Michel Crevoisier

	Michel Crevoisier is retired senior quality assurance expert, ChemOps, at Novartis Pharmaceuticals

	In its 2011 Guidance for Industry on Principles and Practice of Process Validation (1), FDA radically changed its interpretation of process validation. Validation now means “the collection and evaluation of data, from the process design stage through commercial production, which establishes scientific evidence that a process is capable of consistently delivering quality product.” It will no longer be a formalized activity that is limited in time and, typically, to three validation batches.

	For many people, the most challenging consequence of the new approach is that process validation never ends. As long as a process is in use, it is undergoing validation, either in stage 1 (development), stage 2 (qualification) or stage 3 (continued verification), and validation will continue until the process is no longer used.

	Once we have fully understood the new concept of the life cycle approach to process validation, we have to recognize that some typical phrases and words that we have been using for decades are now obsolete.

A common statement so far, but it no longer makes sense. Reading or hearing it should make us raise an eyebrow. As validation never ends, to speak of process validation in the past tense has become as much of a fossil as the old validation concept of “three consecutive batches.”

	Instead, today, we would say, “The process was qualified.” We should further consider replacing the notion of a “validated process” with the concept of “process validity.” Is the process under control? By FDA’s definition, the collected process data and their evaluation should give us the answer.


	These also belong to history. Under the new definition, every run is a validation run. Depending on the maturity of the process, the validation runs can be called development, qualification, or verification runs.


	We cannot redo something that never ends. We can imagine situations where a process needs to be requalified or even redeveloped; but not revalidated.

	Retrospective, concurrent, and prospective validation
	What do they mean today where validation never ends? If process validation is “the collection and evaluation of data,” isn’t process validation simultaneously concurrent (the collection of data), retrospective (the evaluation of data), and prospective (the application of inferential statistics)? Collecting data prospectively, however, is hard to imagine.

	Considering that the new validation concept changes the meaning of many hitherto common terms (think of “qualification” applied to processes, where the term had typically been used for qualifying equipment), one can ask why the FDA decided to still use the term “process validation” at all. What if they had chosen to use “process control” or “process stability” instead? Wouldn’t this wording better express what the new guidance is all about?

	The difficulty may have been that process validation is not only an industrial quality assurance tool (along with keeping process control charts, and evaluations like Six Sigma, for example.)

	The term also enters the domain of the application for and the approval of new drugs. To obtain marketing authorization, the applicant has to demonstrate that he is able to reliably produce commercial quantities of the new drug. The proof of the ability to manufacture is typically submitted in the form of a Process Validation package or dossier (2). Regulators must have decided to keep the term “process validation” to maintain the alignment with the wording governing the application for new drugs.

	Still, it would have been interesting to have a process guidance that did not use the word “validation.” It would have allowed for a clean cut away from the old “three validation runs-based” practice and an easy leap to the modern approach to quality assurance the FDA wants us to embrace.

	So, do we need brainwashing? Six years after the publication of the new FDA guidance, it seems we do. Perhaps it could begin with our quality assurance vocabulary.

Guidance for Industry; Process Validation: General Principles and Practices

	2. EMA, “Guideline on Process Validation for Finished Products - Information and Data to be Provided in Regulatory Submissions,” Chapter 4, November 2016.

Articles

Six years after the guidance, it’s time to change our quality assurance vocabulary.

Process Validation: Do We Need Brainwashing?

At home, when washing the dishes, do we ever consider what meal they will be used for next? When plates and cutlery are taken out of the cupboard to set the table, do we ever ask what they have been used for and whether they are clean enough for the gourmet course about to be served?

	Are there different levels of cleanliness used for the dishes depending on the food served or the guests we entertain? Silly questions, apparently, but the industry’s current approaches to pharmacutical cleaning validation can often seem just as random. Cleaning limits for pharmaceutical manufacturing equipment are computed based on the previous and next products for every product change. As a result, it is often difficult to explain the cleaning and cleaning validation concept. 

	This article is about combining the simplicity of the household cleaning concept with the science needed to protect patients’ safety. To reach this goal, it may be necessary to revisit a few traditional ideas about cleaning validation and to establish new principles. The focus will be on API manufacturing of small molecular weight substances (i.e., with industrial organic chemistry and the corresponding traditional plant equipment such as reactors, separators, crystallizers, centrifuges, filters, dryers, mills, and blenders, plus all the connecting pumps, pipes, and hoses). The size of the vessels typically ranges from 250 to 10000 L, and a manufacturing line can have up to several hundred square meters of product contact surface.

	The change from product A to product B is not a frequent case

	APIs are usually produced in multipurpose equipment. Several individual modules are concatenated to build a product- specific production line or train. After a campaign, the line usually gets disconnected and the individual modules, not always all of them, are reconfigured into a new line for the next product. Some products may use exactly the same production line as others, but most of the time, every product uses a somewhat different configuration of modules. Some equipment is used for many products; other parts are practically dedicated to a few or to only one product.

	When a line is set up for product B, some parts will indeed have been used for product A right before; other parts, however, may have been in contact with product X, and mobile parts may be taken from the storeroom and have been used for product Y or Z months before. There is rarely a pure product changeover from A to B; in reality, the changes are from products A, X, Y ... to product B. 

	Similarly, when cleaning a piece of equipment, it is not always known what the next product will be. The change could be from A to the next products B, C, or D. This is particularly the case with interchangeable mobile equipment such as flexible hoses, filters, or tanks that go back to the storeroom after cleaning. Under such circumstances, a number of questions will arise, including the following:

		What was the previous product manufactured at the facility, and what will be the next one? 

