Full opinion text
Table of Contents $ Pm BACKGROUND . rfi. I.FINDINGS OF FACT . A. The Parties. B. The Regulatory Scheme . CR C. FDA Investigatory Practice. OI D. Drug Testing Overview. Oí 1. Failures. Oi 2. Failure Investigations. 3. Outliers. Retesting. Page 469 Resampling. 470 Remixing. 471 Averaging. 471 Releasing a Batch for Distribution. 471 Blend Testing. 472 (a) Sample Size. 472 (b) Site of Sampling. 474 E. Criticisms of Barr. 474 1. Manufacturing Process Validation . 474 (a) General Requirements. 474 (b) Specific Process Problems . 475 (1) Batch Failures. 475 (2) Retrospective Validation. 476 (a) Omitting Failing Results. 476 (b) Omitting Failing Batches. 476 (c) Insufficient Number of Batches. 477 2. Failure Investigations. 477 3. Release of “Failing” Batches. 478 (a) and (b) Rejecting Failing Results As Laboratory Errors After Retesting. 478 (c) Rejecting Failing Results After Resampling. 479 4. Failure to Control Manufacturing Process Steps. 480 (a) Barr Blend Testing. 480 (1) Site of Sampling. 480 (2) Sampling Procedure. 480 (3) Sample Size. 481 (b) Mixing Time. 481 (c) Particle Size/Distribution. 481 5. Failure to Validate Testing Methods. 481 6. Cleaning Validation Deficiencies. 482 7. Record-Keeping Deficiencies . 483 8. Failure to File ANDA Supplements and Field Alerts. 484 (a) ANDA Supplements. 484 (b) Field Alerts. 484 II. CONCLUSIONS OF LAW. 485 A. Jurisdiction. 485 Standard for Injunctive Relief. B. 485 The Federal Food, Drug and Cosmetic Act C. 485 Evaluating Barr Laboratories. D. 487 Fashioning a Remedy. E. 487 (1) General CGMP Violations. 487 (2) Product Validation. 488 (3) Product Recall. 488 CONCLUSION. 491 OPINION WOLIN, District Judge. Currently before the Court is plaintiffs application for a preliminary injunction directing defendants to suspend, recall or revamp numerous products in their current product line. Plaintiff filed this action in the United States District Court in the Southern District of New York on June 12, 1992, alleging that defendants violated the Federal Food, Drug, and Cosmetic Act. In accordance with the first-filed rule, the case was transferred to the District of New Jersey on June 26, 1992, where it was consolidated with an action defendants brought against plaintiff on April 24, 1992, seeking relief from allegedly ad hoc drug regulations. On July 10, 1992, the Court filed case management and protective orders. Beginning on August 17, 1992, and continuing intermittently to October 12, 1992, through the testimony of inspectors, experts and employees, the parties presented exhaustive but conflicting views of defendants’ business practices. At the Court's direction, on October 26, 1992, the parties submitted proposed findings of fact and conclusions of law. BACKGROUND Each day with confidence and hope millions of people in the United States and other countries reach for pills, powders, capsules and syrups to relieve or prevent an infinite number of physical and mental ailments. The weighty task of ensuring the integrity of these products, frequently unquestioned by most consumers, falls to the Food and Drug Administration, which monitors the practices of the drug industry through a system of approvals and investigations. Built into this maze of often ambiguous rules, however, is the recognition that drug manufacturers are businesses, which must follow efficient as well as effective procedures. The current conflict surrounding these rules is best characterized as a confrontation between a humorless warden and his uncooperative prisoner. Exchanging heavy blows, the parties generated a record of more than twenty-three hundred pages of testimony, almost four hundred exhibits and numerous lengthy declarations. The plaintiff presented two witnesses, a government inspector, David Mulligan, and a regulatory expert, Dr. Robert Gerraughty. Defendants countered with a statistician, Dr. Sanford Bolton, their own regulatory expert, Dr. Christopher Rhodes, an analytical chemist, Dr. Norman Atwater, an expert in pharmaceutical biology, Dr. Murray Cooper, and Barr employees. These witnesses revealed an industry mired in uncertainty and conflict, guided by vague regulations which produce tugs-of-war of varying intensity. The divergent views presented to the Court reflect not only a difference of perspective, but also the changes made at Barr Laboratories since the first threat of this litigation. As a result, the record is a composite of two trials: the case that was and the case that is. (1789:6 (discussion between Court and Ms. Kaswan)); Barr Response ¶ 63. As such, the bases upon which some of the government’s criticisms rest have disappeared during the course of this litigation. Wary of this timing element, the Court has reviewed the lengthy record and the parties’ proposed findings with the dual desire to protect an unsuspecting public and to avoid unnecessarily burdensome rules and now makes the following findings of fact and conclusions of law. I.FINDINGS OF FACT A. The Parties 1. Plaintiff, United States of America, brought this action on behalf of the Food and Drug Administration (“FDA”), an agency within the United States Department of Health and Human Services. 2. Defendant Barr Laboratories, Inc. (“Barr”) is a manufacturer and distributor of drug products in the interstate and foreign commerce of the United States. Barr is incorporated under the laws of the State of New York and is doing business in Northvale, New Jersey and Pomona, New York. 3. Barr currently manufactures sixty drug products. Before this action was commenced, Barr voluntarily suspended the production and distribution of 115 drug products pending further order of the Court. 4. From 1970 until January 5, 1993, defendant Edwin A. Cohen was Barr’s President and Chief Executive Officer and was in charge of the day-to-day operations of Barr. His current title is Chairman of the Board and Chief Executive Officer. 5. Defendant Gerald F. Price is Barr’s Executive Vice President of Operations and is responsible for the performance of Barr’s quality assurance department. He has held this position since 1990. 6. Defendant Ezzel-Din A. Hamza is Barr’s Vice President of Technical Affairs and is responsible for regulatory affairs, research and development, and Barr’s chemistry and microbiology laboratories. He has held this position since 1989. B. The Regulatory Scheme 7. Under the Federal Food, Drug and Cosmetic Act (the “Act”), a drug is adulterated if “the methods used in, or the facilities or controls used for, its manufacture, processing, packing or holding do not conform to or are not operated or administered in conformity with current good manufacturing practice to assure that such drug meets the requirements of this chapter as to safety and has the identity and strength, and meets the quality and purity characteristics, which it purports or is represented to possess.” 21 U.S.C. § 351(a)(2)(B). 8. Current Good Manufacturing Practice (“CGMP”), explained in greater, but by no means sufficient, detail in regulations promulgated by the FDA, see 21 C.F.R. Parts 210 and 211, sets the minimum standards for drug manufacturers. Designed as a quality control measure to prevent super- and sub-potency, product mix-ups, contamination, and mislabeling, (870:20; 872:3 (Gerraughty)), the CGMP regulations outline general rules for all aspects of drug manufacture including buildings and facilities, personnel, equipment, drug components and containers, production, packaging and labeling, and record-keeping. Failure to comply with CGMP regulations renders any resulting drug product “adulterated” and the drug product and its producer subject to regulatory action. 