Pharmaceutical supply chains: key
issues and strategies for optimisation
Nilay Shah
Centre for Process Systems Engineering
Imperial College London
1
Outline
•
•
•
•
•
Some conclusions
Background
The pharmaceutical industry supply and value chains
Supply chain issues
Primary manufacturing:
– Risk management
– Process development
• Challenges for the future
2
Some conclusions
• The pharmaceutical supply chain is very complex, with many
interacting facets
– Difficult to generate radical improvements quickly
– Piecemeal approaches (e.g. improved logistics) will generate
incremental benefits
• Current process technology is one of the main supply chain
bottlenecks
– Many “built-in” inefficiencies that constrain performance
– Not a very responsive system
• Current models in the research community are too “companycentric”
– Future models need to consider a holistic view of an extended
supply chain of specialist agents
•
•
•
•
•
IP generators
Testing specialists
Contract manufacturers
Logistics providers
Healthcare providers/consumers
3
Trends in the pharmaceutical and related
industries
• Time to market is the key metric
• R&D productivity (numbers of new chemical entities registered
per unit amount of investment) is declining
• effective patent lives are shortening
• even while active, patents provide lower barriers to entry
• many cheaper product substitutes in many therapeutic areas
– alternative compounds (“me-too drugs”)
– off-patent generics
• payers of healthcare exerting strong price pressure and
influencing prescribing practices
• for approval, new drugs must:
– address new therapeutic areas; or
– have very significant cost or health benefits over existing
treatments.
4
Value growth
140
119.1
120
100
$ bn
86.3
80
60
50.6
US
Japan
Europe
40
20
14.7
10.1
17.5
0
1981
1999
5
Product pipeline
Discovery
PreClinical
Clinical
Trials
Launch
Commercial
6
Compound success rates by stages
7
Decreasing R&D efficiency
Total Spending on R&D
$bn
New Drug
Applications to FDA
8
The value chain
The “supply chain”
•
•
•
•
Discovery generates candidate molecules
Variety of trials evaluates efficacy and safety
Complex regulation process
Manufacturing:
– Primary manufacture of active ingredient (usually 1-2 sites)
– Secondary manufacture – production of actual doses (up to 20
sites)
– Often geographically separate for taxation, political etc reasons
– Complex logistics
• Distribution
– Supply chains often global
– Many third parties become involved
• Distributors, health authorities etc.
• Retail
– Pharmacies, doctors and hospitals are main outlets for ethical
drugs
9
The value chain
Research & development
15%
Primary manufacturing
5 - 10%
Secondary mfg/packaging
15 - 20%
Marketing/distribution
30 - 35%
General administration
5%
Profit
20%
Total
100%
10
The supply chain
• Difficult to get value from early stage discovery/trials
processes (cf. decreasing R&D efficiency, recent
mergers)
– Can Process Systems Engineering techniques help generate
more focussed searches/libraries?
• Companies view supply chain differently:
– was a means of getting product to where it was needed
– now a means of delivering additional value
• (At least) three interesting problems:
– Process development and design
– Planning of trials/testing and capacity under uncertainty
– “Classical” Supply chain planning and management
11
Process development and design
• Problems:
– Process chemistry, solvent and catalyst choices result in
• Low material efficiencies (of order 1%)
– Inefficient, very traditional batch manufacturing processes result in
• Low velocity ratio or value-added time (of order 1%)
– Sub-optimal design of drug delivery systems results in
• Low bio-availability where required (of order 1% for traditional
formulations e.g. pills)
• 1mg delivered to target area:
– may require 10kg of materials overall!
– ties up a considerable amount of capital!
12
One stage: typical overall mass
balance
Effluent 42%
Landfill 9%
Incineration 6%
Material in
100%
Manufacture
Product 10%
Recovery/recycle 29%
Solvent ‘loss’ 2%
By-product sold 2%
13
Typical batch process
Measure
React
12 hr
Separate
Purify
12 hr
2 days
Working capital for 4 days
How long does the conversion process
really take?
Product
12 hr
Store
Store
18 hr
Separate
12 hr
Low value-added time!
