Jump to content

  • Quick Navigation
Photo

High APC Counts

Share this

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

Charles.C

    Grade - FIFSQN

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

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

Posted 15 January 2016 - 08:13 PM

Hi QAGB,

 

Ultimately i think it comes down to whether the input materials / process adjustments allow the production of material of APC which is consistently compatible with the spec.

 

If the usual results or (more meaningfully) the lot average APC is close to the spec. the failure rate is likely to be high. How much the results need to be below the spec. to yield a failure rate of  x% depends on sigma. The magics of the central limit theorem. I also recall 1 sample procedures being handled in SQC texts for control purposes in special cases but i think it must reduce the capability.

 

But all this obviously needs "reliable" data. So I agree baby steps is good

 

It's a pity there is no methylene blue test for liquid sugar.


Kind Regards,

 

Charles.C


QAGB

    Grade - PIFSQN

  • IFSQN Principal
  • 685 posts
  • 262 thanks
115
Excellent

  • Earth
    Earth

Posted 15 January 2016 - 08:29 PM

Hi QAGB,

 

Ultimately i think it comes down to whether the input materials / process adjustments allow the production of material of APC which is consistently compatible with the spec.

 

If the usual results or (more meaningfully) the lot average APC is close to the spec. the failure rate is likely to be high. How much the results need to be below the spec. to yield a failure rate of  x% depends on sigma. The magics of the central limit theorem. I also recall 1 sample procedures being handled in SQC texts for control purposes in special cases but i think it must reduce the capability.

 

But all this obviously needs "reliable" data. So I agree baby steps is good

 

It's a pity there is no methylene blue test for liquid sugar.

 

 

Hi Charles,

 

I agree, we need to be able to see APC numbers be well within the spec, and not be near the high end to consistently be below limits. We're also going to test residual water from blender cleanings going forward. Once I get some reliable data I should be able to comment more as to how it all pans out.

 

I wish there was a methylene blue test or some magic test strip we could use. Maybe one day someone will come up with it.

 

 

Thanks,

QAGB



QAGB

    Grade - PIFSQN

  • IFSQN Principal
  • 685 posts
  • 262 thanks
115
Excellent

  • Earth
    Earth

Posted 25 January 2016 - 12:42 PM

Hi Charles,

 

I agree, we need to be able to see APC numbers be well within the spec, and not be near the high end to consistently be below limits. We're also going to test residual water from blender cleanings going forward. Once I get some reliable data I should be able to comment more as to how it all pans out.

 

I wish there was a methylene blue test or some magic test strip we could use. Maybe one day someone will come up with it.

 

 

Thanks,

QAGB

 

 

So here's an update: We've done a batch of product, and tested as follows: one typical sample at the beginning of the run approx 600 lbs. into the batch. We also did a composite sample of the run with 1 sample at approx 600 lbs. into the batch, 1 sample at approx. 9000 lbs into the batch, and 1 sample at 18000 lbs into the batch. Remarkably but sadly, the results were exactly the same for APC but way higher than the specifications. This was done after a caustic wash as well.

 

Now we're on to phase 2; which involves disappointment and testing of residual water from tank cleaning to see where we are. We've also done some extra composite sampling on another batch last week, so we'll see how the results go for that. We're going to keep trying the composite sampling method while we try the residual water testing.

 

QAGB



Charles.C

    Grade - FIFSQN

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

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

Posted 25 January 2016 - 01:16 PM

hi QAGB,

 

Not quite sure i understand the composition of the  samples. Do you mean a total of only 2 analysed  samples ?

 

And can you be more quantitative as to the "exactly the same for APC" ?  "Exactly" would require somebody out of Harry Potter IMHO. :smile:

 

PS - i trust you well sanitized the area of sampling and flushed the outlet appropriately.


Kind Regards,

 

Charles.C


QAGB

    Grade - PIFSQN

  • IFSQN Principal
  • 685 posts
  • 262 thanks
115
Excellent

  • Earth
    Earth

Posted 25 January 2016 - 01:27 PM

hi QAGB,

 

Not quite sure i understand the composition of the  samples. Do you mean a total of only 2 samples ?

 

And can you be more quantitative as to the "exactly the same for APC" ?  "Exactly" would require somebody out of Harry Potter IMHO. :smile:

 

PS - i trust you well sanitized the area of sampling and flushed the outlet appropriately.

 

 

Hi Charles,

 

There were 3 samples composited into 1 sample and 1 typical sample for comparison, so all in all, we have two results (one for the normal sample, one for the composite).

