Iic.cas.cz

A rapid quantitative assay of intact paracetamol tablets by
reflectance near-infrared spectroscopy
Andrew D. Trafford,*a Roger D. Jee,a Anthony C. Moffata and Paul Grahamb
a Centre for Pharmaceutical Analysis, School of Pharmacy, University of London,29/39 Brunswick Square, London, UK WC1N 1AXb Sanofi Research Division, Alnwick Research Centre, Willowburn Avenue, Alnwick,Northumberland, UK NE66 2JH Received 24th August 1998, Accepted 3rd December 1998
Near-infrared (NIR) reflectance spectroscopy was used to determine rapidly and non-destructively the content ofparacetamol in bulk batches of intact Sterwin 500 mg tablets by collecting NIR spectra in the range1100–2500 nm and using a multiple linear regression calibration method. The developed NIR method gave resultscomparable to the British Pharmacopoeia 1993 UV assay procedure, the standard errors of calibration andprediction being 0.48% and 0.71% m/m, respectively. The method showed good repeatability, the standarddeviation and coefficient of variation for six NIR assays on the same batch on the same day being 0.14 and0.16% m/m, respectively, while measurements over six consecutive days gave 0.31 and 0.36% m/m, respectively.
Applying the calibration to a parallel test set gave a mean bias of 20.22% and a mean accuracy of 0.45%. Thedeveloped method illustrates how the full potential of NIR can be utilised and how the ICH guidelines whichrecommend the validation of linearity, range, accuracy and precision for pharmaceutical registration purposes canbe applied. Duplicate determinations on bulk batches could be performed in under 2 min, allowing the potentialuse of the method on-line for real time monitoring of a running production process.
Near-infrared (NIR) absorption is mainly due to overtone and sation (ICH) guidelines12 to the assay. The ICH guidelines combination vibrations arising from fundamental vibrations in recommend the validation of linearity, range, accuracy and the mid-infrared region. As these absorptions are weak, NIR precision, both short and long term, for analytical procedures spectroscopy should be ideally suited for the analysis of intact which are to be used for pharmaceutical registration purposes.
tablets. No sample preparation is required and the method isnon-destructive. Surprisingly, the majority of published appli-cations of NIR spectroscopy to tablets have not exploited this Experimental
advantage, but used powdered tablets or solutions. Zappala andPost1 measured meprobamate in tablets after extraction into Materials
chloroform and ranitidine chlorohydrate,2 ranitidine hydro-chloride3 and cefuroxime acetil4 have all been measured as Forty-five batches of paracetamol tablets were used for the powdered tablet samples. An early example of an intact tablet study; 35 were normal production batches of Sterwin 500 mg analysis was a qualitative study5 to identify adulterated aspirin paracetamol tablets (nominal content 84.175% m/m para- tablets. Quantitative assays based on intact tablets have cetamol) with a nominal diameter of 12.7 mm and thickness of included the assay of amiodarone chlorohydrate,6 metoprolol,7 3.80–4.15 mm and 10 were development batches (76–92% m/m SB 216469-S8 and aspirin.9 Recently, we described the paracetamol, i.e., 90–110% of the nominal label content). The application of transmittance NIR spectroscopy to the analysis of development batches were of the same dimensions as the individual intact tablets10,11 and compared the effect of various normal production batches but were specially manufactured for pre-treatments on partial least-squares regression models; this study at a pilot scale on a small scale press, by altering the experiments involved the measurement of 20 intact tablets with content of paracetamol and the major excipient (starch). The an analysis time of 10 min per batch.
raw materials used were Paracetamol Fine PhEur, Potassium A problem with intact tablet assays is that normal production Sorbate PhEur, Povidone K-25 PhEur, Pregelatinised Starch batches do not encompass a sufficiently wide range for setting USNF, Starch Maize PhEur, Stearic Acid BP and Sterilised Talc up a reliable calibration equation. Tablets which are both under- PhEur, which were of the same specification as those used in the and overdosed with respect to the analyte need to be produced tablets and were obtained from Sanofi Research Division, without altering other factors that may affect the assay such as Alnwick Research Centre, Alnwick, Northumberland, UK.
particle size, moisture content and compaction characteristics.
All reagents were of analytical-reagent grade and were Although this complicates the calibration, as compared with obtained from BDH (Poole, Dorset, UK).
powdered tablet assays for which it is easy to prepare standards,the long-term gains of no sample preparation and the potentialfor the use on-line, allowing for real time monitoring of a UV analysis
running production process, make the effort worthwhile.