		How can one make sure that all parts are clean, to the right level for making the next product(s)? 

		Can the cleaning validation concept be built and the cleaning limits computed based on the ideal product. change from A to B? 

	The following principles aim to answer these questions. 

	The cleaning procedure is independent of the following product(s)

	Equipment cleaning should be thought as a reset function. The equipment is reset to a clean ground state form where it can be used for any product. Every piece of equipment labeled clean must be ready for use, no questions asked. 

	Consider the analogy with the household: doing the dishes, one doesn’t ask what will be cooked the next day, and when taking clean plates out of the cupboard, it doesn’t matter what the last meal was. A clean plate is a clean plate.

	Cleaning without asking what the next product will be implies that setting the cleaning limit must be a comprehensive exercise that includes all products of a manufacturing unit. Quite often, the “cleaning validation runs” and the corresponding plans and protocols only consider the actual product change A to B, which leads to cleaning à la carte, asking: How clean does it have to be today? 

	Using the same cleaning procedure for different cleanings, it can happen that different limits must be met from one cleaning to the next, because the limits get recalculated each time for the actual, ideal product change. With such moving targets, it is impossible to properly validate a cleaning process. 

	A standard cleaning limit (SCL) should be set, in mg/m2

	Establishing a standard cleaning limit (SCL) for all the equipment is the core element of the cleaning concept (1). The SCL is the level of cleanliness (expressed in mg/m

) that a production unit must maintain with every cleaning process, every time. The SCL applies to every piece of equipment, fixed or mobile, large or small, independently of the product changes. 

	What does the cleaning practice tell us?

	Most production units will have collected cleanliness data, typically results of swab tests, over quite a long period. An example of such a data set is given in 
	

. The example is from a relatively old manufacturing unit (Unit A), which uses traditional equipment for industrial organic chemistry. The cleaning procedures consist mainly of flushing and boil outs with solvents and manual scrubbing with water and a detergent.

 gives a good idea of the level of cleanliness this particular production unit can achieve. The values vary a great deal and the level of cleanliness that can reasonably be guaranteed is limited by:

		Physicochemical properties of the process residues

 was that this particular production unit could be reliably cleaned down to 25 mg/ m

 . Although much lower values than 25 mg/m

 is considered the unit’s standard cleaning performance and so, for this unit, the SCL was set at 25 mg/m.

	Setting the SCL in this simple and pragmatic way, solely based on experimental data, without any theoretical considerations, reflects acceptance of the variability and limitations inherent in current cleaning practice. However, this approach represents a radical change from tradition. Previously, the limit was calculated in the validation protocol that only considered the effective product change. The quality assurance (QA) department determined to what limit the equipment was to be cleaned at this one time, without considering the big picture of years of cleaning experience.

	The theoretical safety requirements for a changeover from product A to B are known. Based on the permitted daily exposure (PDE) of A, the maximal safe carryover (MSC) of product A into product B can be determined by using 

. Dividing the MSC by the product contact surface used to make B gives the maximum safe surface residue of A (MSSRA) on equipment used for B.

 to all the theoretically possible product changes in a defined manufacturing unit yields the MSSR matrix. The MSSR matrix is the matrix of all possible required levels of cleanliness for a defined portfolio of products. See reference 2 for an example of a computed MSSR matrix.

 shows a graph of more than 250 swab results collected over a period of 18 months in multipurpose API unit A. Swabs were taken after using different cleaning procedures, after campaigns of different lengths, making different products, and from different pieces of equipment made of different material. The lowest values were often limited by the limit of quantitation of the analytical methods.

	The risk of carryover contamination can be assessed by comparing the level of cleanliness that can be reasonably expected (the SCL) with the cleanliness required by the MSSR matrix (2). During risk assessment, the following should be addressed: 

		Are the cleaning processes good enough to ensure safe changeovers? If not, is it reasonable to simply conclude that the cleaning procedures should be improved? Maybe there is room for improvement; but chances are that the cleaning processes have already reached their limits. If this is the case, can risk of the worst changeovers be mitigated by other means than cleaning? 

		Can a few theoretical changeovers be eliminated from the MSSR matrix because they do not share any common equipment? 

		Can manufacturing campaigns be planned in a way to avoid critical changeovers?

		Do certain products call for partially of fully dedicated equipment?

	Such considerations should be summarized in a risk assessment document, which should be part of the cleaning validation master plan. Additional risk assessment can be done based on analysis of the swab results shown in 

 is a safe limit, because the probability of a cleaning giving a swab result >25 mg/m

 is <1%. Should a failure risk of 5% be deemed acceptable, the SCL could even be lowered to ≤10 mg/m

, but, in this case, they were measured in a different manufacturing unit (B) of the same company. 

	The data for unit B are very similar to those for unit A and the SCL was also set at 25 mg/ m

 for unit B. Because units A and B both manufacture small organic molecules, use similar equipment and very similar cleaning processes, it is not really a surprise that similar cleaning results are obtained. 

	Introducing an SCL opens the possibility of having a single cleaning limit for many different manufacturing units, which can help to further standardize cleaning validation. In fact, the SCL of 25 mg/m

 was eventually introduced in five different manufacturing units.

	With its general applicability and ease of use, the SCL concept can be compared to the classification system for cleanrooms, where the room class is independent of the room size or the nature and quantity of product that is handled in the room. The SCL is, similarly, independent of equipment size, products, or batch size.

