Second-Order Statistical Cutoff Methodology for Gas-Liquid Diffusion Integrity Testing

Abstract

The present disclosure relates to methods for reducing background noise during membrane filter integrity testing to increase sensitivity and more effectively handle failed membrane filters.

Description

[Technical field]

[0001] This application claims the benefit of U.S. Provisional Application No. 63 / 131,850, filed December 30, 2020, the entire contents of which are incorporated herein by reference in their entirety.

[0002] The embodiments described herein relate to methods of gas-liquid integrity testing for membrane filters. More specifically, certain embodiments of the present technology relate to a methodology for gas-liquid diffusion integrity testing data analysis that uses population subset normalization to condition the integrity testing data for operating conditions such as temperature and pressure for the purpose of reducing background noise and thereby improving defect detection signal-to-noise ratio. [Background technology]

[0003] High purity filtration of aqueous media used in the fields of biotechnology, chemicals, electronics, pharmaceuticals, and the food and beverage industry is obtained using high performance membrane filter modules capable of high degree of separation. These membrane filters also prevent contamination of the environment, the media being filtered, and the resulting filtrate, in order to prevent unwanted and often dangerous organisms such as bacteria or viruses, as well as environmental contaminants such as dust, dirt, etc., from entering the process stream and the final product.

[0004] Integrity testing is a requirement during critical process filtration applications to ensure that the sterility and / or retention capacity of the membrane filter is not compromised. Membrane filter manufacturers for critical applications also always apply integrity testing to finished filter elements as a lot release or 100% test criteria. Integrity testing detects the presence of oversized pores or defects that may compromise the retention capacity of the porous material. Integrity testing methods include particle challenge testing, liquid-liquid porometry testing, bubble point testing, gas-liquid diffusion testing, and diffusion testing to measure trace constituents. Some of these tests, such as particle challenge testing, are destructive. Thus, these tests cannot be used as pre-use tests. Although liquid-liquid porometry and bubble point testing are useful to ensure that membranes with the proper nominal pore size are installed, the methods lack the sensitivity to identify a small number of small defects.

[0005] Gas-liquid diffusion tests are widely used to evaluate filter integrity. A common gas-liquid pair for integrity testing is air-water because of the pair's safety, resistance, and environmentally friendly properties. The diffusion test measures the rate of gas migration through the filter. At different gas pressures below the bubble point, gas molecules migrate through the water-filled pores of a wet membrane according to Fick's law of diffusion as follows:

number

[0006] A measured gas flow rate above that predicted by Fick's law or higher than the flow rate empirically established for an intact membrane is a signal of a defect. The sensitivity of this test is limited by the minimum detectable excess flow rate. Differences in membrane area, membrane thickness, membrane porosity, and pore tortuosity can result in significant filter-to-filter variation in the gas diffusion flow rate of an intact membrane filter device.

[0007] Variations in test system hardware and instrumentation, as well as operating conditions such as pressure and temperature that affect the diffusivity and solubility of gases in liquids, also contribute to background noise. Background noise can compete or interfere with the defect signal, which is excess gas flow due to convection through the defect(s). Faced with increased noise variability, traditional air diffusion integrity testing must either widen the acceptable specification window (i.e., reduce defect detection capability) or tolerate more false rejects.

[0008] The primary contributing factors to this noise variation are film properties that vary by lot (referred to herein as "date of entry") and by test location on the instrument that runs the integrity test cycle (referred to herein as "bowl"). The sensitivity of the gas-liquid diffusion test to detect defects is also directly limited by background noise. High background noise can cause perfect filters to erroneously fail the test, resulting in additional cost waste. When establishing test pass criteria, including upper and lower diffusion specification limits, a trade-off decision must be made between test sensitivity (product and end user risk mitigation) and the ability to reliably supply product (yield, cost, and safety of supply to market). Summary of the Invention [Problem to be solved by the invention]

