Description
Supply chain optimisation is now a major research theme in process operations and management. A great deal of research has been undertaken on facility location and design, inventory and distribution planning, capacity and production planning and detailed scheduling. Only a small proportion of this work directly addresses the issues faced in the pharmaceutical sector. On the other hand, this sector is very much ready for and in need of sophisticated supply chain optimisation techniques
Computers and Chemical Engineering 28 (2004) 929–941
Pharmaceutical supply chains: key issues and strategies for optimisation
Nilay Shah
?
Centre for Process Systems Engineering, Department of Chemical Engineering, Imperial College of Science, Technology and Medicine, London SW7 2BY, UK
Abstract
Supply chain optimisation is now a major research theme in process operations and management. A great deal of research has been
undertaken on facility location and design, inventory and distribution planning, capacity and production planning and detailed scheduling.
Only a small proportion of this work directly addresses the issues faced in the pharmaceutical sector. On the other hand, this sector is very
much ready for and in need of sophisticated supply chain optimisation techniques.
At the supply chain design stage, a particular problem faced by this industry is the need to balance future capacity with anticipated demands
in the face of the very signi?cant uncertainty that arises out of clinical trials and competitor activity. Ef?cient capacity utilisation plans and
robust infrastructure investment decisions will be important as regulatory pressures increase and margins are eroded. The ability to locate
nodes of the supply chain in tax havens and optimise trading and transfer price structures results in interesting degrees of freedomin the supply
chain design problem. Prior even to capacity planning comes the problem of pipeline and testing planning, where the selection of products
for development and the scheduling of the development tasks requires a careful management of risk and potential rewards.
At the operation stage, it is often dif?cult to ensure responsiveness. Most pharmaceutical products involve primary active ingredient (AI)
production (often multi-stage chemical synthesis or bioprocess) and secondary (formulation) production. Both of the stages are characterised
by low manufacturing velocities and are hampered by the need for quality assurance activities at several points. It is not unusual for the
overall supply chain cycle time to be 300 days. In this environment, supply chain debottlenecking and decoupling strategies together with
co-ordinated inventory management are crucial for quick responses to changing market trends. A good understanding of what actually drives
the supply chain dynamics is also required. As often as not, erratic dynamics are introduced by business processes rather than by external
demand, and may be eliminated by the re-design of internal business processes or supplier/customer relationships.
This paper will consider important issues in supply chain design and operation drawn fromthe literature and fromour collaborative research
projects in this area. The main features of the problems will be reviewed as will the literature to date. Some strategies for solution will be
identi?ed, as will some future research needs.
© 2003 Elsevier Ltd. All rights reserved.
Keywords: Pharmaceutical industries; Supply chain optimisation; Pipeline management
1. Introduction
The pharmaceutical industry can be de?ned as a complex
of processes, operations and organisations involved in the
discovery, development and manufacture of drugs and med-
ications.
The World Health Organisation (WHO) de?nes a drug or
pharmaceutical preparation as:
any substance or mixture of substances manufactured,
sold, offered for sale or represented for use in the diag-
nosis, treatment, mitigation or prevention of disease, ab-
normal physical state or the symptoms thereof in man or
?
Fax: +44-171-594-6606.
E-mail address: [email protected] (N. Shah).
animal; [and for use in] restoring, correcting or modifying
organic functions in man or animal.
This is a very wide de?nition, and correspondingly, there
are number of key players in the pharmaceutical industry,
including:
(i) The large, research and development-based multina-
tionals with a global presence in branded products, both
ethical/prescription and over-the-counter. They tend to
have manufacturing sites in many locations.
(ii) The large generic manufacturers, who produce
out-of-patent ethical products and over-the-counter
products.
(iii) Local manufacturing companies that operate in their
home country, producing both generic products and
branded products under licence or contract.
0098-1354/$ – see front matter © 2003 Elsevier Ltd. All rights reserved.
doi:10.1016/j.compchemeng.2003.09.022
930 N. Shah / Computers and Chemical Engineering 28 (2004) 929–941
(iv) Contract manufacturers, who do not have their own
product portfolio, but produce either key intermediates,
active ingredients (AI) or even ?nal products by pro-
viding outsourcing services to other companies.
(v) Drug discovery and biotechnology companies, often
relatively new start-ups with no signi?cant manufactur-
ing capacity.
Most of the material in this paper is particularly relevant
to the ?rst group. This group dominates the marketplace and,
due to the global nature of the enterprises involved, tends to
have the most challenging supply chain problems.
1.1. The changing circumstances of the industry
In the recent past, the high returns on investment and
high turnovers from “blockbuster” products resulted in the
following regime (Booth, 1999):
• good R&Dproductivity, often creating compounds to treat
previously untreatable diseases;
• long effective patent lives of these compounds;
• ability of these patents to provide technological barriers
to entry;
• a limited number of product substitutes in a given thera-
peutic area; and
• a low price sensitivity; supported by the separation be-
tween prescribing and paying responsibilities.
The resulting corporate strategy was to ensure high mar-
gins by exploiting the price inelasticity and invest a large
proportion of the resultant pro?ts in R&D (approximately
25% of sales), in order to ensure a healthy product pipeline.
The more recent circumstances are much more challeng-
ing:
• R&D productivity (in terms of numbers of new chemical
entities (NCE) registered per unit amount of investment)
is declining;
• effective patent lives are shortening;
• even while active, patents provide lower barriers to entry;
• there are many product substitutes in many therapeutic
areas; either alternative compounds (“me-too drugs”) or
off-patent generics; and
• the payers of healthcare are exerting strong price pressure
and in?uencing prescribing practices; this means that in
order to be approved, new drugs must address new thera-
peutic areas or have very signi?cant cost or health bene?ts
over existing treatments.
On the one hand, the global marketplace has become
more liberalised, exposing products to competition. On the
other, governments and other agencies have tended to in-
tervene more as they become concerned at every increasing
healthcare costs associated with ageing populations. Mea-
sures taken include strict controls on the prices of new drugs,
more cost–bene?t analysis, and encouragement of the use
of generic substitutes or alternatives where possible.
A further weakness that will hamper the large players
in the area is the historical dependence on “blockbuster”
drugs. A recent report by Datamonitor predicts that only 4
of the 19 companies currently selling blockbuster drugs will
be able to maintain double digit growth between 2001 and
2008 (Butler, 2002). This dependence is illustrated by the
following ?gures (Butler, 2002):
• Eli Lilly’s net pro?ts dropped by 20% after Prozac came
off patent. Overall, Eli Lilly’s sales in 2002 were down
4% and the sales excluding Prozac were up 8% (Eli Lilly,
2003).
• BMS’ patent on Glucophage (with previous annual sales
of $2bn) has expired and its sales are down by 87% in
the ?rst 3 months of 2002 after the launch of generic
substitutes.
• Losec represented 34% of AstraZeneca’s sales in 2001
and it might come off patent this year if the company is
unsuccessful in its legal battle—in which case generics
are expected on the market in late 2002.
1.2. Drivers in the pharmaceutical industry
Probably the single most important driver in the pharma-
ceutical industry is the time-to-market. Companies secure
very signi?cant returns in the early life of a successful drug,
before any competition. The competition-free life is short-
ening, typically from 5 to 1–2 years. Competition in this
sense relates to similar (rather than exactly the same) drugs.
For example, Bayer’s anti-cholesterol drug Baycol was
withdrawn in 2001 due to safety concerns, and the two later
entrants Pravachol (from BMS) and Lipitor (from P?zer) are
now the biggest sellers for their companies (Butler, 2002).
Given the signi?cant potential for adverse health effects,
the industry is subject to very stringent regulation. This starts
from the processes used to evaluate the safety and ef?cacy
of the chemical compounds, through to the details of the
process and plant design and manufacturing operations. The
primary regulator that the companies must satisfy is the US
Food and Drug Administration. It may be the case that the
existence of regulatory protocols has hindered innovation in
this sector; with companies blaming regulators for their own
innate conservatism.
The regulatory process tends to be slow and expensive;
both these effects must be borne by the industry. Further-
more, the complex chemical compounds involved have
more complex manufacturing processes, and the activities
of route investigation, process development, scale-up plant
design/retro?t, commissioning and quali?cation are either
increasing in duration or proving stubborn to shorten.
An estimate of £200–400 m is required to launch a new
drug, and an average of 8–12 years elapses from patent ?ling
to ?rst sale (see, e.g. Grabowski, 1997).
There is a general trend for companies to divest excess
capacity that came about from having many local manu-
facturing sites, and move towards a global supply chain
N. Shah / Computers and Chemical Engineering 28 (2004) 929–941 931
management process. This brings with it many complex
co-ordination issues and much tighter capacity constraints.
Currently, the logistics cost in the sector is relatively high
(Booth, 1999).
Research ef?ciency is declining in the sense that the cost
of each new chemical entity is increasing. Although growth
in investment in R&D has exceeded in?ation over the last
30 years, the global trend are as follows: 844 NCEs were
registered in 1961–1970, 665 in 1971–1980 and 506 in
1981–1990 (Ballance, Pogany, & Forstner, 1992). This has
been one of the main drivers behind the recent series of
mergers and acquisitions in the industry, the long-term ben-
e?ts of which will probably not be felt for some time yet as
R&D activities continue to be consolidated.
Historically, most management attention has been paid to
drug discovery and sales and marketing (the extreme ends
of the supply chain), but now much more attention is being
paid to supply chain optimisation as a means of delivering
value. According to Booth (1999):
• there is a welcome move away from viewing the supply
chain as merely having to deliver security of supply at
minimum cost, to a recognition of its ability to generate
both value for the customer and hence to the shareholder;
and
• restructuring of the supply chain along regional and global
lines will require massive reductions in capacity, which
was acquired in many cases to propitiate national interest
in return for sympathetic pricing.
1.3. The life-cycle of a pharmaceutical product
In order to put this paper in the right context, it is im-
portant to describe the life-cycle of a drug; it is somewhat
different from that of other process industry products.
The research or discovery phase tends to use thousands
of more or less random test compounds against therapeutic
targets. It typically takes about 10 years to result in a po-
tential new drug that is registered. From this point onwards
patent protection applies.
The potential new drug must then be tested for both safety
and ef?cacy. This involves a variety of trials; early on for
toxicity and later on for ability to alleviate symptoms and
remove disease. Finally, the process development activity
comes up with a chemical or biochemical route to manufac-
ture and an associated manufacturing process. This set of
activities typically takes 6–8 years and is usually known as
the development activity.
Finally, the more familiar processes of manufacturing and
distribution follow.
2. Components of the pharmaceutical industry
manufacturing and distribution chain
A typical pharmaceutical supply chain will consist of the
one or more of the following nodes:
(i) primary manufacturing (possibly including contractor
sites);
(ii) secondary manufacturing (possibly including contrac-
tor sites);
(iii) market warehouses/distribution centres;
(iv) wholesalers; and
(v) retailers/hospitals.
2.1. Primary manufacturing
The primary manufacturing site is responsible for the pro-
duction of the active ingredient (AI or API). This normally
involves either several chemical synthesis and separation
stages to build up the complex molecules involved, or fer-
mentation and product recovery and puri?cation in the case
of biochemical processes.
The manufacturing process is characterised by long task
processing times, often rounded to multiples of shifts. Where
multistage processes are operated, considerable inventories
are often held between stages. Furthermore, material froman
intermediate stage must often pass some formof quality con-
trol check before being approved for use downstream in the
process. This can introduce additional delays into the system.
The traditional process technology involves batch equip-
ment and ?exible pipework. The relatively low produc-
tion volumes result in multipurpose plants to spread
the capital cost between products. The need to avoid
cross-contamination of products and requirements for val-
idated cleaning and changeovers results in long downtimes
between products. These have been of order 4 weeks in
the past, but the application of techniques similar to the
single-minute exchange of die (SMED) methods (see, e.g.
Moser, Calderari, &Morini, 2000) applied to the car industry
have reduced these somewhat. These downtimes in turn im-
ply that long campaigns are the norm, otherwise equipment
utilisation is too low. It is not unusual for 1 year’s produc-
tion of a product to be produced in a single campaign, and
the material produced being stored until the next campaign
in the following year. Since most complex pharmaceuticals
are produced through multistage processes, the same often
holds true for the stable intermediates (stage products).
Needless to say, this mode of operation does not lend itself
well to responsiveness, and contributes signi?cantly to some
of the poor supply chain metrics exhibited by this industry.
A further source of complexity (and convenience) is the
use of contractors to manufacture some or indeed all of the
active ingredient stages. This process of outsourcing is a
growing one, as research-oriented companies concentrate on
the discovery and development activities and rely on third
parties’ manufacturing competence. This gives rise to ex-
tended supply chain co-ordination problems.
2.2. Secondary manufacturing
This is concerned with taking the active ingredient pro-
duced at the primary site and adding “excipient” inert
932 N. Shah / Computers and Chemical Engineering 28 (2004) 929–941
materials along with further processing and packaging to
produce the ?nal products, usually in SKU form. For ex-
ample, a product that is sold in pill form would undergo:
(i) granulation: with addition of all the excipient materials;
(ii) compression: forming the pills;
(iii) coating;
(iv) quality control; and
(v) packaging.
The secondary manufacturing locations are often geo-
graphically separate from the primary manufacturing lo-
cations. This is frequently the outcome of tax and transfer
price optimisation within the enterprise. There are often
many more secondary manufacturing sites than primary
ones, serving local or regional markets. Transportation be-
tween sites is of the order of 1 or 2 weeks if by ship (usually
the default mode) and of the order of one or two days if by
air.
Wholesalers play a signi?cant role in this sector. They
tend to be large and few. About 80% of demand ?ows
through this channel in the UK (with three large players ac-
counting for almost all the demand), with the large part of
the remainder going to hospitals. In the US another interme-
diary is growing—the managed care organization (MCO) or
healthcare maintenance organization (HMO).
2.3. Operational issues in the pharmaceutical supply chain
Although the processes will vary between companies,
all major pharmaceutical companies will operate ERP sys-
tems and follow a business process along the following
lines:
• Demand management—in each geographical region, for-
ward forecasts (e.g. 3–24 months) are developed, based
on historical data, market intelligence, etc. Tenders for
manufacture may also be evaluated and possibly accepted
at this stage.
