This article examines using experimental design methods to define different procedures for intermediate bulk container cleaning. The authors have evaluated this new approach, in which a highly soluble, low-dose product and a relatively insoluble high-dose product constituted experimental input variables.
Defining cleaning procedures is crucial to ensuring the elimination of product residues from non-dedicated process equipment. This process can be expensive and challenging, however, in facilities where 30-40 different oral solid-dose products may be manufactured each year.
Generally, best cleaning procedures are defined based on monitoring final drug content, pH, and the conductivity of water samples for each product until they are within acceptable ranges.
This article examines a new way of doing this, using experimental design methods to define different procedures for intermediate bulk container (IBC) cleaning.
The authors have evaluated this new approach, in which a highly soluble, low-dose product and a relatively insoluble high-dose product constituted experimental input variables.
Given the number and wide variety of APIs, ingredients, cleaning and processing materials used in pharmaceutical manufacturing, pharmaceutical products could potentially be contaminated with any number of substances. This is why verifying clean liness through cleaning validation is so crucial. Validation identifies potential residues, whether from APIs, ingredients, cleaning agentsand microorganisms, and sets up a process through which potential contaminationis reduced to the lowest acceptable limits (1).
Usually, these limits are defined based on visual, chemical, and microbiological data(2). Chemical limits are expressed as the maximum concentration in the next product (3), amount per surface unit (4), or concentration in the extraction solvent (5). An acceptance limit plus an internal action limit allows pharmaceutical manufacturers to achieve more stringent process control.
Once potential residues have been defined, detection methods must be established.Usually, specific methods such as high-performance liquid chromatography (HPLC)(6) and ultraviolet spectroscopy (UV) (7) are used, but non-specific methods such as total organic carbon (TOC) (8, 9) may also be applied, and pH and conductivity determinations should also be used to evaluate the performance ofthe cleaning procedure.
Disadvantages of specific methods include the need to develop assays for each new API and validate those methods, a process that can be long and expensive (10, 11). However, TOCcan potentially be applied to any product, is sensitive enough to detect quantities down to μg/L or parts per billion (ppb), and typically involves less time-consuming sample preparation than other methods (12).
Although specific methods are usually preferred, non-specific techniques can be used if there is a scientific justification for doing so in continuous process monitoring, whereas specific methods could used only for initial validation.
Sampling methods are then selected by assaying rinse water or by surface swabbing (13).Rinse water sampling is used when dealing with very large pieces of equipment or piping, or in situations where there is limited access to equipment surfaces(14). These samples should be correlated with direct swab sampling to be sure that residues are being detected properly and do not remain undissolved oninaccessible equipment surfaces.
The authors launched a study to apply the principles of experimental design, and recommendations for cleaning validation, to develop a faster way to set andvalidate procedures for cleaning intermediate bulk containers (IBC) for granulated products. They extended this methodology and applied it to set cleaning validation limits for a vertical granulator in a multiproduct oral solid dosage form facility. The experiments were conducted during the performance qualification step of both of these systems.
Plackett-Burman experimental design methods were used to identify the most important cleaning process parameters, with the minimum number of experimental runs, early in the experimentation phase. To evaluate interactions between variables, a two-level fractional factorial design was conducted after the Plackett-Burman modeling.
Although the equation derived from the two-level factorial design explained the cleaning processes as a third-order equation, the initial Plackett-Burman fitting first-order model (which detected only linear effects) was more valuable inpredicting the process parameters in each step of the cleaning process. The process was scaled up for a vertical granulator, and the cleaning design space was verified.
Results suggest that, although factorial design is useful for understanding process behavior and input variables interactions, Plackett-Burman designs allowed definition of linear models to predict process parameters for 13 new cleaning procedures based on product dose. Five of the cleaning procedures (or recipes), predicted from experimental design, were experimentally confirmed using nine different products. Previous analytical detergent characterization helped todefine the acceptance limits equal to the United States Pharmacopeia(USP) purified water pH and conductivity values.
The ratio between water volume and equipment surface allowed the procedure to be scaled up and used for granulator cleaning. The design space for the cleaning procedure was also checked in four independent runs for two high-dose products. Results show that the experimentally-derived models provided a high level of assurance that the cleaning procedures met specifications.