 

  • First sample - at roughly 600 lbs. into the run. This is normally what we do, with no compositing throughout the run.
  • Second sample [ a composite sample] (composed of 3 separate samples - 1 at 600 lbs, 1 at 9,000 lbs, 1 at 18,000 lbs.

The APC results were just that; exactly the same cfu/g. I don't want to divulge exactly what that number is... so we'll say for example, the first sample was 7,000 cfu/g APC, and the second (composited) sample was 7,000 cfu/g APC as well. Yeast, Mold, and Coliform numbers all look great (which is typical). Most of the time the percent error is not much more than 20% on any given micro test we have done from this particular laboratory. Soon we may take a look at some laboratory ring testing, to see what we get.

 

Yes, the sample area was actually sanitized and torched each time a sample was taken. So, for this analysis, the sample area was sanitized and torched 4 times.

 

Thanks,

 

QAGB



Charles.C

    Grade - FIFSQN

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

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

Posted 25 January 2016 - 02:59 PM

Hi QAGB,

 

It's a fascinating result. i wonder if a duplicate would have given a trio.

 

You might consider asking the laboratory for the raw data (ie initial sample size/plate counts/dilutions) which generated the results (would expect, minimally, 4 plate count values). Should be equally fascinating. (But the info will probably be restricted to internal use only.)

 

PS - TBH I would probably have attempted the validation first but i have a naturally suspicious nature. I do understand that this may be more of a challenge.

 

PPS - if the result was near the detection limits it might be slightly less fascinating but yr example should not be in that vicinity assuming the liquid does not look like black treacle.


Kind Regards,

 

Charles.C


QAGB

    Grade - PIFSQN

  • IFSQN Principal
  • 685 posts
  • 262 thanks
115
Excellent

  • Earth
    Earth

Posted 25 January 2016 - 04:30 PM

Hi QAGB,

 

It's a fascinating result. i wonder if a duplicate would have given a trio.

 

You might consider asking the laboratory for the raw data (ie initial sample size/plate counts/dilutions) which generated the results (would expect, minimally, 4 plate count values). Should be equally fascinating. (But the info will probably be restricted to internal use only.)

 

PS - TBH I would probably have attempted the validation first but i have a naturally suspicious nature. I do understand that this may be more of a challenge.

 

PPS - if the result was near the detection limits it might be slightly less fascinating but yr example should not be in that vicinity assuming the liquid does not look like black treacle.

 

 

Hi Charles,

 

My curiosity is right now with the cleaning process. I'm really curious to know what that residual water from the tank tests at after cleaning and rinsing. I feel like that would give a lot of insight.

 

How would you suggest we go about the validation? I'm sort of at a loss at that end.

 

Also, no we weren't at the detection limits. No "too numerous to count" issues, thankfully. I know there are obviously outliers in data, but it would be nice to get some general consistency. :headhurts:

 

Thanks, QAGB



Charles.C

    Grade - FIFSQN

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

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

Posted 25 January 2016 - 05:50 PM

Hi QAGB,

 

Just as an example –

 

If the results  were 7000cfu/g and material is “clear” then assuming duplicate plates are spread for each dilution, density ~= 1g/ml, one would give –

 

Zeroth dilution (1g sample on plate)  >> ca 7000 colonies on each plate >> not usable other than if desperate

One-tenth dilution (0.1g sample) >> ca 700 colonies on each plate >> not usable unless no other options

One-hundredth dilution (0.01g sample) >> ca 70 colonies on each plate >> usable, eg 60 / 80 > average 70

One-thousandth dilution (0.001g sample) >> ca 7 colonies on each plate >> not usable unless no other options

 

(Note - If the material is very dark, then if worst-case scenario, maybe only the one-thousandth dilution is usable  >  statistical problem).

 

So an ideal scenario requires 2 separate samples to give pairs of plates which both exactly average 70. It’s not impossible,  just improbable.

If >1 dilution gives usable plates for the overall average calculation, the likelihood of identical results presumably further reduces.

(I am always cautious over such things because statistics is cunning, just like the chance of 2 people in a room of 30-40 having same birthday, one would think it's minute, in fact it's "amazingly" high)

 

Re–rinsing

Sorry, I have no personal experience on this. How do you normally decide when cleaning is completed ? pH ? (often used for CIP cleaning control from memory).

Micro-wise, it looks like a case of trial/error. eg getting an idea of APC vs flushing time. Presumably the initial level should rapidly tend to normal water level (~zero?) (or whatever lower limit the lab can reliably “measure”). The latter will depend on what their practical options are.


