BINF 5125 CLINICAL PROBLEM SOLVING AND DECISION MAKING

1
BINF 5125CLINICAL PROBLEM SOLVING AND DECISION
MAKING

• Model for the Decision Support

2
Management/Decision Support Systems - An
Overview 9 out of 10 senior executives use
computers for their businesses. 81 say
computers and networks are critical for the
business. 88 say that they are using computers
to increase communications. 87 say that the
use of computers has already cut the time
needed to develop products and increased
productivity. Management Support Systems
Technologies 1. Decision Support Systems (
DSS) 2. Group DSS (GDSS) 3. Executive Information
Systems (EIS)
3
4. Expert Systems (ES) 5. Artificial Neural
Network ( ANNs) / fuzzy logic based Systems
Collectively, these systems are known as
computerized decision support systems. In the
past, the decision making was considered as an
art - a talent acquired over a long period of
time - experience based on judgement, intuitions
and experience. However, in todays environment,
management has to tackle complex problems.
Business is very competitive and there is no room
for the wrong decisions. Technology is changing
fast - it gives a lot of choices for solutions As
a result trial and error approach is out.
Managers have to use new tools and techniques,
which are latest and current in their fields. -
Management Science.
4
Framework for the Decision Support Decision
making process is not simple, It can be very
highly structured to completely
unstructured. Structured Similar problems (
routine) and standard solutions Unstructured
Fuzzy and complex problems with no clear-cut
solutions. The problem of decision making may be
thought of as ( by Simon) Intelligence
Searching for conditions for making
decisions. Design Inventing, developing and
analyzing possible courses of
actions. Choice Selecting a course of
action. These three phases may not be so
clear-cut in some of the problems.
5
Type of Control Type of Decision Operational
Managerial Strategic Support Control Control
Planning Needed Structured Accounts
Budget Financial MIS, Oper Receivable, analy
sis management research Order Entry Short-term
forecast Warehouse locations Trans
Proc Personnel reports distribution
system Semi-Structured Production Credit
evaluation Building new plants DSS
scheduling Budget preparation Mergers
acquisitions Inventory Control Plant layout New
product planning Project scheduling Quality
assurance Unstructured Buying a
software Recruiting executive R D Planning
DSS Approving Loans Buying hardware New
technological ES Selecting
Covers Negotiating developments NN Support
Needed MIS, Management Management science EIS,
ES, Neural Science DSS, ES, EIS Networks
Decision Support Framework
6
To get an optimal solution, we have to ensure
Cost minimization Profit maximization
Optimal productivity Typically unstructured
problems include - Planning for the new services
to be offered. - Choosing a new set of R D
projects - Hiring a new executive. For structured
problems, the techniques or procedures for
obtaining best solutions are very well
established ( known). For example inventory
control, best financial investment
strategy, Semi- structured problems are midway
cases. Decision support systems can improve the
quality of information on which decisions
7

• could be made. Most of the clinical decision
  support systems may fall into semi-structured or
  unstructured problems.
• Why should we go for DSS?
• To get accurate information and consistent
  decisions.
• DSS is viewed as an organizational winner.
• Whenever new and latest and updated information
  is needed.
• Management has mandated DSS.
• Timely information and problems are provided by
  DSS.
• Cost reduction may be achieved.
• The DSS are developed primarily for the following
  objectives
• To serve the information needs of executives.
• Provide extremely friendly user interface for
  executives.

8

• Meet individual executives decision style.
• Provide timely and effective tracking and control
  of important activities/ processes,
• Provide quick access to detailed information in
  text, image or graphic format for easy
  understanding.
• Filter, compress and track critical data and
  information.
• Identify problem areas.
• Expert systems( Decision Support Systems)
• When an organization has a complex decision to
  make or a problem to solve, it has to turn to
  experts, which could be difficult to find and at
  times very costly.
• Expert systems are attempts to mimic human
  experts
• Expert systems are branch of applied artificial
  intelligence. They are used frequently in
  medical diagnostic systems, mineral

