Blend Uniformity Log Out | Topics | Search
Moderators | Register | Edit Profile

NIR Discussion Forum » Bruce Campbell's List » I need help » Blend Uniformity « Previous Next »

Author Message
Top of pagePrevious messageNext messageBottom of page Link to this message

Gustavo Figueira de Paula (gustavo)
Advanced Member
Username: gustavo

Post Number: 22
Registered: 6-2008
Posted on Monday, February 21, 2011 - 4:08 pm:   

Gretchen,

If your flow agent fluoresces, you can use a much cheaper UV-visible spectrometer with much higher sensitivity. Of course, only if the matrix do not fluoresces in the same wavelength range.

I don't have idea about hat flow agent you're using - it's just an idea.

Gustavo.
Top of pagePrevious messageNext messageBottom of page Link to this message

Gretchen Karcher (gkarcher)
New member
Username: gkarcher

Post Number: 1
Registered: 2-2011
Posted on Monday, February 21, 2011 - 10:01 am:   

Hello,

I'm new to the boards, but I'm working on trying to figure out a way to check the blend uniformity of a fine powdered flow agent onto a granule. I've made calibration curves but I can only trust my results to 0.5% and I'm only blending 1-2% flow agent onto my granules, so that's quite an error.

My calibration curves are based on samples in small glass jars on the integrating sphere after being blended for a certain amount of time. But I'm having trouble trusting that the fines are dispersed properly. I've thought of simply tapping all the fines to the bottom of the jar for measuring but I'm not sure how reliable that is.

These samples can't be dissolved easily, so that's out of the question. Do any of you have any suggestions on how best to measure this more accurately?

Any help would be greatly appriciated.

Thanks,
Gretchen
Top of pagePrevious messageNext messageBottom of page Link to this message

venkatarman (venkynir)
Senior Member
Username: venkynir

Post Number: 60
Registered: 3-2004
Posted on Friday, April 25, 2008 - 12:49 pm:   

Dear Atul Karande;
Fine .There are good amount of papers in blending.
Quality test through PCA can be done in on-line.
I do agree special fiber optical sensor with self calibrated is avialble and you can integrate it with DAS.
I have discussed this prof.Paul Hung of your university.
If you send the data and infrenentail method for say moisture and other components we can halp you in this regard.
Top of pagePrevious messageNext messageBottom of page Link to this message

Atul Karande (atul)
New member
Username: atul

Post Number: 1
Registered: 4-2008
Posted on Thursday, April 24, 2008 - 9:10 am:   

Hi Everyone
This is Atul, a new member of this group. I am working on pharmaceutical dry powder blending at its assessment using NIR online monitoring. Well as mentioned in literature and practical experience we generally carry out some pre-treatments of the raw spectra in order to make them more similar with the property of interest and followed by regression for building the calibration models. Here with the property of interest I mean the chemical concentration. From the scores and loadings plot I observed the second derivative treatment was more efficient in removing the unwanted info in spectra and making it more similar with respect to its chemical composition. But afterwards when I used this second derivative data for model building using both PCR and PLS it resulted in inferior model (R2, RMSEC, RMSEP were not quite acceptable) compared to other pretreated spectra. So I am bit confused now regarding the selection of the right choice of pretreatments, what criteria should I adopt.
Thanks
Top of pagePrevious messageNext messageBottom of page Link to this message

Sheelagh Halsey (shalsey)
Junior Member
Username: shalsey

Post Number: 6
Registered: 12-2006
Posted on Wednesday, March 12, 2008 - 12:01 pm:   

Raphael

The best way to validate a qualitative method is just to teach the system 'good' material. If you have 'bad' material, this will help the validation step. With blend analysis this should be possible. If no 'bad' batches are available, you could also use not fully blended samples to prove the method would detect an out of specification blend.

So for the good samples, will need to cover all the natural variability you have from batch to batch. You will then need to build up your good quality library with all these batches. The number of batches is always a difficult question as it will depend on the variability of the process. I would start with soemthing like 10 batches and then compare the next batch to them with whatever qualitative method you have decided to use. You could also look at the spectral variability of the batches you have, probably with principal components. If they are all close to each other, it would indicate you have a well controlled process and maybe you do not need too many batches. If they are very variable, you may need more batches.

