A discussion of the false working principles presented by laser diffraction particle size analysis instrument manufacturers in their descriptions of the advantages of the technology.

1. False Assumptions: Laser Diffraction PSA Systems Exposed Richard N. Kelly Principal Scientist JJPRD

4. Response of Laser Diffraction to Anisometric Particles Nadine Gabas, Noel Hiquily, Claude Laquerie Part. Syst. Charact. II (1994) 121-126 Richard Kelly: It has been known since 1994 that laser diffraction results are shape dependent. It was in that year that Gabas, et. Al. reported the results of their studies with custom made stainless steel shape standards; cubes, plates, and rods. It was reported that the results from measurements of plates were in error by 31% and those of the rods by 70%. Prompted by this report, laser diffraction manufacturers funded an intense research effort into the development of new detector configurations and new analysis algorithms in the hope of finding a solution to the problem posed by the shape dependence of the results.

5. Richard Kelly: However, in 1998 E. Dan Hirleman, presently professor emeritus of the school of mechanical engineering at Perdue, said: OK boys, we’ve given it our best shot. However,

6. LASER DIFFRACTION: INFLUENCE OF PARTICLE SHAPE AND A SHAPE ADAPTATION TECHNIQUE Heike Muhlenweg, E. Dan Hirleman Part. Part. Syst. Charact. 15 (1998) 163-169 Richard Kelly: The best that we can do is provide an algorithm that will transform the results into equivalent spherical volume diameters for those particle systems where the shape is constant.

7. LASER DIFFRACTION: INFLUENCE OF PARTICLE SHAPE AND A SHAPE ADAPTATION TECHNIQUE Heike Muhlenweg, E. Dan Hirleman Part. Part. Syst. Charact. 15 (1998) 163-169 Interpretation: A general algorithm for LD cannot be developed that can simultaneously deconvolute the diffraction pattern for particle systems which vary in both size and shape!!!!!!! Richard Kelly: However, in the general case where shape and size are both variables, no general solution can be provided. There are simply too many variables.

8. CORRELATION ATTEMPTS Image Analysis vs. Laser Diffraction Richard Kelly: During the same period, many tried to find a correlation between image analysis results and laser diffraction results. Many were they, the meek and the mighty, that had come, to try to remove the sword from the stone, but success had they none. It was not until the Kelly lad said; “all of your assumptions about this stone are bad”, that the sword was glad and flew to Kelly’s hand. Interpretation: The table was not meant to be a comprehensive listing of all who over the past 10 years have tried to find a correlation between these two techniques. The examples were chosen to demonstrate the time period over which these attempts were made and the range of talent that has been applied. I believe Brewster and Ramsland were analysts like myself, whereas the others are scientists of international renown; Brian Kaye being the father of fractal geometry. Yet in each case only partial success at best was achieved. In each case, it was stated that either a poor correlation was achieved in the case of acicular particles, or no correlation at all was observed..So, liking a challenge, for the rest of this presentation we will concentrate upon the correlation of results from the measurements of acicular particle systems. DATE JOURNAL AUTHOR/S PLOT 1995 J. Pharm. Sci. E. Brewster & A. Ramsland None Frequency Table IA : Area/Vol. Est. vs. Feret Diameter LD :Volume 1997 Part. Part. Syst. Charact. B. Kaye, D. Alliet, L. Switzer, C. Turbitt-Daoust IA :Area% vs. ESD LD :Volume % 2002 J. Disp. Sci. & Tech. P. Bowen IA :Area% vs. ESD LD :Volume %

9. IMAGE ANALYSIS/LASER DIFFRACTION CROSS VALIDATION NEED Richard Kelly: Even though a correlation was not found, standard setting organizations were suggesting, as shown in this block diagram, that image analysis and laser diffraction be used in tandem. Note: That for the purpose of this conversation that image analysis is considered to be nothing more than computer-aided microscopy. It was suggested that microscopy results be compared to laser diffraction results during laser diffraction method development and that an iterative process of changes be followed until the two sets of data were in substantial agreement.

10. Richard Kelly: But, being without a means of determining when the results were in agreement led to situations like the one shown here. The last column of data are data collected from the measurement of an acicular particle system. When comparing the image analysis area% data and the laser diffraction volume% data, we see that a ca. 300% difference exists between the d10 values and an approximate 100% difference exists between the d50 values. So, how was one to proceed? Image Analysis (Number Probability-microns) Sample ID 656-42-1B Pooled Data Trials 1-2 19376-86B1 19376-120A 656-32-4 Pooled Data Trials 1-2 19376-96A Mean 5.28 10.97 29.80 31.19 80.18 Std Dev. 2.35 9.78 28.42 33.83 78.54 Maximum 20.78 71.00 198.77 227.40 426.37 d15 3.14 2.90 7.96 7.30 9.82 d50 4.81 7.62 19.87 15.91 55.01 d90 8.22 23.30 64.35 83.49 196.52 Image Analysis (*Area Probability-microns) d10 4.02 10.42 26.70 31.42 92.95 d50 6.78 25.17 76.22 91.40 204.25 d90 12.24 48.71 168.57 146.66 323.98 Coulter LS 13 320 Analysis (Volume Probability-microns) d10 0.84 9.91 23.62 26.06 38.70 d50 2.70 27.66 54.40 56.99 108.60 d90 8.17 47.50 127.20 119.80 269.60