21 C.F.R. § 210.1(b). 9. Congress recognizes the United States Pharmacopeia (“USP”), a nonprofit corporation which develops drug product standards with the help of professionals from academia, the medical community, the pharmaceutical industry and the FDA, as an official compendium. (1503:11 (Rhodes)). The USP supplements the CGMP provisions, specifying both the types of tests firms must perform and the range of acceptable results these tests must generate for releasing drug products. As with the CGMP regulations, the USP contains minimum testing requirements, (910:9 (Gerraughty)). These criteria, however, rest on the assumption that the products to which they apply satisfy CGMP. (911:11 (Gerraughty)). 10. Firms outline their chosen standards and procedures in new drug applications (NDAs) or abbreviated new drug applications (ANDAs), which are submitted to the FDA for approval. The ANDA guides a product’s manufacture and release, but does not supersede the overarching CGMP requirements. (448:24 (Mulligan)). 11. Thus, the CGMP regulations provide the yardstick with which FDA investigators, and the Court in the instant action, measure firm behavior. Ironically, the regulations themselves, whose broad and sometimes vague instructions allow conflicting, but plausible, views of the precise requirements, transform what might be a routine evaluation into an arduous task. 12. To the extent that the regulations create ambiguities, the industry can turn for guidance to literature from seminars and pharmaceutical firms, textbooks, reference books and FDA letters to manufacturers, (865:11 (Gerraughty)), or employ scientific judgment where appropriate. The Court, however, cannot rely on industry practice alone to determine whether an individual firm meets the statutory requirements, since industry standards themselves must be reasonable and consistent with the spirit and intent of the CGMP regulations. (1258:24 (Gerraughty)). C. FDA Investigatory Practice 13. FDA investigators conduct both pre-approval and general compliance inspections. During a pre-approval inspection, investigators review pending ANDA applications, (414:7 (Mulligan)), while inspectors conduct compliance investigations to determine whether the firm is following CGMP. After a compliance investigation, FDA inspectors must issue a Form 483 Notice of Inspectional Observations (“Form 483”), in which they record their observations about the firm’s more serious CGMP violations. 14. FDA investigators conducted a general inspection of Barr’s Northvale facility during August and September 1989 as well as separate general inspections of Barr’s Northvale and Pomona facilities from May to September 1991. After each inspection, the investigators issued Forms 483. 15. The 1989 Form 483, which contained six general observations, cited Barr for unvalidated manufacturing and cleaning processes, the lack of failure investigations, incomplete annual reviews and failure to explain retesting. See Exhibit 51. In the 1991 Pomona inspection, the FDA criticized nineteen products extensively and made general comments about Barr’s equipment and complaint logs, stability programs, raw material controls and documentation procedures. See Exhibit 3. The Northvale inspection that same year also censured individual products as well as failure investigations, validation and product mix-ups. See Exhibit 1. 16. In February 1992, FDA investigators returned to the Barr facilities in Northvale and Pomona. The resultant 1992 Forms 483 listed seventy-five and forty-seven criticisms for Northvale and Pomona, respectively, many of which also had been recorded during prior visits to Barr, in the areas of process validation, failure investigations and laboratory practices. See Exhibits 2, 4. D. Drug Testing Overview 17. Testing lies at the heart of a drug manufacturer’s successful operation. Through testing companies validate their processes and ensure the quality of batches for release. As the Forms 483 suggest, much of this current litigation stems from allegedly defective testing practices. With the mechanics of test-taking left undefined by the regulations, before discussing the specific allegations against Barr, the Court will outline the CGMP-required parameters which will guide its evaluation. 1. Failures 18. In the government’s view, a batch failure occurs each time an individual test result does not meet the specifications outlined in the USP or the firm’s ANDA. (132:22 to 133:3 (Mulligan)). In contrast, Barr does not classify initial out-of-specification results as batch failures. Instead, only after confirming out-of-specification results with additional testing, (1394:1 (Bolton); 1978:18 (Cooper)), pursuant to the firm’s predetermined testing procedure, (1979:7 (Cooper)), would Barr conclude that a batch failed. 19. Out-of-specification results obtained in the laboratory fall into three general categories: (1) laboratory error; (2) non-process-related or operator error; and (3) process-related or manufacturing error. (227:23 (Mulligan)). 20. Laboratory error can result from an analyst's mistake or malfunctioning laboratory equipment. (923:18 (Gerraughty)). Examples of analyst error include mistakes in calculations, the use of incorrect standards for comparison, and simple mismea-surement. (923:17 (Gerraughty)). Those human and mechanical errors which occur during the manufacturing process cause nonprocess-related errors. For example, manufacturing equipment may malfunction or an operator may fail to add the proper amount of an active ingredient. In contrast, process-related problems, such as an incorrect mixing time, occur even though the workers and machines function property- 21. While each type of problem is a matter of great concern which requires some form of corrective action, only non-process-related and process-related errors properly are labeled failures. As Inspector Mulligan acknowledged, all failures are not alike. (133:20 (Mulligan) (noting if “reason and logic or science” indicate that out-of-specification result is anomaly, batch need not fail)). An out-of-specification result identified as a laboratory error by a failure investigation or an outlier test, see Ex. 7 at 1503; (202:21, 798:20 (Mulligan)), or overcome by retesting is not a failure. (225:7-14 (Mulligan)). Thus, the Court is unwilling to adopt the government’s view of failure. 2. Failure Investigations 22. Only with an investigation will a firm be able to identify the cause of an out-of-specification result. CGMP requires a thorough investigation following: [a]ny unexplained discrepancy (including a percentage of theoretical yield exceeding the maximum or minimum percentages established in master production and control records) or the failure of a batch or any of its components to meet any of its specifications.... [which] shall extend to other batches of the same drug product and other drug products that may have been associated with the specific failure or discrepancy. A written record of the investigation shall be made and shall include the conclusions and followup. 21 C.F.R. § 211.192; see (924:19 (Ger-raughty) (violation of CGMP to discard out-of-specification results and pass batch on retesting alone)). 