News from Europe's pharmaceutical manufacturing industry coupled with upcoming events, and exclusive articles and interviews from industry experts. WEEKLY
Furosemide is a potent loop diuretic used in the treatment of edematous states associated with cardiac, renal, and hepatic
failure and for the treatment of hypertension. Complications, such as erratic systemic drug availability from the oral route
of administration and from unpredictable responses to a given dosage, appear frequently. Furosemide's exact mechanism of action
is not fully understood, but the drug is believed to act at the luminal surface of the ascending limb of the loop of Henle
by inhibiting the active reabsorption of chloride ions. The response to a given dosage is modulated by the individual's fluid
and electrolyte balance (1). Constant biopharmaceutical quality of furosemide tablets is therefore imperative to minimize
undesirable in vivo variability.
Regulatory agencies around the world require proof of product consistency and a high degree of assurance that a product will
meet all specifications (2–4). Thus, process evaluations need to follow scientific and statistical rationales. Simple test
comparisons might not be sufficient. FDA strongly emphasizes that the pharmaceutical industry must understand process variation,
including all sources and degrees of variation, and ultimately the effect of variation on product attributes (5). Furthermore,
FDA guidance states that these data should be collected from the process-design stage through final-product manufacturing.
This recommendation is a clear shift toward a product life-cycle approach including quality by design (5).
Figure 1: Sampling scheme of the bin blender. (ALL FIGURES ARE COURTESY OF THE AUTHORS)
The adoption of statistical tools to evaluate data from an existing process can expose variability that might reveal that
this process is not robust. On the other hand, a non-optimized process that is on its target value might be modified to improve
productivity or increase robustness. Process variability can easily be revealed using control charts and statistical analysis.
Figure 2: Control chart for individual (I), move range (MR), and standard deviation (StDev). The subgroup size was 10 (I-MR-R/S)
for furosemide content in the powder mixture. LCL is lower control limit, S is standard deviation, UCL is upper control limit,
and X is individual values.
A prerequisite for process evaluation is the determination of the analytical method's variability. This determination can
be accomplished using repeatability and reproducibility (RR) studies. Repeatability, sometimes called "equipment variation," is the ability of the measurement system to provide consistent readings when used
by a single technician or operator. Reproducibility, sometimes referred to as "appraiser variation," is the ability to achieve consistent results for multiple operators. In general,
an interval of 10% < %RR < 30% is considered adequate (6).
Table I: Furosemide content in the powder blend.
Current FDA guidance embraces a risk-based approach, and the International Conference on Harmonization's Pharmaceutical Quality Systems recommends continuous improvement of the process performance and product quality (7–9). Statistical tools can enhance process
understanding and foster innovative approaches to process validation and pharmaceutical development (5, 10). Among the statistical
tools, the process-capability indices (i.e., Cp and Cpk) measure the process's ability to manufacture products that meet specifications
and requirements. These indices greatly simplify the management of statistically controlled processes and have been used with
the fundamental assumptions that the data are distributed normally, that the process is stable, and that its variability is
known (6). The goal of the data evaluation was to assess the potential process capability index (i.e., Cp) and the actual
process capability index (i.e., Cpk) using the content homogeneity of the powder mixture, tablet weight, dosage-unit uniformity,
and dissolution behavior of 40-mg furosemide tablets.
Materials
Table II: Values of the Anderson–Darling statistics (AD) and p-values performed for powder mixture conformity, uniformity
of dosage units, and % released.
Furosemide was supplied by Alcon Biosciences. Crospovidone was provided by ISP. Lactose monohydrate was purchased from Doremus.
Talc was supplied by Indukern, and magnesium stearate by Inbra. The authors bought pregelatinized starch from Colorcon and
maize starch from Cargill. Purified water was supplied by Prati-Donaduzzi. All the materials were of Brazilian Pharmacopoeia
grade, as stated by the suppliers.
Methods
Table III: Estimated variance component for furosemide content in the powder blend.
Manufacturing process of 40-mg furosemide tablets.