9

• exploration, managing assets and liabilities,
  corporate planning, organizational management and
  administration, etc.
• ANN based systems ( or Fuzzy Neural Systems)
• If the decisions are to be made based on partial
  information, under uncertain conditions, based on
  past experiences, or under inexact information,
  such systems are designed using the cutting edge
  technology based tools.
• Systems based on such technology learn by
  experience.
• Sometimes these systems employ pattern
  recognition techniques.
• DSS systems can be used to address ad-hoc and
  unexpected problems also. It can provide valid
  representation of real-world problems.
• A DSS can provide support in a relatively short
  time. For example risk analysis can be performed
  in a few minutes and

10

• Models can be developed in a few hours ( for
  structured problems).
• DSS can be developed by non-data processing
  professionals. It is because the 4-GL packages
  are available and for developing DSS. Usage of
  such tools require very little expertise of the
  programming languages.
• DSS-ES connections
• These two seems similar but are completely
  different. These differences are technological as
  well as managerial.
• Because of different capabilities, they can
  complement each other, creating powerful and
  integrated computer based systems that can
  improve managerial decision making capabilities (
  or process).

11
(No Transcript)
12
The Decision-Making Process. - The process starts
with the intelligence phase, where reality is
examined and the problem is identified and
defined formally. - In the design phase - a model
that represent system is constructed. This is
done by making assumptions that simplify the
reality and by writing relationships among
various variables and parameters. - The model is
then validated and criteria are set for the
evaluation of the model and alternate course of
actions are identified. - The choice phase
includes a proposed solution of the model. Once
the solution seems to be reasonable, it is ready
for the last phase - which is implementation.
13
Sources of Information and Knowledge
People
Sensors
ES, NLP, ANN, EIS, MIS
Newspapers
TV
Scanning
Problem Identification Definition-Intelligence
EIS ( ESS)
DSS MS /or ANN
Quantitative Analysis(design)
Qualitative Analysis (Design)
ES
Decision (Choice)
GDSS (Groups) DSS Individuals
Implementation
Yes
Implementation ?
No
DSS and ES
14
Successful implementation results in solving the
original problem. Failure leads to a return to
the modeling process. What is Decision
Making? It is a process of choosing, among
alternatives, the course of action for attaining
a goal. What should be done? How? By whom?
Where? It would involve 1. Intelligence 2. Design
3. Choice 4. Implementation
15
System ( Decision Support , Expert ) A system is
a collection of objects, such as people,
resources, concepts and procedures intended to
perform a function or goal. There could be
various levels in a system, called
sub-systems. System will have structure Inputs P
atients admitted to a hospital, raw materials in
a plant, Employees in an organization. Outputs
Finished products or consequences of being into
the system Cured patients in a
hospital. Processes All the elements necessary
to convert inputs to outputs. In a hospital -
processes will include conducting tests,
performing surgery, giving medicines, etc.
16

• The system will also have feedback mechanism on
  the quality of decisions and on general
  performance.
• Environment Affects performance of the system,
  ie, suppliers, competitions, interest rate
  change, govt. regulations, etc.
• The boundary Environment is outside the
  boundary of the system. System parameters are
  inside.
• Closed and open systems Open system is
  dependent on the environment, whereas the closed
  system is independent to external parameters and
  environment. Open system accepts inputs from
  environment and may deliver output into it.
• System effectiveness and efficiency Systems are
  evaluated and analyzed with two major classes of
  performance measurements.
• Effectiveness is the degree to which goals are
  achieved. Therefore, concerned with the results
  or the outputs For example patient recovery,
  satisfaction, service quality.