23. The government argues that an adequate failure investigation requires a timely, thorough, and well-documented review of the problem, which yields a written record containing: (1) the reason for the investigation; (2) a summation of process sequences that may have caused the problem; (3) the corrective actions necessary to save the batch and prevent recurrence; (4) a list of other batches and other products possibly affected along with their investigation results; and finally, (5) comments and signatures of production and quality control personnel regarding approval of any material reprocessed after additional testing. Government Findings 1159. Barr advocates a sliding-scale approach, claiming that the nature of the failure should govern the intensity of the investigation. 24.In accordance with the CGMP-re-quired for-cause failure investigation, (1027:13 (Mulligan)), the goal of every such inspection is to determine into which of the three failure categories the problem falls. The degree of inquiry required successfully to complete this task may vary with the object under investigation. As a result, a full investigation of the type the Government outlined always will not be necessary. 25. The issue of failure investigations first arises when testing produces a single out-of-specification result. Before proceeding to retest, (924:11 (Gerraughty)), this unhappy occurrence must be met with a step-by-step review of the suspect laboratory tests. (Id.; 1975:12 (Cooper); 1517:1 (Rhodes)). Specifically, the analyst who performed the test must report the problem to a supervisor, and the two technicians must conduct an informal laboratory inspection, reviewing the notebook which contained the out-of-specification result,. discussing the testing procedure along with any required calculations and examining the instrument used. (295:20 (Mulligan); 1831:8 (Atwater)). Thus, the Court requires that a single out-of-specification result be met with more than “a laboratory investigation consisting principally of retesting.” Barr Response to Government Findings, H 58-59. 26. Such an investigation, along with any conclusions reached, must be preserved with written documentation that enumerates each step of the review, (1519:10 (Rhodes)), in the form of a “computer generated flow sheet” (1517:19 (Rhodes)), or a check-list. (1974:25 (Cooper)). This writing should be preserved in an investigation or failure report, (1517:1 (Rhodes)), and placed in a central file. (1519:19 (Rhodes)). In order to enhance this review process, each analyst conducting a test should follow a written procedure, checking off each step as it is completed. (1974:25 (Cooper)). 27. Any easily identified analyst errors, such as calculation mistakes, should be specified with particularity and supported by evidence. (925:20 (Gerraughty)). In some instances, however, “the subtle influences which can result in test variability are not apparent when such an assay or test investigation is carried out.” (1954:13 (Cooper)). Thus, because it can be difficult to pin down the exact cause of a problem (1833:13 (Atwater)), it is unrealistic to expect that the cause of analyst error always will be determined and documented. (1416:22 (Bolton)). 28. In recognizing the existence of less readily identifiable mistakes and the influence of variables unrelated to the purity or potency of the drug under scrutiny, the Court does not intend to create a means for firms to avoid the performance and documentation of an informal laboratory investigation. The inability to identify an error’s cause with confidence affects retesting procedures, see ¶¶ 38-39 infra, but does not affect the failure inquiry required for initial out-of-specification results. 29. Other problems more serious than single out-of-specification results, from multiple out-of-specification results to product mixups and contamination, require full-scale inquiries involving quality control and assurance personnel in addition to laboratory workers in order to identify the exact process or nonproeess-related errors. 30.Extending beyond the laboratory and often labeled formal investigations, (1563:25 (Rhodes)), these inquiries should follow the outline the Government provided, with firms paying particular attention to any necessary corrective action, whether reprimanding, retraining or firing employees, remixing batches or adjusting processes. Thus, in the failure report firms must: (1) state the reason for the investigation; (2) provide a summation of the process sequences that may have caused the problem; (3) outline the corrective actions necessary to save the batch and prevent a similar recurrence; (4) list other batches and products possibly affected, the results of their investigations, and any required corrective action; and finally, (5) preserve the comments and signatures of all production and quality control personnel who conducted the investigation and approved any reprocessed material after additional testing. 31. The outcome of the failure investigation will determine whether additional batches of the same product and related products also require remedial measures. (1709:10 (Rhodes)). Process-related errors suggest the need to examine other batches of the problem product as well as other products made according to similar procedures. Addressing nonprocess-related errors requires an examination of other batches or products the trouble-making employee or machine may have handled. 32. Thus, the elements of a “thorough” investigation necessarily will vary with the nature of the problem identified. However, all failure investigations must be performed promptly, within thirty business days of the problem’s occurrence, and recorded in written investigation or failure reports. 3. Outliers 33. The outlier test provides an alternative means of invalidating an initial out-of-specification result. If the failure investigation of an initial out-of-specification result proves inconclusive, firms searching for a better explanation can utilize this method. 34. Significant limits accompany the outlier test, however. The USP specifically warns against using outlier tests too often, and thus, as a general rule, firms must be careful not to reject results frequently on this basis. (2036:9 (Baldassarre)). 35. In addition, the utility of the outlier test varies with the type of assay performed. The USP expressly allows firms to apply this test to biological and antibiotic assays, see Ex. 7 at 1503; (202:21, 798:20 (Mulligan)), but is silent on its use with chemical tests. (1239:8 (Mulligan)). Although some experts advocated use of the outlier method for chemical assays, (1395:1 (Bolton)), other testimony suggested that firms generally do not rely on outlier tests to invalidate chemical test results. (1238:21 (Mulligan)). In the Court’s view the silence of the USP with respect to chemical testing and outliers is prohibitory. 36. The substantial innate variability of microbiological assays, (1916:1 (Cooper)), supports this distinction. (1238:11 (Mulligan)). Chemical assays are considerably more precise than biological and microbiological assays, (1915:12 (Cooper)), since only the latter testing is “subject to whims of microorganisms.” (1915:18 (Cooper)). 