• Inventory management and distribution requirements
planning—the demands determined are aggregated and
imposed on the appropriate warehouse/distribution cen-
tre. The impact on ?nished goods inventory is assessed
and if necessary, orders are placed on upstream secondary
manufacturing sites.
• Secondary production planning and scheduling—the or-
ders placed on the secondary sites are planned (typically
using MRP-II type tools) and then scheduled in detail (typ-
ically using APS tools). The impact of production plans
on active ingredient raw material stocks is evaluated and
if necessary, orders for AI are placed on the upstream.
• Primary manufacturing campaign planning and AI inven-
tory management. Here, the demands placed by secondary
manufacturing are satis?ed by careful management of in-
ventory and production planning.
An interesting feature of this process is that the
customer-facing end is effectively a “pull” process (driven
by orders) but the primary manufacturing stage has long
cycle times which make it dif?cult to ensure end-to-end re-
sponsiveness. This means that primary production is effec-
tively a “push” process, driven by medium- and long-term
forecasts. Relatively large stocks of AI must be held to
ensure good service levels and ensure smooth operation
at the interface of these processes. The well-documented
“bullwhip” or Forrester effect is often felt at the primary
manufacturing site, which is unfortunate since this is the
least responsive part of the supply chain as it normally
operates in campaign mode. This makes it dif?cult to ex-
ploit short-term opportunities (e.g. shortages of supply
of a competitor’s product, tenders for national supplies,
epidemics, etc.).
Another feature of this process is an outcome of its large
scale and geographical span. This is the distributed nature of
decision-making, which can lead to tensions and sub-optimal
decisions. Different nodes are not really aware of upstream
nodes’ resource constraints, and orders may be ?lled in order
of receipt, rather than on an economic basis. Of course,
centralised planning would not be without its dif?culties in
this context.
In our experience, the following supply chain performance
measures are typical of the industry:
• The stock levels in the whole chain (“pipeline stocks”)
typically amount to 30–90%of annual demand in quantity,
and there are usually 4–24 weeks’ worth of ?nished good
stocks.
• Stock turns (de?ned as annual sales/average stock) are
typically between 1 and 8.
• Supply chain cycle times (de?ned as elapsed time between
material entering as raw material and leaving as product)
are often between 1000 and 8000 h.
• The value-added time (time when something happens to
material as a percentage of chain cycle time) is of order
0.3–5%.
• Material ef?ciencies (the amounts of product produced
per unit amount of total materials used) are 1–10%.
The relatively high levels of stock are required to buffer
the slow supply chain against market dynamics.
2.4. Strategic and design issues in the pharmaceutical
supply chain
The decisions to be taken at this level include:
• Pipeline and development management—this involves the
selection of potential drugs to develop further, and the
planning of the development activity.
• Process development—the investigation of manufacturing
routes and the generation of manufacturing processes.
• Capacity planning and plant and supply chain network
design.
• Plant design—the selection and sizing of the major equip-
ment and storage units.
N. Shah / Computers and Chemical Engineering 28 (2004) 929–941 933
Some of the key issues are:
• Uncertainty in the demands for existing drugs (due to
competition, uncertainty in the ability to extend the pro-
tected life through new formulations, etc.).
• Uncertainty in the pipeline of new drugs—in particular,
which ones will be successful in trials, what sort of dosage
and treatment regime will be optimal.
• Process development—this is a complex problem, driven
by chemistry and yield optimisation. It often results in
inef?cient processes that are operated much more slowly
than the intrinsic rates—giving rise to batch processes and
long cycle times responsible for some of the problems
seen at the primary production planning stage.
• Capacity planning—the long lead times to make capac-
ity effective mean that decisions often need to be taken at
times of high uncertainty. Waiting for the uncertainties to
be resolved might delay the time to market by an unac-
ceptable amount.
• Network design—often tax implications take precedence
over logistics issues, these result in economic but poten-
tially complicated supply chains.
• Plant design—this tends to be very traditional, with no
real change in manufacturing technology for 50 years
(the workhorse of the primary manufacturing site is the
glass-lined stainless steel batch reactor). There are signif-
icant opportunities for intensi?ed, continuous processing.
3. Overview of some recent work
The recent work in the literature that is relevant here can
be categorised under these headings:
• pipeline and development management;
• capacity planning;
• simultaneous development and capacity planning;
• process development and plant design;
• production planning and scheduling; and
• supply chain simulation and dynamics.
The areas are reviewed in turn below.
3.1. Pipeline and development management
Schmidt and Grossmann (1996) considered the problem
of sequencing of testing tasks where unlimited resources
are assumed to be available. The key feature of the model
that distinguishes it from classical project scheduling is the
that each task has a probability of failure; this affects the
need for successor tasks. They formulated the problem as a
continuous-time MILP and solved the problem of maximis-
ing the overall expected NPV. If many tests for a product
are performed in parallel, the testing activity will be more
expensive, as the effect of failures on successor tests are not
taken into account. On the other hand, the product may come
to market much earlier, resulting in a much better cash ?ow
pro?le. Conversely, sequential test planning might avoid un-
necessary tests and reduce expense, but result in a later ar-
rival of the product in the marketplace.
In practice, the testing activity tends to be quite
resource-constrained, and may also involve outsourcing of
some stages. Jain and Grossmann (1999) develop a method-
ology for the sequencing and scheduling of testing tasks
under resource constraints. In this approach, each product
has a speci?ed set of testing tasks. Each task is charac-
terised by a duration, cost, precedence constraints, resource
requirements and probability of success. A task may be out-
sourced at a higher cost; in this case no internal resources
are required. The income associated with a product is given
as a function of the time of launch in the market.
The formulation developed is conservative and always
feasible in that the resource constraints are always enforced,
regardless of the probability of a task not actually taking
place. The cost component is modelled as an expected cost.
This ensures that the effect of starting tasks earlier than
necessary is modelled, i.e. that later tasks may not actually
take place due to the failure of the earlier one. Two alter-
native formulations are presented, a continuous-time MILP
and a graph-based one. The latter was found to be more ef-
?cient and was able to cope with problems consisting of 30
tasks.
Blau et al. (2000) consider the problem of risk man-
agement at the development stage. As mentioned earlier,
the development phase selects candidate drugs and takes
them through trials and process development. It is a long,
costly and inherently risky process with a large up-front
commitment. The aim of this work is to support the pro-
cess of product selection and test planning while managing
risk effectively. The development activities are modelled
as a probabilistic activity network, where each activity has
a time, precedence relations, resource requirements and
probability of success.
Risk is de?ned as the adverse consequences of exposure
to uncertainty, and in this context is usually related to the
premature withdrawal of a candidate drug. The risk of a
set of decisions must be balanced against the potential re-
ward. In this case, the potential reward is the expected ?-
nancial returns of drugs that do make it through the de-
velopment process. The risk/reward ratio can then be used
to compare different drug candidates. A screening process
removes any obviously unpromising candidates, and then
the remainder must be sequenced through the development
pipleline. A heuristic approach using simulation with lo-
cal rules in response to trigger events (e.g. failure of a
test) is employed. This aims to process tasks as quickly
as possible and although there is no guarantee of not vio-
lating resource constraints, these violations are usually not
large.
Subramanian, Pekny, and Reklaitis (2001) extend this
work to take explicit account of the resource requirements
of the problem. The problem statement is generalised in that
more sources of uncertainty in the problem are considered,
934 N. Shah / Computers and Chemical Engineering 28 (2004) 929–941
and include:
• task processing times;
• task resource requirements;
• task success probabilities;
• task costs; and
• market returns.
They make the point that a single-level mathematical pro-
gramming problem cannot hope to capture all these fea-
tures. On the other hand, discrete-event dynamic systems
(DEDS) techniques cope well with the stochastic elements,
but require local, myopic rules to resolve con?icts or make
choices as they arise. They therefore developed an integrated
optimisation–simulation framework (SIM–OPT), where a
DEDS simulator reverts to an optimisation layer (with dif-
ferent degrees of optimisation) to resolve con?icts or make
choices such as task sequencing. The optimisation layer is
an MILP which is updated by the latest status of the plant.
The results show that using optimisation far outperforms
the typical local rules used in classical DEDS. By repetitive
simulation, the statistical trends can be tracked and answers
to questions about corporate policy (particularly in relation
to risk and resourcing) can be obtained.
3.2. Capacity planning
The capacity planning under clinical trials uncertainty
problem has recently received some attention in the litera-
ture. The deterministic problem of allocating new manufac-
turing capacity to existing or potential sites around the world
is described by Papageorgiou, Rotstein, and Shah (2001).
They describe the features particular to the pharmaceutical
industry, and emphasise the importance of modelling ?nan-
cial ?ows as the taxation regimes affect the rewards associ-
ated with alternative solutions signi?cantly. Indeed, taxation
considerations can easily dominate the location decisions.
The problem of capacity planning under uncertainty was
considered by Rotstein, Papageorgiou, Shah, Murphy, and
Mustafa (1999). They considered the problem where three
products are at the start of clinical trials, and plans for cur-
rent and future manufacturing capacity are to be made. The
key trade-off in the capacity planning decision comes about
due to the lead time between deciding to invest in additional
manufacturing capacity, and that capacity coming on-stream.
Deferring capacity planning decisions until more informa-
tion is available from trials is obviously a lower-risk strat-
egy, but increases the time to market. As mentioned earlier,
this measure is critical. For example, when Tagamet came
to the market in the 1970s, it was free from competition for
at least 5 years, but now this competition-free time can be as
low as only 1–2 years. Rotstein et al. (1999) use a scenario
tree to capture the outcomes of the trials, and use a two-stage
stochastic programming with recourse formulation to model
the problem. The “here-and-now” decisions related to im-
mediate capacity expansions and the “wait-and-see” deci-
sions depend on trial outcomes and include further capacity
expansions, plant or product abandonment and production
and inventory planning. They show how different options
can be compared using a number of metrics, including ex-
pected NPV, the probability of the NPV being negative, the
worst case scenario, the total demand met of all the potential
products and the total demand met of the products chosen
from the portfolio.
Gatica, Shah, and Papageorgiou (2001), Gatica,
Papageorgiou, and Shah (2002a) extend this work to the
case where different products are at different stages in their
life-cycles, and those that are in trials will complete those
trials at different times. This gives rise to a much more
complicated scenario structure—it is a multistage stochastic
optimisation problem, with each stage re?ecting the com-
pletion of a clinical trial. In contrast to most of the work
reported on trials in the literature, this work uses more than
two outcomes (success or failure) for the completion of the
trial. Based on typical practices in industry, four outcomes
(failure, low, target and high) are used. This means that
four scenarios are required per stage, and for a problem
with N stages (i.e. N products in trials), there is one sce-
nario in the ?rst stage (re?ecting products currently in the
market with well-forecasted demands), four scenarios in the
second stage (re?ecting the four possible outcomes for the
?rst pipeline product to come out of trials), 16 in the third
stage (the combinations of outcomes for the two products
to complete trials) and so on, until the ?nal stage which has
4
N
scenarios. Each scenario has associated with it possible
capacity expansions and production and inventory planning
variables and constraints, so overall the problem becomes
a large scale stochastic programming problem with integer
and continuous decisions. It is solved as its deterministic
equivalent, a large MILP. This is relatively straightforward
to solve for the four-product case the authors report.
Clearly, the approach is limited by the complexity of the
scenario tree, so Gatica, Papageorgiou, and Shah (2002b)
extend it using a scenario aggregation procedure similar to
that of Clay and Grossmann (1997) to enable the solution
of larger problems.
3.3. Simultaneous development and capacity planning
The research reviewed above deals either with the problem
of organizing the development (testing) activities or planning
the capacity investments and future production. Maravelias
and Grossmann (2001) consider the problems of planning
of testing tasks and capacity simultaneously. The aim is to
optimize a performance measure (expected NPV) for the
process as a whole.
This bridges the gap between the two problems and aims
to ensure that the company is ready to produce a product
once testing is complete (if the product is successful).
The testing process is modelled as a set of tasks with tech-
nological precedence constraints, durations and resource re-
quirements. The tasks have two possible outcomes, success
or failure. All tasks other than process development may
N. Shah / Computers and Chemical Engineering 28 (2004) 929–941 935
be outsourced if internal resources do not suf?ce. Since the
method is to be applied at any time in the company’s oper-
ation, it takes account of the fact that different products will
be at different stages in their life-cycle. The testing network
is probabilistic in that each test has a probability of being
passed. If at any stage a product fails a test, it is abandoned.
The manufacturing process is assumed to have existing
capacity as well as potential new capacity. So, overall, the
decisions to be taken are:
• the selection of products for testing;
• the assignment of resources to testing tasks and any out-
sourcing decisions;
• task sequencing;
• selection of new plants or expansions of existing ones
(including timings of expansions); and
• production planning.
However, production only takes place if a product success-
fully completes its tests. This results in a stochastic problem.
Since the uncertainties are discrete, this is well represented
by a scenario tree. A large scale MILP results; this is solved
by a Lagrangean decomposition scheme.
3.4. Risk in pharmaceutical supply chain infrastructure
decisions
It is clear that much of the infrastructure-related work (in
particular, the product selection and development and ca-
pacity planning decisions) is subject to considerable risks.
These include product failures during trials, product with-
drawal during sales due to side effects, uncertainties about
?nal dosage and treatment regimes, competition from sim-
ilar products, etc. These are in addition to the background
demand uncertainty associated with the indication.
Most strategic/infrastructural decisions have historically
been based on NPV or some form of expected NPV, which
in turn utilise weighted average costs of capital or some re-
quired return on investment. The problem with the expected
NPV measure in this context is that the risks tend to be few,
signi?cant and discrete in nature; the expected NPV term is
better suited to the situation where uncertainties are many
(each of which is relatively “small”) and continuously dis-
tributed. Keynes succinctly summarises the problem with
the expected NPV approach in this context as being based
on the assumption that “a certain state of mediocrity is as
desirable as an even chance of heaven or hell”.
