Health Sciences 4700 Spring 2009 - Chapter 9

Characteristics of DSS Used in un-/semi-structured decision contexts  Support decision makers, not replace them  Rely on data and models  Generally developed using an evolutionary, i

Health Sciences 4700

Spring 2009

Chapter 9 Decision Support Systems

What are decision support

 Intended to make knowledge more readily

available at the point of care

Simple DSSs

 CPOE system with drug interaction alerts

 EHR with alerts for missing or unknown information

 Database with structured queries to find relevant information

 Differential diagnosis systems (e.g.,

FirstConsult)

Components of an advanced DSS

 Data management system

 Database and query system

 Model management system

 Computations that represent domain models

 “Captured” knowledge and reasoning

 User interface

 Input/output, documentation

 Group support systems

 Enterprise planning and management systems

Characteristics of DSS

 Used in un-/semi-structured decision contexts

 Support decision makers, not replace them

 Rely on data and models

 Generally developed using an evolutionary, iterative process

 Focus should be on complete system

(including people and procedures, not just

computers)

Application domains

 Administrative decision support

 Access and organization of data

 Analysis of data

 Analysis of multiple data sources

 Accounting and modeling of data

 Forecasting from data

 Optimization and comparison of alternatives

 Suggestions of action based on comparisons

Application domains

 Clinical decision support

 Access to information (EHR)

 Analysis of data (single or multiple sources)

 Diagnosis

 Recommendations for treatments/procedures

Applications of DSS

 Reminders and alerts

 Provide complete information at point of care

 Prescribing systems

 Therapy planning and critiquing

 Identify inconsistencies, errors, omissions

 Image recognition

 Identify potential abnormalities, changes

Applications of DSS

 Diagnostic systems

 Help identify conditions based on symptoms

 Among earliest examples of DSS

 Support evidence-based practice

 May provide structured access to research

 Elsevier’s MD/FirstConsult (www.mdconsult.com)

Are DSSs effective?

 In principle, they should be!

 Improved patient safety

 Improved quality of care

 Improved efficiency

Are DSSs effective?

 Research isn’t completely convincing, particularly

in terms of patient outcomes

 Design may not have user in mind

 Reminders and alerts are valuable, but if they interfere with care, they won’t be used (pop-ups)

 May be perceived as de-humanizing care

 Neither caregivers nor patients are willing to turn over decision making to a “machine”

 DSS can reduce non-data driven interaction

Understanding how DSSs work

 Clinical decision making generally involves

“semi-structured” decisions

 Some but not all information is known

 How can we represent knowledge in a DSS?

 Descriptive knowledge: facts and data

 Practical knowledge: steps and instructions

 Inferential knowledge: reasoning from theory and facts (“intelligence”)

 How do we maintain currency of knowledge?

Artificial intelligence (AI)

 AI is field of computer science that attempts

to make computers “reason”

 Study of intelligent behavior and attempts to

create computer systems that behave that way

 Spectrum of AI ranges from “weak” to

Types of expert systems

Rule-based systems

 A patient complains of difficulty breathing

 No fever means the patient probably doesn’t have

a respiratory infection

 A normal chest X-ray will probably preclude

pneumonia (and corroborates the lack of a

respiratory infection)

 A TB skin test and a normal chest X-ray also

strengthens the lack of infection hypothesis

 Patient reports problem occurs after exercise

Rule-based systems

 Patient reports problem occurs after exercise

 No history of cardiac or pulmonary disease

 Nature of the problem is wheezing and shortness

of breath, which is characteristic of asthmatics

 Most likely conclusion: exercise-induced

asthma

How do you get a computer system to act like

an expert?

Rule-based systems

 Knowledge in expert systems is represented

as rules or IF-THEN statements

 Programming takes inputs through the user

Rule-based systems

IF (condition) THEN (result)

ELSE alternative result

-“condition” is a Boolean value that compares inputs to knowledge

IF “condition” is true THEN “result” is executed

ELSE [“condition” is false] “alternative result” is

executed

Alternative can be another IF-THEN, and so forth

Rule-based systems (example)

IF dipstick = pink THEN test <- normal;

IF dipstick = lilac AND symptoms = none

THEN test <- normal;

…

Note that if “dipstick” has the value “pink” then the first condition is true and “test” is assigned the value

“normal”

If “dipstick” is “pink”, then the second condition cannot

be true, no matter what the value of “symptoms” is!User interface would return the value of “test”

Rule-based systems (example)

IF dipstick = pink THEN test <- normal

ELSE IF dipstick = lilac AND

symptoms = none THEN test <- normal ELSE …

Note that once “test” is assigned the value “normal” no other execution is necessary!