4. Retesting 37. In addition to triggering a failure investigation, out-of-specification results also generate the need for retesting. A retest is defined as additional testing on the same sample, and thus it necessarily follows an initial test. An analyst performing a retest takes the second aliquot from either: (1) the sample that was the source of the first aliquot (1022:5 (Bolton), 1533:21 (Rhodes)); or (2) the larger sample previously collected for laboratory purposes. (1534:8 (Rhodes)). These procedures are equivalent. (1839:17, 1840:16 (Atwater)). Thus, whether retesting is performed at the finished product or blend stage, such testing should be performed from the same bottle of tablets or capsules and the same drum or mixer, respectively. (661:2 (Mulligan); 1423:3 (Bolton)). 38. Retesting is proper only after a failure investigation is underway, (235:1, 661:2 (Mulligan)), since the outcome of the failure investigation itself, in part, determines when retesting is appropriate. Retesting is necessary if a failure investigation indicates that analyst error caused an initial out-of-specification result. (144:1, 1023:18 (Mulligan)). A retest is similarly acceptable when review of the analyst’s work is inconclusive. (664:12 (Mulligan)). In these instances, retesting substitutes for or supplements the original tests which have been rejected or questioned, respectively. In the case of nonprocess and process-related errors, however, retesting is suspect. (1024:10 (Gerraughty)). Because the initial tests are genuine, in these circumstances, additional testing alone cannot infuse the product with quality. (1024:15 (Gerraughty)). 39. As a general matter, the amount of retesting required also varies with the problem identified. Out-of-specification results attributed to analyst error require limited retesting. (1529:19 (Rhodes)). Here, retesting merely supplants the first round of initial tests. More extensive retesting should follow an inconclusive failure investigation, since firms need to determine whether the out-of-specification result is a mere anomaly or a reason to reject the batch. (1390:1 (Bolton)). 40. The USP contemplates retesting for quality control purposes (1937:14 (Cooper)), although it does not prescribe or recommend the number of individual tests that must be performed in order to reach a definitive conclusion about the quality of a product. (1929:1 (Cooper)). Thus, the number of retests performed before a firm concludes that an unexplained out-of-specification result is invalid or that a product is unacceptable is a matter of scientific judgment. (488:7, 816:9 (Mulligan); 1959:3 (Cooper)). Yet the goal of retesting is clear; firms must do enough testing to isolate the out-of-specification result, in order to reach the point at which the additional testing overcomes the out-of-specification result. (805:4 (Mulligan)). 41. Nevertheless, retesting cannot continue ad infinitum. Because such a practice is not scientifically valid, a firm’s predetermined testing procedure should contain a point at which testing ends and the product is evaluated. At this time, if the results are not satisfactory, the batch must be rejected. (1958:7 (Cooper)). 42. When evaluating retest results, it is important to consider them in the context of the overall record of the product. Relevant to this review are the history of the product, the type of tests performed, and any results obtained for the batch at other stages of testing. As such, retesting determinations will vary on a case-by-case basis, a necessary corollary of which is that an inflexible retesting rule, designed to be applied in every circumstance, is inappropriate. (1476:9 (Bolton)). 5. Resampling 43. Resampling, in contrast, is a more controversial practice. Typically resam-pling occurs after the initial test and the retests have produced out-of-specification results, thereby indicating a more serious problem. When performing a resample, an analyst leaves the laboratory and takes a new sample from the universe of the batch. (1534:13 (Rhodes)). 44. Resampling is appropriate where provided by the USP (1044:21 (Gerraughty)), as in cases of content uniformity and dissolution testing. (1045:3 (Gerraughty)). Similarly, in the limited circumstances in which a failure investigation suggests that the original sample is unrepresentative, resampling is acceptable. (1024:19 (Gerraughty)). Evidence, not mere suspicion, must support a resample designed to rule out preparation error in the first sample. (1955:8, 1956:1 (Cooper)). 45. Absent these limited exceptions outlined above, however, firms cannot rely on resampling to release a product that has failed testing and retesting. (1045:19 (Ger-raughty)). 6. Remixing 46. The need for remixing arises during the blend stage when testing reveals problems with content uniformity. The regulations themselves allow reworking, which essentially is remixing. (906:11 (Ger-raughty)). As evidenced by the Generic Drug Office directive and the consent agreement between the FDA and Eli Lilly, remixing is allowed in this circumstance. (558:2 (Mulligan)). 47. The instance of remixing, however, is directly related to its propriety. Occasional remixing is acceptable, (674:2 (Mulligan)), but frequent or wholesale remixing is not. The need to remix often provides a clear indication that the process is invalid (907:5, 1021:9 (Gerraughty)), and casts doubt on those batches that passed through testing without incident. (576:14 (Mulligan)). 7. Averaging 48. Although averaging test data can be a rational and valid approach, (1578:4 (Rhodes)), as a general rule, firms should avoid this practice, because averages hide the variability among individual test results. 49.This phenomenon is particularly troubling if testing generates both out-of-specification and passing individual results which when averaged are within specification. Here, relying on the average figure without examining and explaining the individual out-of-specification results is highly misleading and unacceptable. (812:20 (Mulligan); 1255:11 (Gerraughty)); see also (1481:20, 1482:8 (Bolton) (averaging blend assay results invites problems at finished product stage where values of 89, 89, and 92, given range of 90 to 110, should be met with more testing even though average equals 90)). 50. Although averaging camouflages variability, an average may provide more information about the batch’s true assay value that any single test result.' (1872:4 (Atwater)). Thus, in the case of microbiological assay testing, the USP prefers an average, (1932:2 (Cooper)), when reaching an ultimate judgment about the quality of the product. (1929:22 (Cooper)). It is good practice to include out-of-specification results in the average, unless an outlier test indicates that an out-of-specification result is an anomaly. (1579:23 (Rhodes)). 51. Finally, the average of individual content uniformity tests at the finished product stage can act as a proxy for the assay value. (1231-33 (Gerraughty)). Though this estimate cannot substitute for final product assay testing, it can provide some information about a batch. 8.Releasing a Batch for Distribution 52. Section 211.165(a), in relevant part, provides: For each batch of drug product, there shall be appropriate laboratory determination of satisfactory conformance to final specifications for the drug product, including the identity and strength of each active ingredient, prior to release. 21 C.F.R. 211.165(a). And section 211.-165(f) specifies: Drug products failing to meet established standards or specifications and any other relevant quality control criteria shall be rejected. Reprocessing may be performed. Prior to acceptance and use, reprocessed material must meet appropriate standards, specifications, and any other relevant criteria. 