11. POPULAR CROSS VALIDATION STRATEGY WITHIN THE PHARMACEUTICAL INDUSTRY Richard Kelly: A certain fraction of those within the particle size analysis community responded by adopting the behavioral pattern of the no hear, no see, no speak evil apes. As laser diffraction has become their favorite particle sizing technique, if differences between image analysis and laser diffraction results emerged, then they refused to see them, or worse, they would pretend that they were unimportant. If someone attempted to bring the real differences to their attention, they would not listen nor engage in constructive scientific dialogue. The problem with covering your eyes is that you voluntarily enter into a dark age where you are no longer guided by the light of reason and you prevent yourself from seeing even simple truths…if you don’t look for answers you probably will not find them. And by placing your hands over your ears you prevent yourself from hearing the voice of reason. If you are not being guided by that voice, then you make it possible to be led astray by other voices

12. RESULT OF LACK OF SCIENTIFIC CROSS VALIDATION PRINCIPAL Richard Kelly: Such as the voice of flying pigs. I have three daughters; 13, 10, and 2. The two eldest daughters presently are fond of pigs. I had the opportunity to walk my 10 year old to the bus stop this morning, and having shown her the cartoon earlier, she asked me to explain it to her. I told her that the flying pig represents an individual who through the demonstration of his or her talent has distinguished themselves and have therefore risen in position within the community to that of an authority figure. The sheep represent the rest of us, who having great respect for the unique ability of this animal, typically heed the authority figures words. The problem that often occurs is that we often forget to question the verity of the message, especially if there is a long history of the spokesman’s record being sound. The black sheep represents those who don’t forget to question each statement. The black sheep are those responsible for scientific advancement. The black sheep realizes that the authority figure at times is a messenger angel , a conduit through which the true spirit of analytical science flows, whereas at other times the authority figure is just a squealing, flying pig, a conduit through which oozes

13. FALSE A L S E CCURACY DOES NOT MATTER! Accuracy expresses the closeness of agreement between the value which is accepted either as a conventional true value or an accepted reference value and the value found. However, with particle size methods, accuracy cannot be measured because there is no “true” particle size value to compare against and image analysis data cannot be utilized as a reliable standard because the actual measurements are of different physical properties of the particles. POPULAR FALSE MANTRA Richard Kelly: Passages such as the one shown here where the author contends that there is no such thing as a “true” particle size value thereby simply dismissing the applicability of statistics to the characterization of particle systems, and in which the author argues that image analysis and laser diffraction results cannot be compared because “the actual measurements are of different physical properties of the particles”. It should be noted that even the most vocal advocate of laser diffraction, Dr. Brian Scarlett, agrees that the above statements are in error.

14. FALSE A L S E CCURACY DOES NOT MATTER! Different particle size results obtained through various methods are equally accurate, but each method might be expressing the results in different terms, different properties of the particles are being measured. As in the case of this study, if one is more interested in developing profiles of lot-to-lot variability, then the use of any of the available methods that yields “correct” results is appropriate; where a “correct” method is defined as one whose sample was obtained by an appropriate sampling procedure, in which the sample was prepared properly and introduced into the instrument, and in which all instrumental parameters were used correctly for the analysis. POPULAR FALSE MANTRA Richard Kelly: Worse yet it is argued that “Different particle size results obtained through various methods are equally accurate – equally correct”. Whenever I see this statement I am caused to recall the scene in “Boystown” where Father Flanagan played by Spencer Tracey asserts that “There is no such thing as a bad boy”…There is no such thing as a bad particle size analysis system..they are all equally correct. This article was written by a present colleague of mine. It was learned that these passages were written because of following the voice of a flying pig.

17. Position Paper on Particle Sizing: Sample Preparation, Method Validation and Data Presentation Richard Bell, Andrew Dennis, Barry Hendriksen, Nigel North and John Sherwood Pharmaceutical Technology Europe November 1999 NON-POSITION PAPER: METHOD SELECTION Richard Kelly: An acicular particle system is chosen as the example particle system for this paper. The example comment: This system is not spherical, remember this before proceeding with the development of your laser diffraction method…Obviously, no real guidance steps are provided in this document. Also note the absence of a micron bar in the micrograph. I’m certain the now departed Walter McCrone is still very disturbed by this practice.

18. GRAPHICAL CORRELATION OF IMAGE ANALYSIS AND LASER DIFFRACTION DATA: ACICULAR PARTICLES Richard Kelly: Fortunately, a more meaningful practice exists for comparing image analysis and laser diffraction data. We can leave smoke and mirrors behind us and proceed along an actual scientific path. In this practice the so called “volume probability” laser diffraction data is plotted on the same graph as the projected surface area based longest dimension, breadth, area equivalent diameter image analysis data. Then simple arithmetic corrections are applied to the data sets as required. This will be discussed in greater detail later, but first I believe it to be very instructive to explore

20. LASER DIFFRACTION FAIRY TALE Richard Kelly: In other words, the fault was in part due to the inability to break free from the enchantment of the laser diffraction fairy tale. This fairy tale is told in many ways. I have heard it told as recently as last summer at the AAPS particle size analysis workshop. One form of the fairy tale goes as follows: On the first day God created light and found it good. On the second day God created all geometrical forms, including the sphere, and found them good. On the third day, God created all man-kind, including the future laser diffraction manufacturers. God withheld judgment. On that day the future laser diffraction manufacturers said to God: “God, the light you have created is good. And God, of all the geometrical forms you have made, the sphere is muy excellente as it is one of the few that can be characterized by a single parameter, its radius, and because it has a simple volume to mass relationship”. Then the future laser diffraction manufacturers said to God, “OK God, your part in this creation process is over, we will take over from here.

21. LASER DIFFRACTION Richard Kelly: So, on the fourth day the laser diffraction manufacturers created the laser diffraction particle size analyzer. They endowed it with a high intensity monochromatic light source, ie a laser, a sample cell, and a set of concentric ring detectors.