17

• Efficiency is a measure of the use of inputs to
  achieve results., ie, how much money is used to
  generate certain level of sales or treat cancer
  patient completely.
• Model A model is a simplified version of
  reality. It should be simple enough to understand
  and analyze and represent realistic behavior.
  There could be various types of models
• Iconic It is a scale model and it may be replica
  of the actual system and physically looks like
  real system, which is used for testing purposes.
• Analog This type of model does not look like
  real system but behaves like it. It is a symbolic
  representation of reality.
• These could be physical models, but shapes
  could differ These could be two dimensional
  charts/diagrams. For example stock market chart,
  blue print of a house, hospital performance
  chart.
• Mathematical model Sometimes complexity of
  relationship is such

18
that you can not represent it in icon form or
analog form. Therefore, a mathematical model
becomes essential to study its behavior. For
most of the DSS, the analysis is executed
numerically with the aid of mathematical or
quantitative models. The benefits of using models
are 1. Low cost 2. Compression of time in
studying the behavior of the system. 3. Changing
of parameters - manipulation - is very easy. 4.
Cost of making mistake is much lower - for trial
and run type of experiments ( as no hardware is
needed). 5. It can handle uncertainties and risk
factors - can be studied. 6. Mathematical models
are capable of giving a large spectrum of
solutions. There may be a large number of
alternates to choose from.
19

• 7. Models enhance and reinforce learning.
• With recent advances in the computer technology,
  there is an increased use of iconic/analog models
  to complement mathematical modeling.
• The modeling Process
• Trial and Error - Learn from experience. There
  may be too many variables to study in
  mathematical form. The cost of study through
  modeling may be high. Sometimes even environment
  may also be changing. Approximate solution may be
  acceptable.
• Simulation The problem with simulation is that
  you have to supply the parameter values and the
  solution may not be optimal.
• Optimization A more sophisticated approach is
  to use optimization model. It will work if the
  problem is well structured.
• An optimization may be combined with simulation
  to get the

20

• best practical results.
• Heuristics ( rule of thumbs) - mostly by
  experience.
• Now let us look the details of the three major
  phases of developing a DSS
• Intelligence phase
• Design phase
• Choice phase
• An additional phase - Implementation may also be
  added to complete the process.
• 1. Intelligence Phase
• The modeling process starts with the intelligence
  phase. This phase begins with identification of
  the organizational goals and objectives. You
  identify if a problem exists- what we desire and
  what is currently happening?

21

• - One attempts to find the symptoms of the
  problem and its magnitude, ie, excessive cost,
  improper inventory, quality of service.
• - The existence of problem can be appraised by
  monitoring and analyzing the data.
• The critical problem is usually in collection of
  ( estimation of) data for the expectation for
  the future results ( or even current state as
  well).
• Some of the issues and problems during the data
  collection are
• Outcome may occur over extended period of time.
  As a result, cost and revenue accounting may be
  difficult- use present value approach.
• It may be necessary to use subjective approach
  for the data estimation.
• It may be assumed that the future data may be
  similar to the historic ones. If not- predict the
  changes and include the changing trend in the
  current available data points.

22

• If you have identified the problem - see how
  significant is the problem and where is it
  located.
• In this phase classify the problem, decompose it
  and identify the ownership of problem.
• Classification
• Structured - where standard solution exit through
  existing models - programmed problems.
• Poorly structured problems - which are novel and
  non-recurrent type, ie, acquisitions and mergers,
  R D projects, reorganizing a corporation,
  opening of a new hospital.
• In general the problems may be of semi-structured
  type.
• Decomposition
• Break a large problem into smaller and more
  manageable ones.

23

• Sometimes complex problems can not be solved as
  a whole.
• Ownership of problem
• Who can solve? Do you have the expertise? It is
  essential to identify. Otherwise you have to look
  for an external expert.
• The intelligence phase ends with problem
  statement.
• Design Phase
• This phase involves generating, developing and
  analyzing possible courses of action.
• This includes activities such as understanding
  all aspects of the problem, and testing of
  solutions for feasibility.
• In this phase, a model of the problem situation
  is constructed, tested and validated.
• Modeling involves- conceptualization of problem
  in some form.

