OR WAIT null SECS
Process characterization requires a significant commitment of time and resources, but the payoff is better process understanding, improved manufacturing success rates, and avoidance of costly regulatory delays.
As we scale up manufacturing processes, increase production rates, and strive to minimize batch failures in an increasingly stringent regulatory environment, it is valuable to characterize processes at the levels appropriate for each product development stage (preclinical through phase 3 and commercial manufacturing). Amgen currently characterizes several aspects of the manufacturing process comparable to that performed by other biotech companies. The increased understanding of each process that results from these characterizations can vary - from minimal knowledge, during development and non-GMP runs, to detailed knowledge of each unit operation or group of unit operations in late clinical development. Predictably, in-depth process characterization requires a significant commitment of resources and time.
The overall goal of adequate process characterization for commercial manufacturing processes is to ensure efficient and successful validation and the assurance of consistent process performance. More specifically, the characterization should provide:
This article describes a characterization strategy that is cost-effective and consistent.
Process characterization has often been viewed as a "check the box" activity or something that was completed but which required little scientific rigor. This was most likely a result of cost cutting. Thorough process characterization requires a significant resource from process development and analytical departments. Many companies now recognize the benefits of sound process characterization - and, unfortunately, the cost of incompletely understood processes (1).
For compliance, process characterization needs to identify key operating and performance parameters, justify operating ranges and acceptance criteria, recognize interactions between key variables, and ensure that the process delivers a product with reproducible yields and purity. This is the complete and rational approach to process validation. Costly registration delays resulting from poorly understood processes and failed validation batches can cost a company tens of millions of dollars.
From a business standpoint, process characterization can improve run success rates and lead to incremental, but significant, yield improvements. Better process understanding can also minimize - and aid in the investigation of - process deviations.
To conserve resources, waiting until after phase 2 is the best time to start "intensive" process characterization work. At that point, the commercial process should be developed, and you know whether or not the product is likely to make it to market. Unfortunately, the driver for process characterization timing is the start of conformance or validation lots. Therefore the work should be completed by that point so that information gained from it can be used to support operating ranges and acceptance criteria for validation protocols.
At least 12 to 15 months should be allowed for process characterization studies; therefore, they should start no later than 15 months before the conformance runs (see Figure 1). Sometimes phase 2 studies are complete by this point; but if phase 2 data are still pending, process characterization resources may have to be spent "at risk" to ensure completion of the work before the process validation runs. Fortunately, characterization studies can sometimes be ramped up slowly during the first few months, which may provide the time to complete the phase 2 studies, conserving resources until a clearer picture of the product's viability is obtained. Figure 2 shows the "process flow" involved in characterizing the manufacturing steps leading up to the validation studies.
Figure 1. Timing of process characterization studies; it's preferable for work to start at the end of phase 2, but it needs to be completed before the start of conformance or validation lots.
Three steps are involved in precharacterization:
Step 1: Data mining and risk assessment. Data mining - a retrospective review of historical data - and risk assessment analyses determine what operating parameters need to be examined experimentally during process characterization. Data from lab notebooks, technical reports, process histories, run summaries, and manufacturing records are combined with a list of the operating parameters and provisional operating ranges for each unit operation. The data are then compiled and reviewed by the team that developed the process to be characterized. Frequently, information on operating ranges from tests done during process development can help identify the parameters most likely to affect a process (2).
Process Characterization Terminology
Once data mining is complete, risk should be assessed for each unit operation on the effect and likelihood of an excursion from operating ranges. Hazard analysis and critical control points (HACCP) (3), failure mode and effects analysis (FMEA) (4-7), cause-and-effect diagrams (8), and other risk assessment tools can be used for risk analyses. FMEA assigns a numerical rating (typically 1 to 10) to the severity of an excursion from operating parameter ranges, the probability of an excursion, and the likelihood of detecting an excursion before it has an effect on the product (4-7). The combined risk factor - the risk priority number (RPN) - is a multiple of the three variables, rating each parameter from 1 to 1,000 (4-7).
The FMEA exercise should include not only the scientists who developed the process but also quality personnel and plant and process engineers because they bring insight into the likelihood of a process excursion and the difficulty in detecting it. Significant differences in processes may exist between commercial manufacturing of a product and the first-in-human process, so there may be little historical data at commercial levels. Risk analyses may have to draw on development and historical data from both clinical and commercial scales.