There are a number of approaches to extending the ex-
pected NPV metric to deal with risk. Perhaps the most clas-
sical of these is the employment of a “mean–variance” type
objective function (see, e.g. Mulvey, Rosenbaum, & Shetty,
1997), which is of the form:
max Z = ?mean(r) ?(1 ??)var(r)
where r is the reward from the project and ? is a parameter
to trade-off the relative importance of the expected return
and its variability.
This form is not particularly well suited to the pharma-
ceutical industry for at least three reasons:
(i) the focus in this sector is particularly on downside risk
rather than variability of return;
(ii) the distributions of reward associated with a particular
project tend to be bimodal; with one mode re?ecting
failure at some stage and the other success; and
(iii) the one-sided nature of the NPV measure means that
a large variance is not necessarily a bad thing if the
downside risk is low.
Alternative formulations focus on downside risk, and can
be of the form of constraints enforcing a maximum proba-
bility of the reward being less than a particular ?gure (see,
e.g. Kall & Wallace, 1994), i.e. enforcing:
Prob{(r) ? r
0
} ? ?
where r
0
is a minimum threshold return and ? is the max-
imum allowed probability that the actual reward r is below
r
0
. In many cases, the most important sources of uncertainty
are discrete (due to product failure). This gives rise to a dis-
crete outcome space, and a constraint can then be imposed
on the worst case, i.e. the scenario that leads to the lowest
reward.
Eppen, Martin, and Schrage (1987) develop a “risk factor”
based on expected downside risk. This gives a measure of
the failure to meet a certain target pro?t. The risk factor is
easiest understood in a discrete scenario context. Here, it is
calculated as:
RF =
k:r
k
?r
0
Pr
k
(r
0
?r
k
)
where k is the scenario index and Pr
k
is the probability of
scenario k. An upper bound can then be enforced on RF.
By tightening the constraint on expected downside risk, it
is possible to bring alternative solutions to the attention of
the decision-makers. This is signi?cant for problems that
contain many different solutions. The higher the probability
of occurrence of a reward below r
0
, the higher RF is, and
the risk involved in the project is larger. If RF is close to
zero, then the risk associated with each investment decision
is very low, making the investment more attractive from a
risk perspective. This risk model is used in the problem of
capacity planning for products at different stages in clinical
trials by Gatica et al. (2002a).
Applequist, Pekny, and Reklaitis (2000) describe a risk
premium approach which sets out to ?nd the right balance
between the expected value of a set of decisions and the as-
sociated risk (captured by the variance in this case). The ex-
pected return of an investment decision is compared to one
in the ?nancial market with a similar variance (e.g. govern-
ment bonds, large company stocks, small company stocks)
or a model that ?ts the expected return–variance correlation.
For any investment decision to be approved, its expected re-
ward for the associated variance should be better than that
936 N. Shah / Computers and Chemical Engineering 28 (2004) 929–941
possible in the market. They tackle the problem of determin-
ing capital investments and production plans for a process
with uncertain (continuously distributed) product demands.
The problem of evaluating the expected return and its vari-
ance is very complicated and the polytope volume integra-
tion procedure developed is one of the key contributions.
Bhagwat and Griggs (1995) undertook a study of the risk-
ness of the industry—this was prompted by the perceived
above normal rates of return associated with the industry.
They surmised that using a risk premium alone will not suf-
?ce as there are also some systematic risks over and above
those captured by a market measure like risk premium. They
utilise the capital asset pricing model (CAPM) to estimate
the systematic risk. Here, the required rate of return from an
asset i is
K
i
= R +?
i
(E(r
m
) ?R)
where R is the riskless rate of return, E(r
m
) is the expected
market rate of return and E(r
m
) ? R is the risk premium. ?
i
can be thought of as the systematic risk associated with the
ith asset. For a systematic risk to be present, this parameter
should have a value large than 1. In their study of the US
pharmaceutical industry from 1963 to 1992, Bhagwat and
Griggs found an average value of 1.05, with recent values
being higher than in the past.
Booth (1999), on the other hand, questions the use of the
CAPM to set the required return as it is merely a hypothesis
and there is uncertainty about its central premise, namely
that portfolios of investments with higher risks will show
higher expected returns.
Myers (1999) notes that the cost of capital must also be
related to the stage in the life-cycle of the associated prod-
uct(s).
A promising area of research to augment these is the ap-
plication of real options theory to this area. Lerwent (1994)
has considered the case where the initiation of an R&D ac-
tivity can be seen as purchasing an option to continue with
development until further information is obtained. The pric-
ing of the option in this context is a very complicated and
open research issue.
3.5. Process development and plant design
The problem of process development in this sector has re-
cently been reviewed by Shah, Samsatli, Sharif, Borland, and
Papageorgiou (2000). They contend that the current practice
of relying on traditional manufacturing technology means
that processes are designed to be operated in potentially in-
effective ways. A hierarchical approach to designing pro-
cesses that are not constrained by traditional equipment is
recommended by the authors.
Linninger, Ali, and Stephanopoulos (1996) also propose
a hierarchical approach, with the emphasis on the use of
knowledge bases and material balancing at every develop-
ment level to choose and assess options. The focus is partic-
ularly on synthesising routes and developing processes with
low environmental impacts. The integration of these meth-
ods into a software environment supporting process devel-
opment is described by Stephanopoulos, Ali, Linninger, and
Salomone (2000).
The problem of multipurpose and multiproduct batch
plant design is particularly relevant to this area. There has
been a large amount of work from 1970 to the present
date (see, e.g. Barbosa-Povoa & Pantelides, 1999; Henning,
Camussi, & Cerda, 1994; Papageorgaki & Reklaitis, 1990;
Shah & Pantelides, 1991, 1992; Sparrow, Forder, & Rippin,
1975; Voudouris & Grossmann, 1992; Yeh & Reklaitis,
1987). Certainly there are a number of good solutions to
the problem of allocating the best set of equipment to pro-
cessing tasks assuming campaign operation and batch pro-
cessing. A set of interesting challenges in process and plant
design remain ahead of us: the design of novel, intensi?ed
equipment which will allow processes to operate at intrinsic
rates, and increase manufacturing velocities by orders of
magnitude (seehttp://www.britest.co.uk). These will need
to take into account the regulatory constraints which are
well served by relatively slow batch processes. However,
the new FDA process analytical technology initiative will
support the development of such novel processes.
3.6. Production planning and scheduling
This has also received a lot of attention over the period
from 1970 to the present day (see, e.g. reviews by Reklaitis,
1991; Shah, 1998). However, most of this is on short-term,
order-driven scheduling. Most of the work on campaign
planning was done some time ago (e.g. Lazaro, Espuna, &
Puigjaner, 1989; Mauderli & Rippin, 1979, 1980;
Papageorgiou & Pantelides, 1996; Shah & Pantelides,
1991). However the active ingredient production process,
which is effectively the rate-limiting step of the supply
chain still tends to operate this way. The optimal planning
of campaigns within the context of the performance of the
supply chain as a whole (in particular, trading off the cost
and inconvenience of changeovers with overall supply chain
responsiveness and inventory) has not really been studied
for this type of industry.
3.7. Supply chain simulation and dynamics
The long supply chain and the fact that there are many
decision-making agents means that understanding the dy-
namic behaviour will be very important. We have devel-
oped a generic approach to the modelling of supply chain
dynamics and applied it to some pharmaceutical processes
(Gjerdrum, Jalisi, Papageorgiou, & Shah, 2000). We have
not found any other work reported in the literature on this
topic, but a similar simulation study of the food supply chain
was undertaken by van der Vorst, Beulens, and van Beek
(2000).
Our work is concerned with the modelling of both the
physical processes (primary and secondary manufacturing,
N. Shah / Computers and Chemical Engineering 28 (2004) 929–941 937
distribution and warehousing) as well as the business pro-
cesses. By the latter, we mean how decisions are taken
at the different nodes of the chain, who takes them, what
tools/methods are used, etc. The aim is to replicate the be-
haviour of the supply chain in software. This means that the
logic of software tools used for decision-making at various
nodes (e.g. DRP and MRP methods) are replicated in our
simulation tool. It is clear that a purely analytical model can-
not capture this information easily. The aim of this approach
is to suggest non-invasive improvements to the operation of
the supply chain (i.e. neither the physical or IT infrastructure
should be modi?ed). Such improvements may come about
through changes in parameters (e.g. safety stocks) or busi-
ness processes (e.g. relationships between agents).
Once a model has been developed, it is validated against
historical data and then used to perform a variety of what–if
studies. In order to assess future performance, uncertainties
need to be taken into account. At the operational level, these
include product demands, process yields, processing times,
transportation lead times, etc. A stochastic simulation ap-
proach that samples from the uncertain parameters is a use-
ful way of determining expected future performance as well
as con?dence limits on future performance measures (e.g.
service levels). The results of two such studies are described
below.
In the ?rst study, a peculiar dynamic behaviour was seen
in the market warehouse. Although the background demand
for the product was very stable, the manufacturer’s ware-
house experienced highly ?uctuating demands, and needed
to hold considerable inventories to buffer against this. Upon
some investigation, the reason related to a pricing cycle
which caused wholesalers to try to anticipate price increases
and request large pre-emptive orders. Of course, once these
are received, the wholesaler will not order material again
for some time. We used singular-value decomposition tech-
niques to extract the historical dynamics and used them to
generate forward forecasts. We compared the future sup-
ply chain performance using this model against a model
that used collaborative planning between manufacturer and
wholesaler. The key metric was the amount of ?nished goods
safety stock cover the manufacturer required to meet a cer-
tain customer service level. The results may be compared
below (Figs. 1 and 2).
To achieve the target service level, the ?nished goods
stock can be approximately halved in the collaborative case.
It is clear that signi?cant bene?ts are possible through an
alternative way of running the supply chain. Conservative
estimates would put these at $30 m in one-off inventory sav-
ings and $3.6 m p.a. savings. Of course, all the relevant rea-
sons for holding stock (e.g. cycle times, manufacturing facil-
ity reliability, forecast accuracies) must be included in such
models.
Another study considered the effects not of the production
or inventory aspects, but the quality control (QC) procedures.
As mentioned earlier, there is a prevalence of QC activities
in the industry, although they are not really necessary at
Service Level
90
92
94
96
98
100
weeks cover
%
Fig. 1. Variation of service level with ?nished goods safety stock for the
non-collaborative case.
all the points where currently used. These steps account for
signi?cant dead time in the process; often of the order of 1 to
2 weeks, when all the intervening processes are considered.
We developed a model of a process which has ?ve primary
synthesis stages and two secondary manufacturing sites. The
as–is process has QC activities at the end of each primary
stage, and for the ?nal product. The modi?ed process has a
QC step for the AI and a QC step for the ?nal product. The
results for one of the products are compared below.
In Fig. 3, the forward prediction of ?nished goods inven-
tory of pack C is quite smooth. The lower con?dence (95%)
limit on the pro?le is still positive, giving con?dence that
stock-outs are very unlikely. In Fig. 4, the case with QC at all
stages, there is much less certainty in the inventory (the vari-
ance grows signi?cantly with time) and the lower con?dence
level goes to zero. In terms of customer service performance
measures, over a 2-year period, the average service level in
the case with QC is 91% and the probability of a stockout
in any week is 5%. On the other hand, in the low QC case,
these ?gures are 100 and 0%, respectively. Clearly, as pro-
cess development and design advance, the comfort provided
by QC at so many stages in the supply chain will not be
required and the dynamic behaviour will improve markedly.
Overall, the simulation-based approach is very promis-
ing for studying the large and complex supply chains in-
volved. Other studies include when to plan shutdowns in
Service Level
90
92
94
96
98
100
102
weeks cover
%
Fig. 2. Variation of service level with ?nished goods safety stock for the
collaborative case.
938 N. Shah / Computers and Chemical Engineering 28 (2004) 929–941
Base Case for Pack C
0
50000
100000
150000
200000
250000
300000
0 20 40 60 80 100
Time / weeks
mean quantity
safety stock
95% upper confidence
95% lower confidence
P
a
c
k
C
s
t
o
c
k
l
e
v
e
l
/
p
a
c
k
s
Fig. 3. Time pro?le of expected inventory of ?nished goods, including con?dence intervals, for the case with quality control at only two points.
Results for Pack C with QC (Monte Carlo 400 simulations)
0
50000
100000
150000
200000
250000
300000
0 20 40 60 80 100
Time / weeks
mean quantity
safety stock
95% upper confidence
95% lower confidence
P
a
c
k
C
s
t
o
c
k
l
e
v
e
l
/
p
a
c
k
s
Fig. 4. Time pro?le of expected inventory of ?nished goods, including con?dence intervals, for the case with quality control at all stages.
manufacturing sites and strategies for new product introduc-
tion (Gjerdrum, Shah, & Papageorgiou, 2001).
4. Future challenges
The future challenges in this area are broad and complex,
and will provide fertile ground for research. They can be
categorised under three headings:
• improvements to existing processes;
• improvements to the strategic decision-making process;
and
• future scenarios.
4.1. Improvements to existing processes and operations
The supply chain includes many agents, often with dif-
ferent objectives. Their internal dynamics tend to exagger-
ate the external market dynamics and result in detriments
in performance. This is an area where collaborative fore-
casting, planning and inventory management will be very
useful. Here, the different agents in the supply chain will
co-ordinate activities across the chain. One of the main rea-
sons for the current, more distributed practice is the large
scale of the operations, both in terms of activities and geo-
graphical span. Multisite planning and scheduling tools are
required to support a collaborative planning activity. The
supply chain generally contains larger amounts of inventory
N. Shah / Computers and Chemical Engineering 28 (2004) 929–941 939
than might be necessary if a more co-ordinated approach is
followed, with the right supporting tools.
Probably the most important metric to track and try to
improve is that of the overall supply chain cycle time. As
mentioned earlier, ?gures of 100–300 days are common. Our
experience in the industry is that efforts to reduce this are
very effective, but they have been applied in a piecemeal
fashion to certain products, while the measure deteriorates
for others. A systematic analysis of the components of the
cycle time should be undertaken for a wide range of prod-
ucts. Clearly, large scale simulation tools that take account
of both physical and business processes would be useful in
this context.
