PAT and Quality by Design

- A Process Systems Engineering View

Julian Morris1, David Lovett2 and Zengping Chen3
1Centre

for Process Analytics and Control Technology

School of Chemical Engineering & Advanced Materials, Newcastle University UK
2Perceptive

Engineering, Daresbury Innovation Centre, Daresbury, Cheshire UK

3State

Key Laboratory of Chemo/Biosensing and Chemometrics,

Hunan University China
IChemE APC9, York UK 19th / 20th September 2011

Agenda

Overview
z

Process Analytical Technologies:

z

The FDA Process Analytical Technologies (PAT) and Process

Systems Engineering

Challenges in using spectroscopic data in PAT-based process

control and Real Time Release

Application Studies
z

Batch Cooling Crystallisation from Lab to Industrial Pilot scale

Batch Endpoint Control using Process Analytical Data

Closure

The EU provides 32% of the worlds chemicals manufacturing

through some 25,000 enterprises of which 98% are SMEs which
account for 45% of the sectors added value, and 46% of all
employees are in SME

What does PAT, QbD & Real-Time-Release mean to an SME?

Benchmarks for Pharmaceuticals Companies

PHARMACEUTICAL BENCHMARKS
KPIs

Present
Pharmaceutical
Industry

A Winning
Pharmaceuticals
Plant

A World Class
Pharmaceuticals
Manufacturing Plant

Stock Turn

3 to 5

14

50

OTIF

60% to 80%

97.4%

99.6%

RFT

85% to 95%

96%

99.4%

CpK

1 to 2

3.5

3.2

OEE

30%

74%

92%

Stock Turn - this is the total turnover on the site at manufacturing price divided by all the stocks on the site on the
same basis. Stocks include finished goods, work in progress, and purchased raw materials; On Time in Full
(OTIF) delivery - this is the percentage of orders that are satisfied on time in full with zero defects; Right First
Time (RFT) - this is the percentage of the products that at the point of manufacture are delivered right first time
with no defects; CpK - is a statistical process measure on the variability of the product; Overall Equipment
Effectiveness (OEE) - this measures how effectively the manufacturing equipment is used.
Benson R.S, From World Class Research to World Class Manufacturing: the Challenges,
Pharmaceutical Eng. Sept/Oct 2005

Where are we in
Process Analytics and Control Technologies?

Inspection Observations (1)

Courtesy - Des Makohon Senior GMP Inspector,

PharmaIQ, London January 2010

Inspection Observations (2)

Courtesy - Des Makohon Senior GMP Inspector,

PharmaIQ, London January 2010

Consequences of Poor Development

Courtesy - Des Makohon Senior GMP Inspector,

PharmaIQ, London January 2010

Impurities and Polymorphism (Where we were in 1998)

The RITONOVIR aids drug

changed from anhydrous to
hydrate crystal after launch:

Lower solubility and hence bioavailability.

Product was withdrawn for a

year and reformulated.

New FDA approval needed

mega cost implication !
As product purity improved
during process chemistry workup, the stable polymorphic form
changed

Closing the Analytical Control Loop

Some Challenges in using PAT-based
Sensors in Real Time Process Control

Through - Batch NIR Profile

Start of Cooling

Addition of Raw Materials

Main Addition

Some On-line Process Analytics Challenges

Product Quality across lab, pilot & production scales

Spectra from different probes are distinct Interprobe variability is the greatest source

(Staffan Folestad AstraZeneca, APACT09)

Sensing space direction

PCA of spectra collected for over 1 hr

20

3
2

10

t[2]

