by 2 Members:

Charles.C

    Grade - FIFSQN

  • IFSQN Moderator
  • 20,542 posts
  • 5662 thanks
1,544
Excellent

  • Earth
    Earth
  • Gender:Male
  • Interests:SF
    TV
    Movies

Posted 03 October 2016 - 07:05 AM

Hi Inverse,

 

I noticed this approach/example to calculate the sample size which you can try putting some numbers in if interested.

 

Attached File  Sample Size for Zero Nonconforming Items.pdf   170.24KB   209 downloads

 

In reality you are surely best advised to initially inspect/validate the Production Line, eg as per -

 

Attached File  Guidance on Food Allergen management.pdf   3.2MB   139 downloads

 

.

 

 

 

 


Kind Regards,

 

Charles.C


Thanked by 1 Member:

FurFarmandFork

    Food Safety Consultant, Production Supervisor

  • IFSQN Fellow
  • 1,264 posts
  • 590 thanks
206
Excellent

  • United States
    United States
  • Gender:Male
  • Location:Oregon, USA

Posted 04 October 2016 - 07:42 PM

If you're using quantitative limits for micro, remember also to lower your limits when making composite samples.

 

E.g. if I have five barrels that are independent (not continuous run and I could throw away two contaminated ones) and an APC limit of 10,000CFU/g, you can test ten grams of 1 barrel and reject based on that limit. If you instead mix two grams of each barrel, and want to reject individual barrels on that limit, you want to lower the limit to 2000CFU/g, because one dirty barrel at your target limit with 4 clean barrels will produce that result.


Austin Bouck
Owner/Consultant at Fur, Farm, and Fork.
Consulting for companies needing effective, lean food safety systems and solutions.

Subscribe to the blog at furfarmandfork.com for food safety research, insights, and analysis.

Charles.C

    Grade - FIFSQN

  • IFSQN Moderator
  • 20,542 posts
  • 5662 thanks
1,544
Excellent

  • Earth
    Earth
  • Gender:Male
  • Interests:SF
    TV
    Movies

Posted 05 October 2016 - 03:02 AM

If you're using quantitative limits for micro, remember also to lower your limits when making composite samples.

 

E.g. if I have five barrels that are independent (not continuous run and I could throw away two contaminated ones) and an APC limit of 10,000CFU/g, you can test ten grams of 1 barrel and reject based on that limit. If you instead mix two grams of each barrel, and want to reject individual barrels on that limit, you want to lower the limit to 2000CFU/g, because one dirty barrel at your target limit with 4 clean barrels will produce that result.

 

Hi Earth,

 

(slightly OT)

 

Sorry but I didn't quite understand yr example.

 

I presume by independent you meant that the 5 barrels were intended to be representative of different codes within a targetted lot whose acceptability is being assessed regarding APC. The chosen methodology usually relates to the style given in the specification, eg a nmMc format or a value for the estimated mean of the lot. 

 

For estimating the mean, afaik, a composite sample is usable provided that the sampling/splitting/combining/analytical procedure is statistically valid. However i think it is more (micro) usual to average the results of 5 (hopefully duplicated plate) analyses.

 

One difficulty with APC quantitative criteria is that the error in measurements of APC is typically substantial. Hence the x10 factors commonly seen in nmMc specifications.

 

IMEX it is not usually recommended to use a sample as small as 10g for APC measurements.


Kind Regards,

 

Charles.C


FurFarmandFork

    Food Safety Consultant, Production Supervisor

  • IFSQN Fellow
  • 1,264 posts
  • 590 thanks
206
Excellent

  • United States
    United States
  • Gender:Male
  • Location:Oregon, USA

Posted 05 October 2016 - 01:21 PM

Hi Charles,

 

I didn't do a great job explaining within the context of the thread. My recommendation is more applicable for a "skip lot" program where independent lots might be combined to make sure your testing is representative, but not necessarily including a test for each lot.


Austin Bouck
Owner/Consultant at Fur, Farm, and Fork.
Consulting for companies needing effective, lean food safety systems and solutions.