	Via the SCL, one can imagine a classification of API manufacturing equipment (e.g., in class 5, 25, and 50 mg/m2). Such equipment classes, which in fact would reflect validated cleaning commitments, could be criteria to consider when allocating new products to manufacturing sites or outsourcing API manufacturing. Comparing the SCL and the MSSR matrix is a straightforward way to start a cleaning risk assessment and to implement the new EMA guideline on shared facilities (3). In addition, adopting the SCL offers a number of benefits. For example, it:

		Allows for a consistent level of cleanliness to be enforced over a whole production unit

		Simplifies cleaning validation because there is no need to compute and justify the cleaning limit in every single validation plan; the limit can instead be justified once in an appropriate document, (e.g., in the cleaning validation master plan)

		Allows manufacturers to standardize validation activities in the analytical laboratory: Specifications for swab tests are constant, as are the working ranges of high-performance liquid chromatography (HPLC) methods and when spiking coupons for recovery studies.

	Last but not least, the SCL is a simple concept that is easy to understand and to explain.

	The authors would like to thank Sébastien Schuehmacher, Michael Wölfle, and Markus Schaefer for their relentless cleaning efforts and their support in the development of the SCL concept, and to Ian Udri for compiling all the data.

, 31 (1) pp. 74-83, 2007.
	2 M. Crevoisier et al, 

Guideline on Setting Health Based Exposure lLmits for Use in Risk Identification in the Manufacture of Different Medicinal Products in Shared Facilities,


	Vol. 41, No. 9
	September 2017
	Pages: s12-s16

	When referring to this article, please cite it as M. Crevoisier and T. Peglow, " Cleaning Validation for APIs," 

	Michel Crevoisier is former senior quality expert for Novartis Pharma AG, and Thomas Peglow is senior cleaning validation expert at Novartis Pharma AG, Switzerland. The views expressed in this article are those of the authors and not necessarily of their employers.

The industry needs a single standard cleaning limit at 25 mg/m2.


Cleaning Validation for APIs

In a previous article, the authors discussed the lack of scientific justification for the 10-parts-per-million (PPM) limit commonly used in cleaning validation and the lack of any regulatory requirement for it (1). This article discusses how the 10ppm limit, along with the 0.001 dose limit, are not truly risk-based approaches and are also unsound from an operational standpoint as they have caused unnecessary difficulties for many companies. A statistical process-based approach to demonstrate that acceptable cleaning has been achieved compared to a health-based limit is proposed.
	
	The use of a default limit value expressed as a concentration (ppm) has its origins in food safety concerns related to pesticide residues (2). The value of 10 ppm comes from an article written by Gary Fourman and Dr. Michael Mullen more than 20 years ago (3), which suggested a combination of the following for setting acceptance limits for cleaning:
	 

		No more than 0.001 dose of any product will appear in the maximum daily dose of another product

		No more than 10 ppm of a product will appear in another product

		No quantity of residue will be visible on the equipment after cleaning procedures are performed.

	
	Since that time, many pharmaceutical companies have adopted some or all of the criteria suggested in this article, including the 10-ppm value (4). Due to difficulties and inconsistencies in the application of these approaches to the wide variety of pharmaceutical dosage forms, some modifications to these limits were suggested such as using 0.01 or 0.1 (5). Some companies have had issues meeting these limits, in particular the 10 ppm, and have moved to higher values, and some have even published their justifications (6). Consequently, no single, consistent approach exists for setting limits for cleaning validation (2).

	In 2010, the International Society of Pharmaceutical Engineering (ISPE) published its 

Risk-Based Manufacturing of Pharmaceutical Products (RiskMaPP) Baseline Guide

 (7) that introduced the acceptable daily exposure (ADE) as the appropriate value for assessing pharmaceutical manufacturing risks, including for cleaning validation, as it is a value based on all the available pre-clinical and clinical data for a compound. The European Medicine Agency (EMA) uses the term permitted daily exposure (PDE), which it has declared to be “effectively synonymous” with ADE and defines it as “a substance-specific dose that is unlikely to cause an adverse effect if an individual is exposed at or below this dose every day for a lifetime” (8).

	Although the PDE is very conservative and has a superior scientific basis compared with previous approaches, there has been some objection and resistance to the use and implementation of PDE (9) but nothing substantive has ever been published in a peer-reviewed journal. In 2014, EMA issued a guideline that required all companies to develop PDEs by June 2015 for any medicinal product newly introduced into shared manufacturing facilities. It also required that PDEs be used to identify risks by December 2015 for medicinal products already produced in a shared manufacturing facility (8).

	While the guideline was being developed, some biopharmaceutical companies lobbied against it, asserting that their products are denatured or degraded by cleaning and should therefore be exempt from the requirement to establish ADEs for such drug products. EMA subsequently provided language in the final version of the guideline permitting these companies to use “alternative approaches” if they can be “scientifically justified.”

	It has been demonstrated, however, that within the risk-based approach of International Council for Harmonization Q9 (10), the risk associated with any residual biologic (i.e., based on its ADE) still must be addressed (11). Also, some companies have been trying to justify retaining the use of 10ppm and 0.001 dose limits because in many cases these old approaches result in lower limits than the ones based on ADEs. Their rationale is that the most stringent limit of the three limit-setting methods should be imposed.

	The following discussion addresses why retaining either the 10ppm or the 0.001 dose limits is unwarranted. First, neither of them are “health based” and, therefore,  not “risk-based.” Second, they have caused many operational issues, and third, and more importantly, they actually interfere with an appropriate risk assessment, which should now be necessary in accordance with ICH Q9.
	 