Cleaning stations. Automatic cleaning was conducted in a clean-out-of-place (COP) AISI 304 cabin (WB Model, Cosmec,Italy) for IBC and in a clean-in-place (CIP) system for a 600-L Glatt vertical granulator. A sequential process was used, combining an initial rinse phase to remove larger amounts of adhered product, and a second phase in which detergent was applied. In addition, there were two final rinse steps. Samples of the final rinse water were taken to determine external TOC, pH, and conductivity.The systems were designed to combine, in a synergistic mode, the spectrum of critical cleaning parameters (e.g., time, water, pressure, chemical action and temperature), and were automatically controlled to achieve a robust process(15).
TOC was automatically conducted in a GE SIEVERS 900 TOC analyzer, while pH and conductivity were measured in a Mettler-Toledo Seven Multi instrument. Each measurement was repeated at least once. Purified water samples were taken during the last 10 seconds of the final rinse step in ERA ultra-low TOC contentcertified 40-mL TOC flasks.
For experimental design definition, version 6.0.1 of Stat-Ease, Inc.’s Design-Expert software was used. A Plackett-Burman design for six factors(i.e., initial rinse volume, detergent volume, detergent concentration, final rinse volume, purified water rinse volume, and product dose) was conducted as 18 runs in a single block, including two central points.
在实验设计的定义中，使用了Stat-Ease, Inc.公司的 Design-Expert软件的6.0.1版本。Plackett-Burman设计的 6 个因素(即：初始的淋洗体积、清洁剂体积、清洁剂浓度、最终淋洗量、纯化水淋洗量和产品剂量)在一个单一的模块（包括两个中心点）中进行了 18 次运行。
Another two-level fractional factorial design (for the same six factors) was conducted to estimate the effects of all interactions in 18 runs, including two central points. Experiment was conducted as a single block.
Both designs were conducted with two different products, used as categorical variables: Product 1 (20-mg dose, highly soluble in water) and Product 2(400-mg dose with low water solubility).
A neutral detergent (Steris’ CIP 300) was used in the experimental phase and during evaluation of the predicted procedures. Steris’ CIP 200 acid detergent and its CIP 100 basic detergent were also evaluated during process performance qualification.
Cleaning validation constitutes the documented evidence that a cleaning procedure provides equipment ready for manufacturing process, mainly in amultiproduct facility. Activities related to validation studies could be classified in three phases (16). Phase one, commonly called pre-validation, involves research, development, and equipment qualification. Phase two is designed to verify that all the critical parameter limits that have been established are valid, and that the process generates products with sufficient levels of critical quality attributes, even in worst-case situations.
Phase three, the validation maintenance state, involves the frequent review of all documents related to the process performance, to ensure that no deviations, failures, or changes to the production process occurs. A careful design of systems and process controls assures process robustness and quality products.
These phases may be applied to cleaning processes, considering that their developmentand validation are based on defining cleaning procedures and controlling analytical assays, acceptable residue limits, critical sampling points, and methods. All of these elements must be established during phase one. In this article, the authors will summarize the process used to establish cleaning procedures for two critical equipment involved in oral solid dose manufacturing in early pre-validation phase.
In most cases, the first cleaning agent to evaluate in cleaning procedure development is a neutral pH detergent. If it does not provide consistent product removal performance, other acid or basical ternatives are assayed (17). Determining the relationship between detergent dilution in purified water and pH/conductivity values is a prerequisite for using these assays as indicators of cleaning agent removal in cleaning validation studies (18). That is why, in this research, the authors tested three different detergents to evaluate its influence in equipment cleaning procedure performance.
For the CIP 300 detergent, pH values of different concentrations indicated that pH assays could only detect concentrations exceeding 1000 ppm. For lower levels, pH values corresponded to those of purified water (5.0-7.0). Conductivity values regarding detergent concentration showed that this type of assay could detect concentrations above 10 ppm (1.5 μS/cm-2.7 μS/cm), confirming conductivity as a better assay for determining detergent residues in final rinse water samples than pH measures as previously reported (19).
In the case of the CIP 200 acid detergent, 1 ppm corresponded to the lower limit of water pH (5.20-5.76) and conductivity was in the range of 3.0-3.73 μS/cm. For the CIP100 basic detergent, the 1 ppm pH range slightly exceeded the upper limit water pH (6.92-7.31) and conductivity was between 2.29 and 3.98 μS/cm.
