https://www.euro-online.org/enog/inoc2007/Papers/mac-slots.html https://www.euro-online.org/enog/inoc2007/Papers/m https://www.euro-online.org/enog/inoc2007/Papers/mac-slots.html

Calibration model for assay of pharmaceutical product

Ahmed's picture
Forums: 

Hi,

I am building a calibration model to predict the assay of a pharmaceutical product. I have formulated serial concentrations of the product and sampled 5 units of each concentration of the product for analysis by the reference method (HPLC). I further sampled another sample (another 5 units) of each concentration for analysis by NIR. Upon building the calibration data set, I have 5 NIR spectra for each concentration and I set the average of the reference results by HPLC as the reference value for the 5 spectra without considering the established variation within the 5 units.

 

is there a solution to consider the variation of the reference method results when setting the reference values of the calibration set?

Ahmed's picture

example: If the assay of the 5 reference tablets by HPLC is (100,98,97,100 and 101), I set the reference value for all NIR spectra of 5 calibration tablets as the average (99.2) without considering the variation established by the reference method 

Ahmed Ramadan

Natural Wellness Egypt Scientific Office  | Technical Project Assistant Manager

 

miguelG's picture

Hello,

I am not sure if I fully understood the problem but I'll give it a go.

NIR should be a non-destructible technique. So, if you analyse a sample by NIR you should be able to analyse the same sample by HPLC. In that way you could know the refrence value for that specific sample and set it as the reference value.

Unless completely impossible, I would first analyse your samples by NIR then I would analyse that same sample by HPLC.

When you later build your model, for example with PLS or ANN, the model should "automatically" account for the variation of the reference method (the model created accounts for the NIR variation and the reference method variation).

If that is impossible to do now, just properly validate your model and compare the variance of the NIR method with the HPLC method. This way you can have an idea of the variance of the NIR method vs variance of HPLC method.

Ahmed's picture

Dear Sir,

Thank you for your prompt response.

Actually you are totally right. It is better to do NIR first then do the reference analysis by HPLC to use the exact reference value. But I am making analysis for already manufactured and analyzed batches and I don't have the cost to re-analyze it by HPLC, so I was wondering if I can input the variation of the QC reference values of the batches when entering my reference value of each batch (Each batch has a different theoritical conc.) instead of repeating the HPLC analysis 

Ahmed Ramadan

Natural Wellness Egypt Scientific Office  | Technical Project Assistant Manager

 

gabiruth's picture

Hi,

It is truly important that people that are new to any type of secondary calibrations study some fundamental s about it. 5 samples can be used only as a very preliminary approach - if it works, it is necessary to do real scope work. If it fails, it is necessary to investigate why, and if the reasons explain the why, maybe it simply won't work. Since we don't have any clue on the range over which the calibration was done, about the nature of the product I can't comment any more, because any suggestion I would make would be presumptous on my part, and I have only been doing NIR for 21 years. 

 

Gabi Levin, Ph.D.

Brimrose Corp. of America

Baltimore MD

 

Ahmed's picture

Dear Sir,

Thank you for your prompt help.

Actually I was just explaining the case without stating the real number of samples or serial concentrations I am dealing with. I am a pharmacist statistian and I am working on performing multivariate analysis on pharmaceutical products for just 5 years, so defenitly I have a long road to walk and also alot to learn. My point of research for masters degree of pharmaceutics and Industrial pharmacy is about applying NIR chemometrics to evaluate the chemical and physical attributes of solid dosage forms. For physical attributes I made 20 formula of the same product and I made calibration for disintegration and comparative dissolution results with the corresponding NIR spectra. I am now formulating 5 different strengths of the same product (90%, 95%, 100%, 105% and 110%), then I will do NIR analysis for 5 tablets (each tablet will be analyzed 3 times) from each strenght. After this I will make calibration and validation with both leave one out cross validation and true validation with unknown samples to me. 

Ahmed Ramadan

Natural Wellness Egypt Scientific Office  | Technical Project Assistant Manager

 

gabiruth's picture

Dear Mr. Ramadan

 

Thanks, the situation is alittle bit more clear now, but still needs more for any advice. There is a fundamental point I wish to bring out right away. Using the 90 to 110% is a serious misguiding informattion. 

Here is why

1. Tere is a basic rule for calibration of a secondary method (UV, NIR, NMR etc) uses reference data from a primary such as HPLC..The rule is that the uncertainty in the reference data shall not exceed 4% (some would say 5%) from the range over which the calibration is made. 

