Integrated Product-Process Design to Minimize

Expense, Time, and Material

Fernando J. Muzzio Distinguished Professor

Rutgers University March 23/2017

Presented at the 3rd FDA/PQRI Conference on

Advancing Product Quality

CurrentBatchDevelopmentParadigm

Formula7onbatches~1kgFormula7on

ProcessDev.Batches~5kgProcess

Analyt.Dev.batches~5kgAnaly7calMethod

Scaledupbatches–30Kgto1000kgScaledupProcess

Robustnessbatches~5kgRobust/stableForm/Method/process

Valida7onbatches–30Kgto1000kg

Validated/verifiedProcess

Takes2-3yearsandusesthousandsofKgofmaterialsEachstepisperformedatrisk–Formula7onis“op7mized”withoutknowledgeofmanufacturability,processis“op7mized”withoutknowledgeofscaleability,etc.DoesnotuseDOEforanaly7caldevelopment–subop7malaccuracyandrobustnessDoesnotsupportRTR

Integrateddevelopmentparadigm

•  Performsproduct,process,andanaly7caldevelopmentsimultaneouslyandatscale,thusachievingrobustproductandprocessdesignwithlowerrisk

•  Takes<2months,usesmuchlessmaterial(orderofmagnitudedecrease)•  UsesDOEforAnaly7calMethoddevelopment(morereliable)

MasterDOEatmanufacturingscale–formula7onvariables,Processvariables,robustness

Formula7onDev. ProcessDev.

ProcessAnaly7calMethod,includingRTR:non-destruc7ve

transmissionspectroscopy,followedbydissolu7onandhardnesstes7ngOFTHE

SAMETABLETS

Analy7calMethodDev.

Formula7on

Advanced Pharmaceutical Development Thegoalistomodelpharmaceu7calprocessesinsilicoandusethesetoolsforop7miza7on

IntegratedProcessModel“Flowsheets”

ReducedOrderModel

Opera9ngParameters&Design

MaterialProper9es

UnitOpsModels

e.g.,Flow,BulkDensity,AngleofRepose

y = f (x,a,t,m,n)dydt

= g(x,a,t,m,n)M

M

e.g.,Feeders

min f (x)st. h(x) = 0 g(x) ≤ 0

Op9miza9on

Predic9veModeling

20

1 1

k k

i i ii i ij i ji i i j

y x x x xβ β β β ε= = <

= + + + +∑ ∑ ∑∑

5

A Strategy for Minimizing time and materials Maximizing process understanding (as defined as part of the collaboration with Janssen) •  Identify system failure modes •  Define measurements and metrics to predict

impact of failure modes for a given formulation •  Build material property data base and predictive

models for new materials and surrogates in unit ops

•  Use relevant failure mode knowledge to define DOEs and select PAT and control

•  Perform integrated formulation and process optimization

Key Concepts for CM Adoption (as defined as part of the Janssen-Rutgers-Gent Collaboration)

1. Materials (API) Characterization and Prediction (material sparing, faster development)

2. Methods for Improving API Processability (manufacturability, enabling direct compression)

3.  Process Modeling, Control, and PAT for CM RTR

4.  Rapid Development for CM (faster development, process/analytical robustness)

5. Harmonization & Standardization (regulatory clarity) 7

Blend Behavior Flowability, Lubricity, Agglomeration, Blend Uniformity, Bulk Density,

Compactability

Individual Material Behavior Cohesion, Agglomeration, Compressibility, Bulk Density , Compactability

Particle Properties Superficial

Surface Energy, Piezoelectric,

Tribo-electrification

Mechanical Elasticity, Plasticity,

Bonding Mechanism

Crystallographical and Thermal

Form, Crystallinity, Tm, Tg, etc

Geometrical PSD, Shape, Crystal Habit, Surface Area, true density.