Agitator

-10

P1-2343.0
P1-2225.0
P1-2401.0
P1-2303.0
P1-2322.0
P1P1-2629.0
P1-2552.0
P1-2420.0
P1-2648.0
P1-2534.0
P1-2710.0
P1-2746.0
P1-2805.0
P1-2611.0
P1-3341.0
P1-3014.0
P1-3208.0
P1-2728.0
P1-2823.0
P1-2956.0
P1-2900.0
P1-2842.0
P1-3400.0
P1-2919.0
P1-3150.0
P1-3437.0
P1-3708.0
P1-3418.0
P1-3532.0
P1-3227.0
P1-3245.0
P1-3514.0
P1-3131.0
P1-3304.0
P1-3455.0
P1-3840.0
P1-3859.0
P1-3726.0
P1-3744.0
P1-3803.0
P1-4225.0
P1-4646.0
P1-4015.0
P1-4244.0
P1-4705.0
P1-4628.0
P1-4819.0
P1-4742.0
P1-4609.0
P1-4532.0
Position
1
P1-4551.0
P1-4837.0
P1-4514.0
P1-4800.0
P1-4723.0

P2
P2-5649.0 P3
P2-5919.0 P3-3002.0
P2-5507.0
P3-2748.0
P2-5805.0
P2-5420.0
P2-5629.0
P2-5937.0
P2-5610.0
P2-0227.0 P3-3059.0
P2-5440.0
P2-5900.0
P2-5537.0
P 2 P3-3446.0
P3-2651.0
P3-2844.0
P2-0150.0
P2-0208.0
P2-0744.0
P3-5213.0
P2-0858.0
P3-3020.0
P2-0304.0
P3-2806.0
P3-3233.0
P2-0803.0
P2-0033.0
P3-3137.0
P3-2710.0
P2-5708.0
P2-0839.0
P3-2940.0

P1

P3-4941.0
P2-0707.0
P3-2921.0
P2-0322.0
P2-0534.0
P3-2902.0
P3-2729.0
P2-0418.0
P2-0649.0
P2-5400.0
P2-5746.0
P3-3039.0
P2-0359.0
P2-0916.0
P3-4050.0
P3-4401.0
P2-5727.0
P2-0245.0
P2-0821.0
P2-0553.0
P3-3350.0
P2-0725.0
P2-1129.0
P3-3505.0
P2-1206.0
P3-3603.0
P3-5038.0
P3-3155.0
P2-0612.0
P2-1748.0
P2-1843.0
P3-3816.0
P3-4109.0
P3-4730.0
P3-5135.0
P2-1920.0
P3-4535.0
P3-4845.0
P3-3118.0
P3-3621.0
P2-0437.0
P2-0630.0
P3-3408.0
P2-1939.0
P3-4631.0
P3-5232.0
P2-1727.0
P3-4554.0
P3-4420.0
P3-4438.0
P3-4242.0
P3-4127.0
P3-4612.0
P2-1825.0
P3-4321.0
P3-3427.0
P3-5154.0
P3-4301.0
P3-5000.0
P2-2053.0
P3-3544.0
P2-1440.0
P2-1147.0
P2-2148.0
P3-4340.0
P2-2111.0
P2-1650.0
P2-1708.0
P3-4650.0
P3-4708.0
P2-2400.0
P3-3835.0
P3-4922.0
P2-2323.0
P3-3640.0
P2-1421.0
P2-2342.0
P3-5117.0
P3-5019.0
P2-1806.0
P2-2244.0
P3-5058.0
P3-4457.0
P2-1902.0
P3-4516.0
P3-3701.0
P2-2129.0
P2-2034.0
P2-2305.0
P2-1957.0
P2-2225.0
P2-2015.0
P2-2206.0
P4-5854.0
P4-5835.0
P4-0007.0
P4-5912.0
P4-0026.0
P4-0044.0
P4-5949.0
P4-0140.0
P4-0158.0
P4-5931.0
P4-0238.0
P4-0220.0
P4-1217.0
P4-1236.0
P4-1254.0
P4-1621.0
P4-2330.0
P4-1658.0
P4-2006.0
P4-1603.0
P4-2619.0
P4-2849.0
P4-1313.0
P4-1544.0
P4-2157.0
P4-1849.0
P4-2542.0
P4-1409.0
P4-2253.0
P4-1427.0
P4-2446.0
P4-2926.0
P4-1331.0
P4-2216.0
P4-1350.0
P4-2348.0
P4-2409.0
P4-2311.0
P4-2504.0
P4-2523.0
P4-1640.0
P4-2234.0
P4-2945.0
P4-1831.0
P4-2908.0
P4-3040.0
P4-2600.0