Subscribe to the blog at furfarmandfork.com for food safety research, insights, and analysis.

Charles.C

    Grade - FIFSQN

  • IFSQN Moderator
  • 20,542 posts
  • 5662 thanks
1,544
Excellent

  • Earth
    Earth
  • Gender:Male
  • Interests:SF
    TV
    Movies

Posted 05 October 2016 - 05:30 PM

Hi Charles,

 

I didn't do a great job explaining within the context of the thread. My recommendation is more applicable for a "skip lot" program where independent lots might be combined to make sure your testing is representative, but not necessarily including a test for each lot.

 

Hi Earth,

 

Thks for the clarification.

 

Indeed, skip-lots are afaik more into the Dodge-Romig World. :smile:


Kind Regards,

 

Charles.C


Inverse

    Grade - MIFSQN

  • IFSQN Member
  • 53 posts
  • 4 thanks
2
Neutral

  • United Kingdom
    United Kingdom

Posted 21 October 2016 - 08:44 AM

Dear all,

 

Thank you very much for all the information provided!

 

It's not as easy as I thought is going to be... I am still trying to completely understand the meaning of "representative testing" and pass it on to our HACCP team and customers. I hope it all works well!

 

Gerard H., 

 

The answer to your question: The contamination may only occur from raw material, therefore we go through quality control check to assure that all allergens are removed. However, due to their size (I'm talking about crustacean), some of them cannot be removed and we are aware of it. That's why we have precautionary allergen labelling of "May contain" which we want to remove.

 

Thanks for helping me out, much appreciated!

Inga 



Charles.C

    Grade - FIFSQN

  • IFSQN Moderator
  • 20,542 posts
  • 5662 thanks
1,544
Excellent

  • Earth
    Earth
  • Gender:Male
  • Interests:SF
    TV
    Movies

Posted 21 October 2016 - 01:51 PM

Dear all,

 

Thank you very much for all the information provided!

 

It's not as easy as I thought is going to be... I am still trying to completely understand the meaning of "representative testing" and pass it on to our HACCP team and customers. I hope it all works well!

 

Gerard H., 

 

The answer to your question: The contamination may only occur from raw material, therefore we go through quality control check to assure that all allergens are removed. However, due to their size (I'm talking about crustacean), some of them cannot be removed and we are aware of it. That's why we have precautionary allergen labelling of "May contain" which we want to remove.

 

Thanks for helping me out, much appreciated!

Inga 

 

hi Inverse,

 

I agree yr comment. Statistical sampling is a science/art all to itself.

 

You can see by studying a few publications that most official plans for retail products such as issued by the EC make a substantial compromise between "statistical confidence in the result" and logistical capabilities. IMEX this compromise is an unavoidable conclusion for most plants.

 

My suggestion is to initially borrow the methods proposed by experts as far as possible. If nothing else, these can be "claimed" to have been justified for implementation. Further modification can be made depending on the findings. The only obvious caveat is whether some lots are substantially more "critical" than others for some reason.

 

For example USFDA from memory require a minimum of 15 samples from a lot to be found to possess no detectable salmonella in order to designate the lot as "salmonella-free". ( theory suggests that the result indicates any contamination "probably" does not exceed 20% !).

 

Very few organisations IMEX can logistically match FDA. The EC plans referred above usually employ a maximum of 5 (negative) samples for acceptability. IMEX factories do not usually exceed the EC micro. format (or even attain it) because an in-house lab handling capability would likely be overwhelmed or the cost will be phenomenal for outside analysis. 

 

One caveat to the above is the amount of effort/time to carry out the analysis. Micro.via classical procedures is often slow/laborious but allergen testing procedures are likely faster/less labour intensive (?). Of course, if all the analysis is contracted out, it probably simply comes down to cost/time.

 

It is also worth remembering that proper HACCP implementation is supposed to enable/justify minimal analytical requirements !


Kind Regards,

 

Charles.C




Share this


0 user(s) are reading this topic

0 members, 0 guests, 0 anonymous users