53. The USP provides the “established standards” to which the CGMP regulations refer and upon which firms rely to release their drug products for distribution to the public. (486:23 (Mulligan)). These specifications are absolute. (1484:18 (Bolton)) (noting that firms cannot stretch USP — if standard is 93 then cannot release batch on result of 91)). 54. Under section 211.165(f), the government argues, a single out-of-specification result that cannot be invalidated defeats a batch. (312:12 (Mulligan); 1271:22 (Gerraughty) (because variability of assay and instrument built into USP procedure, batch fails if have one failing result)). Barr reads the statute more liberally and refuses to equate a failure to meet specifications with obtaining one out-of-specification result from many quality performance measurements. (1465:14 (Bolton)). 55. Similarly, because the USP specifies the standards for release testing, the government contends that section 211.-165(f) removes any uncertainty that might require the exercise of scientific judgment. (486:23 (Mulligan)). The absolute approach the government recommends, under which a batch fails if one tablet out of one million tablets tested produces an out-of-specification result, (1273:24 (Gerraughty)), is extreme. Instead, in the Court’s view, section 211.165, through the “appropriate laboratory determination” language and the allowance of reprocessing, suggests that scientific judgment can play a role when firms decide whether to release a batch to the public. 56.In light of the case-by-case variability scientific judgment introduces, the Court cannot articulate specific procedures for release decision-making. Instead, in this context, firms should follow the general retesting guidelines. See ¶¶ 37-42. However, firms cannot justify release when only fifty or sixty-six percent of the finished product tests for a particular batch produce passing results. For example, with an assay limit of 90 to 110, test results of 89 and 90, or 89, 91, and 91, or two 89s and two 92s all should be followed by more testing. (1398:1 (Bolton)). The amount of additional tests required would be a matter of scientific judgment, as informed by other relevant data. (816:9 (Mulligan); 1398:21 (Bolton)). The goal is to distinguish between an anomaly and a reason to reject the batch. (1390:1 (Bolton)). 57. While experts disagree about the relative importance of finished product and blend testing, see (1324:9-20, 24, 1325:12 (Bolton) (blend testing serves to enhance confidence in finished product tests, but finished product assay and content uniformity testing more important); (907:17: (Gerraughty) (in-process testing more important because finished product testing very minimal)), it is clear that the release evaluation depends, in part, on the background of the batch and product. Secondary parameters, such as physical properties, blend evaluations, time of mix, weight, thickness, and friability, affect the actual finished product results as well as their reliability. (1317:7 (Bolton)). The lesson for firms and the FDA is that context and history inform many final conclusions. (1486:1 (Bolton) (marginal blend assay results require stricter scrutiny at finished product stage); 1273:10 (Gerraughty) (considering past content uniformity problems); 1368:23 (Bolton) (when reaching conclusions, important to look at all data)); 1419:25 (Bolton) (history of product useful in determining whether error is cause for concern)). 9. Blend Testing 58. An important aspect of drug manufacturing, blend testing gives firms an opportunity to discover and remedy in-process problems before batches reach the final stages of production. Because finished product testing is limited, blend testing is necessary to increase the likelihood of detecting inferior batches. (1181:1, 1221:2 (Gerraughty)); see also (1036:18 (Bolton) (cannot waive blend content uniformity testing and rely solely on finished product results)). (a) Sample Size 59. An element of blend testing which influences the ultimate test results is sam-pie size. No regulation, guideline, or publication requires any specific blend sample size. (1119:15 (Gerraughty)). The CGMP regulations merely provide that “[i]n-process materials shall be tested for identity, strength, quality, and purity as appropriate.” 21 C.F.R. § 211.110(c) (emphasis added). 60. In accordance with the preamble to the CGMP regulations, the Court must construe the “as appropriate” phrase to permit “reasonable, albeit variable interpretations.” (765:16 (Mulligan)). The testing procedures a firm chooses, however, must be logical and effectual. (767:13, 776:14 (Mulligan)). Paraphrasing Inspector Mulligan, the Court must ask whether “what they are saying makes sense.” (769:15 (Mulligan)). 61. Implicit in the “as appropriate” language, the government argues, is a sample size requirement equal to one to three times the product’s run weight. (432:17 (Mulligan)). Driving this theory is the concern that a larger sample will dilute or even negate any nonuniformity in the blend. (672:1 (Mulligan)). Barr argues that a variety of sample sizes comply with CGMP and that the final choice is a matter of scientific judgment. Barr Findings ¶ 122. 62. Although sample size is a question of scientific judgment, the sample chosen must advance the purpose of the test. Thus, what is “appropriate” may vary with the type of test performed. 63. Content uniformity testing, designed to detect the adequacy of the mix by measuring variations in the potency of the blend, (1150:15 (Gerraughty)), should be conducted with a sample that resembles the dosage size. Any other practice likely would blur differences in portions of the blend, (1069:18 (Gerraughty)), and defeat the object of the test. 64. The Court appreciates the difficulty companies experience taking minute samples from large-volume blends. Indeed, testimony revealed that the smallest thief available can retrieve a 250-milligram sample, (1285:1 (Gerraughty)), so in some cases firms cannot obtain a single-run-weight sample. As such, the Court will follow Dr. Gerraughty’s testimony and hold that the appropriate sample for content uniformity testing, in both validation and ordinary production batches, (922:15 (Gerraughty)), is three times the dosage size. (921:4, 922:8 (Gerraughty)). 65. In addition to assuring a more accurate measure of uniformity, this rule accommodates the need for retesting. In order to conduct an initial test and two retests, a standard testing practice in the industry, (671:16 (Mulligan)), analysts need a three-run-weight sample. (671:21 (Mulligan)). Under Inspector Mulligan’s one-run-weight rule, in order to retest the same sample, firms must take additional samples from the same spot in the blend. (681:12 (Mulligan)). Such a requirement would be onerous. 66. Implicit in the sample size determination for content uniformity testing is a prohibition on compositing multiple individual samples taken from different areas of the blend. Again, in order to detect uniformity problems, firms must avoid this practice which would conceal variations in the blend. (1970:6 (Cooper)). 