Determination of Design Space for Oral Pharmaceutical Drugs Kalliopi Chatzizaharia and Dimitris Hatziavramidis School of Chemical Engineering National Technical University of Athens, Greece
Design Space Quality-by-Design initiative by pharmaceutical regulatory agencies of the European Union, Japan and USA enables implementation of changes in formulation and manufacturing processes to manufacture a pharmaceutical product within a multidimensional Design Space, proposed by the manufacturer and having been demonstrated to assure improved process capability and product quality, without the need for further regulatory approval Product Quality Critical Quality Attributes: y = (y 1,,y n ) T Critical Material Attributes (of drug components) & Critical Process Parameters: x = (x 1,,x n ) T
Design Space DS = {x ϵ X E[ y j x] = ŷ j x ϵ L j j = 1,..,m^ L 1 U U L m = L} L: set of specifications for y j Control Space Design Space Distance related to Cpk Knowledge Space
Oral Drugs Drug Quality: efficacy, safety, reliability Efficacy: drug stability, bioavailability membrane permeation, absorption, dissolution rates therapeutic polymorph (crystallization conditions) concentration of A(ctive) P(harmaceutical) I(ngredient) in blood, c db concentration of excipients (disintegrants) in G(astro) I(ntestinal) T(ract), c eg Safety: c dl < c db < c du ; c db < c dl no therapeutic effect c dl < c db toxic effect Reliability: it does what it is supposed to do, according to its label powder blend (API+excipients) composition remains the same throughout processes powder flowability, tablet strength & friability concentration of excipients (glidants, lubricants, binders, etc.), grain size
Bioavailability & Bioequivalence Bioavailability (BA): the rate and extent to which the API is absorbed from a drug product and becomes available at the site of action BA depends on: API release from oral form, balance among dissolution, elimination, metabolism and absorption rates, solubility and membrane permeability Bioequivalence (BE): property of drugs which have similar plasma concentration profiles Biopharmaceutical Classification System: identifies drug classes for which in vivo BE measurements are waived in favor of in vitrodissolution measurements Whenever a waiver can be granted, two drugs are assumed to be bioequivalent if their dissolution profiles are similar EMEA & FDA: Any methods to prove similarity of dissolution profiles are accepted as long as they are justified
Design Space Determination Methods Ignoring data uncertainty: Response Surface (RS) Assumptions: Errors statistically independent & normally distributed Overlapping responses Optimized responses Accounting for data uncertainty & correlation structure: Bayesian Posterior Predictive Approach (BPPA) Assumptions: 1. Prior distribution honestly reflects genuine information 2. Any uncertain quantity is referred to as a random variable p( θ y) p(θ): probability of the parameters θ from prior times p(y θ) = L(θ,y): likelihood of θ parameters for fixed data y p(θ y): posterior probability = p( y θ) p( θ) p( y) = θ p( y θ) p( y θ) p( θ) p( θ)dθ
Design Space Determination Methods Desirability function (d μi ) [(y μi - y μi min )/(T μi - y μi min )] a, y μi min y μi T μi d μi = [(y μi - y μi max )/(T μi - y μi max )] b, T μi min y μi y μi max 0, y μi < y μi min or y μi > y μi max y μi : estimated mean response; y μi min,y μi max, T μi : minimum, maximum desired limits and target for y μi respectively; and a, b are input parameters that determine the shape of the reliability function Composite desirability (D): geometric mean of p individual d μi D = (d μ1 d μ2 d μp ) 1/p 0 d μi and D 1