Three consecutive 380-kg batches of 40-mg furosemide tablets were manufactured by wet granulation. The powders were mixed
in a bin blender (VZD-400, Vanguard) for 20 min. at 18 rpm (see Figure 1). Next, powders were sieved through a 0.5-mm sieve
and loaded into the high-shear mixer (MIC-P, Comasa, Buenos Aires). The granulation liquid was then sprayed tangentially into
the moving powder mixture using a pneumatic atomizer at 1.0 bar atomizing air pressure and a speed of 80 rpm. The powder was
mixed for 5 min. before the granulation was started. The spray rate was 40 g/min. The granules passed through a 2.5-mm sieve,
dried at 50 ą3 °C in an oven for 24 h, passed through a 1.25-mm screen, and finally lubricated with magnesium stearate. The
compression was performed employing a 50-station double rotary tableting machine (2000/50, Lawes, Săo Paulo). The speed was
kept constant at 75,000 tablets/h. The Lawes tableting machine was dedicated equipment to produce furosemide tablets only.
Figure 3: Capability analysis of the content (mg) of furosemide in the powder mixture. CL is control level, CP is process
capability, Cpk is process-capability index, CPL is process capability relative to lower specification limit, CPU is process
capability relative to upper specification limit, LSL is lower specification limit, PPM is parts per million, StDev is standard
deviation, and USL is upper specification limit.
Furosemide assay by ultraviolet-visible spectrophotometer.
The assays were performed in duplicate according to the Brazilian Pharmacopoeia (11). The ultraviolet-visible (UV-vis) spectrophotometer (800XI, Femto, Săo Paulo), UV λ = 271 nm, was used, and the standard
and sample concentrations of furosemide were diluted to 0.6 mg/mL using 0.1 N HCl.
Figure 4: Control charts of individual and moving range (MR) of the individual tablet weights (mg) for the left and right
sides of the tableting machine. LCL is lower control limit, LSL is lower specification limit, UCL is upper control limit,
USL is upper specification limit, and X is individual value.
Dissolution of furosemide tablets.
The dissolution test of furosemide tablets from the different batches (i.e., at the beginning, in the middle, and at the end
of the process) was carried out in triplicate using USP Apparatus 2 (paddle method). The dissolution test was performed using 900 mL of pH 5.8 phosphate buffer at 37.0 ą0.5 °C at
50 rpm. Aliquots (5 mL each) were withdrawn at a predetermined time interval of 60 min. The samples were filtered through
a 0.45-μm membrane filter. The furosemide assay was performed by spectrophotometer UV-vis (Femto) UV λ = 271 nm. The USP 32 specification for dissolution of furosemide tablets is not less than 80% (Q) of furosemide dissolved in 60 min in 900
mL of pH 5.8 phosphate buffer (12).
Table IV: Tablet weights (mg) for left and right sides of the tableting machine for the three batches.
Sampling plan and statistical analysis.
The sampling plan included the collection of 10 g of the powder mixture in 10 specific locations (see Figure 1). For the tablets,
a minimum of 200 units on the right and left side of the double rotary tableting machine were collected. The uniformity of
dosage units of furosemide was calculated using 30 tablets per batch (i.e., 10 tablets in the beginning, 10 in the middle,
and 10 at the end of the tableting process). For the dissolution test, a total of six tablets for each batch—at the beginning,
in the middle, and at the end of the process—were analyzed. The statistical analysis of the three consecutive batches was
evaluated by using Minitab software, version 15 (Minitab, State College, PA).
Figure 5: Capability analysis of the tablet weights for three batches. Cp is process capability, Cpk is process-capability
index, CPL is process capability relative to lower specification limit, CPU is process capability relative to upper specification
limit, LSL is lower specification limit, PPM is parts per million, StDev is standard deviation, USL is upper specification
limit, and X is individual value.
The process stability was evaluated using individual and moving-range charts as well as standard-deviation charts (I-MR-R/S),
considering a subgroup size of 10. The normal distribution was evaluated by the Anderson–Darling test. For non-normal distributions
according to the Anderson–Darling test, the Box–Cox family of power transformations was used to obtain an approximate Gaussian
distribution (13).
Table V: Tablet weight of furosemide (mg) versus left and right side of the tableting machine.