	0.001 dose limits are not truly health-based

	The issue of the 0.001 dose limits not being health-based has been addressed in a previous article (2) and will be revisited briefly here.

 shows how three compounds with widely varying severity of hazards (low to high) will have cleaning limits set using the 0.001 lowest clinical dose approach that are inconsistent with a health-based assessment. The fundamental problem with basing a cleaning limit on the therapeutic dose is that the choice of medication and dose is tailored to an implicit risk-benefit medical decision for individual patients who are suffering from a particular disease or condition, which may be contraindicated in other patient populations. For example, a pregnant woman who takes a Cetirizine hydrochloride tablet should not be exposed to unsafe levels of teratogenic drugs such as Ribavirin or oncology drugs such as Capecitabine. So logically, how could the cleaning limit for Cetirizine hydrochloride be set so much lower than that for a teratogen or anti-cancer drug?


	
	This simple comparison should be sufficient evidence that the use of 0.001 of the lowest therapeutic dose ignores the underlying scientific data for a drug. If the first effect of a substance (also known as lead effect) is identical with the desired therapeutic effect, then there is a direct relationship between the therapeutic dose and the no-observed-effect level (NOEL) of this substance. In many cases though it is sufficient to lower the therapeutic dose by a factor much less than 1000 to reach this NOEL; for a few drugs, however, a factor higher than 1000 may be necessary. If the lead effect of the substance is qualitatively different from the desired therapeutic effect (e.g., teratogenicity or genotoxicity), then this effect may have a different dose-response curve from the dose-response curve of the desired effect in the target patient. In cases where there is no correlation at all between the NOEL for the lead effect and that for the therapeutic dose, the ADE cannot be a simple fraction of the therapeutic dose.

	The principles behind applying risk in pharmaceutical manufacturing were introduced in the ICH Q9 (formally adopted by FDA and the Ministry of Health, Labour and Welfare (MHLW) in 2006) and its applicability to cleaning (including acceptance limits) mentioned in its Annex II.4 and to validation mentioned in Annex II.6. In accordance with ICH Q9, risk can be defined as 


	
	Likewise, risk can be defined as a function of the probability of adverse health effects arising from exposure to a hazard. The types and probabilities of effects from exposure to a drug are API-specific and dose-dependent. Consequently, for cleaning if the identified hazard is an API then the above equation for risk can be expressed as 

 provides the measure of the severity of the hazard that the API presents in long-term exposure. It should be noted that this concept can be applied to any agent, not just APIs.

	The ADE for an API (or any compound) provides a value that can be substituted for the “severity of harm” in Equation 1 (as shown in 

); but what value should be substituted for “probability of exposure”?
	 

 shows the “probability of exposure” to API residues can be determined by comparing the data collected during cleaning to the limits calculated from the ADE of the API. The distance between the data and the ADE-based limit provides the margin of safety and indicates the probability of patient exposure to the API residues resulting from ineffective cleaning. If this margin of safety could be measured, then this value could be used for the “probability of exposure” parameter in 

	A simple statistical method has already been presented for quantifying the “margin of safety” and that is the process capability index (Cpk) or more precisely Cpu (12, 13). The Cpu measures the capability of a process only for an upper specification limit (USL). It is a simple, straightforward comparison of the spread of the data (its variability) to the distance between the mean of the data and the upper specification limit (


	
	This equation can be satisfied using cleaning data such as shown in 

	Readily available software provides powerful statistical and graphical abilities that can calculate the Cpu and quickly help visualize the relationship of data to limits. 

 shows a comparison of two cleaning processes against ADE-based limits using Minitab 17 (14).

, residue data for a cleaning process are distributed well below the ADE-based limit, thereby providing a very wide “margin of safety,” and hence a very low probability of patient exposure. On the right, however, residue data for a cleaning process are distributed relatively close to the ADE-based limit, thereby not providing an adequate “margin of safety.” Consequently, this scenario presents a high probability of patient exposure. It should be noted that although no individual data point failed the limit on the right (they would pass in a traditional cleaning validation), the statistical evaluation demonstrates that failures are likely (PPM > USL = 13,532) and the cleaning process has a risk of failures (i.e., exceeding the ADE-based limit).

	It can be seen that the Cpu can be easily used as a measure for the “probability of exposure” parameter in the cleaning validation risk evaluation.

 shows an actual evaluation of a very low hazard product using total organic carbon (TOC) analysis. Using the ADE, the calculated swab limit was 3179ppm (USL value) in the analytical sample. Because the cleaning procedure was effective, the TOC swab results were far below this limit, resulting in a Cpu value of 6120 with a lower 99% confidence limit for the Cpu of 3084. To put this result into perspective, in a Six Sigma organization, the goal for a process capability is only 2. Clearly, this analysis shows a very low-risk cleaning situation.

	How 0.001 dose limits interfere with the evaluation of risk

	The implementation of the ADE as the starting point in limit calculations for cleaning validation has required revisiting older calculations that use 0.001 dose and 10 ppm. This has revealed situations where the previous limits were not low enough and potentially posed a risk to patients, and situations where the previous limits were lower (often much lower) than those calculated using the ADE.

 shows a comparison of the ADE values vs. 0.001 of the lowest clinical dose for 304 pharmaceutical products (data were provided by several pharmaceutical companies). The data were plotted as a ratio of the ADE to the 0.001 dose, which makes 1.00 on the Y axis the point at which the ADE equals the 0.001 dose (Note: the ratios had to be plotted on a logarithmic scale in order to get all the data on one graph because the values of the ratios varied widely).

