Biomedical Engineering Reference
In-Depth Information
Table 3.2 A summary of dynamic project selection studies related to the pharmaceutical industry
Pipeline
vs.
Portfolio
Cost/
resource
interaction
Outcome/
technical
interaction
Number
of stage
Benefit/impact
interaction
Study
Methodology
Dahan and
Mendelson
( 2001 )
Pipeline
Single
No
No
No
Analytical
Ding and
Eliashberg
( 2002 )
Pipeline
Multiple
No
No
No
Analytical
Kavadias and
Loch ( 2003 )
Pipeline
Multiple
Ye s
No
No
Analytical
Childs and Triantis
( 1999 )
Pipeline
Multiple
Ye s
Ye s
Ye s
Simulation
Loch and Kavadias
( 2002 )
Portfolio
Multiple
Ye s
No
Ye s
Analytical
Blau et al. ( 2004 )
Portfolio
Multiple
Ye s
Ye s
Ye s
Simulation
As an extension of the Gittins index, Kavadias and Loch ( 2003 ) set up a model
in which there are multiple projects but only one scarce resource (could be scien-
tists, lab time, budget, etc). Only one of the project can use this scarce source at a
time. If the projects are independent of one another and equally affected by delays,
this reduces to a multiarmed bandit problem solved by the Gittins index. However,
if projects are affected differently by delays, as is likely the case in a diverse port-
folio, a new policy is needed. The dynamic prioritization policy of Kavadias and
Loch ( 2003 ), called the “Expected Delay Loss Index,” is to work on the project
“with the highest expected delay loss as if the other project was completely fin-
ished first,” and prove it to be optimal if (1) the delay cost increases with the delay
regardless of the performance state, (2) costs are not discounted (or, discounting
is dominated by delay costs), (3) projects are not abandoned based on their perfor-
mance state during processing at the scarce resource, and (4) there are no stochas-
tic delays.
3.3.3.2
Prioritization Using Decision Trees
Another stream of literature on solving project selection and sequencing problems
uses decision trees. Approaches using decision trees consist of analytical methods
(e.g., Dahan and Mendelson 2001 ; Ding and Eliashberg 2002 ) and simulation meth-
ods (e.g., Blau et al. 2004 ; Childs and Triantis 1999 ). Analytical methods provide
closed form solutions which suggest clearer causal relationships, but simulation
methods are able to accommodate complex scenarios which give the model a more
realistic flavor. In Table 3.2 , we categorize the key papers mentioned above.
 
Search WWH ::




Custom Search