在CIP200的酸清洁剂情况下，1ppm相当于水pH的下限（5.20-5.76），电导率在3.0-3.73μS/cm的范围内。对于CIP100的碱清洁剂，1ppm的pH范围稍微超过水pH的上限(6.92-7.31)，电导率在2.29- 3.98 μS/cm之间。
Taking these results into consideration, USP purified water specification based on pH, conductivity, and TOC were adopted as acceptance criteria to evaluate cleaning process development results. Although these limits below 1 ppm for the three detergents tested is lower than the universally recognized limits of 10 ppm(20), the authors initially considered the potential cumulative effect that residues over surfaces in multiple equipment processes could have on the final product (21). The authors opted for an over dimensioned process to mitigate patient risk, but other factors could justify establishing higher limits.
In Plackett-Burman design, the main effects of selected experimental design variables have a complicated confounding relationship with two-factor interactions (22). Therefore, these designs should be used only to study main effects of process parameters when it can be assumed that two-way interactions between them are negligible.
In evaluating the results obtained from Plackett-Burman experiments, a linear model was developed between TOC values and operational cleaning parameters. The Analysis of Variance test showed that the effect dose has a probability value below 0.05, indicating its statistical significance at a confidence level of 95.0% (Table I).
Table I: Variance analysis fortotal organic carbon (TOC) from sample analysis of the Plackett-Burman experimental design. (DF = degrees of freedom; Prob = statistical probability associated with the given F value.)
R-squared statistics indicated that the adjusted model explained only 38.102% of the TOC variability. The adjusted R-squared was 14.8902%, suggesting that the linear equation obtained did not completely explain the system’s behavior, so it is likely that higher order interactions take place between the operational variables.
The Durbin-Watson (DW) statistic was more than 5.0%, proving that there was no serial correlation in residuals. The Plackett-Burman final equation (Equation1), in terms of decoded factors, was:
[Eq. 1] TOC—525.397—0.36299? Initial Rinse — 6.15948 ?Detergent Volume + 1.52647 ? Detergent Concentration — 1.36716 ?Final Rinse — 1.38799 ? Purified +0.5559804 ? Dose
[Eq.1] TOC—525.397—0.36299? 初始淋洗体积 —6.15948 ?清洁剂体积 +1.52647 ? 清洁剂浓度— 1.36716 ?最终淋洗体积— 1.38799 ? 纯化的淋洗体积 +0.5559804 ? 剂量
As expected, this equation indicated that initial rinse, detergent, final rinse,and purified rinse volumes have a negative correlation with TOC values, and increasing dose and detergent concentration also increased the expected TOC results.
Fractional factorial design results. The low Plackett-Burman adjusted R-squared (14.8902%), indicated that linear equation obtained did not completely explain the system’s behavior. To explain the possible interactions between these variables on TOC variability, a fractional factorial design was completed using the results of the common experiments ofthe Plackett-Burman design. Analysis of variance results are shown in Table II.
Table II: Variance analysis for total organic carbon (TOC) from sample analysis of two-level factorial experimental design. (DF = degrees of freedom; Prob = statistical probability associated with the given F value.)
Values of probability less than 0.0500 indicated significant model terms. In this case, D (Industrial water final rinse), AD (interaction of initial industrial water rinse and Industrial water final rinse), AE (interaction of initial industrial water rinse and purified water final rinse) and ABD (interaction of initial industrial water rinse, detergent volume, and Industrial water final rinse) were significant in model terms. Detergent concentration in the range analyzed was not statistically significant.
Adequate precision measures the signal-to-noise ratio. A ratio greater than fouris desirable. The obtained ratio of 10.918 indicated an adequate signal. Residuals were also checked for normality.
The final equation (Equation 2) in terms of decoded factors was as follows:
[Eq. 2] TOC—268.69—55.06● Final Rinse + 97.81 ● Initial Rinse ● Final Rinse + 67.31 ● Initial Rinse ● Purified Rinse—54.09 ● Initial Rinse ● Final Rinse ● Detergen t Volume
[Eq.2] TOC—268.69—55.06●最终淋洗体积+ 97.81●初始淋洗体积●最终淋洗体积+ 67.31●初始淋洗体积● 纯化的淋洗体积—54.09● 初始淋洗体积●最终淋洗体积 ●清洁剂体积
This model explains as much of 70.56% of the process variability, and is more exactin describing the influence of process parameters than the linear model obtained from the Plackett-Burman design.