P4

P4-2427.0

-20
-100 -80 -60 -40 -20 0

t[1]

20 40 60 80 100

Courtesy Ronan OKennedy (GSK), Payal

Roychoudhury, Linda M. Harvey & Brian McNeil
(University of Strathclyde)

Incorporating PAT Sensors into Batch

Cooling Crystallisation Closed Loop
Process Control

The Principles of a Process Systems Base Approach

Courtesy Staffan Folestad AstraZeneca, APACT09

In-Process Analytics & Process Control

FTIR
FTIR

Growth
kinetics

USS
USS

Supersaturation

Size

Video
Video
microscopy
microscopy

Leeds, Heriot-Watt and Newcastle University

Nucleation
kinetics
UVvis
UVvis
MSZW

Shape
Reactant
rheology

LDA/PIV
LDA/PIV

CBB#2 Project: collaboration with

Batch
process
Process
monitoring
& control

Polymorphicform

XRD
XRD

Process
conditions

Mixing &
scale-up

CFD
CFD

Heat
transfer

Reaction
Reaction
Calorimetry
Calorimetry

Partners
AEA Technology
AstraZeneca
Bede Scientific Instruments
BNFL
Clairet Scientific
DTI
EPSRC
GlaxoSmithKline
HEL
Malvern Instruments
Pfizer
Syngenta

Fluctuations in External Variables on Calibration Models

In process analytical applications, spectral measurements can be subject to

changes in process temperature, flow turbulence, compactness, and other
external variations.

Typically, variations of external variables influence spectral data in a nonlinear manner which leads to the poor predictive ability of bilinear
calibration models on raw spectral data.

The influence of external variables on spectral data we classify into two

different modes:
z multiplicative influential mode, and
z composition-related influential mode

A new chemometric method, Extended Loading Space Standardization

(ELSS), has been developed to explicitly model these two kinds of
influential modes.

Spectral Calibration Issues

Spectroscopic measurements in chemical and pharmaceutical processes

are always liable to fluctuations in both control and external process
variables.

This can result in noisy spectra, non-linear shifts, broadening in spectral

bands and multiplicative light scattering perturbations.
Noisy spectra

1450

Multiplicative light
scattering

Shift and broadening

in spectral bands

3.2

0.12
1400

3.0

1300

2.8

0.08

Abs. (AU)

1350

Absorbance (AU)

Intensity

0.10

0.06
0.04
0.02

1250

16.5

17.0

17.5

18.0

18.5

X-ray diffraction profiles of

mannitol-methanol
suspensions with the content
of mannitol varying from
0.0% to 5.0% g/ml.

19.0

2.4

2.2

0.00
1200
16.0

2.6

2.0

-0.02
850

900

950

Wavelength (nm)

1000

1050

1.8
850

900

950

1000

1050

Wavelength (nm)

Five NIR spectra for a ternary 20 NIR spectra of a powder

mixture sample measured at mixture measured under
five different temperatures
different sample compactness

Process Analytics in Reactor Scale-Up

PAT from 1 to 20 to 250l Scales

Pump
Inlet and
return pipes
Flow loop
to flow cell
Water jacket
lines

Reactor
Magnetic
stirrer

Heating/Cooling unit

20 L Crystalliser

Bede MONITORTM In-process XRD

Smoothed
Principal
Component
Analysis
(SPCA)
Crystal
Polymorph
Monitoring
& Control