22. Laser Diffraction – Raw Data Richard Kelly: On the fifth day, the laser diffraction manufacturers created the laser diffraction software. They took the ensemble produced laser diffraction pattern

23. LASER DIFFRACTION INVERSION MATRIX Richard Kelly: And applied advanced mathematical principles

26. LASER DIFFRACTION Example of Graphical Data Representation Richard Kelly: So, the analysis algorithm is able to present the results in a graphical format as shown in the example graph where these results would be interpreted as having been produced by the measurement of a particle system with two subpopulations having d50 values of ca. 2 microns and ca. 6 microns, each population being present with approximately equal mass. On the seventh day God created a black hole and escaped to a different dimension and the laser diffraction manufacturers rested expending only that amount of energy to count their profits.

30. DEMONSTRATION OF LAMINAR FLOW Fibres and Fibre Collectives with Common Laser Diffractometers Part. Part. Syst. Charact. 17 (2000) 113-116 Christoph Berthold, Robert Klein, Jurgen Luhmann, Klaus G. Nickel Richard Kelly: To appreciate my point of view, the following facts must be considered: 1. Berthold, Klein, Luhmann, and Nickel demonstrated in their paper the existence of laminar flow in most LD systems.

31. THEORETICAL PREDICTION OF EFFECT OF ORIENTATION ON DIFFRACTION PATTERN Transformation of Diffraction Pattern due to Ellipsoids into Equivalent Diameter Distribution for Spheres Tatsushi Matsuyama, Hideo Yamamoto, Brian Scarlett Richard Kelly: Secondly, Scarlett et al have reported that in the case of monosized elliptical particles that their mathematical model indicated that in the case of random orientation that diffraction only from the minor chord, ie the breadth of the ellipses, is detected whereas in the case of flow alignment that diffraction from both the major, ie the length of the ellipses, as well as the minor chords are detected such that two peaks rather than one are found in the volume probability distribution graph.

33. MALVERN AEA MEASUREMENT RESULTS Richard Kelly: The volume probability results from the measurements made are characterized by peaks corresponding to diffraction along both the minor and major chords. This response was noted regardless of the flow speed chosen and regardless of the LD system used. Beckman-Coulter, Horiba, Malvern, and Sympatec systems all provided the same response. Therefore, in light of Dr. Scarlett’s paper, it appears that flow alignment occurs for each of these systems, from 12 microns down to 3 microns.

34. An Estimate of Orientation Effects on the Results of Size Distribution Measurements for Oblate Particles Ceramics-Silikaty (2002), 46(2), 41-48 W. Pabst, J. Mikac, E. Gregorova, J. Havrda “ The influence of the orientation of oblate particles on vol.-weighted particle size distributions (PSD) is demonstrated with an idealized model system. The overall est. is based on partial “worse case” ests. found by an explicit calcn. of selected non-random orientations for disk-shaped particles with an aspect ratio of 10 and a wide-range PSD. According to the simulation results the d10, d50 (median) and d90 values for particle orientations of 60 o , 45 o , 30 o , and 0 o deviate by less than 17% from those of the 90 o orientation (corresponding to projections normal to the disk plane), whereby the max. deviation occurs for orientations close to 30 o . For quasi-random orientation the deviation is even smaller (approx. 3-7%). Thus it is shown that, in contrast to the widespread belief, the effect of particle orientation on the resulting vol.-weighted PSD is for oblate particles small enough to be completely smeared out for purely statistical reasons to within typical errors of measurement of the usual sizing methods.” Richard Kelly: These findings coupled with the findings from recent numerical simulations conducted in the Czech Republic that indicate that deviations from perpendicular alignment do not have a statistically significant effect upon the results

35. LESSON 1: FALSE Particles > 1 micron, as a general rule, are not present within the flow cells of laser diffractometers at random orientations with equal statistical weight Laser diffraction-based particle size analysis results are not strongly dependent upon small deviations from the perpendicular laser-particle orientation. Richard Kelly: Lead to the following conclusion: The assumption that particles are present in the LD sample cells in random orientations with respect to the laser each with equal statistical weight is: FALSE. It is instead safer to assume that for particles greater than 1 micron that the particles are oriented 90 degrees with respect to the laser.

37. SERMON ON THE MOUNTS OF MALVERN The Importance of Particle Size Analysis in the Pharmaceutical Industry Dr. Alan Rawle Mie theory allows us to transform the measured intensity distribution data to volume It is applicable for all sizes from 2A gas molecules or atoms to 1 km asteroids in space Just because a particle is non-spherical does not mean that stops scattering light! The interpretation of the scattering data can still be evaluated by the Mie theory and it has been shown that for many shapes (eg cubes, cylinders, rhombohedra) the obtained diffraction pattern can be explained by Mie theory in terms of the volume of the particle A volume distribution is generated directly which is equal to the weight distribution if the density is constant. This is the preferred distribution for chemical engineers. Richard Kelly: This examination brings us to what I call the “Sermon On The Mounts of Malvern”. Dr. Alan Rawle of Malvern several years ago wrote an article entitled: “The Importance of Particle Size Analysis in the Pharmaceutical Industry”. This article served for many years as the primer on LD particle size analysis. Within this article Rawle contends that the advantages of LD include: 1. The Mie theory allows data to be expressed in terms of volume 2. It is applicable to particles of all size from gas atoms to asteroids 3. The diffraction from particles of any shape can be interpreted in terms of spherical equivalent volume diameters

38. SERMON ON THE MOUNTS OF MALVERN The Importance of Particle Size Analysis in the Pharmaceutical Industry Dr. Alan Rawle For the large region d>>5  then Fraunhofer approximation gives a reasonably good correspondence to the exact answer generated by Mie theory The Mie theory assumes the volume of the particle as opposed to Fraunhofer which is a projected area prediction Richard Kelly: To be fair to Dr. Rawle, it must be stated that hidden within the article are the statements that if the diameter of the particle is much greater than the wavelength of the laser, then the Fruanhofer approximation applies which provides data in terms of projected surface area. However, immediately following these statements the Mie theory is again touted as being general and providing volume, and therefore mass, based results.