Kind Regards,

 

Charles.C


QAGB

    Grade - PIFSQN

  • IFSQN Principal
  • 685 posts
  • 262 thanks
115
Excellent

  • Earth
    Earth

Posted 25 January 2016 - 06:31 PM

Hi QAGB,

 

Just as an example –

 

If the results  were 7000cfu/g and material is “clear” then assuming duplicate plates are spread for each dilution, density ~= 1g/ml, one would give –

 

Zeroth dilution (1g sample on plate)  >> ca 7000 colonies on each plate >> not usable other than if desperate

One-tenth dilution (0.1g sample) >> ca 700 colonies on each plate >> not usable unless no other options

One-hundredth dilution (0.01g sample) >> ca 70 colonies on each plate >> usable, eg 60 / 80 > average 70

One-thousandth dilution (0.001g sample) >> ca 7 colonies on each plate >> not usable unless no other options

 

(Note - If the material is very dark, then if worst-case scenario, maybe only the one-thousandth dilution is usable  >  statistical problem).

 

So an ideal scenario requires 2 separate samples to give pairs of plates which both exactly average 70. It’s not impossible,  just improbable.

If >1 dilution gives usable plates for the overall average calculation, the likelihood of identical results presumably further reduces.

(I am always cautious over such things because statistics is cunning, just like the chance of 2 people in a room of 30-40 having same birthday, one would think it's minute, in fact it's "amazingly" high)

 

Re–rinsing

Sorry, I have no personal experience on this. How do you normally decide when cleaning is completed ? pH ? (often used for CIP cleaning control from memory).

Micro-wise, it looks like a case of trial/error. eg getting an idea of APC vs flushing time. Presumably the initial level should rapidly tend to normal water level (~zero?) (or whatever lower limit the lab can reliably “measure”). The latter will depend on what their practical options are.

 

 

Hi Charles,

 

Understood on the plating statistics. I've tried not to think too much into the statistics, because I have to support people here that may or may not understand all the intricacies of statistics and microbiology.  If I try to explain the improbability of the results to people, we'll just fall further into the quagmire. I did find it interesting that both samples would exhibit the same APC results. I have three labs I'll work on submission to for ring testing in the future. For now, it's just baby steps until we can get useful results.

 

A regular rinsing just gets all the syrup remainder off the walls and bottom of the tank. That's more of a general inspection deal. A caustic wash requires removal of syrup remainder and the residual water to be close to source water in terms of pH. We don't have any true CIP systems here. Everything is done manually except some tanks have spray balls installed for water entry. Indeed APC for water should be on the 100/mL level, so it would be interesting to see what we've got.

 

Thanks,

 

QAGB



Charles.C

    Grade - FIFSQN

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

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

Posted 26 January 2016 - 04:58 PM

Hi QAGB,
 
Just a few oddments noted in passing which you are likely already familiar with.
 
It seems  that ATP methods have had "some" success in estimating microbial levels (~ APC) in the sugar industry / syrups. 
 
(And other more high-tech methods, eg PCR, which probably out of the question).
 
 
  ABSTRACT: ATP bioluminescence was evaluated as a method for assessing the level of microbial contamination in sap and predicting maple syrup characteristics. This approach provided results that were strongly correlated with the standard plate count and took less time than the modified resazurin technique. ATP bioluminescence measurement of sap proved to be reliable for predicting physicochemical and sensory characteristics as indicated by the color and flavor of maple syrup. Most of the syrups made from saps with higher ATP bioluminescence values were darker in color and presented off-flavors. Based on these results, ATP bioluminescence could be used to improve sanitary practices associated with collecting and storing maple sap.

 

http://onlinelibrary...8734.x/abstract

 
ATP necessitates a "calibration" graph of course but it is typically a relatively non-skilled /intrusive technique compared to APC.
 
ATP has also been used for microbial counts for hygiene / rinse waters (mention only).
 
Few random examples of alternatives to micro. - 
 
Attached File  cane sugar processing - ATP,1997.pdf   326.6KB   8 downloads
Attached File  syrup - fluorescence spec.2010.pdf   165.69KB   8 downloads
Attached File  syrup - PCR,2011.pdf   682.89KB   7 downloads

Edited by Charles.C, 26 January 2016 - 05:53 PM.
link corrected

Kind Regards,

 

Charles.C


QAGB

    Grade - PIFSQN

  • IFSQN Principal
  • 685 posts
  • 262 thanks
115
Excellent

  • Earth
    Earth

Posted 26 January 2016 - 06:18 PM

 

Hi QAGB,
 
Just a few oddments noted in passing which you are likely already familiar with.
 