[0009] Therefore, there is a need for an integrity testing methodology that maintains the convenience of gas-liquid integrity testing while minimizing the effects of inherent background noise variables that can reduce the ability of the test to detect defects and increase the probability of falsely rejecting a perfect filter. [Means for solving the problem]

[0010] The shortcomings of the prior art are overcome by embodiments described herein, including an embodiment method of reducing background noise and improving the signal-to-noise ratio of an integrity test for processing a membrane filter, the method comprising the steps of performing a gas-liquid diffusion analysis on the membrane filter, identifying whether the membrane filter is an outlier by comparing with an initial fixed cutoff during the gas-liquid diffusion analysis, and if the membrane filter is not an outlier by the initial fixed cutoff, normalizing a value of at least one characteristic of the at least one membrane filter that affects the background noise in the gas-liquid diffusion analysis, based on values ​​from the gas-liquid diffusion analysis of a plurality of membrane filters, identifying whether the membrane filter is an outlier by comparing with a secondary statistical cutoff, and processing the membrane filter as perfect or failing.

[0011] In some embodiments, the gas-liquid diffusion method is air-water. In some embodiments, the at least one characteristic is selected from the group consisting of lot, device test location, water temperature, air temperature, and air pressure. In some embodiments, the values ​​of the at least two characteristics are normalized. Some embodiments of the method further include normalizing the pressure of the two or more membrane filters. Some embodiments of the method further include detecting at least one defect in the two or more membrane filters. In some embodiments, the at least one defect is selected from the range consisting of between about 4 μm and about 9 μm. In some embodiments, the defect is greater than 4 μm. In some embodiments, the defect is selected from the group consisting of about 4 μm, about 5 μm, about 6 μm, about 7 μm, about 8 μm, and about 9 μm. In some embodiments, the initial upper limit cutoff maintains at least a 5% safety margin from the end user diffusion specification criteria. In some embodiments, the initial upper limit cutoff maintains at least a 5% safety margin from the quality control (QC) lot release diffusion criteria.

[0012] In some embodiments, the initial upper cutoff maintains a safety margin selected from the range consisting of about 5% to about 15% from the end user diffusion specification criteria or quality control (QC) lot release diffusion criteria. In some embodiments, the initial upper cutoff maintains a safety margin selected from the group consisting of about 5%, about 6%, about 7%, about 8%, about 9%, about 10%, about 11%, about 12%, about 13%, about 14%, and about 15%. In some embodiments, the membrane filter is an outlier with respect to the initial cutoff and is treated as a failure. In some embodiments, the membrane filter is an outlier with respect to the secondary cutoff and is treated as a failure. In some embodiments, the membrane filter is not an outlier with respect to either the initial or secondary cutoff and is treated as perfect.

[0013] In some embodiments, the plurality is a statistically significant number of membrane filters. In some embodiments, the membrane filter and the plurality of membrane filters are the same type of filtration device. In some embodiments, the filtration device is selected from the group consisting of a sterilizing grade filter, a viral filter, a clarification filter, and an ultrafiltration filter. In some embodiments, the secondary statistical cutoff is established by setting an upper statistical limit 2-5 standard deviations above the normalized median of the values. In some embodiments, the secondary statistical cutoff is established by setting an upper statistical limit about 2, about 3, about 4, and about 5 standard deviations above the normalized median of the values. In some embodiments, the secondary statistical cutoff is established by setting a lower statistical limit 4-6 standard deviations below the normalized median of the values. In some embodiments, the secondary statistical cutoff is established by setting an upper statistical limit about 2, about 3, about 4, and about 5 standard deviations above the normalized median of the values. [Brief description of the drawings]

[0014] [Figure 1A] FIG. 1 is a box plot comparing air diffusion data from a conventional integrity test with air diffusion data from one embodiment of the integrity test method using the quadratic cutoff described herein. [Figure 1B] FIG. 1 is a box plot comparing air diffusion data from a conventional integrity test with air diffusion data from one embodiment of the integrity test method using the quadratic cutoff described herein. [Figure 2A] FIG. 1 is a box plot of the complete data set generated by conventional air diffusion integrity testing. [Figure 2B] FIG. 1 is a box plot of a data set generated by conventional air diffusion integrity testing with outliers removed. [Figure 3A] FIG. 3 is a box plot comparing the data sets of FIG. 1 and FIG. 2 above by subpopulation of test bowls. [Figure 3B] FIG. 3 is a box plot comparing the data sets of the previous FIGS. 1 and 2 by subgroup of membrane input dates. [Figure 4]FIG. 13 is a box plot of the same data set from the previous example normalized for both test location and membrane input date.