Particle Properties and Powder Behavior (determined for specific list of API’s and excipients as part of the Collaboration with Janssen)

-0.4 -0.3 -0.2 -0.1 0.0 0.1 0.2 0.3 0.4

-0.3

-0.2

-0.1

0.0

0.1

0.2

0.3

cBD, g/ml

CPS, % @ 1 5.0kPa

D1 0D50

D90

BFE, mJ

SIFRI

SE, mJ/g

Cohes ion 3kPaCohesion 6k Pa

Cohe sion 9kPaCohe sion 15k Pa

UYS 3kPaUYS 6kPa

UYS 9k Pa

UYS 15 kPa

M PS 3 kPa

MPS 6kPa

MPS 9 kPa

MPS 15k Pa

FFC 3kPaFFC 6k Pa

FFC 9 kPa

FFC 15k Pa

AIF, º @3kPaAIF @6kPa

AI F, º @ 9kPa

AI F, º @15kPa

PC 1 (37%)

PC

2 (1

9%)

Loadings Plot

10

The following materials have similar flow properties :

-  APAP -  API 2 -  API 3 -  API 4

Use of Material Library to Predict Feeder Behavior (as part of the Collaboration with Janssen)

Use similar geometry, screw type, etc.

Score plot of material library with 45 materials

Predicting feeding performance from material flow properties (as defined as part of the Collaboration with Janssen)

11

•  For a given new material, can we compare it to existing materials in the library?

•  Once a new material is included in the material library, can we

predict its feeder performance? •  Can we predict the optimal screw choice for a given new

material?

Similarity scores of the new material

12

0

1

2

3

4

5

6

7

8

Fine

zeolite

Fine

alumina

Sa7n

tone

Zeolite

LactoseMon

ohydrate

Coarsealumina

WeightedEuclideanDistance

Material similarity can be quantified by calculating weighted Euclidean distance. Smaller distance corresponds to higher similarity.

Receivenewmaterial

Characterizenewmaterialproper7es

PerformPCAwithexis7ngmaterials

CalculateweightedEuclideandistance

Ranksimilaritybetweenmaterials

Predicttheop7malfeederscrewbasedsimilarmaterials

Prediction using similarity scores

13

0

0.005

0.01

0.015

0.02

0.025

0.03

0.035

0.04

0.045

0.05

0 1 2 3

RSDorRDM

Screwtype

FinezeoliteRDM

MaterialARDM

FinezeoliteRSD

MaterialARSD

screw type 1 fine concave screw

screw type 2 fine auger screw

screw type 3 coarse concave screw

Prediction using PLS regression

14

X Library of material

properties

X’ Scores of the

principal components

Y

PCA projection

Partial least squares

regression

Alternatively, when material with matching flow properties is difficult to find, a partial least squares (PLS) regression can be used. A PLS regression model relates material flow properties directly to feeder performance, quantified by RSD and RDM.

Predicting feeding performance

15

y=0.994xR²=0.977P<0.05

RMSECV=0.0078RMSEC=0.00150

0.02

0.04

0.06

0.08

0.1

0.12

0 0.05 0.1 0.15

Pred

ictedRSD

MeasuredRSD

FineConcaveScrew

y=0.999xR²=0.994P<0.05

RMSECV=0.0099RMSEC=0.0024

0

0.02

0.04

0.06

0.08

0.1

0.12

0 0.05 0.1 0.15

Pred

ictedRSD

MeasuredRSD

FineAugerScrew

y=0.995xR²=0.978P<0.05

RMSECV=0.0099RMSEC=0.00180

0.02

0.04

0.06

0.08

0.1

0.12

0 0.05 0.1 0.15

Pred

ictedRSD

MeasuredRSD

CoarseConcaveScrew

•  PLS regression helps to answer: 1. For a new material with given properties, can we predict RSD or RDM for a certain screw? 2. For a new material, what is the optimal screw selection?