14
Longer time: higher costs and lower
quality
Cost
Time
Quality
15
Process development and design
• Many research groups now working in relevant fields
• More of a design than operations issue
– Significant improvements in material efficiencies required
• Involvement of process (systems) engineering at early stage
–
–
–
–
Model-based design
True catalysis rather than stoichiometric reagents
Optimise overall material efficiency rather than reaction yields
Large reductions in solvents
» Need for better heat transfer technology
– Improve manufacturing performance
•
•
•
•
Run processes as close to intrinsic rates as possible
Use small-scale continuous processing where possible
Avoid stage-to-stage isolation where possible
Cleaning and changeovers! (see later)
– Improve drug delivery to be more targeted (new field)
16
The supply chain
• Difficult to get value from early stage discovery/trials
processes (cf. decreasing R&D efficiency, recent
mergers)
– Can Process Systems Engineering techniques help generate
more focussed searches/libraries?
• Companies view supply chain differently:
– was a means of getting product to where it was needed
– now a means of delivering additional value
• (At least) three interesting problems:
– Process development and design
– Planning of trials/testing and capacity under uncertainty
– “Classical” supply chain planning and management
17
Testing and capacity planning
• “Traditional” sequential approach:
Tests for
activity
fail
pass
Tests for
toxicity
fail
pass
Tests for
efficacy
Tests characterised by:
ƒ Duration*
ƒ Cost (in-house or outsourced)*
ƒ Resource requirements*
ƒ Hard precedence constraints
ƒ Soft/conditional precedence constraints
ƒ Probability of success
*may be distributions rather than known
fail
Sequential approach conserves resources, but
may increase time to market
18
Optimised planning of tests
(Grossmann & co-workers, Pekny, Reklaitis and co-workers)
• Rather than follow sequential approach, approach
from a resource-constrained scheduling perspective
• Some tasks have conditional dependence
• Degrees of freedom on task precedence
• Optimisation balances:
– Risk of unnecessary expenditure
– Potential rewards of coming to market earlier
– Resource constraints and outsourcing costs
• Resource-constrained stochastic optimisation
problem
– Conservative approaches (always feasible)
– Hybrid simulation-optimisation approaches
19
Capacity and portfolio planning under
uncertainty
• What technology/capacity, where, when (plant
fabrication lead times!), whether to outsource
• Products to prioritise in the R&D pipeline to
structure the future portfolio optimally
• Most severe for pharmaceuticals:
– capacity requirements very dependent on outcome of
clinical trials, registration etc.
• Extreme cases
– pessimistic: no investment and many successful
products: severe capacity limitations
– optimistic: investment → plenty of capacity but no
new products → patent expiry issues
• Need for systematic way to balance risks
20
Product Pipeline and Capacity Plans
demand
materials
entering CT
promising
CT results
current
products
How to:
• allocate capacity between products ?
• plan capacity investment ?
successful
product life-cycle
time
21
Clinical trials
More than two outcomes!
clinical trials
Deterministic
stage
Success
High
0.10
High
Success
Target
0.40
Low
0.9
0.50
0.60
0.95
Launch
Failure
0.05
Minimum
0.10
Failure
Failure
0.1
Phase II (~2 year)
Stage 1
0.30
Not Considered yet in this work
Phase III 3-5years
Stage 2 & 3
Registration
(1 year)
22
Scenario Tree & Clinical Trial
Outcomes
1
Clinical Trial
Outcomes
1
H – High Success
T – Target
L – Low
F – Failure
1
HHT
Scn 2
HH
HHHT
+1 Stochastic
Product C2
2
HT
3
HL
4
HF
HTL
7
+1 Stochastic
Product C4
Periods
10
HTLF
+1 Stochastic
Product C3
Scn 63
HFFL
16
Stage 2
Stage 1
Scn 28
29-32
H
1 Deterministic
Product C1
1
1-4
61-64
HFF
Stage 3
20
Stage 4
30
40
23
Results
NPV profile
24
Strategy matrix: risk analysis
Worst Case ( Exposure ) NPVs
(-100)
Scenarios Legend
Ab
Aa
Bc
Ba
Bb
C
D
Ea
Eb
Fb
Fa
G
H
(-130)
(INCREASING
RISK)
(-160)
30%
15%
0%
Probability of Losses
25
Combined testing and capacity planning
(Grossmann and co-workers)
• Interesting way forward
– Holistic analysis of value chain
• Better synchronisation leads to material being ready if tests
are successful
– Avoid shortage of material for clinical trials
• Basis for future work; extensions:
– Management of risk is a key feature
• Real options techniques should be relevant
– Needs to take account of global trading structures
– Needs to include creative possibilities in model
• e.g. placing of low-commitment options with subcontractors
– Needs to take more extended view of supply chain
26
The supply chain
• Difficult to get value from early stage discovery/trials
processes (cf. decreasing R&D efficiency, recent
mergers)
– Can Process Systems Engineering techniques help generate
more focussed searches/libraries?