Enhancing signal-to-noise ratio

12000
i

X Xr i = i ( I + k Q Q )r
T

F = [r1 , r2 , L rc ] [r1 , r2 , L rc10000

]+

i = 1, 2 , L , m

peaks

x F = Fx = Fx s + Fx n

t = 200 s
t = 600 s
t = 1000 s
t = 1400 s

8000

t = 1800 s

Temperature
monitors
Temperature
readers

Intensity

t = 2200 s
6000

t = 2600 s
t = 3000 s

t = 3400 s
System provides capability to monitor
4000
t = 3800 s
polymorphic form in-process, i.e.
that unaffected by product separation
t = 4200 s
2000
prior to analysis.
t = 4400 s
Typically circa 1 wt % detectable via
t = 4600 s
in-process
much lower
with
Raw (a) XRD,
and Processed
(b) XRD
profiles (by
0 SPCA) for 6 XRD data sets

advanced
chemometric
analysis
15
18
21
24
27
30
peaks
2Theta [degree]
(Smoothed PCA)

Loading Space Standardisation (LSS)

PAT from 1/2 to 20 to 250l Scales

Correcting temperature-induced spectral variations for Heating/Cooling

ATR-FTIRunitdata
in crystallization process monitoring
Turbidity Probe

0.12

Absorbance (AU)

0.10

Temperature Probe

0.08
ATR-FTIR

Control
PC

LSS

0.06

pH Probe
0.04

FTIR
PC
0.02

0.10

Absorbance (AU)

0.12

0.08

20 L Crystalliser

0.06

Stirrer motor Temperature probe ATR-FTIR PC

0.04

0.02
ATR-FTIR

0.00

0.00
850

900

950

1000

Wavelength (nm)

Thermo-stat Bath
1050

850

900

950

Turbidity and
pH Probes
Wavelength

1/2 L Crystalliser
Condenser bath

1000

(nm)

1050

Control PC

Mnchwilen Foxboro Control System as Set-Up for

CBBII Trial on 250 Litre Reactor R-122

PLS quality factor

Baffle temperature

Reflux
condenser
PLS concentration

Valve position

Turbidity
Control temperature
Jacket flow rate

Water
inlet

Control valve
on chiller

To & from
site cryo
services

Power on aggitator
Jacket
temperatures

Heater
power

Heater chiller oil bath unit

Supersaturation Control System Upgrade to IMC Capability

Crystallisation Vessel
FTIR Spectrometer
ENABLIR
FTIR
PC

Spectra

SOFTWARE

WINISO

PLS model
Cal.
model

SOFTWARE

4 20 mA

C(t)
HEL PC

INPUT
Block

Data Processing Block

C(t)

Macro

Solubility
Model

User
Define
Model

S(t)
Control Block
IMC Based
PI Controller
PI Controller

User
Define S
set point

Supersaturation Control of L-Glutamic Acid

250 litre Plant Crystalliser
1.225
1.175
S = 1.1
1.125
1.075
Smin
=
1.075
1.025
Started Supersaturation
0.975
Control
0.925
0.875
0.825
5% seeds
0.775
added
0.725
0.675
Temperature (C)
Concentration (g/500ml)
0.625
0.575
Turbidity (%)
Solubility (g/500ml)
0.525
Supersaturation
Slimit
0.475
0.425
Slimit
0.375
0.325
0.275
0.225
0.175
0.125
0.075
0.025
200
300
400
500
600
700
800
900
1000 1100 1200 1300

Smax = 1.125

90
80
70
60
50
40
30
20
10
0
0

100

Time (min)

Supersaturation (S = C/C*)

Temperature, Concentration, Solubility & Turbidity

100

Variability Modelling and Calibration Challenges

Process Issues:
z

Multiple or changing formulations (recipes)

Cell improvement; cell line changes; media changes

Equipment characteristics; site-to-site process differences, etc

Fluctuations in both control and external process variables

Small data sets are an issue but can be enhanced through Bootstrap
Aggregation and Bootstrap Aggregated Regression.

Analytical Issues:
z

Separating absorbance from multiplicative light scattering effects

caused by the variations in optical path length

Inter probe variability: impact of component variance on PLS calibration

can probe differences be accomodated or eliminated?