39. GRAPHICAL CORRELATION OF IMAGE ANALYSIS AND LASER DIFFRACTION DATA: ACICULAR PARTICLES Richard Kelly: However, within this graph the LD volume data is coincident with either the image analysis projected surface area breadth and/or longest dimension data. When does volume equal projected surface area? In our three dimensional world this never occurs. Coincidence is the result of the fact that what is referred to as volume in the LD data presentation is actually projected surface area.

40. ALTERNATE SOURCE OF STANDARDS Richard Kelly: How about supporting data? Yes, data is good. However, as previously stated, their exist at the moment only three certified shape standards. Therefore, in order to generate supporting data it was necessary to first identify an appropriate LD system. The system chosen was the R8 configured Sympatec LD system as it allows the measurement of particles up to 8 mm in length. Then it was necessary to find a suitable source of pseudo standards. Our source was a local Michaels Arts and Crafts center.

41. MICROGRAPH OF 5 mm PLATE & 6 mm CUBE Richard Kelly: From Michaels a variety of Jewelry & Craft beads were purchased including cubes and plates.

42. EQUIVALENT SPHERICAL DIAMETERS 5mm Plate/6mm Cube Richard Kelly: From the data contained in this table it can be seen that if LD systems provided data in terms of volume that in ESD units the size of the cubes would be 4 to 5 times greater than that of the plates.

43. 5 mm PLATE LD DATA Richard Kelly: Here is the data obtained from the plates

44. 6 mm CUBE LD DATA Richard Kelly: Here is the data from the cubes

45. COMPARISON OF 5mm PLATE AND 6 mm CUBE LD DATA Richard Kelly: And here the data from the 2 sets of experiments are combined and it is clearly seen that nothing close to a 4-5X difference exists between the d50 values. The conclusion is that LD systems operating within the Fraunhofer domain do not provide volume (mass) based data but projected surface area data.

46. Richard Kelly: Similar experiments were performed with other shaped particles

47. 5 mm FLAT STAR LD DATA Richard Kelly: And similar results were obtained.

48. 6 mm SOLID STAR LD DATA

49. LESSON 2: FALSE SURFACE AREA = VOLUME X For particles > 1 micron, laser diffraction PSA system results are not “volume” distribution results and therefore cannot be considered “mass” distribution results. Volume-weighted is not equivalent to “volume” Richard Kelly: The second lesson is that the assumption that LD systems provide volume (mass) based results is FALSE. As seen from the graph presented earlier, although LD results may be volume-weighted as a result of a cubed radius term in the analytical equation, LD results for particles of size above 1 micron are best considered as projected surface area data.

51. MALVERN AEA MEASUREMENT RESULTS Richard Kelly: We have already seen that the LD measurement of the LGC Promochem standards consisting of monosized particles results in volume distribution graphs with peaks corresponding to both diffraction along the minor and major chords. In these graphs the peaks correspond to breadth and length and not to some equivalent spherical measure.

52. AEA MEASUREMENT DATA ESD 2.92 3.40 Richard Kelly: This is clearly seen from a comparison of the calculated ESD and the measurement results. SAMPLE ID AEA 1002 (7.51x1.72x1.02) AEA 1003 (12.13X1.70X1.00) d10 2.024 2.074 d50 5.233 5.341 d90 6.756 12.709

53. LESSON 3: FALSE There is no benefit to be gained from interpreting laser diffraction results in terms of “equivalent spherical volume diameters. Richard Kelly: Lesson 3 is then that the assumption that LD systems provide results in terms of equivalent spherical diameters is FALSE.

54. CORRELATION ATTEMPTS Image Analysis vs. Laser Diffraction Richard Kelly: So, returning to the question as to why it took ten years to find the simple correlation between LD and image analysis results for acicular particles, it should be obvious that the reason is that each of those that had undertaken this challenge failed in part as a result of the enduring enchantment of the laser diffraction fairy tale. DATE JOURNAL AUTHOR/S PLOT 1995 J. Pharm. Sci. E. Brewster & A. Ramsland None Frequency Table IA : Area/Vol. Est. vs. Feret Diameter LD :Volume 1997 Part. Part. Syst. Charact. B. Kaye, D. Alliet, L. Switzer, C. Turbitt-Daoust IA :Area% vs. ESD LD :Volume % 2002 J. Disp. Sci. & Tech. P. Bowen IA :Area% vs. ESD LD :Volume %

55. FALSE FALSE FALSE FALSE Richard Kelly: They failed to realize the extent to which the assumptions provided by the manufacturers were FALSE.

56. THE PERIL OF ADVERTISING CLAIMS Richard Kelly: They had failed to remember the lesson first taught in Homer’s Odyssey, which is; If you are on a voyage in search for the truth, your passage may well take you past the island of the ad man sirens. If you choose to listen, then you must be certain that you remain bound fast to the mast of reason. If not, then you most probably will willingly place yourself forever on the beach of the ad man sirens under the enchantment of their song of advertised claims unwittingly serving as a source of food to them until your scientific ambition remains a well picked carcass.