The process capability indices were calculated when the analyzed parameter was normally distributed or when its distribution
was close to the normal distribution (6, 14). The fully nested analysis of variance (ANOVA) was performed to estimate variance
components for each response variable (i.e., mixture content, tablet weight, dosage-unit uniformity, and dissolution). All
factors were assumed to be random. The mean comparison between the tablet weights from the two sides of the tableting machine
was performed using one-way ANOVA.
Results and discussion
Table VI: Values of the uniformity of dosage units.
Content of furosemide in the powder mixture.
The relative standard deviation of the method was 0.43%. Random 10-g samples of the mixed powder were taken from 10 places
within the bin blender (see Figure 1). Table I shows the data of the three consecutive batches. The powder blends met the
acceptance criteria and the specifications (90.0–110.0%). The control charts did not show a special cause of variation (see
Figure 2). Considering the analysis of locations, the one-way ANOVA was used to test for differences among the locations (see
Figure 1). The analysis revealed no statistical difference between the locations (p value = 0.959).
Figure 6: Control chart for individual, move range (MR) and standard deviation (StDev). The subgroup size was 10 (I-MR-R/S)
for the uniformity of dosage units. LCL is lower control limit, S is standard deviation, and UCL is upper control limit.
The p value for the Anderson–Darling normality test was 0.069 (> 0.05), indicating a normal distribution (see Table II). This test
was developed to be especially sensitive to deviations from normality in the distribution tails. For capability analysis,
the tails are the most critical part of the distribution (15). To understand process variation, the authors performed a fully
nested ANOVA (see Table III). The result showed that 72.90% of the observed variation in the furosemide content resulted from
a batch factor. Thus, to minimize the process variability, the causes of the variation among batches required further investigation.
Figure 7: Capability analysis of the uniformity of dosage units of furosemide tablets. Cp is process capability, Cpk is process-capability
index, CL is control level, CPL is process capability relative to lower specification limit, CPU is process capability relative
to upper specification limit, LSL is lower specification limit, PPM is parts per million, StDev is standard deviation, and
USL is upper specification limit.
The indices of 2.19 and 2.23 for Cpk and Cp, respectively, revealed that this process is statistically centered and robust
(see Figure 3). The estimated nonconformity for the powder mixing process was less than 1 ppm. Cp and Cpk indices equal or
above 1.0 correspond to a satisfactory low proportion of nonconformity (16). However, for a good process under statistical
control, Cpk should be greater than 1.5 (17).
Table VII: Estimated variance component for uniformity of dosage units of furosemide.
Evaluation of the tablet weight.
Figure 4 shows the individual tablet weights for each batch taken from the left and right sides of the tablet machine. The
tablet weight was chosen as a surrogate for the process stability of the compression step. As shown in Figure 4, the two values
above and the five values below the control limits, among 1200 samples analyzed, cannot support the assumption that this process
was unstable. The mean tablet weight was 164.22 ą4.25 mg. In addition, all values were within the specification limits. The
lower specification limit (LSL) was 152 mg, and the upper specification limit (USL) was 177 mg (see Table IV). However, a
few outliers can have a large influence on the process-capability indices, as evidenced in Figure 5. The process capability
indices were 1.00 and 0.98 for Cp and Cpk, respectively, revealing a high proportion of nonconformity (2751.21 ppm). Although
interbatch variability could have contributed to the results, further investigation to address the high nonconformity was
needed. Issues that may cause weight variation are powder flow problems, improper die fill, and powder size distribution.
Table VIII: Values of the % drug released using tablet dissolution of furosemide tablet at the beginning.
In addition, a comparison of the tablet weights at the two sides of the tableting machine was performed (see Table V). The
p value (one-way ANOVA) was 0.522, showing that the tablet weights for both sides did not differ significantly. The residuals
plot indicated a normal distribution and the absence of special causes of variation. Similar results were reported in a study
where a double-station Kilian tableting machine (IMA Kilian, Köln, Germany) was used to manufacture metamizol tablets (18).
Figure 8: Control chart for individual, move range (MR), and standard deviation (StDev). The subgroup size was 18 (I-MR-R/S)
for the % released using tablet dissolution. LCL is lower control limit, S is standard deviation, UCL is upper control limit,
and X is individual value.