67. In contrast, blend assay or potency testing, designed to measure the strength of the blend, can accommodate larger samples. Although averaging the differences in the mixture, here a larger sample also provides a better indication of the overall percentage of active ingredient in the blend. (1150:15 (Gerraughty)). In fact, a blend assay test conducted with a larger sample will be more representative of the final assay. (1328:19 (Bolton)). Similarly, as variation detection is not the object of assay testing, the Court does not object to compositing. (b) Site of Sampling 68. Also important in content uniformity testing are the number of samples taken and how representative they are of the mix. (1329:7 (Bolton)). The government again cites the “as appropriate” language of section 211.110(c) to support its view that blend content uniformity testing should be conducted with samples taken from the mixer and not the drum. Barr maintains that firms are free to sample from either the mixer or the drum and, further, that in some cases, testing from the drum is preferable. Barr Findings 11123. Once again, no regulation, guideline, or publication expresses a preference for blend testing in the mixer or the drum. (1120:1 (Gerraughty)). 69. Expert testimony revealed that firms test from both the drum and the mixer and that either practice is acceptable under CGMP. (1012:19 (Gerraughty), 1336:14 (Bolton)). Thus, the Court is not prepared to prescribe a particular location for blend testing. 70. Rather, the factor of more concern to the experts, and thus the Court, is the representativeness of the sampling technique. (1013:9 (Bolton)). To detect poor uniformity, firms must take samples from “places that might be problems,” (1335:2 (Bolton)), such as weak and hot spots in the blend. (921:4, 1223:13 (Bolton)). Indeed, Barr conceded that Inspector Mulligan’s concern about in-process blend testing in weak spots of the mixer is valid at the process validation stage. (2212:3 (Pierpaoli)). Thus, whether sampling from the mixer or the drum, firms must demonstrate through validation that their technique is representative of all portions and concentrations of the blend. 71.Experts did suggest, however, that sampling from the mixer is preferable, (1012:19 (Gerraughty)), especially if a firm has experienced content uniformity problems. (436:16 (Mulligan)). While this sampling method is not CGMP-required, in the Court’s view, it has advantages. Sampling from the mixer facilitates both remixing and targeting areas of likely nonuniformity- E. Criticisms of Barr 72. Having defined the terms underlying the CGMP-compliance determination, the Court now can turn to the criticisms the FDA has levied against Barr’s practices. Specifically, the FDA contends that: (1) Barr’s manufacturing processes are invalid or not validated; (2) Barr’s failure investigations are inadequate; and that (3) Barr releases drug products that fail their specifications. Also of concern to the FDA are Barr’s cleaning validation and record-keeping deficiencies as well as Barr’s failure to control steps in the manufacturing process, to validate its testing methods and to file ANDA supplements and field alerts. In light of these shortcomings, the government asks the Court for restrictions on Barr’s current product line. The Court will discuss each criticism in turn. 1. Manufacturing Process Validation (a) General Requirements 73. To comply with CGMP, drug manufacturers must develop “written procedures for production and process control designed to assure that the drug products have the identity, strength, quality and purity they purport or are represented to possess.” 21 C.F.R. § 211.100. 74. These written procedures, in turn, must be verified. More specifically, the CGMP provides: To assure batch uniformity and integrity of drug products, written procedures shall be established and followed that describe the in-process controls, and tests, or examinations to be conducted on appropriate samples of in-process materials of each batch. Such control procedures shall be established to monitor the output and to validate the performance of those manufacturing processes that may be responsible for causing variability in the characteristics of in-process material and the drug product. Such control procedures shall include, but are not limited to, the following, where appropriate: (1) Tablet or capsule weight variation; (2) Disintegration time; (3) Adequacy of mixing to assure uniformity and homogeneity; (4) Dissolution time and rate; (5) Clarity, completeness, or Ph of solutions. 21 C.F.R. § 211.110(a). 75. This verification, known as process validation, is a quality control measure through which drug manufacturers become confident that their processes consistently will produce products which meet their predetermined specifications. (873:8 (Ger-raughty)). 76. Process validation requires that “[i]n-proeess materials be tested for identity, strength, quality and purity as appropriate, and approved or rejected by the quality control unit, during the production process, e.g., at commencement or completion of significant phases or after storage for long periods.” Id. § 211.110(c). 77. In its Guideline on General Principles of Process Validation (“FDA Guideline”), published in 1987 and re-noticed unchanged in 1990, the FDA provides further guidance to firms in the area of process validation. See Exhibit 143. 78. Types of in-process testing commonly performed include assay, content uniformity and dissolution tests. Assay testing measures the potency of the active ingredient in the drug product. (21:13 (Mulligan)). Content uniformity testing measures the variability of potency between drug dosages. Id. Dissolution testing determines how the active ingredient will dissolve and work in the human body. (25:7-15 (Mulligan)). 79. Such testing occurs at two stages in the production process, the final blend stage and the finished product stage. (26:9 (Mulligan)). Assay and content uniformity tests are performed at the blend stage, while all three tests are conducted at the finished product stage. 80. The data resulting from this barrage of testing forms the basis for the CGMP-required validation study. Firms may choose a prospective, concurrent or retrospective validation, each an accepted method under CGMP, alone or in combination. 81. Prospective validation requires a manufacturer to make at least three consecutive batches under closely-controlled conditions and to perform the studies and intensive testing necessary to determine that each step in the process is under control and will yield a product that consistently meets its specifications. This method is used for new products that have not been manufactured and marketed. 82. Concurrent validation also requires the manufacture of at least three consecutive batches under closely-controlled conditions followed by intensive testing. This method is used for existing products. 83. Retrospective validation allows a manufacturer to rely on the past production of a series of problem-free batches to show that the process is under control and yields a product that consistently meets its specifications. (223-24 (Mulligan)). Although no single batch is subject to intensified testing, making a product for a long period of time generates a history that contains much useful information about the product. (1314:11 (Bolton)). (b) Specific Process Problems 84. The government argues that Barr’s manufacturing processes are invalid due to a high percentage of batch failures and, further, that Barr’s attempts to validate these processes retrospectively are unavailing since Barr has excluded failing data and failing batches from their studies and relied on an insufficient number of batches. (1) Batch Failures 85. The government first argues that the failure rates associated with Barr products demonstrate the need to revise the underlying manufacturing processes. To the extent that batches included in retrospective studies exhibit a failure rate of ten percent or more, the Court agrees. (1384:10 (Bolton)). 