4.2. Improvements to the strategic decision-making
process
We can consider this in a bottom-up fashion. The current
nature of the process technology is one the main supply
chain bottlenecks. There is a need for more agile equip-
ment which will shorten process cycle times by an order
of magnitude and require minimal time for cleaning and
changeover. This will avoid long campaigns and should
lead to “pull”-based active ingredient manufacturing, and
therefore more responsive supply chains. The underlying
processes will have to change as well, with the focus being
on designing processes that operate at intrinsic rates (e.g.
being limited by reaction kinetics) rather than being limited
by equipment performance (e.g. heat transfer, mass trans-
fer or mixing characteristics) of traditional equipment. In
principle, the low tonnages involved should lead to much
less capital intensive plants if this is achieved. Generally
speaking, signi?cant improvements to manufacturing tech-
nology have not been of the highest priority in this ?eld to
date.
Processes should be designed with a much greater level
of mechanistic understanding and controlled tightly if re-
ductions in quality control activities are to be possible. The
impact of this has been demonstrated earlier.
The integration of development management and capacity
and production planning will be very important. Currently,
capacity issues are often not considered at the development
stage. Booth (1999) lists three undesirable outcomes of not
co-ordinating these activities:
(i) shortages of materials for pre-clinical studies;
(ii) shortages of materials for clinical trials; and
(iii) delays in time-to market—which is not only dependent
on having material available, but also on generating
demand at the late stages of development.
The approach of Maravelias and Grossmann (2001)
is a very promising platform to treat the overall piple-
line/capacity planning problem, but it needs to be integrated
with a sophisticated treatment of risk (e.g. through the use
of real options theory) and economics (e.g. taking account
of local taxation regimes, transfer pricing, duty drawbacks,
etc.), both of which have a very signi?cant impact on in-
vestment decisions. Furthermore, the modelling of testing
and trials needs to be extended to account for variations
in standard outcomes, e.g. very successful trials resulting
in short-circuiting of the approvals process in the case of
life-saving drugs.
Generally speaking, the development of integrated models
of the life-cycle, from discovery through to consumption
would greatly facilitate strategic decision-making.
4.3. Future scenarios
Companies have recently moved away fromproduct diver-
si?cation and locally adapted products. The common pack-
aging and labelling standards in the European Union, for ex-
ample, have supported this. However, there has been much
development in the ?eld of “pharmacoeconomics” which
might generate pressures to reverse this trend. This discipline
helps to make choices in treatment options by consideration
of costs and outcomes (clinical, economic and humanistic).
An important outcome of this type of analysis will be the
insistence on local solutions to local problems (Thwaites &
Townsend, 1998).
Another trend, somewhat further down the line, will
come out of genetic research and which will identify tar-
get sub-populations for different treatment regimes (the
so-called “designer drugs”). These two drivers will give rise
to considerable product and supply chain complexity. The
current manufacturing processes and supply chains are not
well con?gured to cope with this. Primary manufacturing
tends to operate in campaigns; and secondary manufactur-
ing batchsizes are typically 1–4 million tablets. Clearly,
the manufacturing processes and supply chains will have
to be re-designed if product customisation is to increase in
line with these trends. The manufacturers will have to track
supply chain performance measures very carefully in order
to understand the cost-to-serve for a diverse customer base.
An emerging area is that of rapid response vaccines and
other treatments arising out of possible emergencies (e.g.
bioterrorism or very fast developing epidemics). Again the
traditional supply chain (particularly for vaccines) is very
slow and unresponsive. If national governments are to im-
plement emergency preparedness programmes, the entire in-
frastructure must be well designed and tested through sim-
ulation. Decisions such as where to manufacture, in what
quantities, where to hold stocks, where people should report,
etc. need to be taken in a robust fashion. The issue of how
well the supply chain measures up to emergency prepared-
ness concerns is being raised in various fora at the present
(see, e.g. Anon, 2002).
Companies are investing in the development of crops that
are designed in some way to produce pharmaceuticals; this
will give rise to new research activities in process (in par-
ticular recovery steps) innovation, novel equipment design
(e.g. supercritical separations) and of course in a new type
of supply chain to optimise.
940 N. Shah / Computers and Chemical Engineering 28 (2004) 929–941
There are also some changes afoot in the industry struc-
ture as well—the growth in medicinal chemistry, biotech-
nology and genomics will spread IP around and result in
looser, virtual enterprises of joint ventures, alliances, etc.
Including as well the general tendency towards outsourcing
of manufacturing, this will give rise to complex extended
supply chain co-ordination. Lessons may be learnt from the
automotive, PC and consumer electronics world, where such
supply chains already operate.
5. Conclusions
The pharmaceutical supply chain used to be seen as a tool
to supply products to market in an effective way, where the
emphasis was on security of supply. Recent changes in the
operating environment mean that companies are revisiting
the components of their supply chains and identifying ways
of extracting additional bene?ts from them.
In this sector in particular, the supply chain of interest is
not simply the physical processes of conversion and distri-
bution of materials. Equally important is the “value-chain”
perspective of managing the innovation and development
processes through to capacity and production planning.
There are still several exciting research challenges in this
value chain, many of which the process engineering/process
systems engineering community are well placed to address.
Acknowledgements
The author is very grateful to Gabriel Gatica, Jonatan
Gjerdrum and Lazaros Papageorgiou for their contributions
to this work.
References
Anon (2002). Summary of the executive session on emergency pre-
paredness and the pharmaceutical supply chain. American Journal of
Health-System Pharmacy, 59 (3), 247–253.
Applequist, G. E., Pekny, J. F., & Reklaitis, G. V. (2000). Risk and uncer-
tainty in managing chemical manufacturing supply chains. Computers
of Chemical Engineering, 24, 2211–2222.
Ballance, R., Pogany, J. & Forstner, H. (1992). The world’s pharmaceutical
industries. London, UK: Edward Elgar [for UNIDO].
Barbosa-Povoa, A. P. F. D., & Pantelides, C. C. (1999). Design of multi-
purpose production facilities: A RTN decomposition-based algorithm.
Computers of Chemical Engineering, 23, S7–S10.
Bhagwat, Y., & Griggs, F. T. (1995). Analysis of riskiness of pharmaceuti-
cal industry ?rms. Journal of Research in Pharmaceutical Economics,
6, 65–76.
Blau, G., Mehta, B., Bose, S., Pekny, J., Sinclair, G., Keunker, K., &
Bunch, P. (2000). Risk management in the development of new prod-
ucts in highly regulated industries. Computers of Chemical Engineer-
ing, 24, 1005–1011.
Booth, R. (1999). The global supply chain. FT healthcare management
report. London: Financial Times Business Ltd.
Butler, R. (2002). The end of the blockbuster. Chemistry & Industry, 9,
9–10.
Clay, R. L., & Grossmann, I. E. (1997). A disaggregation algorithm for the
optimization of stochastic planning models. Computers of Chemical
Engineering, 21, 751–774.
Eli Lilly (2003).http://www.newsroom.lilly.com/news/Financial/2003-01-
23 q402sales fullyear 2002.html.
Eppen, G. D., Martin, R. K., & Schrage, L. (1987). A scenario approach
to capacity planning. Optical Research, 37, 517–527.
Gatica, G., Shah, N., & Papageorgiou, L. G. (2001). Capacity planning
under clinical trials uncertainty for the pharmaceutical industries. In
Proceedings of the ESCAPE-11 (pp. 865–870) Kondig, Denmark.
Gatica, G., Papageorgiou, L. G., & Shah, N. (2002a). Capacity planning
under uncertainty for the pharmaceutical industry. Transformation of
Institute of Chemical Engineering: Part A.
Gatica, G., Papageorgiou, L. G., & Shah, N. (2002b). An aggregation ap-
proach for capacity planning under uncertainty for the pharmaceutical
industry. FOCAPO-2003.
Gjerdrum, J., Jalisi, Q. W. Z., Papageorgiou, L. G., & Shah, N. (2000).
Dynamic simulation of physical and business processes for supply
chain improvement. In Proceedings of the 5th Annual Conference on
Industrial Engineering Theory, Applications and Practice. Hsinchu,
Taiwan.
Gjerdrum, J., Shah, N., & Papageorgiou, L. G. (2001). New product in-
troduction supply chain planning. In O.-P. Hilmola, editor. Contempo-
rary Research Issues in New Product Introduction. Acta Wasaensia,
77, 25–49.
Grabowski, H. (1997). The effect of pharmacoeconomics on company
research and development decisions. Pharmacoeconomics, 11, 389–
397.
Henning, G. P., Camussi, N. B., & Cerda, J. (1994). Design and plan-
ning of multipurpose plants involving nonlinear processing networks.
Computers of Chemical Engineering, 18, 129–152.
Jain, V., & Grossmann, I. E. (1999). Resource-constrained scheduling
of tests in new product development. Industrial Engineering and
Chemical Research, 38, 3013–3026.
Kall, P., & Wallace, S. W. (1994). Stochastic programming. New York:
Wiley.
Lazaro, M., Espuna, A., & Puigjaner, L. (1989). A comprehensive ap-
proach to production planning in multipurpose batch plants. Comput-
ers of Chemical Engineering, 13, 1031–1047.
Lerwent, J. (1994). The new pharmaceutical paradigm: scienti?c manage-
ment at Merck. Harvard Business Review, Jan-Feb, 88–89.
Linninger, A. A., Ali, S., & Stephanopoulos, G. (1996). Knowledge-based
validation and waste management of batch pharmaceutical process
designs. Computers of Chemical Engineering, 20, S1431–S1436.
Maravelias, C. T., & Grossmann, I. E. (2001). Simultaneous planning for
new product development and batch manufacturing facilities. Industrial
Engineering of Chemical Research, 40, 6147–6164.
Mauderli, A., & Rippin, D. W. T. (1979). Production planning and schedul-
ing for multipurpose batch chemical plants. Computers of Chemical
Engineering, 3, 199–206.
Mauderli, A., & Rippin, D. W. T. (1980). Scheduling production in mul-
tipurpose batch plants: the Batchman program. Chemical Engineering
Progress, 4, 37–45.
Moser, M., Calderari, G., & Morini, P. (2000). Cleaning validation of a
multipurpose plant for active pharmaceutical ingredient bulk produc-
tion. Chimia, 54, 731–733.
Mulvey, J. M., Rosenbaum, D. P., & Shetty, B. (1997). Strategic ?nancial
risk management and operations research. EJOR, 97, 1–16.
Myers, S. (1999). Measuring pharmaceutical risk and the cost of capital.
In J. Sussex, & N. Marchant, editors. Risk and return in the pharma-
ceutical industry (pp. 59–76). London: OHE.
Papageorgaki, S., & Reklaitis, G. V. (1990). Optimal design of multipur-
pose batch plants. Industrial Engineering of Chemical Research, 29,
2054–2062.
Papageorgiou, L. G., & Pantelides, C. C. (1996). Optimal campaign plan-
ning/scheduling of multipurpose batch/semicontinuous tasks. 1. Math-
N. Shah / Computers and Chemical Engineering 28 (2004) 929–941 941
ematical formulation. Industrial Engineering of Chemical Research,
35, 488–509.
Papageorgiou, G. E., Rotstein, G.E. & Shah, N. (2001). Strategic sup-
ply chain optimization for the pharmaceutical industries. Industrial
Engineering of Chemical Research 40, 275–286.
Reklaitis, G. V. (1991). Perspectives on scheudling and planning of process
operations. In Proceedings of the PSE-1991. Montebello, Canada.
Rotstein, G. E., Papageorgiou, L. G., Shah, N., Murphy, D. C., & Mustafa,
R. (1999). A product portfolio approach in the pharmaceutical industry.
Computers of Chemical Engineering, 23, 883–886.
Schmidt, C. W., & Grossmann, I. E. (1996). Optimization models for the
scheduling of testing tasks in new product development. Industrial
Engineering of Chemical Research, 35, 3498–3510.
Shah, N. (1998). Single- and multisite planning and scheduling: current
status and future challenges. AIChE Symposium Series, 320(94), 75–
90.
Shah, N., & Pantelides, C. C. (1991). Optimal long-term campaign plan-
ning and design of batch plants. Industrial Engineering of Chemical
Research, 30, 2308–2321.
Shah, N., & Pantelides, C. C. (1992). Design of multipurpose batch
plants with uncertain production requirements. Industrial Engineering
of Chemical Research, 31, 1325–1337.
Shah, N., Samsatli, N. J., Sharif, M., Borland, J. N., & Papageorgiou,
L. G. (2000). Modelling and optimisation for pharmaceutical and ?ne
chemical process development. AIChE Symposium Series, 323(96),
31–45.
Sparrow, R. E., Forder, G. J., & Rippin, D. W. T. (1975). The choice
of equipment sizes for multiproduct batch plants: heuristics versus
branch-and-bound. Industrial Engineering and Process Design Devel-
opment, 14, 197–203.
Stephanopoulos, G., Ali, S., Linninger, A., & Salomone, E. (2000). Batch
process development: challenging traditional approaches. AIChE Sym-
posium Series, 323(96), 46–57.
Subramanian, D., Pekny, J. F., & Reklaitis, G. V. (2001). A
simulation-optimization framework for Research and Development
Pipeline management. AIChE Journal, 47, 2226–2242.
Thwaites, R., & Townsend, R. J. (1998). Pharmacoeconomics in the new
millenium. Pharmacoeconomics, 13, 175–180.
van der Vorst, J. G. A. J., Beulens, A. J. M., & van Beek, P. (2000).
Modelling and simulating multi-echelon food systems. EJOR, 122,
354–366.
Voudouris, V. T., & Grossmann, I. E. (1992). Mixed-integer
linear-programming reformulations for batch process design with dis-
crete equipment sizes. Industrial Engineering of Chemical Research,
31, 1315–1325.
