Journal Club: Validation of Raman spectroscopy for content determination in tablets

Written by Dr. Janet Thode Posted in Method validation

The various techniques of Raman spectroscopy find a wide range of applications in the different areas of the pharmaceutical industry, be it for the detection of counterfeits, as an in-process control (IPC) during the production of the active pharmaceutical ingredient, for content determinations, as an ID test for vaccines, for the simultaneous determination of several product quality attributes (PQAs) as part of a rapid release strategy, up to sterility tests, to name just a few examples [1-4].

And also for method validation, a wide variety of concepts exist, depending on gusto and regulatory requirements [5], such as e.g. the "accuracy profile" [6]. In today's blog post, however, we’d like to take a look at the (still) classical method validation according to the guideline for method validation ICH Q2(R1) and have therefore scrutinized the publication by Li et al. on a transmission Raman spectroscopic (TRS) method for content determination of niacinamide tablets.


Test principle of the analytical method and background

Raman spectroscopy, like infrared spectroscopy, is one of the vibrational spectroscopic analysis methods, whereby the interaction of light with matter provides insight into the properties of the material under investigation. In Raman spectroscopy, the sample is irradiated by a laser with monochromatic light. When the light hits the material of the sample, the light is scattered, whereby most of the light of the irradiated wavelength remains unchanged (= elastic scattering / Rayleigh scattering), but a very small part changes its wavelength (inelastic scattering / Raman scattering). This emission of light of a different wavelength is based on the interaction of the photons of the light with the molecules of the sample, whereby either energy is transferred from the light to the matter or vice versa. This energy transfer results in the shift of the wavelength of the scattered light with respect to the irradiated light since the wavelength depends on the energy of the light. Plotting the intensity of the inelastically scattered light against the energy change (frequency difference / Raman shift, given in wavenumbers) results in the Raman spectrum.

In contrast to the widely used backscattering Raman spectroscopy, transmission Raman spectroscopy offers the advantage of being able to make a representative statement about the entire sample due to the passage of the laser through the complete sample (instead of only hitting a small punctual surface) and thus also being insensitive to surface coatings.

Various chemometric techniques of multivariate data analysis, such as partial least squares regression (PLSR), can be used for data evaluation. This is a statistical technique that generates a predictive model for linear data. A regression is calculated from many x variables (here: a spectrum with many wavelengths) to one (or more) y variable (response variable, here: content). However, not the original measured data set is used, but the x-data are decomposed into principal components, where also the response variable y is already included... If this sounds too statistical, I feel the same way ???? Conclusion: Good technique for modeling the relationship (à calibration) between spectral measurements and a response variable such as chemical composition or content. In the work here, the calibration was done with calibration samples from which both the Raman spectra were recorded and their respective niacinamide content was determined using HPLC as a reference method. Related to PLSR, the two abbreviations RMSEP and RMSEC should also be mentioned. They stand for root mean square error of prediction and calibration, respectively, and are measures of the quality of the model analogous to RSS in normal linear regression. RMSEP provides information on the predictive ability of the model.

To cover a high variability, the calibration samples were prepared according to the principles of design of experiments (DoE). DoE is a structured procedure for efficiently investigating the relationship between many influencing variables of a process (here: analytical method) and their results, and thus using the cause-and-effect relationship to determine only the influencing factors that are really relevant for the process from a large number of parameters. In contrast to the OFAT (one factor at a time) method, the DoE changes various potential influencing factors simultaneously. In the work presented here, a 3 factor-3 level full factorial design was used to prepare the calibration samples. 3 concentrations of the drug substance (DS) niacinamide (80-120%) and 3 concentrations of the main excipient microcrystalline cellulose (MCC) were considered, as well as 3 concentrations of relative humidity and 2 compression forces for tablet production (à tablet hardness).


Analytical method validation and discussion

Preliminary remark

In addition to the calibration samples already mentioned (n = 100: 90 lab-scale + 10 production-scale), further independent "sets" of tablets with different characteristics were prepared for different purposes. Thus, a set of "test samples" (n = 36, lab-scale, DS and MCC conc.: 90-110% each) was used to verify the evaluation model established with the calibration samples. To evaluate the specificity, a "specificity set" (n = 16, lab-scale, DS conc.: 90-110%) was prepared which also contained the degradant niacin. The last set forms the "validation set" (n = 34, production-scale, DS conc.: 85-115%), where the DS concentration was expanded beyond the range of 90-110% prescribed by the USP for niacinamide tablets.


To evaluate specificity, on the one hand, the Raman spectra of the "specificity set" tablets were compared with the Raman spectrum of pure niacinamide and visually found to be in high agreement. On the other hand, a comparison of the Raman spectra of two variables of the PLSR model shows a clear difference to the degradant niacin. Furthermore, a RMSEP value of 4.31% indicates a good prediction for niacinamide even in the presence of niacin.


To investigate precision, 6 tablets of each concentration of the validation set (85%, 100%, 115%) were used. For repeatability, 2 aspects were investigated: without and with repositioning. The term repositioning refers to flipping (front-to-back) of the tablet in the sample tray and changing the coordinates for the exposure point of the laser on the tablet. For intermediate precision, analysis without repositioning was performed on a second day by a different analyst. For evaluation, the relative standard deviation (RSD) for each concentration level was calculated for all 3 studies and values of less than 2% were obtained. Unfortunately, the raw data are not shown, and the associated table indicates n = 6, so it is not clear whether the intermediate precision really takes into account all 12 results (6 from day 1 and 6 from day 2) or if it refers only to the results of day 2 and thus actually represents only a repeated repeatability precision...