24

• In case of a mathematical model - independent
  and dependent variables are identified.
• Subsequently, equations and their relationship
  is established.
• Simplifications are made, if necessary - through
  assumptions. You should not oversimplify the
  problem. As simpler model leads to easier
  manipulations and faster solutions - but you may
  go away from the real- life behavior.
• In general a model may have the following
  components (quantitative)- with flow-diagram
  relationship

Uncontrollable Variables
Decision var.
Relationship
Result var
25
Result Variables reflect the effectiveness of
the model. How well the model performs or attains
the goal. Result variables are also known as
dependent variables. Decision Variables are the
variables which are involved in the decision
making process. These variables are also known as
independent variables. For example, if you are
planning for an emergency room, number of doctors
and nurses may form decision variables. Uncontroll
able variables ( or parameters) these are
variables which affect result variables, but are
not under the control of decision makers. They
could be fixed or variables, ie, interest rates,
tax regulations, communications technology. These
variables are also independent variables.
However,some of these variables may be outside
the model boundary, while some of them could be
part of the model
26
STRUCTURE OF A QUANTITATIVE MODEL Simple
financial model P R - C, where P profit,
C cost and R revenues or alternately, P
F/(1 i) n where P present value, F future
val. i interest rate, n number of
years. The product mix model Assume that the
company is producing two types of
products PC-7 300 men days 10,000 cost 8,000
profit PC-8 500 men days 15,000 cost 12,000
profit Plant has a capacity of 200,000 working
men days per month and has a budget allocation
of 8,000,000 per month. There is a marketing
requirement of 100 units of product PC-7. Market
can absorb any amount of product. Maximize
return for the company.
27
Model standard - Linear programming Decision
variables X1 number of PC-7, and X2 number of
PC-8 Result variable Z Profit amount. Define
the objective function. Z 8,000 X1 12,000
X2 ( maximize objective function) Uncontrollable
constraints labor cost 300 X1 500 X2 ?
200,000 men days per month budget 10,000 X1
15,000 X2 ? 8,000,000 dollars Marketing
requirement X1 ? 100 units of PC-7 Solution of
this problem is obtained by Linear programming
and gives you X1 666.667, X2 0, Z profit
5.33 M As such there are many possible
solutions. The one which maximizes the profit is
the solution given above and is obtained by
Linear Programming.
28
The evaluation of alternatives and the final
choice depends on the type of criteria we want to
use- best solution or good one? Normative model
the best possible alternate - one should examine
all the alternates and choose the best -
Optimization. Sub-optimization model sometimes
the optimization may lead to unexpected problems,
ie, producing just a few out of many products -
is it good? Such solution may cause high
inventory or marketing problems. So, real
optimization may not be a practical solution. In
such cases go for sub-optimization. Descriptive
Models Such models are extremely useful in DSS
for investigating the consequences of various
courses of actions under different configurations.
29
However, the descriptive analysis check
performance of selected alternates only. There is
no guarantee that the selected alternate will be
optimal. This technique will provide GOOD
ENOUGH solution, which may also be
practical. Simulation is most recognized example
of descriptive modeling. The usual reasons for
selecting such solutions are lack of time or
ability to achieve optimization or in some cases,
unwillingness to pay the high cost for
optimization. Optimization Modeling Techniques
Assignment Dynamic programming Goal
programming Linear programming Non-linear
Programming Transportation model
30
Descriptive Models Describe things as they are -
various types of models are (These models work
under a given set of alternatives.
Information flow Scenario analysis
Financial planning Inventory management
Markov analysis Simulation Technological
forecasting A significant part of the model
building is generating alternatives. In
optimization models, these alternates are
automatically generated and evaluated. In DSS
situations, it is essential to generate practical
alternatives. Alternatives will cost time and
money. When to stop? - This decision require a
lot of considerations. For searching
alternatives? Use creativity, i.e.,
brainstorming sessions, group dynamic sessions,
special training, checklist, etc.
31
Prediction of outcome To evaluate and compare
alternatives. It is essential to predict the
outcome. It can be classified as 1. Certain-
outcome is unique and defined clearly
(deterministic). 2. With risk- outcome is
probabilistic with element of risks. Risk
analysis is needed for the solution optimization.
3. Uncertain- there are many possible outcomes.
It may not be possible to estimate probability of
individual outcomes ( modeling is difficult). The
value of the alternate is judged in terms of
goals attainments. It is also possible to do
Scenario analysis. It is a statement of
assumptions about the operating environment. It
is helpful in simulation and in what-if type
of analyses.
32