Yeh, N. C., & Reklaitis, G. V. (1987). Synthesis and sizing of
batch/semicontinuous processes: single product plants. Computers of
Chemical Engineering, 11, 639–654.
doc_582633281.pdf
Supply chain optimisation is now a major research theme in process operations and management. A great deal of research has been undertaken on facility location and design, inventory and distribution planning, capacity and production planning and detailed scheduling. Only a small proportion of this work directly addresses the issues faced in the pharmaceutical sector. On the other hand, this sector is very much ready for and in need of sophisticated supply chain optimisation techniques
Computers and Chemical Engineering 28 (2004) 929–941
Pharmaceutical supply chains: key issues and strategies for optimisation
Nilay Shah
?
Centre for Process Systems Engineering, Department of Chemical Engineering, Imperial College of Science, Technology and Medicine, London SW7 2BY, UK
Abstract
Supply chain optimisation is now a major research theme in process operations and management. A great deal of research has been
undertaken on facility location and design, inventory and distribution planning, capacity and production planning and detailed scheduling.
Only a small proportion of this work directly addresses the issues faced in the pharmaceutical sector. On the other hand, this sector is very
much ready for and in need of sophisticated supply chain optimisation techniques.
At the supply chain design stage, a particular problem faced by this industry is the need to balance future capacity with anticipated demands
in the face of the very signi?cant uncertainty that arises out of clinical trials and competitor activity. Ef?cient capacity utilisation plans and
robust infrastructure investment decisions will be important as regulatory pressures increase and margins are eroded. The ability to locate
nodes of the supply chain in tax havens and optimise trading and transfer price structures results in interesting degrees of freedomin the supply
chain design problem. Prior even to capacity planning comes the problem of pipeline and testing planning, where the selection of products
for development and the scheduling of the development tasks requires a careful management of risk and potential rewards.
At the operation stage, it is often dif?cult to ensure responsiveness. Most pharmaceutical products involve primary active ingredient (AI)
production (often multi-stage chemical synthesis or bioprocess) and secondary (formulation) production. Both of the stages are characterised
by low manufacturing velocities and are hampered by the need for quality assurance activities at several points. It is not unusual for the
overall supply chain cycle time to be 300 days. In this environment, supply chain debottlenecking and decoupling strategies together with
co-ordinated inventory management are crucial for quick responses to changing market trends. A good understanding of what actually drives
the supply chain dynamics is also required. As often as not, erratic dynamics are introduced by business processes rather than by external
demand, and may be eliminated by the re-design of internal business processes or supplier/customer relationships.
This paper will consider important issues in supply chain design and operation drawn fromthe literature and fromour collaborative research
projects in this area. The main features of the problems will be reviewed as will the literature to date. Some strategies for solution will be
identi?ed, as will some future research needs.
© 2003 Elsevier Ltd. All rights reserved.
Keywords: Pharmaceutical industries; Supply chain optimisation; Pipeline management
1. Introduction
The pharmaceutical industry can be de?ned as a complex
of processes, operations and organisations involved in the
discovery, development and manufacture of drugs and med-
ications.
The World Health Organisation (WHO) de?nes a drug or
pharmaceutical preparation as:
any substance or mixture of substances manufactured,
sold, offered for sale or represented for use in the diag-
nosis, treatment, mitigation or prevention of disease, ab-
normal physical state or the symptoms thereof in man or
?
Fax: +44-171-594-6606.
E-mail address: [email protected] (N. Shah).
animal; [and for use in] restoring, correcting or modifying
organic functions in man or animal.
This is a very wide de?nition, and correspondingly, there
are number of key players in the pharmaceutical industry,
including:
(i) The large, research and development-based multina-
tionals with a global presence in branded products, both
ethical/prescription and over-the-counter. They tend to
have manufacturing sites in many locations.
(ii) The large generic manufacturers, who produce
out-of-patent ethical products and over-the-counter
products.
(iii) Local manufacturing companies that operate in their
home country, producing both generic products and
branded products under licence or contract.
0098-1354/$ – see front matter © 2003 Elsevier Ltd. All rights reserved.
doi:10.1016/j.compchemeng.2003.09.022
930 N. Shah / Computers and Chemical Engineering 28 (2004) 929–941
(iv) Contract manufacturers, who do not have their own
product portfolio, but produce either key intermediates,
active ingredients (AI) or even ?nal products by pro-
viding outsourcing services to other companies.
(v) Drug discovery and biotechnology companies, often
relatively new start-ups with no signi?cant manufactur-
ing capacity.
Most of the material in this paper is particularly relevant
to the ?rst group. This group dominates the marketplace and,
due to the global nature of the enterprises involved, tends to
have the most challenging supply chain problems.
1.1. The changing circumstances of the industry
In the recent past, the high returns on investment and
high turnovers from “blockbuster” products resulted in the
following regime (Booth, 1999):
• good R&Dproductivity, often creating compounds to treat
previously untreatable diseases;
• long effective patent lives of these compounds;
• ability of these patents to provide technological barriers
to entry;
• a limited number of product substitutes in a given thera-
peutic area; and
• a low price sensitivity; supported by the separation be-
tween prescribing and paying responsibilities.
The resulting corporate strategy was to ensure high mar-
gins by exploiting the price inelasticity and invest a large
proportion of the resultant pro?ts in R&D (approximately
25% of sales), in order to ensure a healthy product pipeline.
The more recent circumstances are much more challeng-
ing:
• R&D productivity (in terms of numbers of new chemical
entities (NCE) registered per unit amount of investment)
is declining;
• effective patent lives are shortening;
• even while active, patents provide lower barriers to entry;
• there are many product substitutes in many therapeutic
areas; either alternative compounds (“me-too drugs”) or
off-patent generics; and
• the payers of healthcare are exerting strong price pressure
and in?uencing prescribing practices; this means that in
order to be approved, new drugs must address new thera-
peutic areas or have very signi?cant cost or health bene?ts
over existing treatments.
On the one hand, the global marketplace has become
more liberalised, exposing products to competition. On the
other, governments and other agencies have tended to in-
tervene more as they become concerned at every increasing
healthcare costs associated with ageing populations. Mea-
sures taken include strict controls on the prices of new drugs,
more cost–bene?t analysis, and encouragement of the use
of generic substitutes or alternatives where possible.
A further weakness that will hamper the large players
in the area is the historical dependence on “blockbuster”
drugs. A recent report by Datamonitor predicts that only 4
of the 19 companies currently selling blockbuster drugs will
be able to maintain double digit growth between 2001 and
2008 (Butler, 2002). This dependence is illustrated by the
following ?gures (Butler, 2002):
• Eli Lilly’s net pro?ts dropped by 20% after Prozac came
off patent. Overall, Eli Lilly’s sales in 2002 were down
4% and the sales excluding Prozac were up 8% (Eli Lilly,
2003).
• BMS’ patent on Glucophage (with previous annual sales
of $2bn) has expired and its sales are down by 87% in
the ?rst 3 months of 2002 after the launch of generic
substitutes.
• Losec represented 34% of AstraZeneca’s sales in 2001
and it might come off patent this year if the company is
unsuccessful in its legal battle—in which case generics
are expected on the market in late 2002.
1.2. Drivers in the pharmaceutical industry
Probably the single most important driver in the pharma-
ceutical industry is the time-to-market. Companies secure
very signi?cant returns in the early life of a successful drug,
before any competition. The competition-free life is short-
ening, typically from 5 to 1–2 years. Competition in this
sense relates to similar (rather than exactly the same) drugs.
For example, Bayer’s anti-cholesterol drug Baycol was
withdrawn in 2001 due to safety concerns, and the two later
entrants Pravachol (from BMS) and Lipitor (from P?zer) are
now the biggest sellers for their companies (Butler, 2002).
Given the signi?cant potential for adverse health effects,
the industry is subject to very stringent regulation. This starts
from the processes used to evaluate the safety and ef?cacy
of the chemical compounds, through to the details of the
process and plant design and manufacturing operations. The
primary regulator that the companies must satisfy is the US
Food and Drug Administration. It may be the case that the
existence of regulatory protocols has hindered innovation in
this sector; with companies blaming regulators for their own
innate conservatism.
The regulatory process tends to be slow and expensive;
both these effects must be borne by the industry. Further-
more, the complex chemical compounds involved have
more complex manufacturing processes, and the activities
of route investigation, process development, scale-up plant
design/retro?t, commissioning and quali?cation are either
increasing in duration or proving stubborn to shorten.
An estimate of £200–400 m is required to launch a new
drug, and an average of 8–12 years elapses from patent ?ling
to ?rst sale (see, e.g. Grabowski, 1997).
There is a general trend for companies to divest excess
capacity that came about from having many local manu-
facturing sites, and move towards a global supply chain
N. Shah / Computers and Chemical Engineering 28 (2004) 929–941 931
management process. This brings with it many complex
co-ordination issues and much tighter capacity constraints.
Currently, the logistics cost in the sector is relatively high
(Booth, 1999).
Research ef?ciency is declining in the sense that the cost
of each new chemical entity is increasing. Although growth
in investment in R&D has exceeded in?ation over the last
30 years, the global trend are as follows: 844 NCEs were
registered in 1961–1970, 665 in 1971–1980 and 506 in
1981–1990 (Ballance, Pogany, & Forstner, 1992). This has
been one of the main drivers behind the recent series of
mergers and acquisitions in the industry, the long-term ben-
e?ts of which will probably not be felt for some time yet as
R&D activities continue to be consolidated.
Historically, most management attention has been paid to
drug discovery and sales and marketing (the extreme ends
of the supply chain), but now much more attention is being
paid to supply chain optimisation as a means of delivering
value. According to Booth (1999):
• there is a welcome move away from viewing the supply
chain as merely having to deliver security of supply at
minimum cost, to a recognition of its ability to generate
both value for the customer and hence to the shareholder;
and
• restructuring of the supply chain along regional and global
lines will require massive reductions in capacity, which
was acquired in many cases to propitiate national interest
in return for sympathetic pricing.
1.3. The life-cycle of a pharmaceutical product
In order to put this paper in the right context, it is im-
portant to describe the life-cycle of a drug; it is somewhat
different from that of other process industry products.
The research or discovery phase tends to use thousands
of more or less random test compounds against therapeutic
targets. It typically takes about 10 years to result in a po-
tential new drug that is registered. From this point onwards
patent protection applies.
The potential new drug must then be tested for both safety
and ef?cacy. This involves a variety of trials; early on for
toxicity and later on for ability to alleviate symptoms and
remove disease. Finally, the process development activity
comes up with a chemical or biochemical route to manufac-
ture and an associated manufacturing process. This set of
activities typically takes 6–8 years and is usually known as
the development activity.
Finally, the more familiar processes of manufacturing and
distribution follow.
2. Components of the pharmaceutical industry
manufacturing and distribution chain
A typical pharmaceutical supply chain will consist of the
one or more of the following nodes:
(i) primary manufacturing (possibly including contractor
sites);
(ii) secondary manufacturing (possibly including contrac-
tor sites);
(iii) market warehouses/distribution centres;
(iv) wholesalers; and
(v) retailers/hospitals.
2.1. Primary manufacturing
The primary manufacturing site is responsible for the pro-
duction of the active ingredient (AI or API). This normally
involves either several chemical synthesis and separation
stages to build up the complex molecules involved, or fer-
mentation and product recovery and puri?cation in the case
of biochemical processes.
The manufacturing process is characterised by long task
processing times, often rounded to multiples of shifts. Where
multistage processes are operated, considerable inventories
are often held between stages. Furthermore, material froman
intermediate stage must often pass some formof quality con-
trol check before being approved for use downstream in the
process. This can introduce additional delays into the system.
The traditional process technology involves batch equip-
ment and ?exible pipework. The relatively low produc-
tion volumes result in multipurpose plants to spread
the capital cost between products. The need to avoid
cross-contamination of products and requirements for val-
idated cleaning and changeovers results in long downtimes
between products. These have been of order 4 weeks in
the past, but the application of techniques similar to the
single-minute exchange of die (SMED) methods (see, e.g.
Moser, Calderari, &Morini, 2000) applied to the car industry
have reduced these somewhat. These downtimes in turn im-
ply that long campaigns are the norm, otherwise equipment
utilisation is too low. It is not unusual for 1 year’s produc-
tion of a product to be produced in a single campaign, and
the material produced being stored until the next campaign
in the following year. Since most complex pharmaceuticals
are produced through multistage processes, the same often
holds true for the stable intermediates (stage products).
Needless to say, this mode of operation does not lend itself
well to responsiveness, and contributes signi?cantly to some
of the poor supply chain metrics exhibited by this industry.
A further source of complexity (and convenience) is the
use of contractors to manufacture some or indeed all of the
active ingredient stages. This process of outsourcing is a
growing one, as research-oriented companies concentrate on
the discovery and development activities and rely on third
parties’ manufacturing competence. This gives rise to ex-
tended supply chain co-ordination problems.
2.2. Secondary manufacturing
This is concerned with taking the active ingredient pro-
duced at the primary site and adding “excipient” inert
932 N. Shah / Computers and Chemical Engineering 28 (2004) 929–941
materials along with further processing and packaging to
produce the ?nal products, usually in SKU form. For ex-
ample, a product that is sold in pill form would undergo:
(i) granulation: with addition of all the excipient materials;
(ii) compression: forming the pills;
(iii) coating;
(iv) quality control; and
(v) packaging.
The secondary manufacturing locations are often geo-
graphically separate from the primary manufacturing lo-
cations. This is frequently the outcome of tax and transfer
price optimisation within the enterprise. There are often
many more secondary manufacturing sites than primary
ones, serving local or regional markets. Transportation be-
tween sites is of the order of 1 or 2 weeks if by ship (usually
the default mode) and of the order of one or two days if by
air.
Wholesalers play a signi?cant role in this sector. They
tend to be large and few. About 80% of demand ?ows
through this channel in the UK (with three large players ac-
counting for almost all the demand), with the large part of
the remainder going to hospitals. In the US another interme-
diary is growing—the managed care organization (MCO) or
healthcare maintenance organization (HMO).
2.3. Operational issues in the pharmaceutical supply chain
Although the processes will vary between companies,
all major pharmaceutical companies will operate ERP sys-
tems and follow a business process along the following
lines:
• Demand management—in each geographical region, for-
ward forecasts (e.g. 3–24 months) are developed, based
on historical data, market intelligence, etc. Tenders for
manufacture may also be evaluated and possibly accepted
at this stage.