Figure 2. Process characterization project flow diagram
Developing an FMEA rating system that is not totally subjective - that is consistent and agreed on by everyone - can be challenging. Table 1 (9) provides an example of how parameters in the FMEA rating system can be defined. The resulting data - the RPNs - are usually presented in a Pareto chart (Figure 3). RPNs that fall below a predetermined threshold are not considered key parameters and are not examined during the characterization experiments. A report summarizing FMEA study findings should include the rationale for determining that some parameters do not need to be characterized.
Table 1. A risk priority number (RPN) rating system for rating process characterization parametersfor use in failure mode and effects analysis (FMEA) (9).
Step 2: Scale-down model qualification. Process characterizations require that a representative scale-down model be qualified. Some unit operations, such as homogenization or centrifugation steps, are more difficult to scale down and typically have to be qualified at pilot-plant scale. Other operations such as chromatography, ultrafiltration, and some cell culture and fermentation operations can normally be qualified at bench scale.
In general, scale-down model parameters should be qualified at the center points of clinical or large-scale operating ranges. If the process has not yet been run for clinical manufacturing, the operating parameters should mirror those of the largest pilot-scale runs. If no appropriate large-scale data are available, data from an earlier manufacturing stage of the product can be used. However, the scale-down model has to provide data that are consistent with large-scale runs.
Figure 3. A sample failure mode and effects analysis (FMEA) Pareto chart. Operating parameters are ranked from 1 to 13 by risk, which results in a risk priority number (RPN).
Key performance parameters from small-scale studies should be within the historical range of those observed at large scale.
Scale-down models also ensure that various operating parameters are measured accurately and represent what happens at large scales. Because operating parameters vary during process characterization studies, pH, conductivity, temperature, flow rate, pressure, and other measurements need to be calibrated and read accurately and with adequate precision. Flow cells for bench-scale chromatography should provide the same absorbance readings at large scale, particularly at higher absorbances.
Scale-down experiments should use feedstock representative of that to be used at commercial scale. Storage conditions and feedstock stability should be evaluated at each step to ensure the material used during characterization does not degrade during the course of the study. Doing so will also identify processes that need a constantly replenished supply of representative feedstock. Although not a requirement, released CGMP raw materials, such as buffer salts, media, and resins, should be used when possible.
Table 2. A simple resolution III fractional factorial design for a fermentation process.
Step 3: Protocols. Although not absolutely necessary during precharacterization, having written protocols preapproved for individual unit operations is valuable in planning, executing, and documenting process characterization procedures. Protocols should indicate the operating parameters to be tested and the operating ranges to be examined. Protocols should also indicate which performance parameters will be monitored and what assays will be run. Acceptance criteria are not necessary in the characterization protocols because the outcome of many of the experiments may be unknown.
Process development scientists and the analytical groups who run assays and validations should review and approve the protocols. Process characterization studies draw heavily on the resources of the analytical department, so its personnel need to have the appropriate resources to qualify the assays before the start of characterization work. Protocol approval, at this stage, by the validation group ensures the process characterization package delivered to them includes everything they need for writing validation protocols. For more complex studies, the protocols should be reviewed by statisticians to ensure the studies deliver meaningful results without using excess resources.
Process characterizations require five steps:
Step 1: Characterizing process performance or "process fingerprinting" data may be available before the process characterization studies begin. At the end of this step, however, your team needs to understand what each process contributes, including its yield, impurity clearance, or in-process pool quality. For cell cultures, this step may be a titer or a qualitative assessment of the product (such as the degree of glycosylation or sialylation). For a chromatography process, this step would delineate the impurities cleared and the point in the process that they are cleared (for instance, in the wash step, before product elution, after product elution, or during the regeneration step). For an ultrafiltration method, conductivity or pH after different turnover volumes might be examined.
The outcome of process fingerprinting should be the identification of key performance parameters for each process step, which can then be used to design further characterization experiments. How these performance parameters are affected by excursions from the operating ranges is the objective of Step 2 characterization studies.
Step 2: Screening experiments. FMEA or risk assessments have, by this step, identified the nonkey parameters - parameters with minimal effect on product quality or yield. Screening experiments at this point are designed to eliminate additional parameters from further, more rigorous, process characterization work. The design of experiments (DOE) can make screening a variety of operating parameters easy (10,11). DOE software packages can aid in the design and interpretation of these experiments (12,13).
