presentation 2010 ISSBA , 3 hours presentation

5. Pharmaceutical Development Preclinical From Drug discovery to animal testing (Toxicology) Phase I Phase II Phase III Safety Safety Efficacy Efficacy Dose Commercial Process Development Production Laboratory GMP Validation 3 Batches Traditionnal Development (Minimal Approach) Phase IV PharmacoVigilance Process Design Continuous Verification Enhanced Quality by Design Approach IND/IMPD (First in Man) (e)CTD (AMM) Qualification

6. Enhanced Quality by Design Approach Preclinical Phase I Phase II Phase III Phase IV Process Design Continuous Verification Product Target Quality Profile Potential Critical/Key Quality Attributes Process Design Potential Critical/Key Parameters Design Space Prior Knowledge Science DoE Risk Management (wc)Critical/Key Parameters Critical/Key Quality Attributes Control Strategy Qualification

7. Statistics and Modelisation Preclinical Phase I Phase II Phase III Phase IV Product Life Cycle Process Development DoE : Factorial Design Identification of Critical Parameters

8. Statistics and Modelisation Preclinical Phase I Phase II Phase III Phase IV Product Life Cycle Process Optimisation DoE : Response Surface Model Optimisation and Modelisation In silico modelisation To establish Range and Specifications

9. Statistics and Modelisation Preclinical Phase I Phase II Phase III Phase IV Product Life Cycle Process Characterisation Scale Down Validation DoE : Factorial Design Demonstration of Range and Specifications Definition of Design Space

10. Statistics and Modelisation Preclinical Phase I Phase II Phase III Phase IV Product Life Cycle Process Validation/characterisation Multivariate Analysis Demonstration of scale up and reproducibility 1 L 350 L

11. Statistics and Modelisation Preclinical Phase I Phase II Phase III Phase IV Product Life Cycle Continued Process Verification Multivariate Analysis Graph Plot

12. UMFP - formulation Preclinical Phase I Phase II Phase III Phase IV Product Life Cycle Empirical development Qualitative Justification (//Factorial) Quantitative Justification (//RSM)

13. Analytics/Quality Control Preclinical Phase I Phase II Phase III Phase IV Product Life Cycle Development Validation Qualification (?) (including Robustness by DoE)

17. Regulatory and Science 18th century 1946 1985 (De Moivre) (Placket Burman) FMEA Normal Law DoE 1944 1960 1988 Monte Carlo Simulation Bayesian Statistics (Harry) Six Sigma 2000 Neuronal Network 2005 2006 2009(?) ICH Q8 ICH Q9 FDA validation QbD In Place in LFB Training (SOP, training) To be extended Regulatory Phamaceutical development becomes a modern Science

27. Normal law : average and s.d Average : x position s.d : width Area under the curve = 1

29. Outliers 99 % 1 s.d. : 64 % 2 s.d. : 95 % 3 s.d. : 99 % …… . 6 s.d. : 99.99996% 6  defects/millions

40. Potential CQA : example

42. Tool 1 : IMPACT and UNCERTAINTY AES : Adverses Events, ATA : Anti Therapeutic Antibody No AES ATA not detected No impact on PK/PD No change None = 2 Minor AES ATA detected with minimal in vivo effect Acceptable change with no impact on PD Acceptable change Low = 4 Manageable AES ATA detected with in vivo manageable effect Moderate change with no impact on PD Moderate change Moderate = 12 Reversible AES ATA detected and confers limits on efficacy Moderate change with impact on PD Significant change High = 16 Irrevesible AES ATA detected and confers limits on safety Significant change on PK Very significant change Very High = 20 Safety Immunogenicity PK/PD Efficacy Impact (Score)

43. Tool 1 : IMPACT and UNCERTAINTY GRAS : generally recognised as safe GRAS or studied in clinical trials No impact of specific variant present at higher level in batches used in clinical trials Very low = 1 --- Variant present at same level in batches used in clinical trials Low = 2 Component used in previous process Non clinical or in vitro data with this variant, (data in vitro, non clinical or clinical from similar class Moderate = 3 --- Published external literature for variant in related molecule High = 5 No information (new impurity) No information (new variant) Very High = 7 Description (Process Raw Material) Description (variants and HCP) Uncertainty

45. Tool 2 : SEVERITY AND LIKELIHOOD Very Low – no mesurable impact 1 Low immunogenicity potential or small reduction in efficacy 3 Moderate immunogenicity or reduction in efficacy 5 Bleeding not stopped due to lower efficacy or serious immune response 7 Very High-death, microbiology related infections, hypersensitivity immune reaction 9 Severity (impact to Product Efficacy and Patient Safety) Severity score