	These ratios show that, in more than 15% of the cases, the ADE was lower than the 0.001 dose including two instances in which the ADE was 100 times lower indicating a potential issue for patient exposure. However, these ratios also show that in 85% of the cases, the 0.001 of the lowest clinical dose was lower than the ADE. In fact, of these cases, 47% were 10 times lower, 12% were 100 times lower, and there were six instances that were 1000-8000 times lower than the ADE. This demonstrates that, for the vast majority of drugs, the 0.001 dose limit has been overly conservative.

	While at first it may be startling to realize that most cleaning acceptance limits based on the use of 0.001 dose have been grossly overstated, the most important thing to realize about the increase in cleaning acceptance limits for most APIs as a result of implementation of health-based ADEs is its impact on the ability to evaluate the true risk. 

 shows how the true margin of safety can now be measured using the ADE-based cleaning acceptance limit. Where cleaning data were close to failing using the 0.001 dose-based limits and had a poor Cpu, moving to the ADE-based limits shows that there is a wide margin of safety and an excellent Cpu.

 is a comparison of the ADE/0.001 dose data of one of the products from 

. A hypothetical set of TOC swab data with a mean of 75 parts per billion (ppb) and a standard deviation of 7.5 ppb was evaluated against analytical swab limits calculated using both the ADE and the 0.001 dose limits for that product. The calculation of both analytical swab limits assumed a maximum daily dose of 1000 mg, a batch size of 500kg, and a total surface area of 100,000 in

 for the next product. For sampling, a swab area of 100 in

, a dilution volume of 10 mL, and a recovery factor of 100% were assumed. This resulted in an ADE-based analytical swab limit of 250,000 ppb and a 0.001 dose-based analytical swab limit of only 100 ppb.

, there is a wide margin of safety with a Cpu of 9224 even for the 99% lower confidence level. Because they are calculated from the mean and standard deviation, Cpus can have confidence intervals calculated just as any mean can). Using the 0.001 dose-based limit, there is no margin of safety and a Cpu of only 0.84 for the 99% lower confidence level. The true margin of safety cannot be measured using the 0.001 low-clinical dose-based limit methodology. Applying the 0.001 low-clinical dose based limit, the process capability is artificially low across most of the compounds and could likely lead to false conclusions of cleaning failures for low risk products.

	How 10-ppm limits interfere with the evaluation of risk

	Back in 1993, Fourman and Mullen had recognized that limits calculated based on dose increased as the potency decreased and resulted in higher and higher limits for low potency (i.e., high therapeutic dosage) drugs (3). To account for this effect, they suggested that a 10-ppm limit be imposed when the 0.001 low-clinical dose-based limit went too high and the maximum allowable carryover (MAC) into the next batch would exceed 10 ppm. The only justification these authors provided was that maximum allowable ppm limits were used in the food industry. There was no explanation as to why a maximum value of 10 ppm was selected. In the graphical representation in 

, the implications of imposing the arbitrary lower limit of 10 ppm criterion after applying the 0.001 dose-based limit are seen. The end result is that, after a certain point, all lower potency drugs have the same cleaning acceptance limit, regardless of how low their potency or how low their risk levels are.

, the true margin of safety cannot be measured when imposing the 10-ppm limit as shown in 

. The 10-ppm limit simply cuts off, or truncates, the ability to measure the margin of safety, making a relatively low-risk compound look like a high-risk compound, because the 10-ppm limit is only imposed when the 0.001 dose limit is higher than the 10-ppm limit. 

 illustrates a hypothetical example of a low potency drug where the ADE is higher than the 0.001 dose-based limit, which in turn is higher than the 10-ppm criterion (data not to scale).

	The recent implementation of a regulatory requirement mandating the use of health-based cleaning limits for shared medicinal manufacturing facilities has resulted in a significant increase in the level of cleaning limits for most drugs. It should be understood that the previous use of 0.001 low-clinical dose limit, in the case of most APIs, has been excessively conservative, which has caused significant needless complications with manufacturing operations, especially for low-hazard products. Many companies have gone from using disposable items such as scoops and other utensils, and modifying manufacturing schedules, to dedicating parts, pieces of equipment, or even entire manufacturing lines to deal with these artificially low limits, all the while knowing that their products were low risk. While there have been calls to return to the 10-ppm and 0.001 dose limits as they are more stringent, this reversion would simply have the industry using limits that do not consider all the pre-clinical and clinical data available for each API. And worse, it would perpetuate the many unjustified operational issues for low-hazard products. If the argument for stringency was truly valid, the case should be made to move to a 0.00001 low-clinical dose limit to also cover those 15% of the compounds where the ADE concept has revealed that the 0.001 low-clinical dose limit has been insufficiently protective. Thus, reverting to the 0.001 low-clinical dose limit would be against the science- and health-based evaluation process EMA has sought to institute for cleaning validation. Moreover, such a move would lead to many products being falsely judged as failing cleaning validation, and perhaps ultimately requiring their manufacture in dedicated facilities. The proposal to keep cleaning limits as low as possible is not a valid argument if the real goal is to correctly and objectively identify and control the level of risk to the patient. Maintaining the 10-ppm limit as a default to prevent excessive carryover is equally unjustified, as “visually clean,” a third criterion not discussed herein and which is almost universally implemented to satisfy the inspection criterion required by 21 

) 211.67(b) (15), is more than satisfactory to keep any product carryover low.