As part of your validation you can then compare new batches. If the sample passes both the NIR and reference tests, I would leave the library as is. Similarly is both tests fail, the method is working OK. If it passes the reference tests but fails the NIR, then the library should be updated with the new batch. This would be the same for any qualitative method, such as raw material ID/conformity. I am afraid this will be a bit iterative, but you should get a robust method in the end.

Sheelagh
Top of pagePrevious messageNext messageBottom of page Link to this message

raphael beauget (raph)
Junior Member
Username: raph

Post Number: 7
Registered: 4-2007
Posted on Tuesday, March 11, 2008 - 11:23 am:   

All,

I have read many documents (publications, guidelines,...) about blend uniformity. And I have found many criteria for Qualitative method but only for identification of materials or Qualification of differents grade of a material. there is nothing about Qualitative method for blend uniformity. All criteria don't be applicable for it. For example, I don't know if there are a minimum of batches for the calibration set (is it necessary to integrate variability : different physical properties,...)

It's the same for the validation : I don't believe that it's necessary to analyze "potential challengers" for check the specificity like indicates for Qualitative method.

Have you any indications about it ?

Thanks,

Raph
Top of pagePrevious messageNext messageBottom of page Link to this message

Howard Mark (hlmark)
Senior Member
Username: hlmark

Post Number: 182
Registered: 9-2001
Posted on Tuesday, March 11, 2008 - 4:50 am:   

Yeah, sorry 'bout that, but the copy of the article I looked at had the wrong URL on it.

\o/
/_\
Top of pagePrevious messageNext messageBottom of page Link to this message

Michael C Mound (mike)
Senior Member
Username: mike

Post Number: 48
Registered: 7-2007
Posted on Tuesday, March 11, 2008 - 4:36 am:   

Hi, Raph,

Thanks! Got it now.

Best regards...and thanks, again!!!
Top of pagePrevious messageNext messageBottom of page Link to this message

raphael beauget (raph)
Junior Member
Username: raph

Post Number: 6
Registered: 4-2007
Posted on Tuesday, March 11, 2008 - 2:21 am:   

Hi Mike !

I have had the same issue like you ! If you want the article 24 follow this link :

http://www.aapspharmscitech.org/default/issueView.asp?vol=04&issue=02

Raph
Top of pagePrevious messageNext messageBottom of page Link to this message

Michael C Mound (mike)
Senior Member
Username: mike

Post Number: 47
Registered: 7-2007
Posted on Monday, March 10, 2008 - 10:06 pm:   

...of course, I meant article 24...
Top of pagePrevious messageNext messageBottom of page Link to this message

Michael C Mound (mike)
Senior Member
Username: mike

Post Number: 46
Registered: 7-2007
Posted on Monday, March 10, 2008 - 10:05 pm:   

Hi, Howard,

Must be something I am missing. When going to the referenced URL link cited, I find it impossible to locate article 21. What I do get are references to a Cadillac Seville or Online Dating...seems interesting but a bit "off the mark"...pun unintended (!).

Thanks,

Mike
Top of pagePrevious messageNext messageBottom of page Link to this message

Howard Mark (hlmark)
Senior Member
Username: hlmark

Post Number: 180
Registered: 9-2001
Posted on Monday, March 10, 2008 - 9:08 am:   

Raph - as Sheelagh says, you can use either Mahalanobis Distance or Conformity Index, or both. You may want to check out AAPS PharmSciTech, 4(2), Article 24 (2003) (http://www.pharmscitech.org) for an article comparing the two methods.

\o/
/_\
Top of pagePrevious messageNext messageBottom of page Link to this message

Sheelagh Halsey (shalsey)
New member
Username: shalsey

Post Number: 5
Registered: 12-2006
Posted on Monday, March 10, 2008 - 4:41 am:   

Raphael

Yes, what you describe is what I mentioned as overall conformity (wavelength disctance)in my answer. This algorithm is very good at determining whether or not the whole spectrum conforms to your 'gold standard' blend. This can also be done in other ways with principal components utilising Mahalanobis distance as Howard suggested. These algorithms are very good to help you tell that the blend is finished as they will check that all components are mixed well.