86. Citing assay, content uniformity, dissolution, and stability test failures, among others, the government has calculated batch failure rates ranging from 2.0 to 44.4% for thirty-one of Barr’s products. Because the Court has rejected the government’s definition of “failure,” the Court will not rely solely on the failure percentages calculated on this basis to discredit Barr’s manufacturing processes. 87. Yet, the frequency with which Barr obtains out-of-specification results does give the Court pause. As such, it is with a sharp eye that the Court examines Barr’s validation studies. (2) Retrospective Validation 88. Pointing to its retrospective validation studies, Barr claims that it manufactures each of its sixty products in its current product-line under a validated process and, as a result, that it is in compliance with CGMP. The government challenges this conclusion, pointing to numerous deficiencies in Barr’s testing practice. (a) Omitting Failing Results 89. The government first argues that Barr consistently omits “failing” test results from batches included in its retrospective validation studies. 90. Only those test results determined through an appropriate failure investigation to be caused by analyst or operator error can be excluded. Those results that are not explained, but merely called into question by successful retesting, must be included. 91. Barr’s former practice of rejecting initial out-of-specification results based upon acceptable retest results does not meet this standard. Currently, however, Barr considers all initial results. See ¶ 109 infra. (b) Omitting Failing Batches 92. The government next argues that Barr omits failing batches from its validation studies, thereby obscuring results obtained during the specified time-frame. 93. As a general rule, a retrospective study must contain all batches produced within the time frame chosen for analysis. To avoid a comparison of apples and oranges, however, these batches must be: (1) made in accordance with the process that is being validated; and (2) representative of this process. (884:23 (Gerraughty); 2124:18 (Baldassarre)). Batches that do not meet these criteria should not be included in the retrospective report. 94. Thus, exclusion is warranted in situations in which firms have made the same product under superseded or unapproved processes. This occurs, for example, if a firm alters a procedure, thinking the change does not require prior approval, and later learns from the FDA that approval was needed. (2110:9 (Baldassarre)). Although made under the correct process, those batches that are aborted or rejected due solely to a nonprocess-related error should be excluded as nonrepresentative. (1450:15 (Bolton)). 95. Because the default is inclusion, (2111:15 (Baldassarre) (firm should include all batches unless it has very good explanation)), firms bear the burden of explaining and documenting any gap in the chronological series of batches included in a retrospective study. (2111:15 (Baldassarre); 1456:19 to 1457:4 (Bolton)). Firms also must indicate the process used for each batch. (1003:9 (Gerraughty)). 96. Currently, Barr selects those batches made under the same manufacturing process for its retrospective validations, (2108:12 (Baldassarre)), provided that these batches did not experience failures attributed to operator error. (2124:18 (Baldas-sarre)). Barr does include batches that exhibit process-related failures. (2125:10 (Baldassarre)). Barr explains operator errors and anomalous situations in footnotes, (2127:18 (Baldassarre)), but leaves unnoted any gaps resulting from different manufacturing processes. (2110:22 (Baldassarre)). 97. While Barr’s selection of batches for retrospective studies is acceptable, the government’s request for an explanation is well-taken. (2127:11 (Ms. Yeoh)). Barr must remedy its failure to note and explain gaps in its retrospective validations. Specifically, Barr must document the reason for exclusion, through a failure investigation if the problem is an operator error, or a written explanation of the process and its modifications if different processes are used. (c) Insufficient Number of Batches 98. Finally, the government contends that Barr’s retrospective studies are unreliable, because Barr used fewer than the required minimum number of batches. 99. The CGMP regulations and the FDA Guidelines do not discuss individual validation methods and thus provide no guidance on the issue of minimum batches for retrospective studies. (1118:3, 12 (Ger-raughty)). In the context of both prospective and retrospective validation, the FDA Guidelines express a preference for specific test results, rather than simple pass/fail determinations, since statistical analysis of specific tests generates an expected variance in data. FDA Guidelines at 24. 100. The Government’s expert advocated a ten-batch-minimum rule, a figure the industry allegedly accepts. (1003:20 (Ger-raughty)). Barr contends that with the help of statistical analysis it can validate a process retrospectively with very few batches. Barr Findings If 133. 101. While any number of batches can support a validation study, (1439:19 (Bolton)), because validation is a matter of degree, the degree of assurance about the validity of the process necessarily increases with the number of batches. (1440:11 (Bolton)). More batches produce more reliable statistical conclusions about the process characteristics. (1346:2 (Bolton)). 102. The simple more-is-better principle does not translate easily into a specific minimum number of batches. Although the number of batches required is a matter of scientific judgment, (1345:16 (Bolton)), the hearing testimony provides some parameters in which these decisions must be made. First, the number of batches used in retrospective studies must be greater than those used in prospective validation. Less intensive testing on the retrospective batches and the inability of firms to retest batches examined retrospectively require this rule. (879:4 (Gerraughty)). Second, studies conducted with only five batches, a number Barr’s expert was unwilling “to hang [his] hat on,” (1345:16, 1346:16 (Bolton)), are unacceptable. Instead, as a general rule firms should strive to include twenty to thirty batches in their retrospective studies. (1346:5 (Bolton)). 103. In the absence of any direction from the CGMP regulations or the FDA Guidelines, however, the Court is unwilling to hold that one precise number of batches is needed to complete an adequate retrospective validation study. Any such figure would be arbitrary and might encourage some firms to modify their current practice downward. In addition, an absolute rule would make it impossible to validate retrospectively those drug products with low production rates. The Court is confident that firms will act in accordance with this principle. Any decision to ignore the general rule must be supported by statistical or other evidence that indicates a reliable validation. 