Linearity was evaluated with 5 concentrations as required by the ICH Q2(R1) guideline for method validation. Interestingly, however, in this work data from the calibration set were used for this purpose along with either data from the test set or the specificity set or the validation set. All include the range of 80-120% required for assays, but this is based on the calibration set data. This approach is not quite independent... Actually, in this case I would have liked to see a new study with 5 concentrations of a validation set covering a range of 80-120% and where the content of the validation samples would be determined based on the PLSR model set up with the calibration data... The coefficient of determination R2 for the lines shown ranges from 0.93 (cal. + spec. set), 0.95 (cal. + test set), 0.96 (3-point calibration set) up to 0.98 (cal. + val. set). Residual plots promising good results are mentioned in the publication, but unfortunately not shown.


To determine the trueness, no new experiments were performed, but instead the linearity data of the 4 data sets (calibration set, cal. + test set, cal. + spec. set, cal. + val. set) were analyzed regarding RMSEC, RMSEP and "Bias-Pred". The RMSEs are all less than 5%, indicating very good modeling. "Bias-Pred" is not specified, but the name suggests that it refers to the systematic error in the prediction of the model. Very low values of max. 2.1% were also observed for this, indicating very good trueness of the PLSR model.

Since unfortunately, as already mentioned above, for the precision experiments the determined content values are not shown in the publication, trueness can’t be back-calculated on our own...


In the publication’s introduction, fluorescence / photobleaching, tablet thickness and temperature effects are mentioned as factors influencing the variability of transmission Raman spectroscopic analyses. For the evaluation of robustness, however, only the aspect of photobleaching was examined more closely here. An omission of an investigation on different tablet thicknesses might be justified by the fact that approximately similar and only slightly varying tablet thicknesses can be expected during production under GMP conditions. However, why different temperature effects were not investigated is not clear to me at first glance. Possibly, although performed here in an academic and therefore non-controlled environment, the authors may have had in mind the subsequent application under controlled temperature conditions, as is common in pharmaceutical laboratories... But back to photobleaching: for this, one tablet (without specification of its DS concentration) was irradiated 20x in succession with a laser and the 20 spectra were recorded. Visually, a slight change in the baseline of the Raman spectra was observed with increasing numbers of irradiation. In another figure, the determined niacinamide content of one tablet (unclear whether this is the previous one) was plotted against the number of 20 scans and the content deviates from the target concentration by about 1% at most, i.e., there is no negative influence of photobleaching by 20 scans on the analytical result of sample measurements. Why here, however, the target concentration of the DS of this tablet is 36% instead of 40%, which was assumed to be the 100% level in the DoE, I cannot quite understand...


In addition to its use in release analysis, transmission Raman spectroscopy is particularly suitable as a stability-indicating method for stability studies of drug products, since - as a non-destructive technique - it would allow repeated measurements of the same sample at different times and could thus contribute to save many stability samples.

The basic procedure of the PLSR modeling described here with calibration samples, its fine-tuning with test samples and subsequent validation under consideration of a reference method even corresponds to the current considerations on multivariate analytical methods of the draft ICH Q2(R2), although some weaknesses of the studies published here should be noted:

  • No use of a completely new sample set for validation, but mixing of the calibration and validation sample sets
  • Investigation of robustness using only one single tablet
  • No investigation of specificity with stressed samples, which would have been desirable for a stability-indicating method
  • Unclear or missing data presentation (à intermediate precision, residual plots)
  • Obvious error in the legend of a figure
  • it is not obvious if acceptance criteria were defined

Apart from that, all validation parameters required by the ICH Q2(R1) guideline for method validation for an assay have been covered (more or less, see above). Furthermore, the publication is written in a very understandable manner and clearly explained, so that it was fun to deal with Raman spectroscopy, PLSR and DoE.



[1] Ren J., Shijie M., Lin J., Xu Y., Zhu Q., Xu N. (2022) Research Progress of Raman Spectroscopy and Raman Imaging in Pharmaceutical Analysis, Curr Pharm Des, epub

[2] Silge A., Bocklitz T., Becker B., Matheis W., Popp J., Bekeredjian-Ding I. (2018) Raman spectroscopy-based identification of toxoid vaccine products, NPJ Vaccines (3), p.50

[3] Wei B., Woon N., Dai L., Fish R., Tai M., Handagama W., Yin A., Sun J., Maier A., McDaniel D., Kadaub E., Yang J., Saggu M., Woys A., Pester O., Lambert D., Pell A., Hao Z., Magill G., Yim J., Chan J., Yang L., Macchi F., Bell C., Deperalta G., Chen Y. (2022) Multi-attribute Raman spectroscopy (MARS) for monitoring product quality attributes in formulated monoclonal antibody therapeutics, MAbs 14(1), p.2007564

[4] Grosso R.A., Walther A.R., Brunbech E., Sørensen A., Schebye B., Olsen K.E., Qu H, Hedegaard M.A.B., Arnspang E.C. (2022), Detection of low numbers of bacterial cells in a pharmaceutical drug product using Raman spectroscopy and PLS-DA multivariate analysis, Analyst, epub

[5] Wulandari L., Idroes R., Noviandy. T.R., Indrayanto G. (2022) Application of chemometrics using direct spectroscopic methods as a QC tool in pharmaceutical industry and their validation, Profiles Drug Subst Excip Relat Methodol (47), p.327-379

[6] Mansouri M.A., Sacré P.Y., Coïc L., De Bleye C., Dumont E., Bouklouze A., Hubert P., Marini R.D., Ziemons E. (2020) Quantitation of active pharmaceutical ingredient through the packaging using Raman handheld spectrophotometers: A comparison study, Talanta (207), p.12030 (2022), Detection of low numbers of bacterial cells in a pharmaceutical drug product using Raman spectroscopy and PLS-DA multivariate analysis, Analyst, epub