2. If the active in a tablet is 10 weight %, then the range of 90 to 110 is 2 w%. It means that tablet with 9 to 11 are acceptable. For this  range, the reference data shall be better than 2X0.04 = uncertainty of 0.08 or +/-0.04. HPLC can easily deliver better quality refererence. 

3. If the active is onlly 1%, the range becomes 0.2% and the uncertainty must be better than 0.2X0.04 = 0.008 or +/-0.004 and many times it is questionable if HPLC can do it. 

4. If the active is only 0.2% the range becomes 0.04 and the uncertainty must be better than 0.0016 or +/- 0.0008 which most likely the HPLC will not be able to dleiver. 

So, before any effort into using a secondary method calibration is undertaken, we need to know the: 

Range over which we expect to creaate a clibration

Expected uncertainty in the reference method

Determine if there is real feasibility to achieve useful calibration by the secondary method. 

I can cite a case where I advised a pharma company not to waste time trying to do tablet analysis by NIR for tablets with 0.5 and 1 weight % due to above reasons, after more than 1000 tablets scanned and analyzed - they stopped the effort. I could have used the money they spent in a much better way. 

Important additional point:

Tablets are sold in mg dose - not in w% - NIR doesn't know what is 5 or 10 or 20 mg. It only knows %. A tablet with 10 mg that weighs 100 mg, is 10%, a tablet with 20 mg that weighs 200 mg is same 10%. 

One more point - low % tablets may contain the needed 2 mg - but these 2 mg will not necessarily be evenly distributed within the volume of the tablet. Thus, whne dealing with low % we must be sure that we scanned the whole volume of the tablet, otherwise we will never achieve reliable results.

Gabi Levin

Brimrose Corp. of America 

Home of the Luminar series process AOTF NIR spectrometers

 

Ahmed's picture

Dear Dr. levin,

Thank you for your valuable comments and advices. Please find below comments regarding the above points;

First, reference to the instrument manual (Bruker MPA), the LOD of instrument is 0.1% which differ typically with each molecule and product. However, I am making my research on 100 mg tablet and the API content is 25 mg for which I am not facing that issue.

Second, when performing multivariate analysis one must consider all variables that could affect the dependent variable he gonna predict. in other words, your model must be trained and familiar with all other variables so any change in these variables would not affect the accuracy of the prediction. In my case I am performing content test. The content result unit is mg of API within each tablet. The other varibles that affect the prediction are (Water content, weight of tablet, uniformity, impurites and supplier). from this point, the API comes from one supplier (it is not a variable). (Water content, weight of tablets and Impurities) are aleardy considered in the content result and also the range of these varibles is very resticted.

Finally, All of calibration models are regression analysis between the dependent variable and the spectra of the product which are affected by number of independent variables. the probIem comes from the co-linearity of the independent variables. I think I can put the reference values as mg or mg%(W/W) but if I put mg% (W/W) I have to measure weight of each tablet to get the final result of content test in mg. Moreover, I must train the model on the change of weight of tablet as a variable. so different sample with different weights will be entered with the same content result, and that will consequently exclude the risk of weight change on the accuracy of prediction.

Thank you again for your valuable comments and advices.

Best regards,

Ahmed Ramadan

Ahmed Ramadan

Natural Wellness Egypt Scientific Office  | Technical Project Assistant Manager

 

hlmark's picture

Gabi - Most of what you say is good, sound advice, but some of it is potentially confusing or misleading.

First: how a pharma manuf creates a product is not simply a matter of the concentration of the API in the mixture. True, 10 mg in a 100 mg taablet, or 20 mg in 200 mg tablet both represent 10 wt %, as you said. But the dosage a physician prescribes is normally given by the weight of API in a dose, and the physician typically doesn't care how that's delivered. The manuf can prepare various size doses using the same mixture simply by changing the amount of mixture used for each tablet (dose). This might have secondary effects on the product (e.g., rate of dissolution) but will not affect the dosage, once the tablet is formed.

Secondly, we've learned a lot about the behavior of NIR measurement, and the meaning of "concentration" for NIR analysis, since you and I were beginners in the field. Even without the complications caused by the optical scattering of powdered solids, "concentration", even in liqiuds, turns out to be a very nebulous concept. I recommend you read the following paper: Appl. Spect., 64(9), p.995-1006 (2010) for a discussion of the effects of varying meanings of "concentration" in liquids. The meaning of "concentration" in solids is still an open question, but I'm sure it's not going to be less complicated than for liquids.

Meanwhle, have a happy holiday season

 

\o/

/_\