[0015] The accompanying drawings illustrate certain embodiments of the present disclosure and should not be considered as limiting its scope, as the invention may admit of other equally effective embodiments. It should be understood that elements and features of any embodiment may be found in other embodiments without further recitation, and where appropriate, the same reference numerals have been used to indicate similar elements common to the figures. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0016] The present disclosure describes certain embodiments of methods for integrity testing.

[0017] Certain embodiments of the integrity testing method described herein use normalization of air diffusion data by bowl and loading date to increase signal to noise ratio and improve defect detection capabilities. Additionally, the method better accommodates membrane and instrument variations than traditional integrity testing to provide more rigorous and consistent processing of product without sacrificing yield or reducing sensitivity.

[0018] Certain embodiments of the integrity testing method described herein normalize the air diffusion data by the testing location (bowl) of the equipment as well as by loading date (material and upstream process variations). The resulting variations can be accounted for using normalization by understanding the variations that these subpopulations add to the process in a predictable way. The normalized data sets of the integrity testing method described herein have significantly lower background noise levels without affecting the defect signal, resulting in an increased signal to noise ratio and greater ability to find outliers (defects) within a normal population.

[0019] Certain embodiments of the methods described herein can detect defects as small as 5 microns (μm) in a 10 inch filtration device, compared to 10-15 μm in traditional air diffusion testing.

[0020] Certain embodiments of the integrity testing method described herein comprise the following procedure for setting upper and lower cutoffs for the normalized data set: Certain embodiments of the method include a two-cutoff system with an initial fixed cutoff that identifies all defectives at the time of testing, and a secondary statistical cutoff that is applied to the normalized data set at the end of the lot.

[0021] One embodiment of a method for integrity testing includes the following steps for setting initial fixed upper and lower specification limits. obtaining historical air diffusion data from a wide selection of input dates and determining the degree of diffusion variability; determining a pressure adjustment value that can be directly compared to end user pre- and post-use integrity specifications and QC lot release diffusion standards; determining an upper cutoff from the end user diffusion specifications and QC lot release diffusion criteria that will result in minimal impact while also maintaining a safety margin of at least 15 percent; Determining a lower cutoff that will have minimal impact but also provide an appropriate boundary that will trigger a process re-evaluation if exceeded.

[0022] I. Cutoff Settings In some embodiments, the initial cutoff is placed far enough away from the current air diffusion value that it can withstand a large amount of input day variation without affecting yield. In some embodiments, the initial cutoff maintains a safety margin from end user pre- and post-use integrity testing as well as quality control (QC) lot release testing specifications. In some embodiments, the QC lot release testing specifications are adjusted for pressure. Some embodiments of the integrity testing are more sensitive than both the end user and QC lot release testing specifications.

[0023] In some embodiments, the diffusion specification is set at a reference test pressure (i.e., less than 30.0 sccm at 40 psi). In some embodiments, if the integrity test pressure was not equivalent to an existing reference pressure in an end user or QC specification, a correction factor is defined to compare values ​​from different test pressures.

[0024] In one embodiment, a secondary cutoff is applied to lot data normalized by bowl and / or entry date to significantly reduce two of the largest sources of variation from the data, allowing easier placement of cutoffs for optimal separation of outliers from the normal population.