Evaluation of the uniformity of dosage units.
Table VI shows the dosage-unit uniformity for each batch for three different time periods: the beginning, middle, and end
of the process. No special causes of variability were observed in the control charts (see Figure 6). The p value for the Anderson–Darling test was equal to 0.010, which revealed a non-normal distribution (> 0.05, see Table II).
The Box–Cox method was used to transform the values into a normal distribution (λ = 4.5) (13). In a similar way, non-normal
process capability indices were determined using a generalized λ distribution described by Pal (19). The authors used spreadsheets
to illustrate how easily the necessary calculation can be performed. The transformed values were used to calculate the process-capability
indices. Cp and Cpk were 1.43 and 1.27, respectively. The difference in values indicated that the process was not statistically
centered. However, the predicted nonconformity rate was low (70.32 units), as shown in Figure 7.
Figure 9: Capability analysis of % released using tablet dissolution. Cp is process capability, Cpk is process-capability
index, CPL is process capability relative to lower specification limit, CPU is process capability relative to upper specification
limit, LSL is lower specification limit, PPM is parts per million, StDev is standard deviation, and USL is upper specification
limit.
The estimated source of variance showed that 54.19% of the observed variability was within a sampling group, 27.83% resulted
from the sampling time point, and 17.98% resulted from the batch (see Table VII). To optimize this manufacturing process,
the causes of the variations among the batches and the statistically noncentered profile of this process should be investigated.
Table IX: Estimated variance component for the % drug released using tablet dissolution of furosemide.
Evaluation of the dissolution of furosemide tablets.
The dissolution of furosemide tablets was evaluated for each batch at the beginning, the middle, and the end of the process
(see Table VIII). Figure 8 indicates that the process stability was achieved because no special cause of variation was observed.
The p value for the Anderson–Darling test was equal to 0.006, which confirmed a non-normal distribution (see Table II). The transformed
data (Box–Cox method, λ = 5) were used to calculate Cp and Cpk, which were 3.46 and 2.29, respectively. The estimated nonconforming
proportion was low (0.00 ppm), and the process was slightly not statistically centered (see Figure 9). The estimated cause
of variance showed that 40.34% resulted from a batch factor, and 17.23% from the sampling group (see Table IX).
Conclusion
The statistical approach used in the process evaluation of the blending, tableting, dosage-unit uniformity, weight variation,
and dissolution behavior led to better process understanding of the manufacturing process. The results showed that fully nested
ANOVA is a powerful tool to identify sources of variability. The process capability indices helped the authors to understand
process performance and the potential for process optimization. Although a limited number of batches were investigated, the
statistical methods identified possible approaches for process improvement in the manufacturing of furosemide tablets.
Túlia de Souza Botelho is a student, Vanessa Franco Tavares is a student, Cátia Panizzon Dal Curtivo is a student, and Nádia Araci Bou-Chacra* is an assistant professor of pharmaceutics, all at the Faculty of Pharmaceutical Sciences, University of Săo Paulo, 580 Lineu
Prestes Ave., Butantan, Săo Paulo, SP – Brazil 05508-900, chacra@usp.br . Silvie Rosa Balzan Sarolli is a quality-assurance employee, Márcio Adriano Fernandes is a quality-control employee, and Carmen Maria Donaduzzi is a research pharmacist, all at Prati-Donaduzzi. Raimar Löbenberg is an associate professor of pharmaceutics at the University of Alberta.
*To whom all correspondence should be addressed.
Submitted: Aug. 31, 2010. Accepted: Nov. 29, 2010.
References
1. L.L.B. Ponto and R.D. Schoenwald, Clin. Pharmacokinet. 18 (1), 460–471 (1990).
15. P. Noceti, J. Smith, and S. Hodges, J. Forecast. 22 (6), 447–455 (2003).
16. H.C. Lin and G.J. Sheen, Qual. Eng. 17 (1), 371–390 (2005).
17. T. Pyzdek and P.A. Keller, The Six Sigma Handbook: A Complete Guide for Green Belts, Black Belts, and Managers at All Levels, (McGraw-Hill, New York, 3rd ed., 2009).