It seems  that ATP methods have had "some" success in estimating microbial levels (~ APC) in the sugar industry / syrups. 
 
(And other more high-tech methods, eg PCR, which probably out of the question).
 
 
ATP necessitates a "calibration" graph of course but it is typically a relatively non-skilled /intrusive technique compared to APC.
 
ATP has also been used for microbial counts for hygiene / rinse waters (mention only).
 
Few random examples of alternatives to micro. - 
 

 

 

Hi Charles,

 

Thanks for the links. I couldn't get the Wiley document open, but I did read your quote containing the abstract. Yes, we at times do ATP testing and micro swabbing in parallel for other areas, and we do get mostly correlating results. The trouble is, with our current system, the contact points have to be dry for testing, and we don't always get that; especially with tanks. Also, due to confined space issues, the most we can do is swab the areas around tank doors (if we can get to them), and that in most cases isn't going to be indicative of the tank.

 

I'm not really sure how to test rinse water for ATP. The luminometer just won't work well with it, as it is. I am going to check into it, to see if we can find other options with different types of ATP swabs. That's currently going to be our most affordable option. Thanks for the ATP idea!

 

QAGB



QAGB

    Grade - PIFSQN

  • IFSQN Principal
  • 685 posts
  • 262 thanks
115
Excellent

  • Earth
    Earth

Posted 26 January 2016 - 06:26 PM

Hi Charles,

 

We don't want to cover too many variables at one time. We may do some ring testing on laboratory analysis soon but not just yet. I really wouldn't be shocked to find two different labs supposedly following the same methods come up with totally different results. We have had that happen in the past on simple things such as pH or moisture tests. I was more saddened than amazed by the findings considering most of the labs we use are ISO 17025 certified, and the methods for these things are simple and well-known.

 

We actually have a sample point, which happens to be the filling stations. What you get at the filling stations is representative of what is in going into packaging. Those areas get cleaned and sanitized and heated as well, and the initial product is also run through before taking the first sample.

 

The incoming lots don't come with micro values on COAs but we do get COAs for the material. We do some testing on it, but nothing from a micro standpoint.

 

Thanks,

 

QAGB

 

 

Hi Charles,

 

I'm looking into some options for our luminometer and we might be able to come up with something. The company we get our ATP materials from makes a water test kit and a total viable count kit. I'm going to definitely contact the sales rep and see if these would be options for us.

 

Thanks,

 

QAGB



Charles.C

    Grade - FIFSQN

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

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

Posted 26 January 2016 - 09:02 PM

Hi QAGB,

 

I added the link just to show the source if interested to do a search (I was unable to find a viewable hit). It's a suscribed journal.

 

Perhaps my post was unclear. No connection to swabbing. This is ATP(sample) vs APC(sample) as in the cane sugar example I attached.

 

Here is another one for water (~rinse) -

 

Attached File  ATP method for viable bacteria in water.pdf   140.64KB   10 downloads

 

The point is that in principle you could measure ATP yourself (I deduce you do have a "lab" albeit not micro.capable) or cross-check the external lab result for micro. IMO the method would be too inaccurate for close comparisons but should identify gross discrepancies.

But, as shown in many publications, the success rate for obtaining a linear correlation for ATP vs APC is not high. Seemingly more like pot luck.

 

Regarding yr sampled batch, I don't understand why you make a composite for APC ?. As I understand, you have at least 3 samples (not sure what the "typical" sample is ?) spanning the production which offers a crude monitoring possibility. I deduce the other variables are never a problem so could be composited OK if cost is the difficulty.

IMO, as per typical int.stds, a set of 5 samples across the "batch" would seem logical just for a test run.

 

IMEX, the critical objective from a sampling is to derive as much useful  information as possible. What you do with the information is a different question of course.

 

But I appreciate that the time allocated all depends on the relative importance of the project in the overall scheme of things.


Kind Regards,

 

Charles.C


QAGB

    Grade - PIFSQN

  • IFSQN Principal
  • 685 posts
  • 262 thanks
115
Excellent

  • Earth
    Earth

Posted 26 January 2016 - 09:19 PM

Hi QAGB,

 

I added the link just to show the source if interested to do a search (I was unable to find a viewable hit). It's a suscribed journal.

 

Perhaps my post was unclear. No connection to swabbing. This is ATP(sample) vs APC(sample) as in the cane sugar example I attached.