• ie, what if the demand for hospitalization
  changes- worst case, best case, and
  most-likely case.
• This establishes the evaluation criteria.
• The solution of the model identifies an alternate
  being selected.
• Solving a model is not same as solving the
  problem that the model represents
• Only if the recommended solution is successfully
  implemented ? the problem is successfully solved.
• Choice Phase (search approaches)
• The choice phase involves the search for the
  appropriate course of action, among those
  identified during the design phase. These could
  be analytic or algorithms- step-by-step search.
• For normative models analytic approach or a
  complete exhaustive search is used. ( the search
  is normally analytic or algorithm).

33

• It provides an an optimal solution.
• For descriptive models - a comparison of limited
  number of alternates is used. A set of search
  steps, which lead to the desired goal are tried.
  The search could be blind search leading to
  optimal, complete enumerative, incomplete, or
  good enough types or it can be Heuristic
  search so as to minimize the search time, cost,
  etc.
• Heuristic search are decision rules regarding
  how the problem could be solved. These rules
  could be derived on a basis of rigorous analysis
  of the problem or based on experimentation or
  rules of thumbs or past experience with similar
  situations.
• Heuristic searches are step by step procedures
  which are applied/ repeated until a satisfactory
  solution is found.
• Evaluations ? ( based on) multiple goals,
  sensitivity analysis
• Multiple goals Currently, the trend is away
  from a single goal of

34
profit maximization. Other goals like company
growth, developing new products and services,
serving community, employee welfare are some of
the possible goals. Todays systems are far more
complex and single goal is rare. It is necessary
to analyze each alternate in light of its
potential impact on several goals ( including
share-holders). Sensitivity analysis This
analysis checks - Effects of uncertainty in
estimating external variables. - Effects of
different interaction among decision variables. -
Robustness of decisions under changing
conditions. - Impact of changes in external and
decision variables. This analysis is used to
revise model to eliminate large sensitivities.
35

• - Add details about sensitive variables or
  scenarios.
• - Obtain better estimates of sensitive external
  variables.
• - Alter the real world system to reduce
  sensitivity to critical variables.
• - Live with sensitive real world and monitor the
  results continuously.
• Some of the models do automatic sensitivity
  analysis ( ie, linear programming ).
• This type of analysis can also be done using
• Trial and error analysis.
• What- If analysis.
• Goal seeking analysis, ie, how to adjust an
  input to achieve the desired goal ( how many
  nurses are needed to reduce average waiting time
  in the emergency room to 10 minutes). Or
• How much R D budget is needed for an average
  growth of 15.

36
CRITICAL SUCCESS FACTORS Critical success factors
(CSF) is a diagnostic technique for identifying
the factors that are most critical to achievement
of organizational objectives. This may involve
group discussions, interviews and brainstorming
sessions. Once the critical factors are
determined, it is possible to identify the
problem areas/gaps which are not being adequately
supported by the computerized information
system. Lack of such information on CSF prevents
management from the measurement of the
effectiveness of areas that are critical to the
success of the organization. Therefore, it is
essential to identify such areas/factors before
developing DSS/MSS.
37
Once the choice phase has been completed, the
recommended solution must be implemented.
38

• FORMAL DEFINITION OF A DECISION SUPPORT SYSTEM
• A DSS is an interactive, flexible and adaptive
  system, specifically developed for supporting
  solutions of management decision problems for
  improved decision making or diagnosis of some
  problems.
• It supports all the phases of the decision making
  and always includes a knowledge-base.
• In order for this to be successful, the system
  must be
• Simple Adaptive
• Robust Complete , on important issues
• Easy to control/use Easy to communicate
• COMPONENTS OF DSS
• 1. Data Management Includes a database and is
  managed by the DBMS software.