58. IMAGES OF SCIO 323 HCL LOT 656-42-113 Richard Kelly: The first step is a review of the LD and image analysis data collected from the measurements made on 5 salts of SCIO 323. These salts being selected as their morphology varies in an almost continuous fashion from micronized spherical particles

59. IMAGES OF SCIO 323 HCL LOT 19376-86B1 Richard Kelly: To irregular spherical particles

60. IMAGES OF SCIO 323 FREE BASE LOT 19376-120A Richard Kelly: To a particle system containing acicular and irregular spherical particles

61. IMAGES OF SCIO 323 FREE BASE LOT 656-32-4 Richard Kelly: To a more predominantly acicular particle system with a small population of fines

62. IMAGES OF SCIO 323 FREE BASE LOT 19376-96A Richard Kelly: To finally a particle system consisting of coarse acicular particles

63. DIFFERENTIAL VOLUME % GRAPH SCIO 323 HCL LOT 656-42-113 Richard Kelly: Presented in the next 5 slides is the corresponding LD data. The integration limits were chosen on the basis of the microscopic determination of the largest part. This role for microscopy, I.e. the use of microscopy as a reality check of LD results, is a well accepted role for microscopy.

64. DIFFERENTIAL VOLUME % GRAPH SCIO 323 HCL LOT 19376-86B1

65. DIFFERENTIAL VOLUME % GRAPH SCIO 323 FREE BASE LOT 19376-86B1

66. DIFFERENTIAL VOLUME % GRAPH SCIO 323 FREE BASE LOT 656-32-4

67. DIFFERENTIAL VOLUME % GRAPH SCIO 323 FREE BASE LOT 19736-96A DRIED

68. Richard Kelly: This table provides the opportunity for a numerical comparison of the results. Note that knowing that the LD volume data actually is projected surface area data, the comparison to be made is between the image analysis projected surface area data and the so called LD volume data. The image analysis parameter chosen is longest dimension. From the table it is seen that the IA area data and the LD data compare favorably in the case of the micronized material. A roughly constant difference of ca. 4 microns exists between the data sets in this case. Considering that no attempt was made to make an exact determination of the LD optical parameters and that at the magnification employed the complete particle size range was not covered, this difference can be explained and is not considered significant. In the case of the irregular spherical particles the IA and LD data sets can be considered equivalent. (This again underscores the fact that LD measurements are not volume but projected surface area based). For the remaining three salts where there is an increasing population of acicular particles, it is seen that the d parameters of the IA data increase at a significantly greater rate than that of the LD data which is seen to level off. Image Analysis (Number Probability-microns) Sample ID 656-42-1B Pooled Data Trials 1-2 19376-86B1 19376-120A 656-32-4 Pooled Data Trials 1-2 19376-96A Mean 5.28 10.97 29.80 31.19 80.18 Std Dev. 2.35 9.78 28.42 33.83 78.54 Maximum 20.78 71.00 198.77 227.40 426.37 d15 3.14 2.90 7.96 7.30 9.82 d50 4.81 7.62 19.87 15.91 55.01 d90 8.22 23.30 64.35 83.49 196.52 Image Analysis (*Area Probability-microns) d10 4.02 10.42 26.70 31.42 92.95 d50 6.78 25.17 76.22 91.40 204.25 d90 12.24 48.71 168.57 146.66 323.98 Coulter LS 13 320 Analysis (Volume Probability-microns) d10 0.84 9.91 23.62 26.06 38.70 d50 2.70 27.66 54.40 56.99 108.60 d90 8.17 47.50 127.20 119.80 269.60

69. REINTEGRATED AEA MEASUREMENT DATA Richard Kelly: The reason for the leveling of the LD data is that it is the result of the increasing contribution of diffraction from the minor chord (breadth) to the total sum of the diffraction events. But as the diffraction events are additive, in certain cases the minor chord contributions can simply be subtracted from the total leaving, in the case of acicular particle systems, primarily the diffraction events from diffraction along the major chord (longest dimension). STAT BEFORE (7.51x1.72x1.02) AFTER (7.51x1.72x1.02) d10 1.951 4.944 d50 4.810 6.473 d90 7.745 8.232

70. REINTEGRATED SCIO MEASUREMENT DATA LOT 19736-96A Richard Kelly: Applying this procedure to each of the three data sets containing acicular particles, we see that the result of the procedure are d50 and d90 values that are closer in agreement to the IA derived d50 and d90 values.

71. REINTEGRATED SCIO MEASUREMENT DATA LOT 19736-96A STAT IMAGE ANALYSIS COULTER BEFORE RI COULTER AFTER RI %ERROR BEFORE RI %ERROR AFTER RI d10 92.95 38.70 112.00 58.36 20.49 d50 204.25 108.60 174.50 46.83 14.56 d90 323.98 269.60 330.40 16.78 1.98

72. REINTEGRATED SCIO MEASUREMENT DATA LOT 656-32-4

73. REINTEGRATED SCIO MEASUREMENT DATA LOT 656-32-4 STAT IMAGE ANALYSIS COULTER BEFORE RI COULTER AFTER RI %ERROR BEFORE RI %ERROR AFTER RI d10 31.42 26.06 57.31 17.06 82.40 d50 91.40 56.99 83.93 37.65 8.17 d90 146.66 119.80 142.00 18.31 3.17

74. REINTEGRATED SCIO MEASUREMENT DATA LOT 19736-120A

75. REINTEGRATED SCIO MEASUREMENT DATA LOT 19736-120A STAT IMAGE ANALYSIS COULTER BEFORE RI COULTER AFTER RI %ERROR BEFORE RI %ERROR AFTER RI d10 26.70 23.62 57.92 11.54 116.93 d50 76.22 54.40 86.66 28.63 13.70 d90 168.57 127.20 156.30 24.54 7.28

76. GRAPHICAL CORRELATION OF IMAGE ANALYSIS AND LASER DIFFRACTION DATA: ACICULAR PARTICLES Richard Kelly: The change in the d10 values do not follow this trend, however, it is not to be expected that they would. The LD d10 values predominantly reflect the minor chord contribution to the sum of the diffraction events. This is clearly seen in the graph of the results. So, the LD d10 values to be compared to the IA d10 values in these cases would be those from the uncorrected data.