• Inventory management and distribution requirements
planning—the demands determined are aggregated and
imposed on the appropriate warehouse/distribution cen-
tre. The impact on ?nished goods inventory is assessed
and if necessary, orders are placed on upstream secondary
manufacturing sites.
• Secondary production planning and scheduling—the or-
ders placed on the secondary sites are planned (typically
using MRP-II type tools) and then scheduled in detail (typ-
ically using APS tools). The impact of production plans
on active ingredient raw material stocks is evaluated and
if necessary, orders for AI are placed on the upstream.
• Primary manufacturing campaign planning and AI inven-
tory management. Here, the demands placed by secondary
manufacturing are satis?ed by careful management of in-
ventory and production planning.
An interesting feature of this process is that the
customer-facing end is effectively a “pull” process (driven
by orders) but the primary manufacturing stage has long
cycle times which make it dif?cult to ensure end-to-end re-
sponsiveness. This means that primary production is effec-
tively a “push” process, driven by medium- and long-term
forecasts. Relatively large stocks of AI must be held to
ensure good service levels and ensure smooth operation
at the interface of these processes. The well-documented
“bullwhip” or Forrester effect is often felt at the primary
manufacturing site, which is unfortunate since this is the
least responsive part of the supply chain as it normally
operates in campaign mode. This makes it dif?cult to ex-
ploit short-term opportunities (e.g. shortages of supply
of a competitor’s product, tenders for national supplies,
epidemics, etc.).
Another feature of this process is an outcome of its large
scale and geographical span. This is the distributed nature of
decision-making, which can lead to tensions and sub-optimal
decisions. Different nodes are not really aware of upstream
nodes’ resource constraints, and orders may be ?lled in order
of receipt, rather than on an economic basis. Of course,
centralised planning would not be without its dif?culties in
this context.
In our experience, the following supply chain performance
measures are typical of the industry:
• The stock levels in the whole chain (“pipeline stocks”)
typically amount to 30–90%of annual demand in quantity,
and there are usually 4–24 weeks’ worth of ?nished good
stocks.
• Stock turns (de?ned as annual sales/average stock) are
typically between 1 and 8.
• Supply chain cycle times (de?ned as elapsed time between
material entering as raw material and leaving as product)
are often between 1000 and 8000 h.
• The value-added time (time when something happens to
material as a percentage of chain cycle time) is of order
0.3–5%.
• Material ef?ciencies (the amounts of product produced
per unit amount of total materials used) are 1–10%.
The relatively high levels of stock are required to buffer
the slow supply chain against market dynamics.
2.4. Strategic and design issues in the pharmaceutical
supply chain
The decisions to be taken at this level include:
• Pipeline and development management—this involves the
selection of potential drugs to develop further, and the
planning of the development activity.
• Process development—the investigation of manufacturing
routes and the generation of manufacturing processes.
• Capacity planning and plant and supply chain network
design.
• Plant design—the selection and sizing of the major equip-
ment and storage units.
N. Shah / Computers and Chemical Engineering 28 (2004) 929–941 933
Some of the key issues are:
• Uncertainty in the demands for existing drugs (due to
competition, uncertainty in the ability to extend the pro-
tected life through new formulations, etc.).
• Uncertainty in the pipeline of new drugs—in particular,
which ones will be successful in trials, what sort of dosage
and treatment regime will be optimal.
• Process development—this is a complex problem, driven
by chemistry and yield optimisation. It often results in
inef?cient processes that are operated much more slowly
than the intrinsic rates—giving rise to batch processes and
long cycle times responsible for some of the problems
seen at the primary production planning stage.
• Capacity planning—the long lead times to make capac-
ity effective mean that decisions often need to be taken at
times of high uncertainty. Waiting for the uncertainties to
be resolved might delay the time to market by an unac-
ceptable amount.
• Network design—often tax implications take precedence
over logistics issues, these result in economic but poten-
tially complicated supply chains.
• Plant design—this tends to be very traditional, with no
real change in manufacturing technology for 50 years
(the workhorse of the primary manufacturing site is the
glass-lined stainless steel batch reactor). There are signif-
icant opportunities for intensi?ed, continuous processing.
3. Overview of some recent work
The recent work in the literature that is relevant here can
be categorised under these headings:
• pipeline and development management;
• capacity planning;
• simultaneous development and capacity planning;
• process development and plant design;
• production planning and scheduling; and
• supply chain simulation and dynamics.
The areas are reviewed in turn below.
3.1. Pipeline and development management
Schmidt and Grossmann (1996) considered the problem
of sequencing of testing tasks where unlimited resources
are assumed to be available. The key feature of the model
that distinguishes it from classical project scheduling is the
that each task has a probability of failure; this affects the
need for successor tasks. They formulated the problem as a
continuous-time MILP and solved the problem of maximis-
ing the overall expected NPV. If many tests for a product
are performed in parallel, the testing activity will be more
expensive, as the effect of failures on successor tests are not
taken into account. On the other hand, the product may come
to market much earlier, resulting in a much better cash ?ow
pro?le. Conversely, sequential test planning might avoid un-
necessary tests and reduce expense, but result in a later ar-
rival of the product in the marketplace.
In practice, the testing activity tends to be quite
resource-constrained, and may also involve outsourcing of
some stages. Jain and Grossmann (1999) develop a method-
ology for the sequencing and scheduling of testing tasks
under resource constraints. In this approach, each product
has a speci?ed set of testing tasks. Each task is charac-
terised by a duration, cost, precedence constraints, resource
requirements and probability of success. A task may be out-
sourced at a higher cost; in this case no internal resources
are required. The income associated with a product is given
as a function of the time of launch in the market.
The formulation developed is conservative and always
feasible in that the resource constraints are always enforced,
regardless of the probability of a task not actually taking
place. The cost component is modelled as an expected cost.
This ensures that the effect of starting tasks earlier than
necessary is modelled, i.e. that later tasks may not actually
take place due to the failure of the earlier one. Two alter-
native formulations are presented, a continuous-time MILP
and a graph-based one. The latter was found to be more ef-
?cient and was able to cope with problems consisting of 30
tasks.
Blau et al. (2000) consider the problem of risk man-
agement at the development stage. As mentioned earlier,
the development phase selects candidate drugs and takes
them through trials and process development. It is a long,
costly and inherently risky process with a large up-front
commitment. The aim of this work is to support the pro-
cess of product selection and test planning while managing
risk effectively. The development activities are modelled
as a probabilistic activity network, where each activity has
a time, precedence relations, resource requirements and
probability of success.
Risk is de?ned as the adverse consequences of exposure
to uncertainty, and in this context is usually related to the
premature withdrawal of a candidate drug. The risk of a
set of decisions must be balanced against the potential re-
ward. In this case, the potential reward is the expected ?-
nancial returns of drugs that do make it through the de-
velopment process. The risk/reward ratio can then be used
to compare different drug candidates. A screening process
removes any obviously unpromising candidates, and then
the remainder must be sequenced through the development
pipleline. A heuristic approach using simulation with lo-
cal rules in response to trigger events (e.g. failure of a
test) is employed. This aims to process tasks as quickly
as possible and although there is no guarantee of not vio-
lating resource constraints, these violations are usually not
large.
Subramanian, Pekny, and Reklaitis (2001) extend this
work to take explicit account of the resource requirements
of the problem. The problem statement is generalised in that
more sources of uncertainty in the problem are considered,
934 N. Shah / Computers and Chemical Engineering 28 (2004) 929–941
and include:
• task processing times;
• task resource requirements;
• task success probabilities;
• task costs; and
• market returns.
They make the point that a single-level mathematical pro-
gramming problem cannot hope to capture all these fea-
tures. On the other hand, discrete-event dynamic systems
(DEDS) techniques cope well with the stochastic elements,
but require local, myopic rules to resolve con?icts or make
choices as they arise. They therefore developed an integrated
optimisation–simulation framework (SIM–OPT), where a
DEDS simulator reverts to an optimisation layer (with dif-
ferent degrees of optimisation) to resolve con?icts or make
choices such as task sequencing. The optimisation layer is
an MILP which is updated by the latest status of the plant.
The results show that using optimisation far outperforms
the typical local rules used in classical DEDS. By repetitive
simulation, the statistical trends can be tracked and answers
to questions about corporate policy (particularly in relation
to risk and resourcing) can be obtained.
3.2. Capacity planning
The capacity planning under clinical trials uncertainty
problem has recently received some attention in the litera-
ture. The deterministic problem of allocating new manufac-
turing capacity to existing or potential sites around the world
is described by Papageorgiou, Rotstein, and Shah (2001).
They describe the features particular to the pharmaceutical
industry, and emphasise the importance of modelling ?nan-
cial ?ows as the taxation regimes affect the rewards associ-
ated with alternative solutions signi?cantly. Indeed, taxation
considerations can easily dominate the location decisions.
The problem of capacity planning under uncertainty was
considered by Rotstein, Papageorgiou, Shah, Murphy, and
Mustafa (1999). They considered the problem where three
products are at the start of clinical trials, and plans for cur-
rent and future manufacturing capacity are to be made. The
key trade-off in the capacity planning decision comes about
due to the lead time between deciding to invest in additional
manufacturing capacity, and that capacity coming on-stream.
Deferring capacity planning decisions until more informa-
tion is available from trials is obviously a lower-risk strat-
egy, but increases the time to market. As mentioned earlier,
this measure is critical. For example, when Tagamet came
to the market in the 1970s, it was free from competition for
at least 5 years, but now this competition-free time can be as
low as only 1–2 years. Rotstein et al. (1999) use a scenario
tree to capture the outcomes of the trials, and use a two-stage
stochastic programming with recourse formulation to model
the problem. The “here-and-now” decisions related to im-
mediate capacity expansions and the “wait-and-see” deci-
sions depend on trial outcomes and include further capacity
expansions, plant or product abandonment and production
and inventory planning. They show how different options
can be compared using a number of metrics, including ex-
pected NPV, the probability of the NPV being negative, the
worst case scenario, the total demand met of all the potential
products and the total demand met of the products chosen
from the portfolio.
Gatica, Shah, and Papageorgiou (2001), Gatica,
Papageorgiou, and Shah (2002a) extend this work to the
case where different products are at different stages in their
life-cycles, and those that are in trials will complete those
trials at different times. This gives rise to a much more
complicated scenario structure—it is a multistage stochastic
optimisation problem, with each stage re?ecting the com-
pletion of a clinical trial. In contrast to most of the work
reported on trials in the literature, this work uses more than
two outcomes (success or failure) for the completion of the
trial. Based on typical practices in industry, four outcomes
(failure, low, target and high) are used. This means that
four scenarios are required per stage, and for a problem
with N stages (i.e. N products in trials), there is one sce-
nario in the ?rst stage (re?ecting products currently in the
market with well-forecasted demands), four scenarios in the
second stage (re?ecting the four possible outcomes for the
?rst pipeline product to come out of trials), 16 in the third
stage (the combinations of outcomes for the two products
to complete trials) and so on, until the ?nal stage which has
4
N
scenarios. Each scenario has associated with it possible
capacity expansions and production and inventory planning
variables and constraints, so overall the problem becomes
a large scale stochastic programming problem with integer
and continuous decisions. It is solved as its deterministic
equivalent, a large MILP. This is relatively straightforward
to solve for the four-product case the authors report.
Clearly, the approach is limited by the complexity of the
scenario tree, so Gatica, Papageorgiou, and Shah (2002b)
extend it using a scenario aggregation procedure similar to
that of Clay and Grossmann (1997) to enable the solution
of larger problems.
3.3. Simultaneous development and capacity planning
The research reviewed above deals either with the problem
of organizing the development (testing) activities or planning
the capacity investments and future production. Maravelias
and Grossmann (2001) consider the problems of planning
of testing tasks and capacity simultaneously. The aim is to
optimize a performance measure (expected NPV) for the
process as a whole.
This bridges the gap between the two problems and aims
to ensure that the company is ready to produce a product
once testing is complete (if the product is successful).
The testing process is modelled as a set of tasks with tech-
nological precedence constraints, durations and resource re-
quirements. The tasks have two possible outcomes, success
or failure. All tasks other than process development may
N. Shah / Computers and Chemical Engineering 28 (2004) 929–941 935
be outsourced if internal resources do not suf?ce. Since the
method is to be applied at any time in the company’s oper-
ation, it takes account of the fact that different products will
be at different stages in their life-cycle. The testing network
is probabilistic in that each test has a probability of being
passed. If at any stage a product fails a test, it is abandoned.
The manufacturing process is assumed to have existing
capacity as well as potential new capacity. So, overall, the
decisions to be taken are:
• the selection of products for testing;
• the assignment of resources to testing tasks and any out-
sourcing decisions;
• task sequencing;
• selection of new plants or expansions of existing ones
(including timings of expansions); and
• production planning.
However, production only takes place if a product success-
fully completes its tests. This results in a stochastic problem.
Since the uncertainties are discrete, this is well represented
by a scenario tree. A large scale MILP results; this is solved
by a Lagrangean decomposition scheme.
3.4. Risk in pharmaceutical supply chain infrastructure
decisions
It is clear that much of the infrastructure-related work (in
particular, the product selection and development and ca-
pacity planning decisions) is subject to considerable risks.
These include product failures during trials, product with-
drawal during sales due to side effects, uncertainties about
?nal dosage and treatment regimes, competition from sim-
ilar products, etc. These are in addition to the background
demand uncertainty associated with the indication.
Most strategic/infrastructural decisions have historically
been based on NPV or some form of expected NPV, which
in turn utilise weighted average costs of capital or some re-
quired return on investment. The problem with the expected
NPV measure in this context is that the risks tend to be few,
signi?cant and discrete in nature; the expected NPV term is
better suited to the situation where uncertainties are many
(each of which is relatively “small”) and continuously dis-
tributed. Keynes succinctly summarises the problem with
the expected NPV approach in this context as being based
on the assumption that “a certain state of mediocrity is as
desirable as an even chance of heaven or hell”.