 

Here is another one for water (~rinse) -

 

attachicon.gifATP method for viable bacteria in water.pdf

 

The point is that in principle you could measure ATP yourself (I deduce you do have a "lab" albeit not micro.capable) or cross-check the external lab result for micro. IMO the method would be too inaccurate for close comparisons but should identify gross discrepancies.

But, as shown in many publications, the success rate for obtaining a linear correlation for ATP vs APC is not high. Seemingly more like pot luck.

 

Regarding yr sampled batch, I don't understand why you make a composite for APC ?. As I understand, you have at least 3 samples (not sure what the "typical" sample is ?) spanning the production which offers a crude monitoring possibility. I deduce the other variables are never a problem so could be composited OK if cost is the difficulty.

IMO, as per typical int.stds, a set of 5 samples across the "batch" would seem logical just for a test run.

 

IMEX, the critical objective from a sampling is to derive as much useful  information as possible. What you do with the information is a different question of course.

 

But I appreciate that the time allocated all depends on the relative importance of the project in the overall scheme of things.

 

Hi Charles,

 

I speak on general terms of ATP. We have a luminometer that detects ATP. We do swabbing for ATP with it, because the kit comes with swabs. There are other types of ATP test kits that don't involve swabbing (using the same luminometer), which is partly what I am looking into. I was also stating that, regardless of sampling method, we have seen that increased ATP numbers often times match up with high APC numbers. That's not always the case, but generally speaking it can trend in that direction. I used swabbing as the point of reference there, because that's the current method for testing food contact surfaces in our plant. I'm working with our sales rep to get a kit for ATP that tests the residual water (no swabbing needed). I'll be trying that out soon, and then comparing results with outside micro testing (we do have an internal lab, it's just not set up for micro testing).

 

As far as the sampling methods, I thought we were looking at sampling size. We normally (as part of procedure) only take one sample (close to the beginning of the batch) to send out for testing which accounts for what I have been calling the "typical sample" [at roughly 600 lbs. into the run in this case]. Based on previous posts, I gathered  maybe one sample at the beginning of the run wasn't enough; as the information you sent to me stated that micro isn't always uniformly distributed throughout. In response, I wanted us to do a composite of product throughout the run to see if the results would be better if the sample represented more of the entire batch, and not just the beginning.

 

Thanks,

 

QAGB



Charles.C

    Grade - FIFSQN

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

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

Posted 27 January 2016 - 07:42 AM

Hi QAGB,

 

Yes, I agree, the sample size is (particularly initially) critical.

 

With >=3 points one could start to apply some simple statistics although 5 is probably a lot better, especially if there is a weird underlying distribution. Despite the latest lab results, Nature says there has to be some distribution :smile:

 

In fact, regarding yr already accumulated (single point) data, if one hypothesised that the lots so far received are randomly derived from some infinite mother pile in the sky, one can test this hypothesis which, if found tenable, permits combining the data to generate a working sigma and hence make predictions. From memory, some people combine the data regardless, perhaps after removing gross outliers.

 

Alternatively, one can simply do like the USFDA treat imports - every lot has no history. The trade-off, at least theoretically, is in sample size. Practice is more likely manpower/cost  constraints.


Kind Regards,

 

Charles.C


QAGB

    Grade - PIFSQN

  • IFSQN Principal
  • 685 posts
  • 262 thanks
115
Excellent

  • Earth
    Earth

Posted 27 January 2016 - 02:09 PM

Hi QAGB,

 

Yes, I agree, the sample size is (particularly initially) critical.

 

With >=3 points one could start to apply some simple statistics although 5 is probably a lot better, especially if there is a weird underlying distribution. Despite the latest lab results, Nature says there has to be some distribution :smile:

 

In fact, regarding yr already accumulated (single point) data, if one hypothesised that the lots so far received are randomly derived from some infinite mother pile in the sky, one can test this hypothesis which, if found tenable, permits combining the data to generate a working sigma and hence make predictions. From memory, some people combine the data regardless, perhaps after removing gross outliers.

 

Alternatively, one can simply do like the USFDA treat imports - every lot has no history. The trade-off, at least theoretically, is in sample size. Practice is more likely manpower/cost  constraints.

 

 

Hi Charles,

 

I agree that sample size is important. No matter how many sampling points we choose, the end result can be somewhat complex. For example, say we are doing 20 totes (3,000 lbs. each) to be tested for micro. The following is all hypothetical.