77. IMAGE ANALYSIS/LASER DIFFRACTION CROSS VALIDATION NEED FULFILLED? Richard Kelly: Having shown that a procedure exists that can be used to meaningfully compare LD and IA data, let us explore an example of its use in LD suitability testing and method development.

78. IMAGES OF ACICULAR PARTICLES + FINES Richard Kelly: Consider a particle system consisting of both acicular particles and fines. Let us for the purpose of this example imagine that it has been determined that to control product quality and/or manufacturability that monitoring of the fines is required. The suitability question is whether LD can be applied in a QC lab for this purpose for this particle system.

79. GRAPHICAL COMPARISON OF IMAGE ANALYSIS & LASER DIFFRACTION DATA ACICULAR PARTICLES + SPHERICAL FINES Richard Kelly: Obtaining the IA and LD data and plotting the data as previously described, it is seen that in this case the LD data does not follow the IA breadth data for the entire particle population. Instead it is observed that the LD data follows the breadth line for only the acicular particles (aspect ratio greater than 2) up to almost 40% by area. This is actually to be expected in as much as LD intensity is proportional to the square of the cross sectional area of the particle. Therefore, the breadth contribution from the acicular particles renders the contribution from the fines statistically insignificant.

80. ANOTHER TRIP TO MICHAELS Richard Kelly: To provide some supporting evidence to the previous statements, let’s make another trip to Michaels.

81. Richard Kelly: 2.5 mm spheres and 4.5 mm x 2.5 mm rods were purchased. Samples were prepared in which the weight contribution of the rods varied from 0% to 100%. These samples were again analyzed using the R8 configured Sympatec LD system.

82. Richard Kelly: Presented is the LD volume-weighted probability distribution graph obtained from the measurement of the rods

83. Richard Kelly: And this graph presents the results from the measurement of the spheres

84. LASER DIFFRACTION RESULTS CYLINDER/SPHERE W/W% MIXTURES Richard Kelly: Three measurements of each weight percent mixture were analyzed. The results were averaged and plotted resulting in this graph which presents the average d10, d50, and d90 values for each combination. The leveling effect is clearly seen as the d10 and d50 lines have close to 0 slope values over almost the entire measurement range. Only the d90 value is seen to be sensitive to the change, but only after exceeding the limit of detection of large particles in the presence in fines which in this case was close to 10% by weight. The reason often stated for the choice of LD as the PSA measurement technique is the techniques reproducibility. Last summer at the AAPS workshop, a presentation was given on the development of a LD method for a particle system characterized by an average aspect ratio of 10:1. In this presentation a table of IA and LD data was presented containing the measurement results for different lots. It was noted that the LD data was more reproducible and so LD was chosen as the measurement technique. From the graph shown here, I believe it appropriate to ask whether the LD data just indicated the inability to differentiate between samples with real size differences.

85. MIXTURE RESULTS 4 . 5mm CYLINDERS / 2.5 mm SPHERES Richard Kelly: This table is a numerical comparison of the results. From these data it is seen that the d10 value barely rose above a total of 6% difference from the starting value, that the d50 value did not rise above 5% difference from the starting value until the rods were present in excess of 50% by weight, and again that the d90 values were sensitive to the change only after the limit of detection of ca. 10% by weight had been exceeded.

86. FDA SCIENCE POSTER FEB 2001 COMPARATIVE ANALYSIS OF COMMON PARTICLE SIZING TECHNIQUES FOR PHARMACEUTICAL POWDERS H.R. Prasanna, E.H. Jefferson, J.S. Taylor, A.S. Hussain, R.F. Karuhn, R.C. Lyon Richard Kelly: The FDA is aware of this limitation of LD. The FDA conducted research into the extent of comparability of various common particle size analysis techniques. The sample chosen was nitrofurantoin. It was chosen because of its acicular morphology and because it contained significant fines. In the poster they presented in Feb 2001 they concluded that

87. FDA SCIENCE POSTER FEB 2001 COMPARATIVE ANALYSIS OF COMMON PARTICLE SIZING TECHNIQUES FOR PHARMACEUTICAL POWDERS H.R. Prasanna, E.H. Jefferson, J.S. Taylor, A.S. Hussain, R.F. Karuhn, R.C. Lyon Richard Kelly: “ The large population of fine particles was only detected by the particle counting techniques (image analysis and light obscuration) and not by the ensemble technique (laser diffraction).

88. EXERPT FROM MAY 2003 AAPS PSWS SPEECH GIVEN BY BRIAN SCARLETT The control of an existing product or process requires less stringency than the design of a new product or process. For control, the minimum requirement is that the instrument should be reproducible. Practically, this means that the instrument variance is much less than the sample variance. The more vital requirement is that the instrument should be sensitive to those changes in the particle size distribution which cause a change in the product properties. There is no virtue in an instrument being very reproducible if it is so insensitive so that it does not detect the relevant changes. The accuracy of the measurement is more difficult to achieve since it concerns the actual size of the particles and there arises the question of definition. Nevertheless, in product design, and consequently also process design, the actual size of the particles may be crucial. If the particles must actually match the size of a capillary or pore in the body, then an arbitrary scale of measurement is insufficient and the question of what is to be measured must be addressed. Richard Kelly: Therefore, in the case presented of acicular particles in the presence of fines we might conclude that LD would not be the particle size analysis technique of choice. In this case we would recall Dr. Brian Scarlett’s words that for product quality control and manufacturability that “The more vital requirement is that the instrument should be sensitive to those changes in the particle size distribution which causes a change in the product properties. There is no virtue in an instrument being very reproducible if it is so insensitive so that it does not detect the relevant changes”. It is also to be noted that Dr. Scarlett directly rebutted the popular false mantra “Accuracy does not matter.”