39
2. Model Management It is a software package
that includes financial, statistical, management
science, or other suitable mathematical models,
and other decision making tools for systems
in-depth analytical capabilities. You would also
require an appropriate software model
manager. 3. Communications Subsystem( dialog
subsystem) User can communicate and command the
DSS through this module. It provide user
interface. It should be very user friendly. 4.
Knowledge Management Subsystem This is an
optional module. It can help in giving an advice
to decision making process and other related
issues.
40

• 1. Data Management Subsystem It consists of
• DSS Database
• DBMS
• Data Directory
• Query Facility
• The database is a collection of data which are
  organized for the the organization. It can be
  used by many users.

Data internal and external
Other computer based Systems
Model Management
Data Management
Knowledge Manager
Manager (user)
Conceptual Model of DSS
Dialog Management
41
Internal data Sources
External data sources
Finance
Marketing
Production
Personnel
Other
Dialog Management
Model Management
Extraction
Private, Personal data
Query Facility
Knowledge Management
Decision Support Database
The Data Management Subsystem
DBMS Retrieval Inquiry Update / Report
generation Delete
Data Directory
42

• Data information is stored in many files
• A file contains information regarding an
  application. In a large computerized system, one
  file can be very large and it can reside on an
  auxiliary storage device. Example Payroll system
  file for a large organization.
• The data in a DSS database may include internal
  data, external data and private data.
• - Internal data are generated through
  transaction processing or data processing. These
  data are the major source of information for DSS
  systems. It may include accounting, financial,
  marketing, production, patient, and diagnostic
  data. Items like sales forecast, maintenance
  data, etc may also be included.
• - External data may include marketing research
  data, census data regional employment data,
  central medical database, tax laws, etc. These
  data are normally provided by the service
  providers.

43
on online basis. - Private data May include
many rule of thumbs, assessment about the value
of the stored data and the confidence level of
information stored. Databases can be organized in
many different configurations.- relational,
hierarchical, network types. DSS may have its own
database or can use the existing organizational
database.
DBMS manages several databases
DBMS
DSS DB 1
DSS DB 2
Application 1
DSS 1
DSS 2
User
44
There could be many advantages of a separate
database 1. A greater control over data 2. A
better fit exists with the DSS software 3. DSS
may be cross-functional, requiring data from many
sources. One dedicated database may be more
efficient. 4. Changes and updates are faster,
easier and cheaper. 5. Easier access and data
manipulation. 6. Can adopt a database structure
that is optimal for a particular Decision Support
System. Disadvantages 1. An additional database
is more expensive to build, secure and manage. 2.
Common data in databases are to be updated in
multiple
45

• locations . Access rights, security, updates are
  complicated issues.
• A DBMS performs
• Data Extraction
• Data capture from various sources, importing
  files, filtration, and condensation.
• Processing of data, customize outputs, generate
  reports and graphs.
• Control ( invisible to user ) - user ask for
  information and DBMS gets it.
• It has to screen requests and determine, if the
  person making request is authorized.
• User has access to the requested file.
• User has access to data items in the file.
• Micro-based systems may perform only a few
  functions.

46
- Manager can obtain information from the DSS in
the forms of reports, charts, graphs, and output
of the model, through DBMS - DBMS serves as a
gate-keeper and makes the data available. - The
periodic reports are prepared by the application
programs and these programs request data from the
database through the DBMS. - An effective
database and its management can provide support
to many activities, ie, forecasting, problem
identification, general navigation among records,
report generation, etc. - The real power of the
DSS is provided when the database is integrated
with the model management module. Query
Facility It accepts requests for data ( from
the DSS components) and determines how such
requests can be filled. Some of the
47
important functions include- selection and
manipulation operations. The Directory The
data directory is a catalog of all types of data
in the database. It contains data definitions.
Its main function is to answer about the
availability of data items, their source, and
their meanings. The directory is important for
supporting the intelligence phase of the decision
making process by helping to scan the data and
identifying the problem areas. It also supports
the addition, deletion, and retrieval of
information on entries.