Three-stage tableting process P(owder) W(et) Drying T(ableting) G(ranulation) G(ranule) Westerhuis, J.A. et al. (1997) Multivariate modelling of the tablet manufacturing process with wet granulation for tablet optimization and in-process control. Int. J. Pharm. 156(1), 109 117 Box-Behnken design: corners and midpoints of edges of a cube Design factors - CPPs & CMAs: water amount for wet granulation - Water (ml), granulation time - Time (min), moisture of granules Moisture (%), compression force - CompF (kn); excipients: HPC (%) & MCC (%) Response variables - CQAs: tablets crushing strength (CS), disintegration time (DT) and ejection force (EF)
Regression Model logcs= 0.215*MCC+0.219*Moisture+0.102*CompF+0.006*Water+ 2.834*(Water) 2-0.001(CompF) 2-0.008*Moisture*CompF-4.691e- 04*MCC*Water-8.959 logdt= 0.190*MCC+0.095*HPC-1.280*Moisture+0.074*CompF+0.004*Water+ 0.184*(Moisture) 2 +3.070e-05*(Water) 2-4.442e-04*MCC*Water-5.475 EF shows no significant terms CQAs Data Range Target [Specifications] CS (N) 4.00-75.00 66.07 [30.90-75.86] DT (s) 2.00-421.00 288.40 [69.18-416.87] EF (N) 41.00-358.00 274.00 [218.55-329.45]
Three-stage tableting Determination of Design Space Overlapping Responses Optimized Responses Optimal D High Cur 0.99955 Low MCC HPC Water Time Moisture CompF 90.0 5.0 650.0 7.0 5.40 30.0 [68.5120] [2.0] [414.2345] [3.2828] [4.9863] [30.0] 65.0 2.0 400.0 3.0 2.80 10.0 Composite Desirability 0.99955 logcs Targ: 1.820 y = 1.8200 d = 1.0000 logdt Targ: 2.460 y = 2.4600 d = 1.0000 EF Targ: 274.0 y = 273.9254 d = 0.99865
Design Space determination by Optimized Responses method CPPs- CMAs values Response Surface Analysis Bayesian Method MCC -optimum 68.51 - HPC -optimum 2.00 - Water -optimum 414.24 - Time -optimum 3.28 - Moisture -optimum 4.98 - CompF -optimum 10.00 - MCC l 77.94 75.00 HPC l 5.00 5.00 Water l 650.00 400.00 Time l 3.00 5.00 Moisture l 2.80 3.10 CompF l 15.15 30.00 MCC u 69.28 65.00 HPC u 2.00 3.00 Water u 400.00 450.00 Time u 3.00 3.00 Moisture u 5.20 4.20 CompF u 30.00 30.00
Desirability of Methods for Design Space determination Specifications CS (N) DT (s) EF (N) Target 66.07 316.23 274.00 yl 30.90 69.18 218.55 yu 75.86 416.87 329.45 Response Surface Design Analysis CS DT EF 0.85 Target 66.07 288.40 273.93 yl 41.69 69.18 218.55 yu 75.86 416.87 293.19 Bayesian Method CS DT EF 0.74 yl 57.54 234.42 269.15 yu 72.44 416.87 281.84 Method's Composite Desirability
In vitro Dissolution of an oral drug Experimental data on in vitro dissolution of an oral drug by ELPEN (Greek Pharmaceutical Company) Tablet mass = 312 mg Formulation: API and 3 excipients: main x 1 and x 2 ; x 3 : 2-5% w/w Comparison of dissolution profiles Dissolution kinetics models: zero and first order kinetics, Hixson Crowell, Weibull, Higuchi, Baker Lonsdale, Korsmeyer Peppas and Hopfenberg Model-independent comparison of dissolution profiles Ratio tests: ratios of parameters (e.g., %drug dissolved, AUC, mean dissolution time) from release assays of the reference and test products at the same time Pair-wise procedures: Rescigno index (ξ i ), difference factor (f 1 ) and similarity factor (f 2 )
Comparison of Dissolution profiles Dissolution efficiency (DE) ratio of area under the dissolution curve up to a testing time point to the area of the rectangle that describes 100% dissolution up to the same time point DDDD = 100 d: function of % drug dissolved at time t Dissolution Area Difference (DAD) factor d exp : dissolution function of each run; d ref : dissolution function of reference drug dissolved throughout time t min DAD best similarity of the dissolution profiles Calculation method: multiple segment trapezoidal rule (steep changes in dissolution-vs-time curves) tt dd dddd 0 dd 100 tt DDDDDD = 1 ( tt dd eeeeee dddd 0 tt 0 dd rrrrrr dddd )