[0025] One embodiment of a method for integrity testing includes the following steps to set secondary statistical upper and lower specification limits. selecting an integrity test data set that represents a statistically significant number of membrane filter units, entry dates, and any other testing or filter production variables that may affect air diffusion test background noise; testing the filters using air diffusion and removing any gross rejects; performing a Ryan-Joiner (RJ) normality test; removing outliers until the RJ score is an acceptable value; performing a normalization of the air diffusion data according to the methodology described herein; calculating a proposed upper specification limit between 3 and 4 standard deviations above the normalized median and a proposed lower specification limit between 4 and 6 standard deviations below the normalized median; evaluating the bacterial retention performance of units below, at, and above the upper available cutoff; taking into account the defect mode, defect size, and log reduction value (LRV) of retention failure below a cutoff, and sample size; reviewing the retention data and determining a final upper cutoff; assessing the proximity of the diffusion test result to a lower cutoff; and The step of reviewing the results and determining the final lower specification limit.

[0026] In some embodiments, the integrity testing methods described herein allow for the identification of erroneous raw materials used during the manufacturing process that existing techniques cannot identify during in-process testing. Some embodiments of the integrity testing methods have identified improperly processed materials (e.g., misoriented membranes) that existing techniques cannot identify during in-process testing.

[0027] Some embodiments of the normalization method adjust for test location (bowl) and lot (date of application) variations. Some embodiments include additional sources of variation such as water and air temperature, second membrane layer application date, support material roll, and actual achieved prestress or test pressure. Tradeoffs to statistical confidence must be acknowledged when evaluating the inclusion of additional factors resulting from reduced subgroup sample sizes.

[0028] Some embodiments of the integrity testing method can be used for any filtration device that employs gas-liquid diffusion testing, including, but not limited to, sterilizing grade filters, viral filters, clarification filters, and ultrafiltration filters. Some embodiments of the integrity testing method described herein maintain adequate yield and improve defect detection.

[0029] II.Equipment An embodiment of the integrity testing method is implemented on a control system that is built to operate 24 hours a day and 365 days a year. In an embodiment, a combination of instrumentation and software infrastructure allows for the automated collection of both structural and integrity testing material and process data. In an embodiment, process and material information is collected simultaneously by a data acquisition system.

[0030] In some embodiments, the membrane filter integrity test is performed in a single piece flow, or alternatively, a batch process of testing may be performed.

[0031] In one embodiment of the integrity testing method, an automated system uses the initial (overall) diffusion specification limits to process failed product in real time. In one embodiment, a lot typically comprises about 1000 to about 1800 10 inch membrane filters.

[0032] An embodiment of the method of integrity testing includes flagging any test bowls with anomalous results for review of potential influences from factors that exceed the expected median and / or standard deviation results compared to the recipe set or grand lot population. An embodiment of the method of integrity testing includes flagging any membrane rolls (dates of incorporation) with anomalous results for review of potential influences from factors that exceed the expected median and / or standard deviation results compared to the recipe set or grand lot population.

[0033] III. Definition Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0034] As used herein, the singular forms "a," "an," and "the" include plural references unless the content clearly indicates otherwise.

[0035] As used herein, "CF1" refers to a correction factor for normalization of a data set based on a first characteristic. In an embodiment, CF1 is a correction factor for a normalized data set with a bowl around a median of 0.

[0036] As used herein, "CF1MF" refers to a correction factor for mass flow normalization.

[0037] As used herein, "CF2" refers to a correction factor for normalization of a dataset based on a second characteristic. In one embodiment, CF2 is a correction factor for a normalized dataset with a master roll (entry date) around a median of 0.

[0038] As used herein, "CF2MF" refers to the correction factor for mass flow normalization.

[0039] As used herein, "initial fixed cutoff" refers to failures at the time of testing, similar to conventional integrity testing cutoffs. Initial fixed cutoffs are upper specification limits (USL) and lower specification limits (LSL) applied at the time of testing to identify moderate to total rejects. Units that fail the initial cutoffs are rejected and removed from the lot during the testing process.

[0040] As used herein, "perfect" refers to a membrane filter that is free of defects.