In previous example, you would have to ensure that

“dipstick” was NOT pink for every subsequent

statement in order to make a correct diagnosis

Rule-based systems

 Problems with rule-based systems

 Extracting complete knowledge is difficult

 There is always some measure of uncertainty in a rule

 Complex systems require a large number of rules

 Domains tend to be narrowly focused

 Maintaining currency is difficult

 Input range is often large

 Interaction with system tends to be difficult

Case-based reasoning

 In case-based reasoning (CBR), systems

expertise is embodied in a library of past

cases, rather than being encoded in classical rules

 Each case typically contains a description of the problem, plus a solution and/or the outcome

 The knowledge and reasoning process used by

an expert to solve the problem is not recorded,

but is implicit in the solution

Case-based reasoning

 To solve a current problem, it is matched

against the cases in the case base, and

similar cases are retrieved

 Retrieved cases are used to suggest a solution which is reused and tested for success

 If necessary, the solution is revised

 Finally the current problem and the final solution are retained as part of a new case

Case-based reasoning

 Case-based reasoning is liked by many

people because they feel happier with

examples rather than conclusions separated from their context

 A case library can also be a powerful corporate resource, allowing everyone in an organization to tap into the corporate case library when handling

a new problem

http://www.aiai.ed.ac.uk/links/cbr.html

 The rankings are used to determine the next step

“How do you feel today?” (0.1 = bad, 0.9 = great)

Fuzzy logic

 Fuzzy logic has the capacity to deal with “real world” problems that cannot be expressed in Boolean terms

 Can be combined with Boolean logic to produce a recommendation (e.g., medication dosage)

http://www.austinlinks.com/Fuzzy/tutorial.html

Bayesian networks

 Bayesian (belief) networks are based on

probability distributions

 Graphical/computational model represents a set

of variables and the probability that they are

independent (or dependent, conversely)

 Distribution can represent the likelihood that event

A is the cause of condition B

What is the likelihood that a person who doesn’t

smoke has lung cancer?

Neural networks

 Artificial Neural Network (ANN): an information

processing paradigm inspired by the way biological nervous systems, such as the brain, process

information

 The information processing system is composed of a large number of highly interconnected processing elements

(neurones) working in unison to solve specific problems

 ANNs, like people, learn by example

 An ANN is configured for a specific application, such as

pattern recognition or data classification, through a learning process

Neural networks

 An artificial neuron has many inputs and one output

 Two modes of operation; training mode and using mode

 Neuron can be trained to “fire” (or not), for particular input patterns

 In the using mode, when a taught input pattern is detected at the input, its associated output becomes the current output.

Neural networks

 If the input pattern does not belong in the

taught list of input patterns, the firing rule is used to determine whether to fire or not

 Patterns not in the collection cause the node to fire if, on comparison, they have more input

elements in common with the 'nearest' pattern in firing set than with the 'nearest' pattern in the non-firing set

 If there is a tie, then the pattern remains in the

undefined state

Neural networks

 Neural networks can derive meaning from

complicated or imprecise data

 Used to extract patterns and detect trends too complex to

be noticed by either humans or other computer techniques

 A trained neural network can be thought of as an "expert"

in the category of information it has been given to analyze

 This expert can then be used to provide projections given new situations of interest and answer "what if" questions http://www.doc.ic.ac.uk/~nd/surprise_96/journal/vol4/cs11/report.html

 Rule-based system for capturing knowledge

about internal medicine

 100,000 rules

Issues in decision support

systems

 Standard vocabularies are needed

 Chance of error is hard to calculate

 Garbage in, garbage out

 Programming errors or omissions

 Knowledge is almost certainly incomplete

 Use of DSS may emphasize symptoms and data rather than social/psychological

information that are harder to quantify

Issues in decision support

systems

 Major focus in DSS development has been

on diagnosis and therapy planning

 Some suggest that real support is needed in disease management

 DSSs have not addressed this area

 Artificial intelligence is still an area of

research, not practically viable (yet)