By Angelo De Palma, Ph.D., Contributing Editor
The U.S. FDA's Process Analytical Technology (PAT) initiative promises to bring pharmaceutical manufacturing into the 21st century by automating real-time measurement, integrating data with control systems, and providing levels of process understanding comparable to those that the chemical and semiconductor industries have enjoyed for decades.
The initiative lags leading companies' efforts in this area by years, and real-time monitoring in other process industries by decades. Large, established manufacturers enjoy a comfortable lead in PAT deployment, but the pharmaceutical industry as a whole has a long way to go.
Strictly speaking, PAT encompasses more than just the adaptation of traditionally laboratory-bound analytical techniques to manufacturing environments,but that's where much of the initiative's emphasis has been placed to date. Today, these PAT installations come in three basic flavors.
- In-line sensors or probes are in direct contact with process materials inside the reactor and transmit signals outside.
- On-line analysis diverts process material to a connected analysis loop, analyzes it, and returns it to the vessel.
- At-line measurements employ laboratory-type analyzers, but require sampling and sometimes human intervention.
For already-validated processes, PAT presents drugmakers with something of a dilemma: revalidate an existing process and take on all costs, delays, regulatory uncertainty—and potential benefits—inherent therein, or stick with the safe, fuzzy warmth of an approved process already known to work.If process understanding is insufficient incentive for testing PAT waters, FDA offers regulatory incentives within its risk-based approach to validation and compliance: Show us the science, the Agency says, and we'll review you fairly and consistently. Within the risk-based framework, drug manufacturers that are starting out in PAT typically continue using traditional offline analysis alongside process analytics until they can demonstrate equivalence between the two. Once equivalence can be shown, they can cut the cord and reap the benefits. However, with respect to revalidation, a number of questions remain. Will it be enough to prove that PAT provides equal or better performance than off-line analysis? Will a formal revalidation be necessary, or will demonstrating equivalence serve as the revalidation? "The mechanisms for switching need to be flushed out," says Andrew Malcolmson, director of business development at Malvern Process Systems (Southborough, Mass.). So far, he says, not many drug manufacturers have worked through this exercise for existing processes. "It's definitely easier to implement PAT for new processes," he says.Normally, when process engineers decide to use PAT with an existing process, it's for a product or operation that has posed problems in the past,in such cases, the potential benefits of installing PAT clearly outweigh the regulatory uncertainties. Pfizer (New York), a leader in applying pharmaceutical process analytics, has converted dozens of older processes--some developed more than 50 years ago--but believes that any operation within any process is fair game. "Although PAT is certainly much easier to deploy early in development--and we are doing that for new processes--we had to start somewhere, and the logical place was existing processes," says Norman Winskill, Ph.D., Pfizer vice president of global manufacturing services.From the Ground UpTime is what distinguishes PAT deployment for new processes vs. retrofits of older lines--new processes have it, while existing ones don't, says Chris Ambrozic, senior consultant at Umetrics, Inc. (Kinnelon, N.J.) Manufacturers are reluctant to shut down profitable lines to retrofit and revalidate, especially if they've just spent five or ten years perfecting a manufacturing process.The scientific and engineering justifications for PAT are most obvious when process analytics are adopted at the lab bench and scaled upwards, through pilot and production scales. Earlier deployment helps iron out process wrinkles, but, perhaps most importantly, it allows analytics, control, feedback, and data processing functions to be tested on nonessential preclinical batches. "What companies do with PAT data at the back end needs to be discussed during development," says Howard Hartley, manager for engineered systems at Pfaudler (Rochester, N.Y.). "Companies looking for real process optimization, continuous improvement, and statistical process control need to get PAT data into their plant control system," he says. "And they need to do this early on because it's expensive. Just plunking an analyzer out there won't give you the full benefit."Although PAT was originally announced for small-molecule manufacture, biotech companies have employed process analytics--without the capital letters--from the industry's beginnings. Despite the inefficiencies in biomanufacturing, the sector is a model for new-process PAT deployment. "What is a pH probe built into a fermenter but an online, real-time measurement device?" asks Justin Neway, chief scientific officer with Aegis Analytical (Lafayette, Colo.). "Biotech has a huge lead in implementing PAT because they already deal with huge quantities of data and know how to correlate them with process outcomes," he says. "Biotech figured out long ago that it's one thing to get a probe into a process, but if you can't cut out the significant piece of the strip chart that's most useful, the information is useless."Part of a Larger Picture