• Companies view supply chain differently:
– was a means of getting product to where it was needed
– now a means of delivering additional value
• (At least) three interesting problems:
– Process development and design
– Planning of trials/testing and capacity under uncertainty
– “Classical” Supply chain planning and management
27
“Classical” supply chain planning and
management
• Some performance measures
–
–
–
–
–
Pipeline stocks may be 30-110% of annual demand
Finished good stocks 10-50% (4-26 weeks) of annual demand
Supply chain cycle times of order 1000s of hours
Value added times 0.3-5% of cycle times
Supply chain costs overtaking R&D costs
• What can better operations deliver?
– 30% stock reduction
– 30% increase in value-added time
– 7% reduction in supply chain costs
• Benefits of improved operation for one large drug:
– $30m one-off
– $8-16m p.a.
28
Issues/Structure
• Primary production processes usually “slow” and
“unresponsive”
–
–
–
–
–
lowish yield
labour- and time-intensive
can take 30-200 days from end to end
many QA steps along the way
long changeovers (one to four weeks) force campaign operation
• Secondary processing often geographically separate from
primary
– transportation lags, but sensible use of API storage can mitigate
• Secondary processing sometimes serves market directly, but
more commonly:
– regional storage locations/wholesalers and other agents
• Long supply chain cycle times (60-300 days)
– Many delays in process
– poor responsiveness to changes in demands
– High service levels required → high stocks
29
Dynamic supply chain analysis
• Dynamic analysis of existing supply chains generates
considerable insight
– Understand relationship between policies/parameters and
performance measures
• Use a generic modelling approach which captures
physical and business processes
• Library of supply chain objects based on generic
node:
Inbound
material
management
Handling/
processing/
conversion
Outbound
material
management
30
One SKU: forward look
3.50E+06
100
3.00E+06
2.50E+06
99.8
Average Inventory
2.00E+06
Safety Stock
High Confidence
1.50E+06
Low Confidence
99.6
1.00E+06
99.4
5.00E+05
4
Nov-01
5
6
7
8
Weeks Safety Stock
Dec-01
Oct-01
Sep-01
Jul-01
Aug-01
Jun-01
Apr-01
May-01
Mar-01
Jan-01
Feb-01
Nov-00
Dec-00
Oct-00
Sep-00
Jul-00
Aug-00
Jun-00
Apr-00
May-00
Mar-00
Jan-00
Feb-00
0.00E+00
Customer service level
0.7
2.50E+06
0.6
2.00E+06
0.5
1.50E+06
0.4
0.3
1.00E+06
0.2
5.00E+05
0.1
0
0.00E+00
4
5
6
7
Weeks Safety Stock
Expected number of stock out (per period)
8
4
5
6
7
8
Weeks Safety Stock
Average inventory kept
31
A particular market
•
•
•
•
End-user demand fairly steady, but
Internal processes create additional dynamics
Need for high safety stocks to buffer
Aim for high service levels – high perceived
opportunity cost
• What if demand was more in line with end-use ?