[0041] As used herein, "failed" refers to a defective membrane filter. In some embodiments, a failed membrane filter is instead referred to as "defective."

[0042] As used herein, a "secondary statistical cutoff" refers to the use of normalized test data applied at the end of a lot to identify additional defective items that are outliers relative to the normal population. The secondary statistical cutoff is a fixed USL and LSL that is applied to the data set at the end of the lot after normalization by bowl and master roll. Units that fail the secondary statistical cutoff are rejected and removed from the lot during the accountability step. The secondary statistical cutoff is established by setting the USL to be within 2-4 standard deviations above the normalized median. In some embodiments, the USL is at least 2 standard deviations or more above the normalized median. In some embodiments, the USL is about 2 standard deviations or more above the normalized median. In some embodiments, the USL is at least 3 standard deviations or more above the normalized median. In some embodiments, the USL is about 3 standard deviations or more above the normalized median. In some embodiments, the USL is at least 4 standard deviations or more above the normalized median. In some embodiments, the USL is about 4 standard deviations or more above the normalized median.

[0043] In some embodiments, the secondary statistical cutoff is established by setting the LSL to be within 4-6 standard deviations below the normalized median. In some embodiments, the LSL is at least 4 standard deviations below the normalized median. In some embodiments, the LSL is about 4 standard deviations below the normalized median. In some embodiments, the LSL is at least 5 standard deviations below the normalized median. In some embodiments, the LSL is about 5 standard deviations below the normalized median. In some embodiments, the LSL is at least 6 standard deviations below the normalized median. In some embodiments, the LSL is about 6 standard deviations below the normalized median. The standard deviation calculation may be from the characteristics of the membrane filter used in development, such as a launch day variation lot or bowl.

[0044] As used herein, "MRMinQty" refers to the minimum sample size per master roll required for standard processing. If the minimum quantity is not met, alternate processing rules apply. This is a recipe-defined parameter.

[0045] As used herein, "BowlMinQty" refers to the minimum sample size per bowl per lot required for standard processing. If the minimum quantity is not met, alternate processing rules apply. This is a recipe-defined parameter.

[0046] As used herein, "MaxBowlVar" refers to the limit of allowable variation between the bowl median for a standard process and the average lot bowl median. It is a recipe-defined parameter.

[0047] As used herein, "MaxMRVar" refers to the limit of acceptable variation between the master roll for standard processing and the average lot master roll median. It is a recipe-defined parameter.

[0048] As used herein, "MaxMRStDev" refers to the limit of the acceptable standard deviation of the normalized data for each master roll for standard processing. The sample standard deviation is used in this calculation. This is a recipe definition parameter. EXAMPLES

[0049] Example 1. Experimental Procedure In an additional example, an initial fixed cutoff and secondary statistical cutoff processing procedure for air diffusion data in an in-line integrity test was performed as follows: Calculations may be performed manually and / or using a database.

[0050] The mass flow rate variables USL, LSL, and CF2MF were rounded to the nearest tenth of a percentile. The CF1, CF1MF, CF2, and any standard deviation (StdDev) variables were rounded to the nearest hundredth of a percentile. Normalization was used to align the medians of the subpopulations of the data set. This process normalized the data set by bowl around a median of 0. The corrected data set was then normalized by master roll (date of input) around a median of 0.

[0051] Air diffusion data was collected at the end of the lot. Data from membrane filters that did not pass the initial cutoff, including aerosol failures, wet failures, and wet retests, were removed. One data set was used per membrane filter. If a membrane filter had more than one air diffusion test value, the final diffusion test value was used. Sample size N and mass flow median (BOWL_MEDIAN) were calculated for each bowl from the remaining data set. A correction factor CF1 for normalization was then calculated for each bowl B.

[0052] CF1 B =0-BOWL_MEDIAN B

[0053] Example: CF1 343 =0-BOWL_MEDIAN 343

[0054] The previous steps have been completed for all bowls.