This paper describes a rapid quantitative assay for the UV measurements were made using a Perkin-Elmer (Beacons- determination of paracetamol in bulk batches of intact tablets field, Buckinghamshire, UK) Lambda 15 UV/VIS spec- using diffuse reflectance NIR spectroscopy and multiple trophotometer and 1 cm pathlength matched silica cells. Tablet wavelength linear regression. Where possible we applied the batches were assayed in duplicate using the following procedure recommendations of the International Conference on Harmoni- based on the British Pharmacopoeia 199313 assay. The Analyst, 1999, 124, 163–167
Pharmacopoeia method was modified because out of specifica- Reference analysis
tion tablets were being analysed. For the assay, 20 tablets wereground to a fine powder and accurately 0.15 g were measured Reference values (‘true’ values) for the paracetamol content of and transferred into a 200.0 ml calibrated flask. A 50 ml volume all batches of tablets were determined in duplicate using the UV of 0.1 m sodium hydroxide solution was added and the sample assay procedure. The precision of the UV assay procedure was was shaken for 10 min, then 100 ml of distilled water were estimated by pooling the results from the duplicate measure- added and the sample shaken for a further 10 min. The solution ments made on the 35 calibration and validation batches and was diluted to volume with distilled water, mixed and filtered calculating the standard deviation (s) using the equation and 5.0 ml of the solution were pipetted into a 500.0 ml calibrated flask. 50 ml of 0.1 m sodium hydroxide solution were added and the solution made to volume with distilled water. The absorbance was measured at 257 nm using 0.01 m sodium hydroxide solution as a blank. The content of paracetamol was calculated taking 715 as the value of A1% at the maximum at Where n is the number of duplicates, k1 and k2 are the individualduplicate results and ¯xk is the mean of the duplicates. Thestandard deviation was 0.31% m/m, giving a standard error (SE) NIR analysis
for a UV determined reference value of 0.22% m/m using theequation NIR reflectance spectra were measured using a NIRSystems 6500 near-infrared spectrophotometer (Foss NIRSystems, Maidenhead, Berkshire, UK) fitted with a reflectance detector(NR-6503), sample transport module (NR-6511) and coarsesample cell (NR-7080). The instrument was governed by NSAS(Near-Infrared Spectral Analyses Software) version 3.52 (Foss Method development and calibration
NIRSystems). Each measured spectrum was the average of 32scans and measured over the wavelength range 1100–2500 nm.
The NIR spectra were transferred to the Vision software along Batches were measured by pouring approximately 100 tablets with the mean UV assay values for the paracetamol content of into the coarse sample cell and scanning on the quarter full each batch. Absorbance spectra were treated mathematically by setting. Measurements were made in duplicate, the coarse performing a standard normal variate (SNV) transformation14 to sample cell being refilled with the same tablets between scans.
remove mutiplicative interferences of scatter and particle size Spectra were processed using Vision beta software (Foss effectively, followed by calculating the second derivative (segment size of 20 and a gap size of zero data points) tomaintain the peak locations but enhance the resolution. Fifteennormal production batches were assigned to a calibration set Results and discussion
and 10 to a validation set and forward search multiple linearregression (MLR), using the Vision program, was applied to the Feasibility study
data. A poor multiple correlation coefficient (R2) between theUV and NIR values of 0.694 was obtained at 1926 nm, which NIR reflectance spectra of all the tablet ingredients were only improved to R2 = 0.762 when a second wavelength was recorded as log (1/R), where R is the reflectance, and their added. This suggested that it was not possible to generate second derivative absorbance spectra were examined to identify acceptable calibration equations using only normal production any unique spectral features of the active constituent (para- batches owing to the limited range of paracetamol concentra- cetamol). The second derivative spectrum of paracetamol tions in normal production batches. To improve the calibration showed characteristic spectral features at about 1525 and equation, five development batches, which covered the ex- 1625–1675 nm; the minimum at about 1525 nm appeared to be tremes of concentration, were added to the calibration set and a the least affected by peaks from the excipients. When the second further five development batches added to the validation set.
derivative spectra of three development batches (76, 84 and Forward search MLR gave R2 = 0.971 at 1426 nm, which 93% m/m paracetamol, equivalent to 90, 100 and 110% of the improved to 0.974 when a second wavelength, 1528 nm, was nominal tablet content, respectively; Fig. 1) were compared, added. This second wavelength corresponded to the character- they had a minimum in this wavelength region which correlated istic spectral feature of paracetamol identified in the feasibility well with the paracetamol content in the tablets.
When the first wavelength was selected manually from the characteristic spectral region, 1530 nm gave the best correlationwith R2 = 0.860. When a second wavelength of 1426 nm,selected by the Vision software, was added, R2 improved to0.974, the same as the initial two wavelength MLR selected bythe Vision software. On addition of a third wavelength, R2 onlyimproved to 0.976, a minor improvement considering thesubsequent risk of overfitting the data.