	The pharmaceutical industry has been moving toward a science and risk-based approach since 2001 with the roll out of FDA’s CGMPs for the 21st Century (16), the process analytical technology (PAT) initiative (17), the process validation guidance (18), and the ICH Q series guidelines, among others. While this movement has been slow, the authors believe that the time is right for cleaning validation to embrace a science and risk-based approach to acceptance limits, in conjunction with the process capability methods identified here.

	According to ICH Q9, “The level of effort, formality, and documentation of the quality risk management process should be commensurate with the level of risk.” Consequently, cleaning validation efforts should be focused on where the risks are, with the science behind the ADE informing us which products have the greatest risk (see 

). This was one of the stated goals of the Risk-MaPP Guide. The Cpu then informs us what the probability of patient exposure is and can be used to complete the calculation shown in 

	Continuing the use of 10-ppm and 0.001 low-clinical dose criteria obscures the true patient risk and presents a potential compliance gap as well as operational difficulties, which is why both of them should be abandoned. With the requirement to have established API-specific ADEs, the 10-ppm and 0.001 low-clinical dose should now be relegated to a chapter of history. The ADE must now be used for risk assessment for all products. In concert, statistical process control (SPC) methodologies should be used for establishing ongoing cleaning validation limits based on actual process data as has previously been described (12).

	The authors wish to thank Osamu Shirokizawa, Mariann Neverovitch, and Joel Young for reviewing this article and for their insightful comments and helpful suggestions.

, 40 (1) pp. 52-56 (January 2016).
	2. A. Walsh, 

, 31 (4): 74-83, (July/August 2011).
	3. G. Fourman and M. Mullen, 

 17 (4): 54-60, (April 1993).
	4. D. LeBlanc, “Residue Limits for Cleaning Validation in Finished Dosage Form Manufacturers,” PDA Letter, Parenteral Drug Association 2006 Survey on Cleaning Acceptance Limits, May 2007.
	5. Parenteral Drug Association, “Points to Consider for Cleaning Validation,” 

, Bethesda, MD, 1998.
	6. ECA Academy, “Justification of Limits for Cleaning Validation in the Manufacture of Active Pharmaceutical Ingredients,” May 14 2007, www.gmp-compliance.org/eca_news_928.html
	7. ISPE, 

ISPE Baseline Guide: Risk-Based Manufacture of Pharmaceutical Products (Risk-MaPP)

, First Edition (ISPE, September 2010).
	8. EMA, 

EMA Guideline on Setting Health Based Exposure Limits for Use in Risk Identification in the Manufacture of Different Medicinal Products in Shared Facilities

, EMA/CHMP/CVMP/ SWP/169430/2012, 20 November 2014, www.ema.europa.eu/docs/en_GB/document_library/Scientific_guideline/2014/11/WC500177735.pdf.
	9. D. LeBlanc, Risk-MaPP-Gate Website, https://sites.google.com/site/riskmappgate/home
	10. ICH, 

, Step 4 (ICH Nov. 9 2005), www.ich.org/fileadmin/Public_Web_Site/ICH_Products/Guidelines/Quality/Q9/Step4/Q9_Guideline.pdf.
	11. A. Walsh, 

 31 (5) 44-49 (September/October 2011).
	13. D. C. Montgomery, 

Introduction to Statistical Quality Control

,  7th Edition (John Wiley & Sons, Inc., Hoboken, NJ, 2013).
	14. Minitab Inc.  Minitab 17. State College, PA. (2016).
	15. 21 

) 211.67(b)
	16. FDA, Pharmaceutical cGMPs for the 21st Century Industry-A Risk-Based Approach: Final Report (FDA, September 2004 (FDA), www.fda.gov/downloads/Drugs/DevelopmentApprovalProcess/Manufacturing/QuestionsandAnswersonCurrentGoodManufacturingPracticescGMPforDrugs/UCM176374.pdf.
	17. FDA, 

Guidance for Industry: PAT-A Framework for Innovative Pharmaceutical Development, Manufacturing, and Quality Assurance 

(FDA, September 2004), www.fda.gov/downloads/Drugs/.../Guidances/ucm070305.pdf.
	18. FDA, 

Guidance for Industry: Process Validation: General Principles and Practices

 (FDA, January 2011), www.fda.gov/downloads/Drugs/.../Guidances/UCM070336.pdf.

	When referring to this article, please cite it as A. Walsh et al., "Cleaning Limits-Why the 10-ppm and 0.001-Dose Criteria Should be Abandoned, Part II," 

Conventional limit-setting techniques are not health-based and can make risk assessment more difficult.

Cleaning Limits—Why the 10-ppm and 0.001-Dose  Criteria Should be Abandoned, Part II

	The revised Chapter 5 of Part 1 of the European Union’s GMP guidelines and the European Medicines Agency’s (EMA) 

Guideline on Setting Health Based Exposure Limits for Use in Risk Identification in the Manufacture of Different Medicinal Products in Shared Facilities

 came into effect in 2015 (1, 2). Both publications expect the industry to use a “quality risk management process, which includes a potency and toxicological evaluation” when deciding whether to use dedicated or shared facilities to produce pharmaceuticals. Consequently, the same comprehensive approach based on health risks should be followed to determine the acceptable limits set for cleaning shared equipment. The authors believe it is the right time to abandon the 10-ppm rule in favor of more scientific and data-driven health based limits.