There are a number of approaches to extending the ex-
pected NPV metric to deal with risk. Perhaps the most clas-
sical of these is the employment of a “mean–variance” type
objective function (see, e.g. Mulvey, Rosenbaum, & Shetty,
1997), which is of the form:
max Z = ?mean(r) ?(1 ??)var(r)
where r is the reward from the project and ? is a parameter
to trade-off the relative importance of the expected return
and its variability.
This form is not particularly well suited to the pharma-
ceutical industry for at least three reasons:
(i) the focus in this sector is particularly on downside risk
rather than variability of return;
(ii) the distributions of reward associated with a particular
project tend to be bimodal; with one mode re?ecting
failure at some stage and the other success; and
(iii) the one-sided nature of the NPV measure means that
a large variance is not necessarily a bad thing if the
downside risk is low.
Alternative formulations focus on downside risk, and can
be of the form of constraints enforcing a maximum proba-
bility of the reward being less than a particular ?gure (see,
e.g. Kall & Wallace, 1994), i.e. enforcing:
Prob{(r) ? r
0
} ? ?
where r
0
is a minimum threshold return and ? is the max-
imum allowed probability that the actual reward r is below
r
0
. In many cases, the most important sources of uncertainty
are discrete (due to product failure). This gives rise to a dis-
crete outcome space, and a constraint can then be imposed
on the worst case, i.e. the scenario that leads to the lowest
reward.
Eppen, Martin, and Schrage (1987) develop a “risk factor”
based on expected downside risk. This gives a measure of
the failure to meet a certain target pro?t. The risk factor is
easiest understood in a discrete scenario context. Here, it is
calculated as:
RF =
k:r
k
?r
0
Pr
k
(r
0
?r
k
)
where k is the scenario index and Pr
k
is the probability of
scenario k. An upper bound can then be enforced on RF.
By tightening the constraint on expected downside risk, it
is possible to bring alternative solutions to the attention of
the decision-makers. This is signi?cant for problems that
contain many different solutions. The higher the probability
of occurrence of a reward below r
0
, the higher RF is, and
the risk involved in the project is larger. If RF is close to
zero, then the risk associated with each investment decision
is very low, making the investment more attractive from a
risk perspective. This risk model is used in the problem of
capacity planning for products at different stages in clinical
trials by Gatica et al. (2002a).
Applequist, Pekny, and Reklaitis (2000) describe a risk
premium approach which sets out to ?nd the right balance
between the expected value of a set of decisions and the as-
sociated risk (captured by the variance in this case). The ex-
pected return of an investment decision is compared to one
in the ?nancial market with a similar variance (e.g. govern-
ment bonds, large company stocks, small company stocks)
or a model that ?ts the expected return–variance correlation.
For any investment decision to be approved, its expected re-
ward for the associated variance should be better than that
936 N. Shah / Computers and Chemical Engineering 28 (2004) 929–941
possible in the market. They tackle the problem of determin-
ing capital investments and production plans for a process
with uncertain (continuously distributed) product demands.
The problem of evaluating the expected return and its vari-
ance is very complicated and the polytope volume integra-
tion procedure developed is one of the key contributions.
Bhagwat and Griggs (1995) undertook a study of the risk-
ness of the industry—this was prompted by the perceived
above normal rates of return associated with the industry.
They surmised that using a risk premium alone will not suf-
?ce as there are also some systematic risks over and above
those captured by a market measure like risk premium. They
utilise the capital asset pricing model (CAPM) to estimate
the systematic risk. Here, the required rate of return from an
asset i is
K
i
= R +?
i
(E(r
m
) ?R)
where R is the riskless rate of return, E(r
m
) is the expected
market rate of return and E(r
m
) ? R is the risk premium. ?
i
can be thought of as the systematic risk associated with the
ith asset. For a systematic risk to be present, this parameter
should have a value large than 1. In their study of the US
pharmaceutical industry from 1963 to 1992, Bhagwat and
Griggs found an average value of 1.05, with recent values
being higher than in the past.
Booth (1999), on the other hand, questions the use of the
CAPM to set the required return as it is merely a hypothesis
and there is uncertainty about its central premise, namely
that portfolios of investments with higher risks will show
higher expected returns.
Myers (1999) notes that the cost of capital must also be
related to the stage in the life-cycle of the associated prod-
uct(s).
A promising area of research to augment these is the ap-
plication of real options theory to this area. Lerwent (1994)
has considered the case where the initiation of an R&D ac-
tivity can be seen as purchasing an option to continue with
development until further information is obtained. The pric-
ing of the option in this context is a very complicated and
open research issue.
3.5. Process development and plant design
The problem of process development in this sector has re-
cently been reviewed by Shah, Samsatli, Sharif, Borland, and
Papageorgiou (2000). They contend that the current practice
of relying on traditional manufacturing technology means
that processes are designed to be operated in potentially in-
effective ways. A hierarchical approach to designing pro-
cesses that are not constrained by traditional equipment is
recommended by the authors.
Linninger, Ali, and Stephanopoulos (1996) also propose
a hierarchical approach, with the emphasis on the use of
knowledge bases and material balancing at every develop-
ment level to choose and assess options. The focus is partic-
ularly on synthesising routes and developing processes with
low environmental impacts. The integration of these meth-
ods into a software environment supporting process devel-
opment is described by Stephanopoulos, Ali, Linninger, and
Salomone (2000).
The problem of multipurpose and multiproduct batch
plant design is particularly relevant to this area. There has
been a large amount of work from 1970 to the present
date (see, e.g. Barbosa-Povoa & Pantelides, 1999; Henning,
Camussi, & Cerda, 1994; Papageorgaki & Reklaitis, 1990;
Shah & Pantelides, 1991, 1992; Sparrow, Forder, & Rippin,
1975; Voudouris & Grossmann, 1992; Yeh & Reklaitis,
1987). Certainly there are a number of good solutions to
the problem of allocating the best set of equipment to pro-
cessing tasks assuming campaign operation and batch pro-
cessing. A set of interesting challenges in process and plant
design remain ahead of us: the design of novel, intensi?ed
equipment which will allow processes to operate at intrinsic
rates, and increase manufacturing velocities by orders of
magnitude (seehttp://www.britest.co.uk). These will need
to take into account the regulatory constraints which are
well served by relatively slow batch processes. However,
the new FDA process analytical technology initiative will
support the development of such novel processes.
3.6. Production planning and scheduling
This has also received a lot of attention over the period
from 1970 to the present day (see, e.g. reviews by Reklaitis,
1991; Shah, 1998). However, most of this is on short-term,
order-driven scheduling. Most of the work on campaign
planning was done some time ago (e.g. Lazaro, Espuna, &
Puigjaner, 1989; Mauderli & Rippin, 1979, 1980;
Papageorgiou & Pantelides, 1996; Shah & Pantelides,
1991). However the active ingredient production process,
which is effectively the rate-limiting step of the supply
chain still tends to operate this way. The optimal planning
of campaigns within the context of the performance of the
supply chain as a whole (in particular, trading off the cost
and inconvenience of changeovers with overall supply chain
responsiveness and inventory) has not really been studied
for this type of industry.
3.7. Supply chain simulation and dynamics
The long supply chain and the fact that there are many
decision-making agents means that understanding the dy-
namic behaviour will be very important. We have devel-
oped a generic approach to the modelling of supply chain
dynamics and applied it to some pharmaceutical processes
(Gjerdrum, Jalisi, Papageorgiou, & Shah, 2000). We have
not found any other work reported in the literature on this
topic, but a similar simulation study of the food supply chain
was undertaken by van der Vorst, Beulens, and van Beek
(2000).
Our work is concerned with the modelling of both the
physical processes (primary and secondary manufacturing,
N. Shah / Computers and Chemical Engineering 28 (2004) 929–941 937
distribution and warehousing) as well as the business pro-
cesses. By the latter, we mean how decisions are taken
at the different nodes of the chain, who takes them, what
tools/methods are used, etc. The aim is to replicate the be-
haviour of the supply chain in software. This means that the
logic of software tools used for decision-making at various
nodes (e.g. DRP and MRP methods) are replicated in our
simulation tool. It is clear that a purely analytical model can-
not capture this information easily. The aim of this approach
is to suggest non-invasive improvements to the operation of
the supply chain (i.e. neither the physical or IT infrastructure
should be modi?ed). Such improvements may come about
through changes in parameters (e.g. safety stocks) or busi-
ness processes (e.g. relationships between agents).
Once a model has been developed, it is validated against
historical data and then used to perform a variety of what–if
studies. In order to assess future performance, uncertainties
need to be taken into account. At the operational level, these
include product demands, process yields, processing times,
transportation lead times, etc. A stochastic simulation ap-
proach that samples from the uncertain parameters is a use-
ful way of determining expected future performance as well
as con?dence limits on future performance measures (e.g.
service levels). The results of two such studies are described
below.
In the ?rst study, a peculiar dynamic behaviour was seen
in the market warehouse. Although the background demand
for the product was very stable, the manufacturer’s ware-
house experienced highly ?uctuating demands, and needed
to hold considerable inventories to buffer against this. Upon
some investigation, the reason related to a pricing cycle
which caused wholesalers to try to anticipate price increases
and request large pre-emptive orders. Of course, once these
are received, the wholesaler will not order material again
for some time. We used singular-value decomposition tech-
niques to extract the historical dynamics and used them to
generate forward forecasts. We compared the future sup-
ply chain performance using this model against a model
that used collaborative planning between manufacturer and
wholesaler. The key metric was the amount of ?nished goods
safety stock cover the manufacturer required to meet a cer-
tain customer service level. The results may be compared
below (Figs. 1 and 2).
To achieve the target service level, the ?nished goods
stock can be approximately halved in the collaborative case.
It is clear that signi?cant bene?ts are possible through an
alternative way of running the supply chain. Conservative
estimates would put these at $30 m in one-off inventory sav-
ings and $3.6 m p.a. savings. Of course, all the relevant rea-
sons for holding stock (e.g. cycle times, manufacturing facil-
ity reliability, forecast accuracies) must be included in such
models.
Another study considered the effects not of the production
or inventory aspects, but the quality control (QC) procedures.
As mentioned earlier, there is a prevalence of QC activities
in the industry, although they are not really necessary at
Service Level
90
92
94
96
98
100
weeks cover
%
Fig. 1. Variation of service level with ?nished goods safety stock for the
non-collaborative case.
all the points where currently used. These steps account for
signi?cant dead time in the process; often of the order of 1 to
2 weeks, when all the intervening processes are considered.
We developed a model of a process which has ?ve primary
synthesis stages and two secondary manufacturing sites. The
as–is process has QC activities at the end of each primary
stage, and for the ?nal product. The modi?ed process has a
QC step for the AI and a QC step for the ?nal product. The
results for one of the products are compared below.
In Fig. 3, the forward prediction of ?nished goods inven-
tory of pack C is quite smooth. The lower con?dence (95%)
limit on the pro?le is still positive, giving con?dence that
stock-outs are very unlikely. In Fig. 4, the case with QC at all
stages, there is much less certainty in the inventory (the vari-
ance grows signi?cantly with time) and the lower con?dence
level goes to zero. In terms of customer service performance
measures, over a 2-year period, the average service level in
the case with QC is 91% and the probability of a stockout
in any week is 5%. On the other hand, in the low QC case,
these ?gures are 100 and 0%, respectively. Clearly, as pro-
cess development and design advance, the comfort provided
by QC at so many stages in the supply chain will not be
required and the dynamic behaviour will improve markedly.
Overall, the simulation-based approach is very promis-
ing for studying the large and complex supply chains in-
volved. Other studies include when to plan shutdowns in
Service Level
90
92
94
96
98
100
102
weeks cover
%
Fig. 2. Variation of service level with ?nished goods safety stock for the
collaborative case.
938 N. Shah / Computers and Chemical Engineering 28 (2004) 929–941
Base Case for Pack C
0
50000
100000
150000
200000
250000
300000
0 20 40 60 80 100
Time / weeks
mean quantity
safety stock
95% upper confidence
95% lower confidence
P
a
c
k
C
s
t
o
c
k
l
e
v
e
l
/
p
a
c
k
s
Fig. 3. Time pro?le of expected inventory of ?nished goods, including con?dence intervals, for the case with quality control at only two points.
Results for Pack C with QC (Monte Carlo 400 simulations)
0
50000
100000
150000
200000
250000
300000
0 20 40 60 80 100
Time / weeks
mean quantity
safety stock
95% upper confidence
95% lower confidence
P
a
c
k
C
s
t
o
c
k
l
e
v
e
l
/
p
a
c
k
s
Fig. 4. Time pro?le of expected inventory of ?nished goods, including con?dence intervals, for the case with quality control at all stages.
manufacturing sites and strategies for new product introduc-
tion (Gjerdrum, Shah, & Papageorgiou, 2001).
4. Future challenges
The future challenges in this area are broad and complex,
and will provide fertile ground for research. They can be
categorised under three headings:
• improvements to existing processes;
• improvements to the strategic decision-making process;
and
• future scenarios.
4.1. Improvements to existing processes and operations
The supply chain includes many agents, often with dif-
ferent objectives. Their internal dynamics tend to exagger-
ate the external market dynamics and result in detriments
in performance. This is an area where collaborative fore-
casting, planning and inventory management will be very
useful. Here, the different agents in the supply chain will
co-ordinate activities across the chain. One of the main rea-
sons for the current, more distributed practice is the large
scale of the operations, both in terms of activities and geo-
graphical span. Multisite planning and scheduling tools are
required to support a collaborative planning activity. The
supply chain generally contains larger amounts of inventory
N. Shah / Computers and Chemical Engineering 28 (2004) 929–941 939
than might be necessary if a more co-ordinated approach is
followed, with the right supporting tools.
Probably the most important metric to track and try to
improve is that of the overall supply chain cycle time. As
mentioned earlier, ?gures of 100–300 days are common. Our
experience in the industry is that efforts to reduce this are
very effective, but they have been applied in a piecemeal
fashion to certain products, while the measure deteriorates
for others. A systematic analysis of the components of the
cycle time should be undertaken for a wide range of prod-
ucts. Clearly, large scale simulation tools that take account
of both physical and business processes would be useful in
this context.