The Plackett-Burman-derived equation was evaluated for different products doses using Microsoft Excel’s Solver program, in which only process variables were changed. This procedure allowed recipes to be defined for each group of products based on product dose. Because detergent concentration was determined not to be a significant parameter from both experimental designs, it was restricted as a constant to the minimal value assayed. All restrictions were constricted to the range of cleaning process parameters included in the experimental design.
The resulting procedures were evaluated in at least two containers for seven other products. Results of this testing demonstrated that predicting cleaning process parameters, using the statistically obtained model, was adequate for meeting the acceptance criteria for different doses and for solubility products of batches ranging from 90 to 180 Kg (Table III).
Table III: Results of applying container-cleaning procedures, using recipes developed with Plackett-Burman methods.
When used correctly in the right circumstances, visual inspection is a powerful detection method (23, 24). Using the approach outlined in this article, visual inspection of equipment surfaces revealed no residues, suggesting that the method is effective.
Cleaning procedures developed for product dose groups allowed for reducing process time because the worst-case cycle approach requires 11 minutes. For lower doses, the process times were as follows:
Product 3-10 mg: 6 minutes
Product 1-20 mg: 7 minutes
Product 5-200 mg: 8 minutes
Product 2-400 mg: 9 minutes.
The Plackett-Burman model was more practical because it allowed the authors to define operational parameters for 13 groups of products based only on the results obtained with two products of different dose and solubility. This method could be easily applied in situations where new automatic systems are installed in a facility to establish cleaning process conditions faster and at lower cost than developing cleaning process conditions for each product separately.
Other critical elements that should be validated are the maximum delay before cleaning (i.e., the maximum amount of time that the equipment should be dirty before it is cleaned (25) and the maximum time that it could remain clean after the applied procedure (26). These values could be included as input and output variables, respectively, in the experimental design.
In order to speed development of a cleaning procedure for a vertical granulator, a relationship between the granulator and the container product contact surface area and cleaning volumes was calculated. This approach facilitated the process of develop and evaluating cleaning recipes previous to validation.
For scale-up, a ratio was obtained between the total volume of water required to clean 500-mg dose containers and the container area. This value was related to the correspondent 600-L granulator product contact area, and the relationship used to determine the volume required for cleaning.
The cleaning procedure was adjusted sequentially, in order to reach the total calculated volume. This volume was then used as a central point, and variation around it was determined, using a fractional factorial design for three variables with no central point.
Two categorical variables were used (two different 500-mg products and two different detergents). Robustness of the design space for the cleaning procedure was determined in eight runs with two different 500 mg product doseand CIP 200 and CIP 100 detergents.
Results indicated (see Table IV) that in all cases, actual residue limits were below the levels calculated, and were in the parts-per-million of TOC for Product 7 and Product 8.
Table IV: Results of a design space cleaning procedure developed for a 600-L Glatt vertical granulator, using the ratio of water volume to equipment surface. CIP is clean-in-place. Visual inspection confirmed that the procedure development was robust. No residues could be detected on any of the equipment surfaces.
Evaluations suggest that the mathematical model obtained from a Plackett-Burman design can predict, with a high degree of accuracy, the optimal conditions for the processof IBC cleaning. Results using the model were checked in nine products in atleast two independent replicates. In all cases, the results were obtained within the acceptance limits of product and detergent removal.
Application of different cleaning process parameters (in terms of rinse and detergent volumes) for each product allowed water consumption to be reduced by 320 L per container, resulting in better use of installed capacities for clean-out-of-place cleaning of IBCs, because each type of product had a defined process time. If the common approach of determining the “worst-case” procedureis used for all containers, time and water wastes are significantly increased.
When the process was scaled up to clean a vertical granulator, results of the predicted design space were higher than those obtained for IBC, but all results were below 10 ppm, demonstrating that experimental design can be a powerful tool for developing more robust cleaning procedures.