  • I choose 5 sampling points throughout the batch: Tote 4, Tote 8, Tote 12, Tote 16, Tote 20.
  • I decide to have these samples composited.
  • I also pull a singular sample from Tote 4 to be tested as per typical procedures.
  •  There's no way to see a true distribution at this point.
  • If the results are both good on the composite and singular sample, then we can  see that the trend is that the APC is somewhat uniform, and product can ship.
  • If the results are both bad on the composite and singular sample, we can still see the trend is that APC is somewhat uniform, but product goes on hold.
  • If the APC on the composite is higher than the limit, but the singular sample was good, everything still goes on hold because the average APC of the batch is higher than the limit .The trend in this case would show APC not to be uniform.
  • If the APC on the singular sample is higher than the limit, but the composite sample is good, the product could ship based on average APC of the batch. The trend in this case would show APC not to be uniform.

 

or alternatively,

 

  • I choose 5 sampling points throughout the batch: Tote 4, Tote 8, Tote 12, Tote 16, Tote 20.
  • I decide not to composite these samples and just test them all singularly.
  • Perhaps we get good results on Tote 4, Tote 8, and Tote 20. Tote 12, and Tote 16 come back higher than limits.
  • If the average APC of the 5 samples is below the limits, the product could ship.
  • If the average APC of the 5 samples is above the limits, you would theoretically be ok to ship Totes #1-8, and #20. However Totes #9-19 would have to be tested again (maybe another 3-5 samples) to determine if any of those could ship.
  • If the APC results on the 5 samples are similar, this would suggest APC uniformity.
  • If the APC results on the 5 samples are not similar, this would suggest APC is not uniform.

 

The conclusions we can draw here would be whether or not sample size has been an issue all along, and whether or not APC is distributed uniformly. The same conclusions should be able to be drawn with 3 samples of a smaller batch size, regardless of the sampling method (whether composited or not). At this point, we just want to know if the sampling is truly representative of the batch, or whether testing product throughout the run would help us figure out some of the issues.

 

PS: This all makes sense in my head, but it's probably going to be confusing. Hopefully, I explained it decently enough.

 

QAGB



Charles.C

    Grade - FIFSQN

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

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

Posted 27 January 2016 - 04:05 PM

Hi QAGB,

 

I appreciate yr detailed presentation. The difficulty i have is that it seems more based on intuition/sample  than statistics/lot. However perhaps in yr case this does not matter.

 

We may well be conversing at cross-purposes.

 

IMEX The significance of the APC "value"  depends on the terms of the business, and sometimes regulatory factors (eg at import). Just as an example, a one-off accept/reject  shipment scenario is rather different  from a commercial POV as compared to a long-term arrangement where "corrective" adjustments can be made.

 

IMEX, the evaluation is effected by a mutually agreed sampling / analysis / decision procedure. Such procedures are available in the literature, notably designed to provide specified risks for supplier/receiver/both as may be agreed. In reality, some sales managers choose to ignore such fundamentals and risk their intuitive arm. It's their business, they have been warned by statistics.

 

In the absence of any particular penalties, the data becomes basically informative so one can select any mix of statistical support / inferential logic as preferred. If it works for you, that's fine.

 

It all depends on the requirements.


Kind Regards,

 

Charles.C


QAGB

    Grade - PIFSQN

  • IFSQN Principal
  • 685 posts
  • 262 thanks
115
Excellent

  • Earth
    Earth

Posted 27 January 2016 - 04:26 PM

Hi QAGB,

 

I appreciate yr detailed presentation. The difficulty i have is that it seems more based on intuition/sample  than statistics/lot. However perhaps in yr case this does not matter.

 

We may well be conversing at cross-purposes.

 

IMEX The significance of the APC "value"  depends on the terms of the business, and sometimes regulatory factors (eg at import). Just as an example, a one-off accept/reject  shipment scenario is rather different  from a commercial POV as compared to a long-term arrangement where "corrective" adjustments can be made.

 

IMEX, the evaluation is effected by a mutually agreed sampling / analysis / decision procedure. Such procedures are available in the literature, notably designed to provide specified risks for supplier/receiver/both as may be agreed. In reality, some sales managers choose to ignore such fundamentals and risk their intuitive arm. It's their business, they have been warned by statistics.

 

In the absence of any particular penalties, the data becomes basically informative so one can select any mix of statistical support / inferential logic as preferred. If it works for you, that's fine.

 

It all depends on the requirements.

 

 

Hi Charles,

 

Based on what you're saying, can I summarize that as to mean, there should be an agreed procedure between customer and supplier as to how the process for sampling (and decision) should be done? 