Nearly every pharmaceutical regulatory expert has his or her pet PAT caveat. Most advise that process analytics requires a plan, entails a learning curve, and that no one technique will serve all processes equally well.Process analytics alone won't resolve every process or product inconsistency. Fully half of all the process variations that lead to rejected batches can be traced to variations in raw material, says Chris Ambrozic of Umetrics. Excipients, he points out, introduce tremendous variation into the formulation and dissolution properties of final drug product. "When we start projects with clients, we won't proceed unless the actual QA testing of raw materials is included. You could drive a bus sideways through some companies' raw material specs."Manufacturers who expect process analytics to give them black/white, yes/no answers right out of the box are deluding themselves, warns Nancy Mathis, Ph.D., CEO of Mathis Instruments (Fredericton, New Brunswick). "Most deployments require that you correlate instrument readouts with known measurements before even attempting to interpret the resulting tons of instrumental data." In other words plan, check against off-line measurements, and keep your fingers crossed.PAT requires short-term investments in instrumentation, consulting services, internal human resources, down-time for existing processes, or development delays for new ones. Companies adopting process analytics must embrace FDA's brave new world of risk-based compliance and validation. Within that world, one would expect that the most risk-averse manufacturers, especially contractors and manufacturers with tight capacities, would be the last to adopt process analytics. However, the contract manufacturer Patheon (Mississauga, Ont.), views PAT as an opportunity to streamline operations, improve quality, and offer related services. "Problems begin during development," says Anil Kane, Ph.D., associate director for formulations development, "especially when processes are transferred without fully understanding them, which leads to quality issues."Last but not least, PAT begs for changes in an organization's quality philosophy. Despite nearly universal recognition of PAT's benefits, today's manufacturing paradigm is still skewed towards finished product testing. "Process analytics," Kane observes, "offers the hope of designing quality in."PAT the Pfizer Way
As one of the most PAT-savvy manufacturers, Pfizer (New York) relies on core spectroscopic process analytics (infrared, near-infrared, ultraviolet/visible, fluorescence, mass, and Raman), plus high-performance liquid chromatography (HPLC), rapid microbial testing, gas chromatography (GC), and spectroscopic imaging. Another Pfizer favorite is focused beam reflectance (FMBR), which uses laser light reflection to monitor particle characteristics. Pfizer also employs on-line HPLC and GC, but does not consider these core applications. Pfizer constantly seeks analytic ideas from food, semiconductor, aerospace, and chemical industries to apply to pharmaceutical manufacturing. "Virtually anything goes," says Norman Winskill, Ph.D., Pfizer vice president of global manufacturing services.For example the company borrowed acoustic analysis, useful in granulation and tablet pressing, from the materials processing industry. Spectroscopic imaging of solids, another Pfizer specialty borrowed from materials analysis, uses spectroscopy plus a camera to obtain complete spectra of defined areas and assemble those spectra into a spectral image.Vendors are fair game for PAT collaborations, particularly when on- or in-line analytics require extensive equipment modifications. "We'll even work with several vendors on the same technology, like NIR, depending on the application and the suitability of their instruments," Winskill says.Pfizer's reliance on core technologies reflects, in part, what is available in on-line or in-line format, but as vendors recognize FDA's seriousness about process analytics, availability has become a moving target. "It's amazing how fast technology is advancing," Winskill observes.As one of the industry's process analytics leaders, Pfizer has maintained a dedicated PAT effort for more than two decades. "Initially, we developed PAT applications to improve process understanding and control of fermentation processes," Winskill says. "About fifteen years ago we started to apply those same techniques to [small molecule] pharmaceutical manufacturing." PAT has become one of the cornerstones of Pfizer's "Right First Time" (RFT) program (Pharmaceutical Manufacturing, June 2004, p. 37). "Externally, as we increase process understanding and show we're in control of our processes, we would expect that FDA will consider Pfizer operations to be low-risk, and under its risk-based GMP initiative give us less scrutiny and more leeway to implement continuous change ourselves," Winskill says. "Doing things right just because regulators want you to is not the right attitude, Winskill adds. "If you do it for the right business reasons you should be in compliance anyway. We would have adopted PAT with or without FDA because it's the right thing to do, an important component of our RFT program. As long as companies and FDA understand PAT in light of that risk-based approach, compliance issues should not get out of hand," he says.PAT's real objective is to create an environment which encourages creativity and innovation in pharmaceutical manufacturing, Winskill says. "For whatever reason, the pharmaceutical industry has been slow to adopt innovative manufacturing techniques and real or perceived regulatory hurdles are at least one contributing factor. Both FDA and industry--certainly Pfizer--want to change that."Is Your Process Data PAT-able?