However, if you want to compare the result to the %RSD of the active, these methods will not be specific. I would start by overlaying all the ingredients of the blend and see if you can find a unique peak for the active compared to the excipients. You can usually find an aromatic peak around 1680nm that is pretty well separated from everything else in derivative. If this is the case, you can use the absorbance of this peak to calculate %RSDs. As you say, you can do this over time and compare to thief samples.

As my final method, I would use a combination of these approaches as they will give you different information. I would plot out some sort of qualitative conformity result to check the overall spectrum is good. In addition, plot a running standard deviation of the active to check that as well. Remember it does not stop there, you could also plot out lubricant (Mg stearate)since many components can be analysed simultaneously.

Sheelagh
Top of pagePrevious messageNext messageBottom of page Link to this message

raphael beauget (raph)
New member
Username: raph

Post Number: 5
Registered: 4-2007
Posted on Monday, March 10, 2008 - 4:08 am:   

Howard, sheelagh,

Firstly, thank you for yours answers !
As sheelagh says, I think perform %RSD of 10 samples from 10 différents point of the blender. I want realize this at different times (for example : 1,2,4,8,10 min). After, I can follow %RSD in relation to time.

For compute %RSD, I need values but active isn't always easy to isolate following the nature of the blend ! I have read this following protocol, I want to know what's your opinion on that :

1. Make a reference set with 10 spectra of final blend (the uniformity of blend have been controlled by HPLC)

2. Calculate CI= A(ref,i)-A(sample,i)/s(ref,i)
with "s" is the standart deviation
"ref" is the average of 10 reference spectra

3. The %RSD is calculted from CI.

Have you some experiment with CI ?

Thank you

Regards,
Top of pagePrevious messageNext messageBottom of page Link to this message

Howard Mark (hlmark)
Senior Member
Username: hlmark

Post Number: 179
Registered: 9-2001
Posted on Friday, March 07, 2008 - 1:20 pm:   

Raphael - you could also compute the Mahalanobis Distance, and when that gets small and doesn't change much you should be done.

With any of these computational approaches, you (and your computer!) are going to have to learn what parameters for the computation (basic calculation, threshold, time below threshold, amount of averaging (if any), how "small" is small, how "constant" is constant, etc.) to use for your stopping criterion.

Howard

\o/
/_\
Top of pagePrevious messageNext messageBottom of page Link to this message

Sheelagh Halsey (shalsey)
New member
Username: shalsey

Post Number: 4
Registered: 12-2006
Posted on Friday, March 07, 2008 - 11:55 am:   

Typically you can follow the homogeneity of a blend by plotting a standard deviation of the active peak for example. Once the standard deviation has reached a minimum the blend should be finished. You could also look at overall conformity of the spectrum using algorithms like wavelength distance or principal components.

To specifically compare %RSD of the blend and the refence method, I have compared the actual peak height of the active (pretreated with e.g. SNV and second derivative) on the last 10 rotations. The %RSD can be calculated for these 10 values. You would then take your 10 samples from different points of the blender and perform your reference analysis for the active. You would then calculate the %RSD of these results. Hopefully both sets of values would be below 2% for example, or what ever acceptance criteria you have for the reference method method.

Sometimes you can get odd results if your absorbance value is close to zero. This can happen with derivative spectra.

You could also calculate the %RSD of the 10 results for each method to prove you had similar variance in each one.
Top of pagePrevious messageNext messageBottom of page Link to this message

raphael beauget (raph)
New member
Username: raph

Post Number: 4
Registered: 4-2007
Posted on Friday, March 07, 2008 - 9:43 am:   

Hello everybody,

Blend uniformity analysis by NIR can be performed qualitatively or quantitatively. I want control blend uniformity by fiber optic probe but I need create models quickly and I don't need quantitative informations. I think qualitative method is the better for my application. For develop the model I want follow the RSD(%) during the blending but i don't know how can i do this !!

Have you some experiments for this kind of method ?

Thank you,

Regards

Add Your Message Here
Posting is currently disabled in this topic. Contact your discussion moderator for more information.