The two wavelength calibration was therefore chosen and the Y = 87.22 2 35.11A1530 + 169.65A1426 where Y is the predicted paracetamol content (% m/m) and A1530and A1426 are the ordinate values of the transformed spectra(SNV plus the second derivative) at 1530 and 1426 nm,respectively. The fit had a residual sum of squares (RSS) of 3.89 Second derivative absorbance NIR spectra of intact paracetamol [eqn. (4)], giving a standard error of calibration (SEC) of 0.48% tablets: (a) 76, (b) 84 and (c) 93% m/m paracetamol content.
Analyst, 1999, 124, 163–167
versus UV assay values is shown in Fig. 3. Again, the 95% confidence intervals for the intercept (211.52 to 11.92) and slope (0.93–1.07) suggest that there is no evidence for relative Parallel test set
where y is the ‘true’ paracetamol content (i.e., mean UV assayvalue), n the number of batches and p the number of coefficients Further validation was performed on batches which were in eqn. (3). Fig. 2 shows a plot of predicted values versus independent of both the calibration and validation sets.15 The reference values for the calibration set. Ideally, the intercept (a) validation set used above was not totally independent of the and slope (b) should be 0 and 1, respectively, if there is no fixed calibration set as they were scanned at the same time. Validation systematic error or relative systematic error in the calibration batches are often used to tune a calibration procedure by, for equation. Linear regression was applied and the 95% con- example, choosing optimum wavelengths for a multiple regres- fidence intervals for the intercept [eqn. (6)] and slope [eqn. (7)] sion equation so the use of a parallel test set avoids overfitting.
Also the performance of the calibration on the parallel test set isa more reliable indicator of its future performance on normal production batches. A month after the calibration and validation sets had been measured, 10 totally independent production batches were used to evaluate the performance of the calibra- tion. The UV and NIR results (Table 1) were then compared using the paired sampled Student’s t-test [eqn. (10)] to show ifthe results were statistically equivalent.
where t is Student’s t at the 95% probability level and n 2 2degrees of freedom, RSD is the residual standard deviation xx is the sum of squares [eqn. (9)].
where d is the mean residual (NIR 2 UV) and sd is the standard deviation of the residuals. A value of 1.069 for tcalc wasobtained. The two-sided critical value of Students’ t at the 5% significance level for n 2 2 degrees of freedom was 2.262, suggesting that there was no evidence for a difference between The confidence interval for the intercept (23.56 to 8.08) included 0, and there was therefore no evidence to suggest anon-zero intercept. Similarly, the confidence interval for theslope (0.90–1.04) included 1, suggesting that there was no Accuracy
evidence for a relative systematic error in the calibrationequation.
For the purposes of this study, the accuracy was taken as howclose the NIR values were to the UV assay values. The standarderrors of calibration (0.48% m/m) and prediction (0.71% m/m) Validation of calibration equation
gave an indication of the accuracy of the NIR determination anda further estimate of accuracy was given by the standard The calibration equation was validated by using it to calculate deviation of the residuals obtained in the parallel test set (0.53% the paracetamol content of each of the batches in the validation m/m). As expected, these values were greater than the standard sample set. The standard error for prediction (SEP) was 0.71% error (0.22% m/m) for the UV measurement itself, although m/m (equation as for SEC with p = 0). A plot of predicted acceptable and therefore suitable for batch release purposes.
Plot of NIR predicted versus UV determined paracetamol content Plot of NIR predicted versus UV determined paracetamol content (% m/m) for the calibration set (r = 0.989).
(% m/m) for the validation set (r = 0.962).
Analyst, 1999, 124, 163–167
The mean bias [eqn. (11)] and the mean accuracy [eqn. (12)] for the parallel test set (n = 10) were determined to be 20.22 and0.45% with standard deviations of 0.63 and 0.48%, re- NIR determinations of paracetamol content (% m/m) there was no evidence that the analysts were not equallyprecise.
Repeatability
The short term precision (within-day) was determined bymeasuring the paracetamol content of a single batch six times Conclusions
within one day by both UV and NIR methods. The standarddeviation and coefficient of variation (CV) for the NIR assay The developed one step NIR method for the quantitative assay procedure were 0.14% m/m and 0.16%, respectively, and for the of paracetamol in intact Sterwin 500 mg tablets was rapid and UV assay procedure they were 0.41% m/m and 0.48%, statistical analysis of the data indicated that the NIR method was respectively. The mean (±95% confidence limit) UV and NIR comparable to the British Pharmacopoeia 1993 reference assay values were 84.53 ± 0.43% and 84.38 ± 0.14% m/m, respectively. The confidence intervals overlap, further suggest- The NIR procedure had the advantages over the reference ing that there was no evidence for a difference in values technique of requiring no sample preparation or the use of obtained by the two procedures (cf., parallel test set).