16

0

0.005

0.01

0.015

0.02

0.025

0.03

0.035

0.04

0.045

0 1 2 3

RSDorRDM

ScrewType

MeasuredRSD

PredictedRSD

MeasuredRDM

PredictedRDM

Receivenewmaterial

Characterizenewmaterialproper7es

ProjectdatatoPLSRmodelstopredictfeeder

performance

Selectop7malscrewbasedonmodelpredic7on

Runconfirmatoryfeederexperiment

Addconfirmatorydataandupdatemodelknowledge

Prediction using PLSR models

screw type 1 fine concave screw

screw type 2 fine auger screw

screw type 3 coarse concave screw

Predicting feed factor from material properties

17

y=-0.9761x+9.1761R²=0.93556

y=-1.0269x+10.124R²=0.91921

y=-1.9553x+18.606R²=0.97741

0

5

10

15

20

25

30

35

-8 -6 -4 -2 0 2 4 6

Ini7alfe

edfactorkg/hr

PC1

Fineconcavescrew

Fineaugerscrew

Coarseconcavescrew

0

5

10

15

20

25

30

35

Coarsealumina

Fine

alumina

Fine

Zeo

lite

Sa7n

tone

Zeolite

Lactose

MaterialA

Ini7alfe

edfactorkg/hr

Materials

Fineconcavescrew

Fineaugerscrew

Coarseconcavescrew

•  Initial feed factor reflects maximal feeding capacity for a material. •  Results show that using scores of the first principal component, the

initial feed factor can be predicted based on the linear correlation. •  The feed factor using different screws can also be predicted.

M

M

EX

M

M

API

M

M

Cri9calQues9onsinDCCM(as defined as part of the Collaboration with Janssen)

Q1:Canwefeedeachingredientattherequiredflowrate?

Q2:Dotheingredientss9ckand/oragglomerate?

Q3:Canweachieveblendhomogeneity?

Q5:Areblendflowproper9esgoodenoughtosupportWeightUniformity?

Q4:Doestheblends9ckoragglomerate?

Q6,Q7:Canwemeettabletdissolu9on(Q6)andhardness(Q7)atareasonableflowrate?

Lub

Improving API processability Minimizing amounts of API needed in development

Strategy (as defined as part of the Collaboration with Janssen)

19

FindDesignSpace

FindFailureSurfaceEnvelope

TabletPress

Lubricant

Feeders

M M

M

API

M

M

Mill

Proper9es

Blender

Density

Agglomera7on,Cohesion(surfaceenergy,PSD)

PoorMixing/segrega7on(contentUniformity)

FailtocomplyHardness

Cohesion,Adhesion

DryCoa7ng,

20

M M

EXM

EX

FailureMode Metrics

FlowRatevariability

FlowRateSetPointdivergence

TapDensityBulkDensityPermeability

ShearCellElectrosta7cs

Bonding,Plas9c,Elas9cFailtocomplyDissolu7on

Chocking,jamming,discon7nuousflow

SurfaceEnergyPSD

IGC-Droplet

LaserDiffract.

SurfaceEnergyOrContactAngle

IGC-Droplet

SurfaceEnergyPSD

FimngCompac7onModel

UnitOp Inputs ProcessingParameters Responses OutputMaterialProper9es

Feeder(i)

RMCohesion(i)[Coh1i]RPMImpeller(RPM1)

FeederFlowRate(FFR)=f[Coh1,Cps1,SE1,E1,RPM1,

RPM2,FR,HSG,ST]Cohesion2(Coh2)=f[Coh1,Cps1,SE1,

E1,RPM1,RPM2,FR,HSG,ST]RMCompressibility(i)[Cps1i]

RMSurfaceEnergy(i)[SE1i] RPMScrew(RPM2)

FeederFlowRateVariability(σFFR)=f[Coh1,Cps1,SE1,E1,RPM1,RPM2,FR,HSG,ST]

RMElectrosta9c(i)[E1i]PowderBulkDensity2(DB2)=f[Coh1,Cps1,SE1,E1,RPM1,RPM2,FR,HSG,ST]