	Health authorities do not require shared equipment to be cleaned to 10 ppm of the next product. Careful reading of GMP regulatory texts (3-11), FDA’s 

Guide on Inspections--Cleaning Validation 

(12), and the Pharmaceutical Inspection Convention and Pharmaceutical Inspection Co-operation Scheme’s (PIC/S) Recommendations on Cleaning Validation (13) yields no indication that any of these health authorities or PIC/S ever explicitly prescribed numerical cleaning limits. When it comes to cleaning limits, these guidelines all follow a similar discourse--they expect companies to abide by a general duty of care, sometimes adding that the limits should be “based on the knowledge of the materials involved.” Some of the texts mention the three traditional rules originally proposed by Fourman and Mullen (14) (i.e., visually clean equipment, residue levels ≤ 1/1000th daily dose, and ≤10 ppm of next product) as examples but never as binding rules. In this case, consistent with other performance-based regulations, the responsibility to establish and justify cleaning limits for shared facilities remains with pharmaceutical manufacturers.

	The 10-ppm limit goes against the basic principle of toxicology that it is the dose that makes the poison. Parts per million (ppm) is a concentration unit (i.e., a relative number and not an absolute one). In chemistry parlance, it is an intensive property (such as density or melting point) that is independent of the amount of the material, and as such, it is not suitable to describe a dose, which is an extensive property (such as mass or volume) that is dependent on the amount of material. Defined as a concentration, the limit dose will depend on the maximum amount of something else, in this case, on the amount of the following product (or API, the approaches vary) administered to the patient. Assuming that therapeutic doses range from 0.1 mg to 5000 mg per day, the amount of a contaminating substance corresponding to 10 ppm will span the same range of 4 logs. Using a limit of 10 ppm without risk characterization of exposure is equivalent to pretending that it makes no difference if a drug is contaminated by 1 ng (the equivalent of 10 ppm of 0.1 mg) or by 50 mcg (the equivalent of 10 ppm of 5000 mg) of the same substance.

	Whether the contaminant is Botox or a mild antacid, the limit amount computed by 10 ppm will be the same. Therefore, involvement of a risk assessment scientist (e.g., toxicologist or physician) is crucial for the characterization and differentiation of health risks and the overall risk management process. 

 compares the weight of scientific justification for three different approaches to compute the maximum safe surface residue (MSSR). In contrast to approaches targeting 10 ppm and 1/1000th of the daily dose, the permitted daily exposure (PDE) approach ensures that available preclinical and clinical data are considered in computation of the MSSR.

Does the 10-ppm limit ensure good cleaning?


	No; 10 ppm is a concentration unit that gives no indication on cleanliness. Operationally, a sensible physical unit for cleanliness must express the amount of residue per unit surface area (e.g., mg/m

). By switching from carryover (amount of residue [g]) to cleanliness [mg/m

], the surface is normalized, and the cleanliness of all pieces of equipment of a facility can be compared; cleaning results between facilities and even between companies can also be compared. When speaking of cleanliness, the size of the equipment does not matter; “cleanliness” is an intensive property of the equipment. Therefore, until the 10 ppm has been converted into a cleanliness value in [mg/m

], it is not possible to say what level of cleanliness the 10 ppm eventually represent. Depending on the particular situation, 10 ppm may mean 1 mg/m

Can 10 ppm be used to identify risks according to the EMA guideline on shared facilities?


	The EMA guideline (2) describes primarily a method for computing health-based limits based on all available toxicological and pharmacological information, which are termed permitted daily exposure (PDE). The 

guideline’s focus is not on how to set cleaning limits. The PDEs are only one element needed to set cleaning limits, which are calculated as MSSR. The other element is the knowledge of the efficiency of a cleaning process, which is independent from the next product. Therefore, while there is a link between PDEs and cleaning limits, they are absolutely not the same. This is expressed in the EMA guideline, which says, “these levels can be used as a risk identification tool and can also be used to justify carryover limits used in cleaning validation.”

	An evaluation of the effect of using 10 ppm in addition to PDE to identify risks was performed with the example of an actual product portfolio of a multiproduct API manufacturing facility making 11 different APIs. The basic data of the APIs are given in 

	All changeovers from a preceding product (α) to a following product (β) were considered. The matrix of maximum safe carryovers (MSC) was first calculated according to 

. In a second step, the matrix of MSSR was calculated by dividing the MSC by the equipment surface area used to make each following API (see 

	Each cell of the MSSR matrix shown in Table II reflects an individual product change from an API α to an API β and contains the MSSR for the particular change (i.e., the cleanliness required to make the corresponding API β when it is made after API α).

 shows a vivid landscape with strong contrasts, with values from 5 to 146,970 mg/m

. It is easy to identify the changeovers that pose the greatest risks. Comparing the calculated MSSRs with 

the results of approximately 250 swab results that were gathered over two years, 99% of all swabs gave results <25 mg/m

; high-risk changeovers are identified as the ones that require a MSSR of < 25 mg/m

	For example, the changeover from product D to product E was calculated as follows:
	

	Five products C, G, I, J, and K are potentially at risk, but only when they follow certain products, like A, G, I, and K. Similarly A, G, I, and K represent a risk for the other products, but not for all of them.

	The comparison of the matrix of MSSR based on PDE with the known efficiency of the cleaning processes allows for a sound identification of risks in line with the spirit of the EMA guideline. The comparison may give the concept of “worst-case approach to cleaning validation” a new orientation by shifting the focus from the hazards of worst-case products (with regard to solubility, cleanability, toxicity, etc.) to the worst-case risks, (i.e., to the “risky” changeovers between two products). Such risks will have to be mitigated, for example, by development of better cleaning procedures, avoiding high-risk changeovers by careful campaign scheduling and dedication of certain process lines or parts of the equipment to a group of products. It is important, however, to accept that cleaning will have its limits: There will be cases where the risk cannot be mitigated by cleaning alone.