potential environmentally harmful reagents. Although the ICHguidelines were developed mainly for the validation of Intermediate precision
analytical procedures primarily based on the analyte being insolution, it was found possible to apply them successfully to the The between-day precision was measured by assaying a single validation of a reflectance NIR intact bulk tablet assay. This batch on six consecutive days by both UV and NIR methods.
paper illustrates how the full potential of NIR can be utilised and The standard deviation and CV for the NIR assay procedure the ICH guidelines applied to the quantification of active were 0.31% m/m and 0.36%, respectively, and for the UV assay ingredients in bulk samples of intact tablets and it is hoped that procedure they were 0.52% m/m and 0.61%, respectively. As in this will lead to others developing rapid tablet assays. Although the repeatability study, the mean (±95% confidence limit) the calibration process requires the preparation of intact tablets UV and NIR results, 84.52 ± 0.54% m/m and 84.15 ± 0.32% covering the calibration range and analyses by a reference m/m, respectively, gave no evidence for a difference in the method, once the calibration has been established duplicate quantitative determinations on bulk batches can be performed in The between analyst precision with the NIR procedure was determined by six different analysts testing the same batch six The mean bias (20.22%) and mean accuracy (0.45%) were times on the same day (Table 2). Reference scans were taken comparable to those for the assay on individual intact tablets by between the different analysts’ scans and the coarse sample cell transmission NIR spectroscopy,10 which had values of 20.08% was refilled between each scan. The hypothesis that all the and 0.59%, respectively, but was five times quicker and analysts’ results were equally precise was tested using Co- therefore ideally suited to the analysis of bulk tablets on-line, chran’s test.16 This test compares the largest variance (s2) with allowing real time monitoring of a running production proc- the other variances by dividing the largest variance by the sum of all the variances. This was determined to be 0.413, which was A method similar to the above using a sample transport less than the critical value of 0.445 (5% significance level), so module, coarse sample cell and a reflectance detector has been Analyst, 1999, 124, 163–167
accepted by the Medicines Control Agency for the assay of S. M. Han and P. G. Faulkner, J. Pharm. Biomed. Anal., 1996, 14,
paracetamol in intact Sterwin tablets.
P. Merkle and K. A. Kovar, J. Pharm. Biomed. Anal., 1997, 17,
The authors thank Foss NIRSystems for the loan of the NIRSystems 6500 spectrophotometer and Sanofi, Eli Lilly and A. Eustaquio, P. Graham, R. D. Jee, A. C. Moffat and A. D. Trafford, Bristol-Myers Squibb for funding a research studentship.
Analyst, 1998, 123, 2303.
M. Blanco, A. Eustaquio, R. D. Jee and A. C. Moffat, Anal. Chim.
Acta, in the press.
International Conference on Harmonisation (ICH) of Technical References
Requirements for Registration of Pharmaceuticals for HumanUse, Validation of Analytical Procedures: Methodology,Recommended for Adoption at step 4 of the ICH Process on A. F. Zappala and A. Post, J. Pharm. Sci., 1977, 66, 292.
6 November 1996 by the ICH Steering Committee. Internet address: P. Corti, E. Dreassi, G. Corbini, S. Lonardi, R. Viviani, L. Mosconi and M. Bernuzzi, Pharm. Acta Helv., 1990, 65, 28.
British Pharmacopoeia 1993, H.M. Stationery Office, London, 1993, E. Dreassi, G. Ceramelli, P. Corti, P. L. Perruccio and S. Lonardi, Analyst, 1996, 121, 219.
R. J. Barnes, M. S. Dhanoa and S. J. Lister, Appl. Spectrosc., 1989, E. Dreassi, G. Ceramelli, L. Savini and P. Corti, Analyst, 1995, 120,
S. A. Halsey, Technical Note, NIRSystems, Silver Spring, MD, R. A. Lodder and G. M. Hieftje, Appl. Spectrosc., 1988, 42, 518.
R. Jensen, E. Peuchant, I. Castagne, A. M. Boirac and G. Roux, W. Caulcutt and R. Boddy, Statistics for Analytical Chemists, Spectrosc. Int. J., 1988, 6, 63.
Chapman and Hall, London, 1983, p. 70.
J. Gottfries, H. Depui, M. Fransson, M. Jongeneelen, M. Josefson,
F. W. Langkilde and D. T. Witte, J. Pharm. Biomed. Anal., 1996, 14,
1495.
Analyst, 1999, 124, 163–167

Source: http://www.iic.cas.cz/~grygar/mapl-R_Paracetamol.pdf