Fractional factorial, Plackett Burman, and D-optimal DOE designs can be used for screening (10,11). Table 2 shows an example of a simple Resolution III fractional factorial design for a fermentation process (11). In this design, only nine experiments are required to screen six operating parameters. Results can be analyzed using Pareto charts or regression analysis. Although these screening experiments do not allow you to rule in or out any interactions between parameters, they allow you to determine the main effects and identify the factors with the greatest effect on key performance parameters.
Sometimes a "one-off" study - where only a single operating parameter is tested and all other parameters are held at their center points - is simpler. Although this approach can require more experiments, its results are more easily interpreted.
Typically, the performance response throughout the operating parameter range tested is linear, therefore only two-level (a high and a low value) studies are used (Table 2). The two levels we prefer to test are at about 1.5 to 2 times the preferred operating range during manufacturing. Table 3 shows several common operating parameters and their respective ranges for screening experiments. This approach accomplishes several things:
Table 3. Examples of some operating parameter test ranges for initial screening experiments.
Step 3: Interactions between key parameters. From screening experiments results, the operating parameters tested during Step 3 are known to affect process performance. Therefore, these parameters are typically tested only to the edge of their normal operating range. As in the screening experiments, DOE approaches are useful. Depending on the number of variables to be tested, full factorial, fractional factorial, or another DOE can be used (10,11). In general, however, you want to use a design from which the effect of any suspected interaction can be determined. This means higher resolutions (IV or V) are needed in the experimental design (10,11). Although performance response is typically assumed to be linear throughout the operating range tested, the validity of that assumption needs to be judged. When nonlinearity is suspected, multilevel experimental designs should be used.
Table 4. Anion-exchange chromatography design of experiment (DOE) in which operating parameters are combined to give maximum and minimum retention times, as well as maximum and minimum pool volumes.
An outcome of these experiments may be the identification of other weak spots in the process resulting from interactions that were not observed during the initial screening experiments. In some instances, operating ranges may have to be readjusted. The final information delivered by these experiments is an assurance that the process runs "as advertised" within the confines of the various combinations of operating ranges and that there is no point where combinations of two or more operating parameters, run to the edge of their respective ranges, cause the process to fail.
Variables can sometimes be combined to more quickly explore the "operating space" of a given unit operation (10), minimizing the amount of experimentation required. Combined variables, however, may offer results that are difficult to interpret when a process fails, and combining assumes there are additive effects between operating parameters, which may not always be the case.
The best time to combine variables is when you have confidence in your operating ranges, and you need only confirmatory information that your process performs properly over those ranges. An example of this is shown in Table 4. The initial screening experiments identify operating parameters that affect either pool volume or peak retention time on an anion-exchange chromatography step. In our example, we combined variables to determine maximum and minimum pool volumes and the earliest and latest retention times. (Another way to look at this is that we set up a full factorial experiment using pool volume and retention time as input parameters.) The performance parameters (purity and yields) were within historical ranges for this set of experiments. Therefore, by combining the six variables in the left-hand column in this way, we confirmed that the process performs as advertised throughout all of its prescribed operating ranges for these key parameters.
Figure 4. An example of resetting column 1 impurity acceptance criteria based on process robustness studies. Running column 1 under worst case conditions leads to high levels of impurity, which are subsequently cleared to within historical ranges by column 2.
Step 4: Process redundancy experiments - using feed quality as a process input. Until this point, all experiments have been performed with representative feed material. However, to test the process's "top-to-bottom" robustness, the effect of feed quality on each unit operation should be tested. This provides an understanding of the downstream sensitivity to upstream excursions.
All other operating parameters (pH, temperature, and the like) are run at the center of their respective ranges, because the likelihood of having both an operating parameter excursion and a feed quality excursion at the same time is remote.
Process redundancy experiments can be used to set acceptance criteria for performance parameters for each unit operation. A unit operation can be run under conditions that cause it to fail to perform adequately. The pool from the failed unit operation is used to determine if downstream operations can make up for poor upstream performance, and thus keep the product within specifications.