[0055] If N for any bowl (B) is less than BowlMinQty, the following formula was used instead to calculate CF1 for the low sample size bowl(s): BOWL_MEDIAN AVG is the average bowl median of all bowls with sample sizes greater than BowlMinQty.

[0056] CF1 B =0-BOWL_MEDIAN AVG

[0057] Example: Bowl 346 had N=20, which was less than the BowlMinQty of 24.

[0058] CF1 346 =0-BOWL_MEDIAN AVG

[0059] Example: Four bowls were used in the test, MaxBowlVar=0.9 sccm (standard cubic centimeters).

[0060] BOWL_MEDIAN 343 =12.8sccm

[0061] BOWL_MEDIAN 344 =14.4sccm

[0062] BOWL_MEDIAN 345 =13.8sccm

[0063] BOWL_MEDIAN 346 =14.2sccm

[0064] Calculated BOWL_MEDIAN AVG =(12.8+14.4+13.8+14.2) / 4=13.8

[0065] Mass flow rate value (MF) tested in bowl (B) X For each membrane filter x in the data set with m m , a correction factor for normalization of the mass flow rate (CFMF) values ​​was calculated using the following equation:

[0066] CF1MF X =MF X +CF1 B

[0067] Example: Series 1001 was tested in bowl 345 with a recorded mass flow rate value of 14.3 sccm. CF1 345 was calculated to be -13.8sccm.

[0068] CF1MF 1001 =14.3+(-13.8sccm)=0.5sccm

[0069] The CF1MF data set was reviewed and the sample size N and median CF1MF value were calculated for each master roll (MR_MEDIAN) used within the lot. A correction factor CF2 for normalization was then calculated for each master roll (R) using the following: CF2 R =0-MR_MEDIAN R .

[0070] Working Example: CF2 3135UE =0-MR_MEDIAN 3135UE

[0071] CF2 3004UD =0-MR_MEDIAN 3004UD

[0072] The previous steps have been completed for all master rolls.

[0073] Limit validation was performed on the master roll sample size and master roll median. If the N of any master roll (R) was less than the MRMinQty, the following formula was used instead to calculate the CF2 of the low sample size master roll(s): CF2 R =0-MR_MEDIAN AVG . MR_MEDIAN AVG is the average master roll median of the remaining master rolls with sample sizes exceeding MRMinQty.

[0074] Example: Master roll 3162UE had N=15, which is less than the MRMinQty of 24.

[0075] CF2 3162UE =0-MR_MEDIAN AVG

[0076] Example: Four master rolls were used in one lot, MaxMRVar=1.5sccm.

[0077] The data is as follows:

[0078] MR_MEDIAN 5240UE =0.0sccm

[0079] MR_MEDIAN 5241UE =-1.0sccm

[0080] MR_MEDIAN 5242UE =1.7sccm

[0081] MR_MEDIAN 5243UE =-0.7sccm

[0082] Calculated MR_MEDIAN AVG =(0.0+-1.0+1.7+-0.7) / 4=0.0

[0083] For each group of membrane filters (x) in the data set, the CF1MF value with membranes from the master roll (R) (CF1MF X ) to calculate the CF2MF value using the following formula:

[0084] CF2MF X =CF1MF X +CF2 R

[0085] Example: Group 2001 contained membranes from master roll 3452UE and had a calculated CF1MF value of 1.1 sccm. CF23452UE was calculated to be -0.8 sccm.

[0086] CF2MF 2001 =1.1+(-0.8sccm)=0.3sccm

[0087] A limit validation was performed on the master roll standard deviation of the normalized mass flow values. Outliers were removed from the CF2MF data set and then the standard deviation of the CF2MF was calculated for each master roll. The CF2MF values ​​of each unit were compared to the secondary statistical cutoffs USL and LSL. Any units above the USL or below the LSL were treated as rejected units. Rejects from the secondary statistical cutoffs were removed from the production lot.