89. DARK AGE: PRIMARY MOTIVATIONAL FORCE Richard Kelly: Findings such as these might be considered by the true analytical scientist as photons of the light of reason and gladly accepted as further aid in finding ones way to truth. However, those that love the dark hate the light. So, many LD devotees see these findings as possibly being part of the content of a Pandora’s box . As we know fear is one of the primary forces in a dark age and LD devotees fear that these findings might disrupt their unchecked use of their favorite particle size analysis technique.

90. POPULAR CROSS VALIDATION STRATEGY WITHIN THE PHARMACEUTICAL INDUSTRY Richard Kelly: This fear reinforces no see, no hear, no speak evil behavior making the drafting of truly useful guidance documents difficult if not impossible.

91. LD: LIMIT OF DETECTION Richard Kelly: From personal experience I can tell you that well intentioned individuals have attempted to have findings such as those discussed today included in some useful fashion into guidance documents. For instance, David Driscoll of the Beth Israel, Deaconess hospital of the Harvard Medical School found that the LD limit of detection weakness could actually put patients at risk of death if LD were used in a QA environment for the monitoring of IV lipid emulsions. Therefore, an attempt was made to have such information included in a guidance document. Shortly before the document was to be presented for approval, the LD forces rose up and scrubbed the document clean of any statement that might cause the unchecked use of LD to be questioned. A good guidance document, recognizing the widespread use of LD, should make it very clear what the “actual” versus “advertised” operating principles of this technique are and present some form of prescription for a rational procedure by which the decision as to whether LD is a suitable technique for the problem at hand is to be made. The mistakes of the European position paper on LD method development should not be repeated. As mature scientists we should be willing to admit what is fact and what is but the lingering remnant of the LD fairy tale.

92. QUIZ Richard Kelly: Let’s see whether I have been successful as a teacher. Let’s take the “TRUE/FALSE Quiz”.

93. <429> LIGHT DIFFRACTION MEASUREMENT OF PARTICLE SIZE PHARMACOPEIAL FORUM VOL. 29(4) [July-Aug. 2003 Light diffraction is one of the most widely used techniques for measuring the size of a wide range of particles from very fine to very coarse. The method is popular because it is quick and easy to use, flexible, and it can be adapted to measure samples in various physical forms. The method depends on the analysis of the diffraction pattern produced when particles are exposed to a collimated beam of light. As the patterns are characteristic of the particle size, mathematical analysis can produce an accurate, repeatable picture of the size distribution. Question 1: (True or False) Richard Kelly: This quiz consists of a review of statements extracted from an article from the Pharmacopeial Forum Volume 29 (4) 2003. 1. True/False

95. FALSE: A L S E CCURACY DOES NOT MATTER! NOT IN AGREEMENT WITH POPULAR MANTRA

96. <429> LIGHT DIFFRACTION MEASUREMENT OF PARTICLE SIZE PHARMACOPEIAL FORUM VOL. 29(4) [July-Aug. 2003 For non-spherical particles, an equivalent-sphere size distribution is obtained because the technique uses the assumption of spherical particles in its optical model. Question 2: (True or False)

98. 429> LIGHT DIFFRACTION MEASUREMENT OF PARTICLE SIZE PHARMACOPEIAL FORUM VOL. 29(4) [July-Aug. 2003 It is assumed that the recorded scattering pattern of the particle ensemble is identical to the sum of the patterns from all individual single scattering particles presented in random relative positions. Question 3: (True or False)

99. FALSE A L S E S demonstrated by the work of: R. KELLY, E. STRANZL, K. ALLEN

100. <429> LIGHT DIFFRACTION MEASUREMENT OF PARTICLE SIZE PHARMACOPEIAL FORUM VOL. 29(4) [July-Aug. 2003 System Suitability The system suitability test is used to verify that the precision and accuracy are adequate for the analysis to be done. The test is based on the concept that the equipment, electronics, and analytical operations constitute an integral system that can be evaluated as such. This can be done by measuring at regular intervals a control material of known size distribution. In general, unless otherwise specified in the individual monograph, the mean values of three measurements must deviate from the established value by less than 10% for x 50 , and by less than 15% for x 10 and x 90 . Above 10 um, these maximum values are doubled. Question 4: (True or False)

101. FALSE A L S E CCORDING TO MOST ASER DIFRACTION PSA UPPORTERS MPHATICALLY!!!!!!!!!

102. FALSE A L S E CCURACY DOES NOT MATTER! Accuracy expresses the closeness of agreement between the value which is accepted either as a conventional true value or an accepted reference value and the value found. However, with particle size methods, accuracy cannot be measured because there is no “true” particle size value to compare against and image analysis data cannot be utilized as a reliable standard because the actual measurements are of different physical properties of the particles. POPULAR FALSE MANTRA

103. FALSE A L S E CCURACY DOES NOT MATTER! Different particle size results obtained through various methods are equally accurate, but each method might be expressing the results in different terms, different properties of the particles are being measured. As in the case of this study, if one is more interested in developing profiles of lot-to-lot variability, then the use of any of the available methods that yields “correct” results is appropriate; where a “correct” method is defined as one whose sample was obtained by an appropriate sampling procedure, in which the sample was prepared properly and introduced into the instrument, and in which all instrumental parameters were used correctly for the analysis. POPULAR FALSE MANTRA

106. Position Paper on Particle Sizing: Sample Preparation, Method Validation and Data Presentation Richard Bell, Andrew Dennis, Barry Hendriksen, Nigel North and John Sherwood Pharmaceutical Technology Europe November 1999 NON-POSITION PAPER