An example of process redundancy experiments is shown in Figure 4. In this experiment, column 1 was run in such a way that a product-related variant was present at values higher than the historical range. When the column 1 pool was processed through column 2, however, the amount of this variant fell to within the historical range for the column 2 pool. Therefore, rather than setting the acceptance criteria for the column 1 pool based on historical ranges for the product-related variant, a wider acceptance criteria can be set because of the process redundancy between columns 1 and 2. There may be other instances in which a pool has to be processed through more than one additional downstream step to determine the process redundancy, but the principle is the same.
Setting acceptance criteria by determining process redundancy is scientifically based and can give a more realistic indication of what a process actually delivers than statistical analyses of historical data (such as standard deviation or tolerance intervals) (14), particularly because at the time the process characterization is carried out, there may be little or no historical data from which to set statistically valid acceptance criteria. Setting acceptance criteria for process validation studies based on what a process can actually deliver results in fewer validation failures from criteria being set incorrectly. Of course, as the process matures and more lots are run, the acceptance criteria can be more easily based on statistical analyses.
Step 5: Finishing up - reports and follow-up. Process characterization reports, which include the results from the clearance, screening, interaction, and process redundancy studies, should be written for each unit operation. Key operating parameters and respective ranges should be identified, as should acceptance criteria for all key in-process performance parameters. A rationale for why certain parameters are identified as nonkey should be included as well. Data from these reports will be used to support validation studies, and the reports should be completed before writing validation protocols.
After the characterization work is completed, it may be valuable to go back and repeat the FMEA exercises. The severity factor for each operating parameter is known by this step, and repeating the exercise could dramatically change the RPN. In this way, facility engineers can devote their time to those unit operations and operating parameters that require the greatest control and detection.
Process characterization requires a significant commitment of time and resources, but the payoff is better process understanding, improved manufacturing success rates, and avoidance of costly regulatory delays, which make the investment worthwhile. The process characterization approach we have described will furnish you with information that can be used for setting operating ranges and acceptance criteria for performance parameters, as well as defining the overall robustness of your process. At Amgen and at other companies, however, our process characterization strategies continue to evolve, and we fully expect even more efficient and consistent approaches to be developed in the future.
The authors wish to thank Tim Tressel, Dan Weese, Dave Smiley, and Mike Covington for their suggestions and input.
(1) Bobrowicz, G., "The Compliance Costs of Hasty Process Development,"
12(8), 35 38 (August 1999).
(2) Gardner, A. and Smith, T., "Identification and Establishment of Operating Ranges of Critical Process Variables," Pharmaceutical Process Validation, G. Sofer and D. Zabriskie, Eds. (Marcel Dekker Press, New York, 1999, pp. 61 76).
(3) Armbruster, A. and Feldsien, T., "Applying HACCP to Pharmaceutical Process Validation," BioPharm 13(10), 170 178 (October 2000).
(4) Kieffer, R., et al., "Applications of Failure Mode and Effects Analysis in the Pharmaceutical Industry," Pharm. Technol. Eur., 36 49 (September 1997).
(5) Nobel, P., "Reduction of Risk and the Evaluation of Quality Assurance," PDA J. Pharm. Sci. Technol. 55(4), 235 239 (2001).
(6) Sahni, A., "Using Failure Mode and Effects Analysis to Improve Manufacturing Processes," Med. Dev. Diagn. Ind., 47 51 (July 1993).
(7) McDermott, R. et al., The Basics of FMEA (Productivity, Inc., Portland, OR, 1996).
(8) Burr, J.T., SPC Tools for Everyone (ASQC Quality Press, Milwaukee, WI, 1993).
(9) Rath and Strong's Six Sigma Pocket Guide (Rath & Strong, Inc., Lexington, MA, 2001), pp. 26 31.
(10) Kelley, B., "Establishing Process Robustness Using Designed Experiments," Pharmaceutical Process Validation, G. Sofer and D. Zabriskie, Eds. (Marcel Dekker Press, New York, 1999), pp. 29 60.
(11) Haaland, P., Experimental Design in Biotechnology (Marcel Dekker, Inc., New York, 1989).
(12) Juran, J.M. and Godfry, A.B., Juran's Quality Handbook, 5th ed. (McGraw-Hill, New York, 1999), pp. 47.1 47.77.
(13) Montgomery, D.C., Design and Analysis of Experiments, 5th ed. (John Wiley and Sons, New York, 2001).
(14) Seely, R., Munyakazi, L., and Haury, J., "Statistical Tools for Setting in Process Acceptance Criteria" BioPharm, 14(10), 28 34 (October 2001).BPI