 

If so, I'd say that makes sense. We've just never had any customers ask us for sampling methods, or provide us with any specific sampling methods. I don't know if there are assumptions being made, or they don't really know much about the sampling process.

 

In reference to this statement: "The difficulty i have is that it seems more based on intuition/sample  than statistics/lot." I don't know that I totally understand what it means. When you are given statistics, historical data, etc. I would think a bit of inference or intuition has to come as a result of the numbers given. For example, if I test 12 samples (one sample every 5 minutes) from a batch, and I get a pH of 4.4-4.5 each time, and all these numbers are 100% within the required range, I would then infer the rest of the product before, in between each of those 12 samples, and after those 12 samples would have a pH of 4.4-4.5. There's always a chance that something could have happened in a one of the product lines, and a trace of another product causes a small rise or lowering in pH at some point during the process, but based on the statistics....I would infer the risk to be low, and that the total product should have a pH of 4.4-4.5. In turn, back to the APC issue, if I do larger sampling sizes, get results, and can see trends in those results (based on statistics), would I not be able to infer certain characteristics? Pardon my confusion, I'm just trying to understand.

 

QAGB



Charles.C

    Grade - FIFSQN

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

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

Posted 27 January 2016 - 06:13 PM

Hi QAGB,

 

Based on what you're saying, can I summarize that as to mean, there should be an agreed procedure between customer and supplier as to how the process for sampling (and decision) should be done?

 

Yes, precisely. IMEX it's laid down in the contactual specifications. An occasional practical problem IMEX is that some labs, including commercial ones, shortcut standard procedures in the interests of time, convenience, whatever. I deduced from yr earlier posts that  this set-up is not typical of  yr own business situation.  It happens.

 

Unfortunately, regardless of contracts,  the buyer in some cases simply refuses to pay. Checkmate.

 

Referring to last paragraph, I meant it seems to me that that your logic is more focussing on the "micro" situation whereas statistics tends to look at the "macro" (with some notable exceptions). Not a criticism, just a difference in viewpoint. :smile: As i said, if it works for you......

 

IMO the statistical equivalent of yr last paragraph relates to the sampling theory of distributions of (averages of [original] distribution) in order to characterize the original distribution. Ideally the central limit theorem guarantees a (hypothesised) normal curve of averages from a whole range of weird original distributions, eg uniform, triangular, etc. The degree of non-homogeneity in the original distribution is reflected in the variance of the (hypothesised) distribution of averages. The variance then  enables the prediction of the risk of a disparity between buyer/seller (assuming both are using similar procedures). This is textbook sampling of materials / contractual spec. designing, eg by ASTM.

Only problem IMEX is that the text rapidly becomes gibberish to a non-statistician.


Kind Regards,

 

Charles.C


QAGB

    Grade - PIFSQN

  • IFSQN Principal
  • 685 posts
  • 262 thanks
115
Excellent

  • Earth
    Earth

Posted 27 January 2016 - 06:27 PM

Hi QAGB,

 

 

Yes, precisely. IMEX it's laid down in the contactual specifications. An occasional practical problem IMEX is that some labs, including commercial ones, shortcut standard procedures in the interests of time, convenience, whatever. I deduced from yr earlier posts that  this set-up is not typical of  yr own business situation.  It happens.

 

Unfortunately, regardless of contracts,  the buyer in some cases simply refuses to pay. Checkmate.

 

Referring to last paragraph, I meant it seems to me that that your logic is more focussing on the "micro" situation whereas statistics tends to look at the "macro" (with some notable exceptions). Not a criticism, just a difference in viewpoint. :smile: As i said, if it works for you......

 

IMO the statistical equivalent of yr last paragraph relates to the sampling theory of distributions of (averages of [original] distribution) in order to characterize the original distribution. Ideally the central limit theorem guarantees a (hypothesised) normal curve of averages from a whole range of weird original distributions, eg uniform, triangular, etc. The degree of non-homogeneity in the original distribution is reflected in the variance of the (hypothesised) distribution of averages. The variance then  enables the prediction of the risk of a disparity between buyer/seller (assuming both are using similar procedures). This is textbook sampling of materials / contractual spec. designing, eg by ASTM.

Only problem IMEX is that the text rapidly becomes gibberish to a non-statistician.

 

 

Hi Charles,

 

I get what you are saying now. I'd really be interested in knowing the sampling methods of the supplier, but unfortunately they aren't forthcoming. I'd also be interested in knowing sampling methods of other companies that have APC specifications and how they use the statistics to formulate reject/accept procedures. I've never had a customer ask us how we go about doing this, and during audits, our methods have never been noted as an issue (so I wonder if this is sort of becoming the norm in most places). There are just so many questions I have regarding this topic...