107. GET THE WORD AROUND!! Richard Kelly: AS can be seen, it is not to be expected that documents authored by representatives of the pharmaceutical industry will free themselves of all elements of the laser diffraction fairy tale any time soon. So, how is this message to be circulated….Well, one thought of mine is shown here. I had a lot of jewelry and craft supplies remaining after my experiments. Knowing that my message would not necessarily be effective in preventing lapses of memory when even those that had sat through presentations such as this one were forced to discuss particle size analysis, I thought maybe, just maybe if I made bracelets like the one shown, that individuals feeling themselves falling again under the enchantment of the laser diffraction fairy tale could look down and gather the strength to again bind themselves to the mast of reason and in this way I could prevent them from becoming well picked carcasses on the isle of the sirens of false advertising claims.

109. TRUE HEROS JJPRD MANAGEMENT FORMER PENN STATE CO-OPS KIM ALLEN JACKIE KAZANJIAN ELLIOTT STRANZL Special Thanks to: Joe Wolfgang - Sympatec Jon Hilden - Bristol Meyers Squibb Beth A. Sarsfield - Bristol Meyers Squibb Richard Kelly: At this time I want to put my hands together to applaud JJPRD management who funded the co-op program that allowed this work to be conducted over a two year period. I want to applaud the efforts of my former Penn State co-ops. And special applause to Joe Wolgang of Sympatec for directing me to Sympatec customers who were willing to allow me to run my Jewelry & Craft samples through their LD system. And of course special thanks to Jon Hilden and Beth Sarsfield of BMS who assisted me in running the Sympatec experiments. And I applaud each of you for having the patience to listen to this presentation. Thank you!!!!!

113. Particle Segmentation Your image as it comes from the microscope needs to be prepared before its features can be analyzed. The PGT Image Display itself can start the process with its easy-to-use Histogram Binary Threshold function, but what if there are other problems? The Particle Segmentation program includes features for noise removal and hole filling. And what if the particles touch? Feature Analysis would consider such particles as one feature. That's where artificial intelligence shines. The segmentation part of this program keeps &quot;in mind&quot; each step as it recursively applies shrink/split/regrow operations until there is a split. And it does so in within a few minutes, even when there are many particles. .

114. IMAGE ANALYSIS MEASUREMENTS

115. IMAGE ANALYSIS MEASUREMENTS

116. IMAGE ANALYSIS MEASUREMENT NUMBER PROBABILITY TABLE 1 SUMMARY OF IMAGE ANALYSIS RESULTS LONGEST DIMENSION BASED RESULTS SCIO 323: HCL SAMPLES: (656-42-1B, 19376-86B1) FREE BASE SAMPLES: (19376-120A, 656-32-4, 19376-96A)

117. IMAGE ANALYSIS MEASUREMENT AREA PROBABILITY TABLE 2 SUMMARY OF PARTICLE SIZE ANALYSIS RESULTS AREA % VERSUS LONGEST DIMENSION SCIO 323: HCL SAMPLES: (656-42-1B, 19376-86B1) FREE BASE SAMPLES: (19376-120A, 656-32-4, 19376-96A)

118. FALSE SURFACE AREA = VOLUME X

123. FALSE IF PARTICLE SHAPE NOT SPHERICAL, THEN x LD PSA RESULTS = CORRECT / ACCURATE

124. Image Analysis (Number Probability-microns) Sample ID 656-42-113 Pooled Data Trials 1-2 19376-86B1 19376-120A 656-32-4 Pooled Data Trials 1-2 19376-96A Mean 5.28 10.97 29.80 31.19 80.18 Std Dev. 2.35 9.78 28.42 33.83 78.54 Maximum 20.78 71.00 198.77 227.40 426.37 d15 3.14 2.90 7.96 7.30 9.82 d50 4.81 7.62 19.87 15.91 55.01 d90 8.22 23.30 64.35 83.49 196.52 Image Analysis (*Area Probability-microns) d10 4.02 10.42 26.70 31.42 92.95 d50 6.78 25.17 76.22 91.40 204.25 d90 12.24 48.71 168.57 146.66 323.98 Coulter LS 13 320 Analysis (Volume Probability-microns) d10 0.84 9.91 23.62 26.06 38.70 d50 2.70 27.66 54.40 56.99 108.60 d90 8.17 47.50 127.20 119.80 269.60

125. JJPRD LD RANDOMIZATION RESULTS R.Kelly, E. Stranzl, K. Allen Spring House, Pa. This data was collected using 0.7 centistoke silicon oil and 90% pump speed. Under turbulent flow conditions, there is only one distinct peak corresponding to the minor chord (breadth). This data was collected using 50 centistoke silicon oil and 50% pump speed. Under less turbulent flow conditions, there are two distinct peaks, corresponding to the minor chord (breadth) and the major chord (longest dimension).

127. JJPRD LD RANDOMIZATION RESULTS R.Kelly, E. Stranzl, K. Allen Spring House, Pa.

128. JJPRD LD RANDOMIZATION RESULTS R.Kelly, E. Stranzl, K. Allen Spring House, Pa. Recommended Max.  

129. JJPRD LD RANDOMIZATION RESULTS R.Kelly, E. Stranzl, K. Allen Spring House, Pa.

130. JJPRD LD RANDOMIZATION RESULTS R.Kelly, E. Stranzl, K. Allen Spring House, Pa.

131. ALTERNATE EXPLANATION: LIMIT OF DETECTION

132. JJPRD LD RANDOMIZATION RESULTS R.Kelly, E. Stranzl, K. Allen Spring House, Pa. Recommended Max.  