[0088] Example 2. Comparing standard data with normalized data Figure 1A (CGEP = 10 inch SHF filter, CSTVARLT1 = lot name) shows box plots of air diffusion values ​​from a lot composed of complete filters (0.2 μm PES (polyethersulfone) sterilizing grade 10 inch cartridge membrane filter; diffusion specification; less than 30.0 sccm at 40 psi) from the 11th introductory date. Typical diffusion data showed a population with a median of 13.4 sccm and a standard deviation of 0.45 sccm. Potential test specification limits, typically existing within a range of 3-4 standard deviations, were placed at 14.8-15.2 sccm. An upper cutoff was established at 15.0 sccm, which is located 1.6 sccm (3.5 standard deviations) from the population median. Devices with a nominal defect signal of 1.7 sccm or greater were treated as defective. At a test pressure of 35 psi, this excess flow rate corresponds to a single cylindrical defect size of 7-8 um calculated for an orifice type defect and was experimentally confirmed.

[0089] Figure 1B (CF2MF = normalized mass flow rate) is a box plot of the same data from the same membrane filter, normalized by test bowl and membrane loading date using the integrity testing methodology herein. The data was normalized to have a median of 0, with a population standard deviation of 0.16 sccm. For the normalized data set, an upper cutoff of 3 standard deviations was placed at 0.5 sccm from the population median. Devices with a nominal defect signal of 0.6 sccm or greater were treated as defective. At a test pressure of 35 psi, this excess flow rate corresponded to a defect size of 4-5 μm.

[0090] It has been observed that normalization of the data set significantly reduces the overall variability of the complete filter diffusion population. Thus, the completeness test method herein provides a high degree of sensitivity, applying a constant cutoff for each device, since its final processing is judged against subpopulation medians rather than absolute diffusion values.

[0091] Example 3: Treatment of Sterilizing Grade Filters It has been observed that air diffusion integrity testing using a secondary cutoff can adequately treat sterilizing grade filters as defective, which may fail the bacterial retention test but pass the conventional air diffusion integrity test. The lot shown in Figure 2A has air diffusion data plotted at the conventional upper and lower cutoffs for this product, which are 9.5 sccm and 5 sccm, respectively. Perfect filters are observed as a normally distributed population that falls between the upper and lower cutoffs, and filters with elevated diffusion (i.e., outliers to the normal population) are treated as failed units. Figure 2B shows a passing filter from this lot with all failed units removed.

[0092] Example 4: Comparison of population normalization for test bowls and membrane injection dates Looking at the data set from the previous example by subpopulation of test bowls (FIG. 3A) and membrane loading dates (FIG. 3B), additional outliers were observed. CVGL Lot C0BB87266 represents a 0.2 μm PVDF (polyvinylidene fluoride) sterilizing grade 10 inch DURAPORE® cartridge membrane filter.

[0093] Example 5: Treatment of Membrane Filters that Fail the Retention Test The box plot in Figure 4 shows the same data set from the previous example normalized for both test location and membrane insertion date. Certain additional outliers to the full population were observed, all of which passed the traditional air diffusion integrity test. One of these additional outliers also failed the retention test demonstrating the benefit to product quality (product and end user risk) from the application of this method of integrity testing.

[0094] Equivalent All ranges of the formulations recited herein include the ranges therebetween and can include or exclude the endpoints. Optionally included ranges are from the integer values ​​between (or including one original endpoint) of the recited scale or the next smaller scale. For example, if the lower limit is 0.2, the optionally included endpoints can be 0.3, 0.4, ... 1.1, 1.2, etc., as well as 1, 2, 3, etc., and if the higher range is 8, the optional included endpoints can be 7, 6, etc., as well as 7.9, 7.8, etc. One-sided boundaries such as 3 or greater also include a consistent boundary (or range) starting from the recited scale or the next smaller integer value. For example, 3 or greater includes 4 or 3.1 or greater.