RefillRate(RR)RMPSD(i)[PSD1i]

RMBulkDensity(i)[BD1i]HopperSize/geometry(HSG)

Screwtype(ST)

Mill

Coh2(i) RPMBlade(RPM3)

MillHoldup BlendHomogeneity3(BH3)=f[Coh2,σFFR2,RPM3,FR5,C

%,MSSG,ST]σFFR2(i)

Composi9on(PSD,(i)[C%] MillScreen(MS)Agglomera9on3(Ag3)=f[Coh2,σFFR2,

RPM3,FR5,HSG,ST,C%]BulkDensity2(i)[BD2i]Hip:Densityhasnoeffectbeyondwhat’scapturedby

cohesion

Screwtype(ST)SpacersGeometry(SG)

Cohesion3(Coh3)=f[Coh2,σFFR2,RPM3,FR5,HSG,ST,C%]

Density3(D3)=f[Coh2,σFFR2,RPM3,FR5,HSG,ST,C%]

Blender

BlendHomogeneity3[BH3] RPMBlade(RPM4)

Holdup(inves7gatecomposi7oninblender)

Lubricity4(L4)=f[FR5,Ag3,RPM4,BG,STBP,C%]

Agglomera7on3[Ag3] BlenderResidenceTime(BRT) Compa9bility(Cpt4)=f[L4,FR5,Ag3,RPM4,BG,STBP,C%,PSD3(i)]Composi9on[C%] BlenderGeometry(BG) DispersionCoefficient(BDC)

Cohesion3[Coh3]

Screwtype/Bladeparern(STBP)

Cohesion4(Coh4)=f[Coh3,L4,RPM4,FR5,Ag3,HSG,ST,C%]

BladePasses(BBP)BulkDensity3[BD3]

Agglomera9on4(Ag4)=f[RPM4,FR5,Ag3,HSG,ST,C%]

MillHoldup

BlendHomogeneity4(BH4)=f[BH3,RPM4,FR5,Ag3,HSG,ST,

C%]BulkDensity4

(BD4)=f[BD3,L4,RPM4,FR5,Coh3,HSG,ST,C%]

Feeder,Mill,Blender:InputsandOutputs(aspartlydefinedaspartoftheJanssen-Rutgers-GentCollabora7on)

UnitOp Inputs ProcessingParameters Responses ProductProper9es

TabletPressandFeedFrame

Lubricity[L4]RPMFeedFrame(RPM5)

Compac7onForce(CF)TabletThickness(TH5)=f[FR,L4,RPM5,

CH,ThG,FC,C%,FFG,TT,#S,PC]Compac7bility[Cpt4] Ejec7onForce(EF)

RPMTurret(RPM6=FlowRate(FR))

DwellTime(DT)WeightVariability(WV5)=f[L4,FR,Ag4,RPM5,Coh4,BH4,BD4,CH,C%,FFG,TT,

#S,PC]ChuteHeight(CH)Composi9on[C%]

TabletDensity(porosity)5(TD5)=f[Coh4,L4,RPM4,FR,Ag3,PSD(i),C%]

ThicknessGap(ThG)

Cohesion4[Coh4] FillCam[FC] ContentUniformity5(CU5)=f[BH4,

RPM5,Ag4,WV5,C%,FFG,TT,#S,PC]FeedFrameGeometry(FFG)Agglomera7on4[Ag4] TabletTooling(TT)

Hardness5(H5)=f[Cpt4,RPM5,FR,L4,WV5,C%,FFG,TT,#S,PC]

BlendHomogeneity4[BH4]#sta7ons(#S)

BulkDensity4[BD4] Dissolu9on5(Diss5)=[Cpt4,RPM5,FR,

L4,WV5,C%,FFG,TT,#S,PC]PreCompression(PC)

FeedFrameandTabletPress:InputsandOutputs(aspartlydefinedaspartoftheCollabora7onwithJanssen)