32
Demand profiles
700000
600000
500000
400000
P a xil2 01 00(Ac tu a l)
(Fitting)
300000
Prediction
200000
100000
Sep-00
Jul-00
May-00
Mar-00
Jan-00
Nov-99
Sep-99
Jul-99
May-99
Mar-99
Jan-99
Nov-98
Sep-98
Jul-98
May-98
Mar-98
Jan-98
Nov-97
Sep-97
Jul-97
May-97
Mar-97
Jan-97
0
includes our forecasting algorithms for future forecasts – these pick up key
dynamics
33
Current policy: performance
Service Level
Average stock
100
800000
98
700000
600000
%
packs
96
94
92
500000
400000
300000
90
200000
2
3
4
5
weeks cover
6
7
8
2
3
4
5
6
7
weeks cover
34
8
Smoothed performance
Average stock
102
800000
100
700000
98
600000
packs
%
Service Level
96
94
500000
400000
92
300000
90
200000
2
3
4
5
weeks cover
6
7
8
2
3
4
5
6
7
8
weeks cover
35
Comparison of two supply chain
responses
• Pharmaceutical process
– primary production has five synthesis stages
– two secondary manufacturing sites
• Two different process principles
– Case A: QC at the end of each synthesis stage and the final
products
– Case B: QC for the product of the primary process (AI), and
the final product
36
Inventory variation for one SKU
Base Case for Pack C
Results for Pack C with QC (Monte Carlo 400 simulations)
300000
300000
mean quantity
mean quantity
safety stock
safety stock
250000
95% upper confidence
250000
95% upper confidence
95% lower confidence
Pack C stock level / packs
Pack C stock level / packs
95% lower confidence
200000
150000
100000
200000
150000
100000
50000
50000
0
0
0
0
20
40
60
80
100
20
40
60
80
100
Time / weeks
Time / weeks
Case A
Case B
37
SCM more generally…
• Re-emphasise need for better process technology for 10
manufacture
• Currently push-based at back-end of supply chain:
– hard to be responsive
– complicated dynamics
– can’t exploit short-term opportunities (e.g. tenders)
• aim for much faster processes
• avoid too many quality control interventions and isolations
• aim for easy to clean/reconfigure (disposable?) plants
– Move towards short-term scheduling
• Interim:
– optimise campaign planning (cf. literature of 1980s!)
– optimise changeovers using SMED concepts
38
SCM more generally (cont’d) …
• Current SCM methodologies involve a degree of
decentralisation
–
–
–
–
Regional demand management
Primary or secondary planning
Subcontractors
Robust, but built-in inefficiencies
• Need for more integrated, seamless planning
– Very large-scale multi-site problems with large geographical span
– Looser alliances may arise (e.g. semiconductors, computers)
• Increased co-ordination problems
– Tailored optimisation algorithms will be required
– Need to build in robustness (cf. work of Maranas and co-workers,
Sahinidis and co-workers)
39
Some conclusions
• The pharmaceutical supply chain is very complex, with many
interacting facets
– Difficult to generate radical improvements quickly
– Piecemeal approaches (e.g. improved logistics) will generate
incremental benefits
• Current process technology is one of the main supply chain
bottlenecks
– Many “built-in” inefficiencies that constrain performance
– Not a very responsive system
• Current models in the research community are too “companycentric”
– Future models need to consider a holistic view of an extended
supply chain of specialist agents
•
•
•
•
•
IP generators
Testing specialists
Contract manufacturers
Logistics providers
Healthcare providers/consumers
40
Future perspectives
• Industry
– at crossroads?
• “Big is beautiful” v. alliances of specialists
• Latter will need next generation of supply chain tools
• Products
– More complex, more synthesis stages, more chiral, more active,
smaller lot-sizes
– Better drug delivery mechanisms
• Smaller dosages
– Likely to become more specialised
• Local solutions to local problems
• Genetic research leading to target sub-populations
• Current manufacturing and supply chain poorly suited to this
– Economies of scale 1-2 orders of magnitude out
– Rapid response vaccines (civilian and military)
– More crop-derived products – new supply chains
41
Acknowledgements
• Members of Britest project (www.britest.co.uk)
• UK Engineering and Physical Sciences Research
Council
• Fellow researchers: Gabriel Gatica, Jonatan
Gjerdrum, Wing Yan Hung, Rodney Johnson, Lazaros
Papageorgiou, Nouri Samsatli, Mona Sharif, Trevor
Tsang
42