 

QAGB



Charles.C

    Grade - FIFSQN

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

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

Posted 27 January 2016 - 06:51 PM

Hi QAGB,

 

Actually I regarded yourself as the supplier delivering APC <= 5000/10,000 cfu/g to MrXYZ

 

I deduce yr customers don't operate an accept/reject procedure.

 

 I also deduce you never query the customers sampling/analysis procedures. Because there is no actual  penalty for exceeding the APC  limits ?.

 

And since you don't monitor the tanker inputs, I assume no (APC) penalties there either. Interesting set-up.


Kind Regards,

 

Charles.C


QAGB

    Grade - PIFSQN

  • IFSQN Principal
  • 685 posts
  • 262 thanks
115
Excellent

  • Earth
    Earth

Posted 27 January 2016 - 07:36 PM

Hi QAGB,

 

Actually I regarded yourself as the supplier delivering APC <= 5000/10,000 cfu/g to MrXYZ

 

I deduce yr customers don't operate an accept/reject procedure.

 

 I also deduce you never query the customers sampling/analysis procedures. Because there is no actual  penalty for exceeding the APC  limits ?.

 

And since you don't monitor the tanker inputs, I assume no (APC) penalties there either. Interesting set-up.

 

 

Hi Charles,

 

We get product from a supplier, we blend, heat, etc. and then ship to customers.Yes we are a supplier, but we also get those materials from a supplier that also has set APC specifications. We try to stay at or around all supplier specifications, including APC (as the resulting specifications are industry standard).

 

The penalties for not meeting APC specifications would be putting product on hold, changing disposition, losing packaging, and not meeting expected dates. Sometimes customers requiring testing will accept certain APC levels when they are able to do so. However, many of them want a certain level, and will not accept anything above it.

 

We don't have APC penalties for tanker inputs because with the volume we produce, we can't afford to hold tanker loads until we get incoming APC results (3-5 business days). 

 

QAGB 



Charles.C

    Grade - FIFSQN

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

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

Posted 27 January 2016 - 08:04 PM

Hi QAGB,

 

Thks. I now have another 2 queries.

 

(1) Do you get many instances (eg >10% delivered lots)  where the final receiver's claimed APC is significantly different to yr own ? eg outside +/- 50% of yr own value  ? (the reason for my choice of 50% is that IMEX it is not rare to get duplicate APC plates in such a range)

 

(2) Do you ever know how the customer samples/analyses to get their APC ? (eg is the AOAC procedure mentioned earlier somehow stipulated for both ends?


Kind Regards,

 

Charles.C


QAGB

    Grade - PIFSQN

  • IFSQN Principal
  • 685 posts
  • 262 thanks
115
Excellent

  • Earth
    Earth

Posted 27 January 2016 - 08:28 PM

Hi QAGB,

 

Thks. I now have another 2 queries.

 

(1) Do you get many instances (eg >10% delivered lots)  where the final receiver's claimed APC is significantly different to yr own ? eg outside +/- 50% of yr own value  ? (the reason for my choice of 50% is that IMEX it is not rare to get duplicate APC plates in such a range)

 

(2) Do you ever know how the customer samples/analyses to get their APC ? (eg is the AOAC procedure mentioned earlier somehow stipulated for both ends?

 

Hi Charles,

 

1) In most cases, the customers don't (to my knowledge) do testing on the product if we do the testing for them. If they are doing the testing as well, then we haven't ever had any issues where APC results on our end and theirs didn't match up.

 

2) As a result of not knowing whether they do additional testing, we don't have that information.

 

QAGB



Charles.C

    Grade - FIFSQN

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

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

Posted 27 January 2016 - 09:10 PM

Hi QAGB,

 

Thks again, i rather expected yr response. In other words, the customers are trusting yr data. So no claims. Unfortunately it also means that zero feedback to compare lab. performance.

 

Frankly, IMO a result of 10,000cfu/g could easily show 5000cfu/g on a repeated analysis of the same sample. This means that the 2 limits are within the error range of the measurement. It's slightly apples and oranges but for comparison, the typical micro.specs use  M/m = 10:1

 

Well, the only systematic approach I know for predicting an APC level of prescribed safety margin below the maximum limit is a statistical one as per my preceding posts. I guess the choice is up to you. :smile:


Kind Regards,

 

Charles.C




Share this


0 user(s) are reading this topic

0 members, 0 guests, 0 anonymous users