Can calibration models be made generic for different production unit

operations / production lines?

Closing the Analytical Control Loop

Incorporating PAT Sensors into Real
Time Process Control

Batch Endpoint Control

using Process Analytical Data

Batch Endpoint Control Concept

Process Analytical Technology - Calibration PLS Model

Controlling Scores

Industrial Examples
z

Drug Product Granulation Control

Results

Batch Endpoint Control Concept

Controller calculates future

MV moves over the
whole future window
to the end of the batch

Trajectory of CTQ parameter

CTQ end-point target
Manipulated variable (MV)

TIME
Start of batch

1.
2.
3.
4.

Current time

End of batch

End-point value of CTQ parameter continuously estimated throughout the batch

MV trajectories modified by controller to make estimated end-point hit target
Unfolded PLS model - predicts end-point value based on MV and process variable
trajectories over the entire batch
At each point in the batch, computes trajectory of MV moves over entire batch to
minimise the error between the predicted end-point value and end-point target

Batch Endpoint Control Concept

Evaluation of End-Point Model

End-point value is estimated during the batch as the batch progresses

the uncertainty associated with the estimate decreases.

ACTUAL
END-POINT

CTQ
Parameter

Continuous
measurement shown
for illustration (not
usually measurable in
real time)

MV
Trajectory

batch start

batch end

Process Analytical Technology

Calibration PLS Model

Using PAT sensors in control offers some exciting possibilities:

z

Using modern PAT devices capable of 1-2 second or sub-second

measurement rates, real-time control based on PAT measurement is a reality.

Richness of measurement: not just a single data point per sample but a vector
of data per sample that exposes a broad amount of information about the
process State.

Sensor calibration models can give real time inference of product property
Evolution of absorbance at a selected wave number

NIR absorbance
Spectrum

Batch Evolution (25 mins)

Process Analytical Technology

Calibration PLS Model

AND also throws some challenges:

z

Real time pre-processing Management of outliers in real time.

No control system is going to control a spectrum of several hundred

simultaneous values. So must define which aspect of the spectrum
provides the most important / critical information.

Ideally there could be a calibration model to determine the product

property?

Or Alternatively - Are there particular features/segments of the spectrum of

interest?
z

Can the scores of the PCA/PLS calibration model be used to control

Industrial Example Drug Product Granulation Control

Control System Overview

Source: Example from Published Industrial Applications: Pfizers HSWG Real-time Control
System (IFPAC 09 Mojgan Moshgbar).

Industrial Example: Controlling Scores

Source: Example from Published Industrial Applications: Pfizers HSWG Real-time Control
System (IFPAC 09 Mojgan Moshgbar).

Industrial Example: Results

Source: Example from Published Industrial Applications: Pfizers HSWG Real-time Control
System (IFPAC 09 Mojgan Moshgbar).

Conclusions

PAT is not just Analytics it is sophisticated sensors and analysers together

with smart chemometrics and modelling.
z

Variability issues around process plants, process analytics and multiple

or changing formulations (recipes) A Process Systems Challenge.

The Quality by Design initiative within the Pharmaceutical Industry has used
Advanced Control techniques, combined with chemometric models to monitor
and control Critical Quality and Process-ability parameters effectively.
1. Spectral data contains much information that may be converted into new
variables that capture the key aspects of the process character as it progresses
through a batch.
2. Process insight into unit operation may be obtained using Process Analytical
Devices.
3. Periodic adjustment of the process variables at discrete decision points to
nudge the process is enough to modify the final quality parameters effectively.
4. Robust data quality and control of the whole system not just analytical but
conventional measured systems must be quality monitored to ensure that
behaviour outside the bounds of understanding is identified in real time.

Thanks and Acknowledgements

Many thanks to the organisers

for their kind invitation
and of course you, the audience,
for your kind attention

Julian Morris and Zengping Chen acknowledge their CPACT research

colleagues and the CPACT member companies for their R&D challenges