Process Modeling and Control

Feedback control

Feedforward control

Flowsheet model

General RTR Sensing Approach

TabletPress

Blender

Feeders

mill

API

M

M

M

M

M

M

M

M

Content/DensityBlendUniformity

NIR,Raman

LT

Thickness

Density

US

Hardness

NIR

Dissolu9on(check)

Feedforwardcontrol

Force

Weight

CompressionGap

Cross-check

Content(check)

Goals accomplished

APAP:•  Dissolu7onModelbasedonNIRspectraoftablets •  HardnessModelbasedonultrasoundanddensitytodeterminemassandthickness

•  RTRmethodologystrategysuccessfullydesignedandvalidated

25

General Dissolution Prediction Methodology

Definetargetcondi7ons

0

50

100

150

0 20406080100120

Drugrelease%

Time/min

y = 0.9833x R² = 0.86308

0

50

100

150

200

250

300

0 50 100 150 200 250 300

α p

red

icte

d

α reference

Reference vs predicted

020406080

100

0 20 40 60 80100120%DrugDissolved

Time(min)

referencepredic9onf2=79.13

Iden7fydissolu7onmechanism

Prediction of Individual Dissolution Profiles: Model-dependent Approach

27Reference:dissolu+onprofilesPredicted:NIRPCs

Dissolution Prediction Flowchart

Integrateddevelopmentparadigm

•  Performsproduct,process,andanaly7caldevelopmentsimultaneouslyandatscale,thusachievingrobustproductandprocessdesignwithlowerrisk

•  Takes<2months,usesmuchlessmaterial(orderofmagnitudedecrease)•  UsesDOEforAnaly7calMethoddevelopment(morereliable)

MasterDOEatmanufacturingscale–formula7onvariables,Processvariables,robustness

Formula7onDev. ProcessDev.

ProcessAnaly7calMethod,includingRTR:non-destruc7ve

transmissionspectroscopy,followedbydissolu7onandhardnesstes7ngOFTHE

SAMETABLETS

Analy7calMethodDev.

Formula7on

MaterialProper9es-ProcessParametersInterrela9on(asdefinedaspartoftheCollabora7onwithJanssen)

30

LubricatedBlendRawMaterial

Proper9es(RMP)

Cohesion1.CompressibilityElectrosta9csSurfaceEnergy

PSD

ProcessParameters

RPM2(Mill)

ScreenSize

ProcessParameters

RPM3(Blender)

IntermediateMaterialProper9es(IMP)

Lubricity4Compactability4.

Cohesion4Agglomera9on4

BlendHomogeneity4BulkDensity4

M

M

API

M

M

Lubricant

Feeders

Mill

ProcessParameters

RPM5(FeedFrame)

RPM6(flowRate)

ChuteHeight(CH)

ThicknessGap(ThG)

PreCompac7on(PC)

FillCam[FC]

ProductMaterialProper9es(RMP)

ThicknessWeightVariabilityDensity(porosity)ContentUniformity

HardnessDissolu9on

Product

M

EXM

EX

Conclusions•  Integratedproduct/process/analy7caldevelopment:

–  Faster–  Lessmaterial–  Bererprocess–  Bererformula7on–  Bereranaly7cs

•  RTRrequiresRTQA•  Level3ProcesscontrolrequiresRTQA•  Typically,RTQArequiresdissolu7onprodic7on

31

Wayforward

1.  Standardiza7onofMaterialPropertytes7ng2.  Performancestandardsforequipment3.  Standardiza7onoftracerselec7onforRTDstudies4. Model-basedacceleratedprocessdevelopment5.  Standardiza7onofBHPATverifica7on6.  Sta7s7calmethodsforacceptanceandreleasedecisions7.  Failuremodetaxonomy–standardizedtes7ng8.  GeneralmethodologyforRTR–dissolu7onpredic7on

32

ENGINEERING RESEARCH CENTER FOR

STRUCTURED ORGANIC PARTICULATE SYSTEMS