By now you know that process analytics involves more than buying an instrument and plugging it in. Following is a checklist to help with early PAT planning. Ask yourself the following questions:
- Can you collect data from an appropriate or relevant component or operation? Can data be acquired in-line? On-line? Will you need to retrofit sampling ports, or can you use existing ports?
- What is the appropriateness and quality of data you're most likely to collect? Will you need more than one type of measurement to determine, for example, whether product is dry enough or a reaction has gone to completion?
- Where will data reside after collecting it? Will they be stored on a standalone instrument or fed into a programmable logic controller (PLC)? Does the PLC communicate with a supervisory control and data acquisition (SCADA) system or database? Are the data available to multiple instruments, controllers, and individuals throughout the organization?
- Are datapoints of the right frequency? Is information emerging rapidly enough to permit detection of process upsets or make pass/fail quality decisions?
Without hands-on knowledge of how data are generated and what they mean, PAT becomes an empty exercise, says Steve Ambrozic, senior consultant with the instrumentation and software vendor Umetrics, Inc. (Kinnelon, N.J.). He recommends performing off-line multivariate analysis on PAT data independent of the PAT set-up to get a "feel" for the information coming out of the process. "This allows you to analyze multiple variables and how they relate."
Putting PAT in Context
How will the rest of your site, and organization, deal with the fundamental changes demanded by process analytics? Following are some basic guidelines:
- Make sure that process engineers are experienced in the proposed analytic technology, or that the required expertise can be acquired from outside sources. Different analytic techniques require vastly different skill sets, for example calibration services. While larger firms maintain adequate in-house expertise, start-ps often rely almost entirely on vendors for PAT know-ow, says Dan Klevisha, vice president at Bruker Optics (Billerica, Mass.) "You have to get the expertise from somewhere," he notes. "However, there's always a portion of deployment that must come internally."
- Remember that process analyticals cannot be driven by one group alone. Saying that PAT deployment is cross-functional, involving IT, management, QA, process engineering, and process development, is almost cliche. Projects where everybody has something to say are usually the most successful.
- Consider the special operational requirements and optimal location for equipment. For example, attempting to monitor bin blenders, which rotate, with fiber optic probes, will result in twisted, broken probes. In such situations, engineers often turn to wireless sensors operating through a sight glass.
- Develop good quantitative models early in process development to help make sense of PAT data. Models should reflect likely product variations and use samples that mimic that variation. For example modeling a tablet analysis requires manufacturing "off" product of varying shape, density, and composition.
- Remember that PAT needs to be part of an existing quality program and budget. As such, it is a shared responsibility between quality and operations groups, says Justin Neway, Chief Science Officer at Aegis Analytical (Lafayette, Colo.). It can be most useful to view PAT as FDA sees it: one tool among many to assure product quality and process efficiency.
- Specify sample or sampling interfaces as early as possible in process development. For spectroscopic techniques relying on in-process probes, developers must come up with a way to deal with probe fouling.
- Remember to ask whether the given process or operation requires open- or closed-loop control.Does an operator need to say "yes/no" or will the process automatically right itself?