4.2. Improvements to the strategic decision-making
process
We can consider this in a bottom-up fashion. The current
nature of the process technology is one the main supply
chain bottlenecks. There is a need for more agile equip-
ment which will shorten process cycle times by an order
of magnitude and require minimal time for cleaning and
changeover. This will avoid long campaigns and should
lead to “pull”-based active ingredient manufacturing, and
therefore more responsive supply chains. The underlying
processes will have to change as well, with the focus being
on designing processes that operate at intrinsic rates (e.g.
being limited by reaction kinetics) rather than being limited
by equipment performance (e.g. heat transfer, mass trans-
fer or mixing characteristics) of traditional equipment. In
principle, the low tonnages involved should lead to much
less capital intensive plants if this is achieved. Generally
speaking, signi?cant improvements to manufacturing tech-
nology have not been of the highest priority in this ?eld to
date.
Processes should be designed with a much greater level
of mechanistic understanding and controlled tightly if re-
ductions in quality control activities are to be possible. The
impact of this has been demonstrated earlier.
The integration of development management and capacity
and production planning will be very important. Currently,
capacity issues are often not considered at the development
stage. Booth (1999) lists three undesirable outcomes of not
co-ordinating these activities:
(i) shortages of materials for pre-clinical studies;
(ii) shortages of materials for clinical trials; and
(iii) delays in time-to market—which is not only dependent
on having material available, but also on generating
demand at the late stages of development.
The approach of Maravelias and Grossmann (2001)
is a very promising platform to treat the overall piple-
line/capacity planning problem, but it needs to be integrated
with a sophisticated treatment of risk (e.g. through the use
of real options theory) and economics (e.g. taking account
of local taxation regimes, transfer pricing, duty drawbacks,
etc.), both of which have a very signi?cant impact on in-
vestment decisions. Furthermore, the modelling of testing
and trials needs to be extended to account for variations
in standard outcomes, e.g. very successful trials resulting
in short-circuiting of the approvals process in the case of
life-saving drugs.
Generally speaking, the development of integrated models
of the life-cycle, from discovery through to consumption
would greatly facilitate strategic decision-making.
4.3. Future scenarios
Companies have recently moved away fromproduct diver-
si?cation and locally adapted products. The common pack-
aging and labelling standards in the European Union, for ex-
ample, have supported this. However, there has been much
development in the ?eld of “pharmacoeconomics” which
might generate pressures to reverse this trend. This discipline
helps to make choices in treatment options by consideration
of costs and outcomes (clinical, economic and humanistic).
An important outcome of this type of analysis will be the
insistence on local solutions to local problems (Thwaites &
Townsend, 1998).
Another trend, somewhat further down the line, will
come out of genetic research and which will identify tar-
get sub-populations for different treatment regimes (the
so-called “designer drugs”). These two drivers will give rise
to considerable product and supply chain complexity. The
current manufacturing processes and supply chains are not
well con?gured to cope with this. Primary manufacturing
tends to operate in campaigns; and secondary manufactur-
ing batchsizes are typically 1–4 million tablets. Clearly,
the manufacturing processes and supply chains will have
to be re-designed if product customisation is to increase in
line with these trends. The manufacturers will have to track
supply chain performance measures very carefully in order
to understand the cost-to-serve for a diverse customer base.
An emerging area is that of rapid response vaccines and
other treatments arising out of possible emergencies (e.g.
bioterrorism or very fast developing epidemics). Again the
traditional supply chain (particularly for vaccines) is very
slow and unresponsive. If national governments are to im-
plement emergency preparedness programmes, the entire in-
frastructure must be well designed and tested through sim-
ulation. Decisions such as where to manufacture, in what
quantities, where to hold stocks, where people should report,
etc. need to be taken in a robust fashion. The issue of how
well the supply chain measures up to emergency prepared-
ness concerns is being raised in various fora at the present
(see, e.g. Anon, 2002).
Companies are investing in the development of crops that
are designed in some way to produce pharmaceuticals; this
will give rise to new research activities in process (in par-
ticular recovery steps) innovation, novel equipment design
(e.g. supercritical separations) and of course in a new type
of supply chain to optimise.
940 N. Shah / Computers and Chemical Engineering 28 (2004) 929–941
There are also some changes afoot in the industry struc-
ture as well—the growth in medicinal chemistry, biotech-
nology and genomics will spread IP around and result in
looser, virtual enterprises of joint ventures, alliances, etc.
Including as well the general tendency towards outsourcing
of manufacturing, this will give rise to complex extended
supply chain co-ordination. Lessons may be learnt from the
automotive, PC and consumer electronics world, where such
supply chains already operate.
5. Conclusions
The pharmaceutical supply chain used to be seen as a tool
to supply products to market in an effective way, where the
emphasis was on security of supply. Recent changes in the
operating environment mean that companies are revisiting
the components of their supply chains and identifying ways
of extracting additional bene?ts from them.
In this sector in particular, the supply chain of interest is
not simply the physical processes of conversion and distri-
bution of materials. Equally important is the “value-chain”
perspective of managing the innovation and development
processes through to capacity and production planning.
There are still several exciting research challenges in this
value chain, many of which the process engineering/process
systems engineering community are well placed to address.
Acknowledgements
The author is very grateful to Gabriel Gatica, Jonatan
Gjerdrum and Lazaros Papageorgiou for their contributions
to this work.
References
Anon (2002). Summary of the executive session on emergency pre-
paredness and the pharmaceutical supply chain. American Journal of
Health-System Pharmacy, 59 (3), 247–253.
Applequist, G. E., Pekny, J. F., & Reklaitis, G. V. (2000). Risk and uncer-
tainty in managing chemical manufacturing supply chains. Computers
of Chemical Engineering, 24, 2211–2222.
Ballance, R., Pogany, J. & Forstner, H. (1992). The world’s pharmaceutical
industries. London, UK: Edward Elgar [for UNIDO].
Barbosa-Povoa, A. P. F. D., & Pantelides, C. C. (1999). Design of multi-
purpose production facilities: A RTN decomposition-based algorithm.
Computers of Chemical Engineering, 23, S7–S10.
Bhagwat, Y., & Griggs, F. T. (1995). Analysis of riskiness of pharmaceuti-
cal industry ?rms. Journal of Research in Pharmaceutical Economics,
6, 65–76.
Blau, G., Mehta, B., Bose, S., Pekny, J., Sinclair, G., Keunker, K., &
Bunch, P. (2000). Risk management in the development of new prod-
ucts in highly regulated industries. Computers of Chemical Engineer-
ing, 24, 1005–1011.
Booth, R. (1999). The global supply chain. FT healthcare management
report. London: Financial Times Business Ltd.
Butler, R. (2002). The end of the blockbuster. Chemistry & Industry, 9,
9–10.
Clay, R. L., & Grossmann, I. E. (1997). A disaggregation algorithm for the
optimization of stochastic planning models. Computers of Chemical
Engineering, 21, 751–774.
Eli Lilly (2003).http://www.newsroom.lilly.com/news/Financial/2003-01-
23 q402sales fullyear 2002.html.
Eppen, G. D., Martin, R. K., & Schrage, L. (1987). A scenario approach
to capacity planning. Optical Research, 37, 517–527.
Gatica, G., Shah, N., & Papageorgiou, L. G. (2001). Capacity planning
under clinical trials uncertainty for the pharmaceutical industries. In
Proceedings of the ESCAPE-11 (pp. 865–870) Kondig, Denmark.
Gatica, G., Papageorgiou, L. G., & Shah, N. (2002a). Capacity planning
under uncertainty for the pharmaceutical industry. Transformation of
Institute of Chemical Engineering: Part A.
Gatica, G., Papageorgiou, L. G., & Shah, N. (2002b). An aggregation ap-
proach for capacity planning under uncertainty for the pharmaceutical
industry. FOCAPO-2003.
Gjerdrum, J., Jalisi, Q. W. Z., Papageorgiou, L. G., & Shah, N. (2000).
Dynamic simulation of physical and business processes for supply
chain improvement. In Proceedings of the 5th Annual Conference on
Industrial Engineering Theory, Applications and Practice. Hsinchu,
Taiwan.
Gjerdrum, J., Shah, N., & Papageorgiou, L. G. (2001). New product in-
troduction supply chain planning. In O.-P. Hilmola, editor. Contempo-
rary Research Issues in New Product Introduction. Acta Wasaensia,
77, 25–49.
Grabowski, H. (1997). The effect of pharmacoeconomics on company
research and development decisions. Pharmacoeconomics, 11, 389–
397.
Henning, G. P., Camussi, N. B., & Cerda, J. (1994). Design and plan-
ning of multipurpose plants involving nonlinear processing networks.
Computers of Chemical Engineering, 18, 129–152.
Jain, V., & Grossmann, I. E. (1999). Resource-constrained scheduling
of tests in new product development. Industrial Engineering and
Chemical Research, 38, 3013–3026.
Kall, P., & Wallace, S. W. (1994). Stochastic programming. New York:
Wiley.
Lazaro, M., Espuna, A., & Puigjaner, L. (1989). A comprehensive ap-
proach to production planning in multipurpose batch plants. Comput-
ers of Chemical Engineering, 13, 1031–1047.
Lerwent, J. (1994). The new pharmaceutical paradigm: scienti?c manage-
ment at Merck. Harvard Business Review, Jan-Feb, 88–89.
Linninger, A. A., Ali, S., & Stephanopoulos, G. (1996). Knowledge-based
validation and waste management of batch pharmaceutical process
designs. Computers of Chemical Engineering, 20, S1431–S1436.
Maravelias, C. T., & Grossmann, I. E. (2001). Simultaneous planning for
new product development and batch manufacturing facilities. Industrial
Engineering of Chemical Research, 40, 6147–6164.
Mauderli, A., & Rippin, D. W. T. (1979). Production planning and schedul-
ing for multipurpose batch chemical plants. Computers of Chemical
Engineering, 3, 199–206.
Mauderli, A., & Rippin, D. W. T. (1980). Scheduling production in mul-
tipurpose batch plants: the Batchman program. Chemical Engineering
Progress, 4, 37–45.
Moser, M., Calderari, G., & Morini, P. (2000). Cleaning validation of a
multipurpose plant for active pharmaceutical ingredient bulk produc-
tion. Chimia, 54, 731–733.
Mulvey, J. M., Rosenbaum, D. P., & Shetty, B. (1997). Strategic ?nancial
risk management and operations research. EJOR, 97, 1–16.
Myers, S. (1999). Measuring pharmaceutical risk and the cost of capital.
In J. Sussex, & N. Marchant, editors. Risk and return in the pharma-
ceutical industry (pp. 59–76). London: OHE.
Papageorgaki, S., & Reklaitis, G. V. (1990). Optimal design of multipur-
pose batch plants. Industrial Engineering of Chemical Research, 29,
2054–2062.
Papageorgiou, L. G., & Pantelides, C. C. (1996). Optimal campaign plan-
ning/scheduling of multipurpose batch/semicontinuous tasks. 1. Math-
N. Shah / Computers and Chemical Engineering 28 (2004) 929–941 941
ematical formulation. Industrial Engineering of Chemical Research,
35, 488–509.
Papageorgiou, G. E., Rotstein, G.E. & Shah, N. (2001). Strategic sup-
ply chain optimization for the pharmaceutical industries. Industrial
Engineering of Chemical Research 40, 275–286.
Reklaitis, G. V. (1991). Perspectives on scheudling and planning of process
operations. In Proceedings of the PSE-1991. Montebello, Canada.
Rotstein, G. E., Papageorgiou, L. G., Shah, N., Murphy, D. C., & Mustafa,
R. (1999). A product portfolio approach in the pharmaceutical industry.
Computers of Chemical Engineering, 23, 883–886.
Schmidt, C. W., & Grossmann, I. E. (1996). Optimization models for the
scheduling of testing tasks in new product development. Industrial
Engineering of Chemical Research, 35, 3498–3510.
Shah, N. (1998). Single- and multisite planning and scheduling: current
status and future challenges. AIChE Symposium Series, 320(94), 75–
90.
Shah, N., & Pantelides, C. C. (1991). Optimal long-term campaign plan-
ning and design of batch plants. Industrial Engineering of Chemical
Research, 30, 2308–2321.
Shah, N., & Pantelides, C. C. (1992). Design of multipurpose batch
plants with uncertain production requirements. Industrial Engineering
of Chemical Research, 31, 1325–1337.
Shah, N., Samsatli, N. J., Sharif, M., Borland, J. N., & Papageorgiou,
L. G. (2000). Modelling and optimisation for pharmaceutical and ?ne
chemical process development. AIChE Symposium Series, 323(96),
31–45.
Sparrow, R. E., Forder, G. J., & Rippin, D. W. T. (1975). The choice
of equipment sizes for multiproduct batch plants: heuristics versus
branch-and-bound. Industrial Engineering and Process Design Devel-
opment, 14, 197–203.
Stephanopoulos, G., Ali, S., Linninger, A., & Salomone, E. (2000). Batch
process development: challenging traditional approaches. AIChE Sym-
posium Series, 323(96), 46–57.
Subramanian, D., Pekny, J. F., & Reklaitis, G. V. (2001). A
simulation-optimization framework for Research and Development
Pipeline management. AIChE Journal, 47, 2226–2242.
Thwaites, R., & Townsend, R. J. (1998). Pharmacoeconomics in the new
millenium. Pharmacoeconomics, 13, 175–180.
van der Vorst, J. G. A. J., Beulens, A. J. M., & van Beek, P. (2000).
Modelling and simulating multi-echelon food systems. EJOR, 122,
354–366.
Voudouris, V. T., & Grossmann, I. E. (1992). Mixed-integer
linear-programming reformulations for batch process design with dis-
crete equipment sizes. Industrial Engineering of Chemical Research,
31, 1315–1325.
Yeh, N. C., & Reklaitis, G. V. (1987). Synthesis and sizing of
batch/semicontinuous processes: single product plants. Computers of
Chemical Engineering, 11, 639–654.
doc_582633281.pdf