[0095] Throughout this specification, references to "one embodiment," "a particular embodiment," "one or more embodiments," "an embodiment," or "an embodiment" indicate that the described feature, structure, material, or characteristic is an embodiment included in the present disclosure. Thus, appearances of phrases such as "in one or more embodiments," "in a particular embodiment," "in one embodiment," "an embodiment," or "in one embodiment" throughout this specification are not necessarily all referring to the same embodiment.

[0096] The patent applications and patent publications and other non-patent literature cited herein are incorporated by reference in their entirety in the entirety of the portion cited as if each individual publication or reference was specifically and individually indicated to be incorporated by reference as if fully set forth. Any patent application to which this application claims priority is also incorporated by reference herein in the manner described above for publications and references.

Claims

1. A method of integrity testing for a membrane filter, comprising: performing a gas-liquid diffusion analysis on the membrane filter; comparing the diffusion value to an initial fixed cutoff during the gas-liquid diffusion analysis to identify whether the membrane filter is an outlier; If the membrane filter is not an outlier relative to an initial fixed cutoff, normalizing the value of at least one characteristic of the at least two membrane filters that affects background noise in the gas-liquid diffusion analysis, the at least one characteristic being selected from the group consisting of lot, device test location, water temperature, air temperature, and air pressure; based on values ​​from the gas-liquid diffusion analysis of the plurality of membrane filters, comparing to a secondary statistical cutoff to identify whether the membrane filter is an outlier; processing the membrane filter as full or rejected; A method comprising: The initial fixed cutoff was within 3-4 standard deviations of the diffusion value; The secondary statistical cutoff is within three standard deviations of the normalized diffusion value.

2. The method of claim 1, wherein the gas-liquid diffusion method is air-water.

3. 3. The method of claim 1, wherein the at least one characteristic is selected from the group consisting of lot, equipment test location, water temperature, air temperature, and air pressure.

4. 4. The method according to claim 1, wherein the values ​​of at least two characteristics are normalized, the at least two characteristics being selected from the group consisting of lot, equipment test location, water temperature, air temperature and air pressure.

5. The method of any one of claims 1 to 4, further comprising normalizing the pressure of two or more membrane filters.

6. The method according to any one of claims 1 to 5, further comprising the step of detecting at least one defect in two or more membrane filters, the defect being a too large pore.

7. The method according to any one of claims 1 to 6, wherein the membrane filter is a sterilizing grade filter of 10 inch size.

8. The method of claim 7, wherein the at least one defect is between 4 μm and 9 μm in size.

9. 9. The method of claim 7, wherein the defects are greater than 4 μm.

10. The method of any one of claims 1 to 9, wherein the membrane filter is an outlier relative to an initial cutoff and is treated as a failure.

11. The method of any one of claims 1 to 10, wherein the membrane filter is an outlier relative to a secondary cutoff and is treated as a failure.

12. The method according to any one of claims 1 to 11, wherein the membrane filter is treated as being perfect and not an outlier for either the initial cutoff or the secondary cutoff.

13. The method according to any one of claims 1 to 12, wherein the plurality is between 1000 and 1800.

14. The method of any one of claims 1 to 13, wherein the membrane filter and the plurality of membrane filters are the same type of filtration device.

15. 15. The method of any one of claims 1 to 14, wherein the filtration device is selected from the group consisting of a sterilizing grade filter, a virus filter, a clarification filter, and an ultrafiltration filter.

16. 16. The method of any one of claims 1 to 15, wherein the secondary statistical cutoff is established by setting an upper statistical limit 2 to 5 standard deviations above the normalized median of the values.

17. 17. The method of any one of claims 1 to 16, wherein secondary statistical cutoffs are established by setting upper statistical limits at 2, 3, 4 and 5 standard deviations above the normalized median of the values.

18. 18. The method of any one of claims 1 to 17, wherein the secondary statistical cutoff is established by setting a lower statistical limit 4 to 6 standard deviations below the normalized median of the values.

19. 19. The method of any one of claims 1 to 18, wherein secondary statistical cutoffs are established by setting upper statistical limits at 2, 3, 4 and 5 standard deviations above the normalized median of the values.

Images