2. Failure Investigations 104. The government next attacks Barr’s failure-investigation practice. Citing FDA criticism and investigations dating back to 1985, it alleges that until recently Barr rarely performed for-cause failure investigations. Government Findings Ml 63-64. Even under the Court’s more generous interpretation of the section 211.192-pre-scribed failure investigation, the evidence clearly indicates that Barr’s past failure investigations fall far short of the CGMP mandate. (1513:4 (Rhodes) (concern about failure investigation documentation and product mix-ups); 1536:8 (Rhodes) (Barr not documenting out-of-specification results and decision to resample); 1991:19 (Cooper) (inadequate documentation prevents judgment about whether procedures and investigations proper)). At best, Barr carried out some investigations, but did not document its efforts. (1953:25 (Cooper)). 105. The testimony of Barr experts and employees, however, suggests that Barr recently has begun to conduct for-cause inquiries in a timely and adequate fashion. (1048:12 (Gerraughty); 1515:1, 1551:8, 1152:24, 1553:2,20, 1555:6, 1557:18, 1573:17, 1574:20 (Rhodes)); see also (1519:19) (Rhodes) (Barr has established central file for failure investigation reports)). A step-by-step critique of Barr's current failure investigation practice follows. See ¶¶ 107-115 infra. 3. Release of “Failing” Batches 106. The government, in a three-part argument also contends that Barr has and continues to. release batches of product which fail their predetermined specifications. Specifically, it claims that Barr: (a) discards failing results based on an assumption of laboratory error; (b) without exception discards failing results based on passing retest results, a procedure that is statistically and scientifically invalid; and (c) discards initial results after retesting different parts of the batch. Barr denies each allegation. (a) and (b) Rejecting Failing Results As Laboratory Errors After Retesting 107. The government first focuses on Barr’s alleged procedure of rejecting initial out-of-specification results as laboratory error based solely on passing retest results. Although separate criticisms, the complaints aimed at Barr’s failure investigations and retest procedure are closely related and best considered together. The Court will conduct a step-by-step analysis of Barr’s procedures for chemical and microbiological assays in turn. 108. Currently, Barr meets out-of-specification results from chemical tests with a laboratory investigation. Barr Findings 1111 54-55. If this limited inquiry successfully identifies a laboratory error, Barr rejects the initial results. Id. ¶ 55. Consistent with the Court’s interpretation of section 211.192, this procedure complies with CGMP. 109. If the laboratory investigation is inconclusive, as is sometimes the case, two additional analysts retest the same sample. If these retests are within specification, Barr averages the three test results and releases the batch if the average also is within specification. Id.; Ex. 14 at 5; (405:22, 685:13, 1040:18 (Mulligan); 2157:23 (Baldassarre)). Although retesting is preceded by an investigation as CGMP requires, this procedure nevertheless is in direct conflict with expert testimony. As the Court noted previously, the amount of retesting required, a matter of scientific judgment, will vary on a case-by-case basis depending on the history of the product and the batch. See ¶¶ 39-42 supra. Thus, Barr cannot rely on its two-out-of-three retest/release procedure. 110. Next, “[i]f the results appear to be inconsistent the laboratory performs the appropriate outlier ■ test.” Whether this phrase contemplates a comparison between the retest results and the initial result, or between the retest results themselves, the use of an outlier test in the chemical testing context is impermissible. See 1111 35-36 supra. 111. Finally, if the average of the results is out-of-specification, Barr proceeds to resample in order to determine whether there was a sampling error and begins a formal investigation. Barr Findings 11 56. Again, this reflexive approach is highly suspect. (1244:21 (Mulligan)). Because re-sampling is appropriate only in limited circumstances, the Court cannot condone this resample rule. (1245:5 (Mulligan) (not practice of industry)). No action can be taken with respect to the batch until the results of the formal investigation are known. 112. The problems with Barr’s retesting procedure are not erased by the protocol’s popularity at other firms. See Barr Findings ¶ 57. As the Court indicated earlier, industry practice while instructive also is subject to the CGMP-compliance test. See ¶ 12 supra. With the exception of the testimony of Dr. Rhodes, (1542:1-11; 1544:22; 1666:15; 1669:7 (Rhodes)), the record contains resounding condemnation of Barr’s retesting protocol. (1041:6 (Gerraughty) (two of three chance that passing result is correct insufficient); 1044:14 (Gerraughty) (this practice not standard in industry or in compliance with CGMP); 1476:1 (Bolton) (cannot discard out-of-specification result based on two acceptable retests of same sample); 1476:9 (not a good practice in every case as need to consider all data); 1477:15 (Bolton) (specifically rejecting TM-196C); 1244:13 (Gerraughty) (that no discussion of results themselves problematic in that procedure would allow firm to pass batch based on results of 89, 91, 91, which is inappropriate; 1246:16, 22 (Mulligan) (if all results are within same precision of instrument, “real gamble” to assume that true value is 91)). 113. Barr’s microbiology protocol provides for additional testing using aliquots from each of the three original containers if the initial results are out-of-specification. Barr Findings 11 58. If all individual retests are within specification, and the average of the initial and retest results is within specification, the batch passes. Id. ¶ 59; (1946:9 (Cooper)). 114. If, however, any retest results are out of specification, Barr proceeds to re-sampling. (1949:1-12 (Cooper)). If the re-sampling results and the average value of the initial tests, retests and resamples is within specification, the batch passes. Id. 115. Given the variability inherent in microbiology testing, Barr’s retesting procedure is appropriate, (1955:8 (Cooper)), provided that Barr continues to use the outlier test to discredit out-of-specification results, (2192:7 (Pierpaoli)), and does not return to its former practice of simply disregarding initial out-of-specification results. (2191:21 (Pierpaoli) (Barr previously threw away initial result without evaluating it); 1951:1 (Cooper) (Barr previously would not include average of initial test results)). However, as the Court has noted, firms cannot remedy test and retest problems through reflexive resampling. See till 44-45 supra. (c) Rejecting Failing Results After Re-sampling 116. Here, the government focuses its concern on Barr’s practice of conducting retests using samples from a different 20-tablet grinds taken from the same laboratory sample. (2191:19 (Pierpaoli)). Instead, the government contends, Barr should retest the sample used in the initial test. Expert testimony, however, determined that both of these procedures properly are termed a “retest.” 117. Of more concern to the Court is Barr’s practice of resampling, discussed supra. Although Barr changed its operating procedure to alleviate FDA criticisms, leaving out the paragraph that allowed resam-pling when a reconstitution error or a sample integrity problem was suspected, (2079:15 (Baldassarre)), a Barr employee testified that resampling, a necessary element of a failure