	How does the risk identification work with the 10-ppm rule? Because the calculation is independent of the preceding products, the matrix collapses to a list of batch sizes divided by 105 (see 

based MSSRs all come out in a narrow range, between 9 and 59 mg/m2 and they do not really allow clear differentiation between the high- and low-risk changeovers. Compared to PDE, the 10-ppm calculation does not improve the detection of high risks.

	In practice, the 10-ppm methodology gives the impression that there are no low-risk changeovers. If the 10-ppm row in Table III is followed, whenever C, D, I, or K is made, the manufacturer has to meet the MSSR dictated by 10 ppm. The assumed cleaning limit of 25 mg/m

 will not meet the requirement to make C, D, I, or K in any case, independently of what the preceding product was.

 merely reflect the low risk associated with the corresponding changeovers. They should not be read as cleaning limits. The suspicion that production will set the cleaning limit at the level of the PDE-derived MSSR is not justified. Even if the MSSR values for particular changeovers are high, this does not justify reducing the cleaning effort, because cleaning will still have to fulfill the requirement of visually clean equipment. This will be sufficient to ensure a suitable minimal hygiene standard.

	The cleaning limit should also be based on the knowledge of the capability and the reliability of the cleaning processes, and it will ideally be set as far below the MSSR as reasonably possible.


	There is no scientific justification for a limit concentration that ignores the hazards presented by potential contaminants. Implementation of the 10-ppm criterion in cleaning validation does not ensure safe cleaning. and it may distort the identification and analysis of cross contamination risks. Moreover, it may impose more stringent limits for cleanliness than are required by health authorities without adding to the patient safety. Eliminating an unnecessary requirement serves to focus attention and resources on high-risk situations. The 10-ppm criterion offers no scientific or practical benefit and can be discontinued in cleaning validation practice.


	The authors would like to thank Ed Sargent and Courtney M. Callis for their views during compilation of this manuscript.


	1. European Commission, The Rules Governing Medicinal Products in the European Union; Volume 4 EU Guidelines for Good Manufacturing Practice for Medicinal Products for Human and Veterinary Use (Brussels, August 2014).
	2. EMA, Guideline on Setting Health Based Exposure Limits for Use in Risk Identification in the Manufacture of Different Medicinal Products in Shared Facilities (London, November 2014).
	3. European Commission, Volume 4 Good Manufacturing Practice (GMP) Guidelines, Parts I & II, http://ec.europa.eu/health/documents/eudralex/vol-4/index_en.htm, accessed Dec. 1, 2015.
	4. European Commission, Volume 4: EU Guidelines for Good Manufacturing Practice for Medicinal Products for Human and Veterinary Use, Annex 15: Qualification and Validation (Brussels, February 2014).
	5. FDA, Code of Federal Regulations, Part 211 Current Good Manufacturing Practice for Finished Pharmaceuticals, www.accessdata.fda.gov/scripts/cdrh/cfdocs/cfcfr/CFRSearch.cfm?CFRPart=211, accessed Dec. 1, 2015.
	6. Ministério Da Saúde Agência Nacional De Vigilância Sanitária, Dispõe sobre as Boas Práticas de Fabricação de Insumos Farmacêuticos Ativos (Brazil, December 2014).
	7. Agência Nacional de Vigilância Sanitária--Anvisa, Validação de limpeza para farmoquímicas, http://portal.anvisa.gov.br/wps/wcm/connect/2443e9804f1af51ea138bdc88f4b6a31/Guia+de+valida%C3%A7%C3%A3o+de+limpeza+para+Farmoqu%C3%ADmicas.pdf?MOD=AJPERES, accessed Dec. 1, 2015.
	8. Health Canada, Cleaning Validation Guidelines, Guide-0028 (Ontario, January 2008).
	9. South Korea, Good Manufacturing Practice for Medicinal Products for Human Use, Enforcement Regulation of Pharmaceutical Affairs Act [Annex 2], http://wenku.baidu.com/view/42de6028ed630b1c59eeb53e.html, accessed Dec. 1, 2015.
	10. South Korea, Provisions for Conducting Validation for Drugs, etc, The KFDA Notification No. 2009-173 (December 2009)
	11. China FDA, Good Manufacturing Practice for Drugs (2010 Revision), http://eng.sfda.gov.cn/WS03/CL0768/65113.html, accessed Dec. 1, 2015.
	12. FDA, Validation of Cleaning Processes (7/93), Guide to Inspections: Inspections of Cleaning Processes (7/93), www.fda.gov/ICECI/Inspections/InspectionGuides/ucm074922.htm, accessed Dec. 1, 2015.
	13. PIC/S, Recommendations on Validation Master Plan, Installation and Operational Qualification, Non-Sterile Process Validation, Cleaning Validation; PI 006-03 (September 2007).
	14. G.L. Fourman and M.V. Mullen, Pharmaceutical Technology, April issue, 54-60 (1993).

:
	When referring to this article, please cite it as M. Crevosier, "Cleaning Limits-Why the 10-ppm Criterion should be Abandoned," 

The 10-ppm criterion for the acceptable concentration of potential API in cleaning validation to minimize cross contamination into next product has been employed for many years. This article describes why the 10-ppm criterion, which was established based on analytical limitations and estimates of acceptability, is no longer necessary and why a risk-based approach should be universally adopted.