46. Tool 2 : SEVERITY AND LIKELIHOOD Very Low or never observed 1 Low 3 Moderate 5 High 7 Very High 9 Likelyhood of severity Likelihood score

48. Tool 2 : SEVERITY AND LIKELIHOOD Very Low or never observed 1 Low 3 Moderate 5 High 7 Very High 9 Likelyhood of severity Likelihood score

50. Product Quality Risk Assesment Summary AEX 2 viral clearance IEX, Ca elution / Potential process step 10 -2.45 … / / / Tool #3 Quantification of detergent / / Detergent … … … Validation of viral clearance 49 100 Virus free a/Ag, SDS 12 36 Des-Gla variants Pooling strategy MS-HPLC 49 100 Glycosylation variants Control strategy Tool #2 Tool #1 pCQA

52. From pCQA to process development

60. FMEA : example

63. Measure : Temperature variation Useful to detect shift

64. Measure : output (yield x-bar Chart)

65. Measure : Radar/Spider Plot

66. Measure : Matrix Plot

69. Capability indices and statistics

70. Analyse : multivariate analysis Input 1 (pH) Monovariate Outp u t Input 2 (°C) multivariate Input 1 (pH) Multivariate

71. What is DoE Classical approach DoE Process understanding

72. DoE : nearly a century

73. For me, only 20 years

74. DoE - interactions - real optimum - quality of information X Interaction  Optimum

75. From simple (fractionnal) design to RSM Factorial Box Behnken Doehlert X1 X2 X3

76. From linear to Surface Response Model (non linear)

80. High collinearity : regression by least square not efficient Use of Ridge statistics allows to analyse non orthogonal Design Lack of Orthogonality is not a problem -0.2684 -2.9287 Factor 3 -0.1870 -1.5614 Factor 2 0.6741 0.01 (ridge regression) 4.2637 0.0 (classical regression) Factor 1 Ridge parameter

81. OLS Ridge Same overall topology, but completely different precision of the model (Monte Carlo simulation…)

82. DoE : number of experiments

85. Example of semi factorial design

87. 3D View of critical factors and interactions C A B A B C AB AC BC ABC

88. Identification of critical factors and interactions Identification of Critical Factors and interaction Critical factors Place for improvement Reproducibility Half Normal Plot

89. Identification of Critical Factors and interaction Pareto Chart

90. Factorial - semi factorial Design Factorial Semi factorial

92. Factorial - semi factorial Design Factorial analysis : N L Semi Factorial analysis : N (L-X) Loss of resolution (aliase)

93. But how many factors to select?

94. Recherche des paramètres critiques :

95. Analyse statistique : Are the factors selected significant ? Ttest (P95) Comparison of A low –A high 5% Comparison of B low –B high 5% Comparison of C low –C high 5% Comparison of AB low –AB high 5% … Ttest not applicable - + Factor X Anova

96. Analyse statistique : ANOVA (Table) Factors Variation degree of SS/df MS/ residual associated selected associated freedom probability

97. Demonstration of PAR

98. If a model is found significant, estimation of the impact on product quality can be studied by in silico simultation

99. Significant model found… But what is its accuracy/validity ?

102. Box – Cox Plot If heterodiasticity,  = f (   ) The transformation  = 1-  will reduce that effect :  = -1 : inverse,  = 0 : Log  = 0.5 : square root,  = 1 : no transformation

104. Accuracy/validity of the model

105. Weight of runs on the model

109. 25 + 5 mS/cm NOR Definition of Critical Factors PAR/NOR Access to C P PAR Edge of failure ?

112. Model for yield and HCP clearance

115. Optimisation of chromatographic conditions Current conditions Optimised conditions Wash 1 : 25 % A 0.2 M B in 8 % A Wash 2 : 0.5 M B Elution : 0.5 M B in 25 % A 0.75 M B in 22.5 % A Only a mathematical model, results must be controled (C in 6sigma) Yield : 68 – 85 % 85 % HCP Clearance : 2.9 – 3.1 Log 4.1 Log

116. Quality of results depends also of analytics

120. Monte Carlo theory Y = f(a,b,c)

121. Example 1 : area calculation Precision increased with number of shoots Only valid if shoots randomized

122. Example 2 : NovoNordisk

149. Design Assays from world class company

150. Graphical interpretation

151. Présentation d’un concurrent, QbD, Dusseldorf, Octobre 2008 Factorial Design RSM

161. Bayes limit : Bias/variance dilemna Best Model

162. DoE models : which is the best ?

166. QbD strongly requested by Authorities, lack of implementation may lead not only to a Dossier Assessment Refusal Report but to the discontinuation of GMP authorisation for Manufacturing of Facility