Comparison of Dissolution profiles Similarity factor (f 2 ) NN 1/2 ff 2 = 50 log 1 + (1/NN) (xx tttt xx ) 2 rrrr 100 ii=1 N: number of time points; xx tttt : mean % drug dissolved of test product : mean % drug dissolved of reference drug xx rrrr 50% f 2 100%: similar dissolution profiles Drug dissolved within 15 min > 85% similar dissolution profiles without further mathematical evaluation
Mixture Design for experimental data Mixture Design of a generic oral tablet formulation o Design factors: mixture components proportions o For a 3-component mixture: 0 x i 1 and x i = 1, i = 1, 2, 3 o Simplex Centroid design points: vertex points (x i = 1), centroid (x i s equal) and axial points (x i =0) Two replicates of the center point (process variability and curvature evaluation) Response variables - CQAs: tablet weight, hardness and dissolution profile
Reference - Generic drug Dissolution profiles Comparison 100 90 %dissolved 80 70 60 50 40 30 20 10 0 0 10 20 30 40 50 60 t reference drug run2 run4 run5 run1 run6 run3 Reference dissolution profile: first order kinetics equation d(t)=a + b e kt d: drug percent dissolved at time t; a=91.05, b=-91.29, k=-0.57, R 2 = 0.987
Reference - Generic drug Dissolution profiles Comparison Similarity factor (f 2 ) and Dissolution Area Difference (DAD) Experimental Runs DAD f2_15 (%) f2_60 (%) run 1 0.391 66 69 run 2 0.043 78 79 run 3 0.185 54 57 run 4 0.236 33 40 run 5 0.083 60 67 run 6 0.049 62 67
Mixture Regression Model weight = 314.29x 1 +309.06x 2 +315.64x 3-2.81x 1 x 2 hardness = 17x 1 +10.73x 2 +359.87x 3 +588.22x 1 x 2 f2_15 = -296.78x 1-195.01x 2 +753.42x 3 +1076.85x 1 x 2 f2_60 =-208.94x 1-110.04x 2 +646.15x 3 +755.41x 1 x 2 CQAs Data Range Target [Specifications] weight (mg) 310.15-312.26 312 [304.20-319.80] hardness (N) 154.400-169.40 170 [165.75-174.25] f2_15 33 78 100 [50-100] f2_60 40 79 100 [50-100]
Design Space determination Overlapping responses Optimized responses
Design Space determination Component values Mixture Design Analysis Bayesian Method x1-optimum 38.66 - x2-optimum 38.66 - x3-optimum 9.67 - x1-l 43.33 50 x2-l 36.67 30 x3-l 7.00 7 x1-u 37.67 35 x2-u 39.67 45 x3-u 9.66 7
Composite Desirability for Design Space determination methods Specifications weight (mg) hardness (N) f 2_15 (%) f 2_60 (%) Target 312.00 170.00 100.00 100.00 yl 304.00 165.75 50.00 50.00 yu 320.00 174.25 100.00 100.00 Mixture Design Analysis weight hardness f 2_15 f 2_60 0.64 Target 311.56 168.50 77.85 79.24 yl 311.60 165.42 56.66 60.12 yu 311.50 168.35 78.89 80.29 Bayesian Method weight hardness f 2_15 f 2_60 0.57 yl 311.90 168.30 50.44 63.11 yu 312.30 169.20 77.71 79.00 Method's Composite Desirability
Conclusions DS determination: multi-response optimization and overlapping responses Multi-response optimization: DS bounds in neighborhood of optimum conditions Overlapping responses: DS bounds from the global lower and upper specification limits Method effectiveness measure: Composite Desirability Best performance: Experimental Design analysis - proposed when the data are complete and there is little uncertainty Bayesian approach: proposed if data involve high uncertainty and correlation structures Bioequivalence introduced in Experimental Design and DS determination as a CQA with f 2 values
Acknowledgement I would like to acknowledge financial support in the form of a scholarship from the National Technical University of Athens
Thank you for your attention!