Clinical Decision Support Systems: Current Trends, Emerging Paradigms

1
Clinical Decision Support Systems Current
Trends, Emerging Paradigms

• Leong Tze Yun, PhD
• Medical Computing Laboratory
• School of Computing
• National University of Singapore

2
What is Clinical Decision Support?

• To provide the right information, in the right
  format, to the right person, at the right place,
  and at the right time to improve health care
  decisions and outcomes
• To facilitate decisions about risk, diagnosis,
  therapy, and follow-up in patient care
• Clinical practice is clinical decision making!

3
Clinical Decision Support Systems
The input Experts domain knowledge, Information
from literature, databases
The output Clear Insights, Smart
choices, Better outcomes
The engine Networked, distributed
systems Advanced modeling and analysis
tools Multiple interfaces GUIs, system
interfaces, web interfaces
4
Common Tools and Applications

• Clinical workflow tools
• Document templates
• Data and imaging reports and dashboards
• Computerized alerts and reminders
• Risk indices
• Diagnostic advices and critiques
• Clinical guidelines

5
Current Trends

• Global mandate on reducing errors, improving
  quality, and lowering cost in health care
• Reports from Institute of Medicine (1999, 2001)
• 58th World Health Assembly (WHA) eHealth
  Resolution (2005)
• Toward personalized medicine
• Integrating clinical, imaging, genomic,
  molecular, socio-economic information to improve
  healthcare outcomes
• Clinical decision support (CDS)
• No longer a why question, but how to put into
  practice?
• Major global CDS initiatives
• US, Asian, Australian, European initiatives

6
Some US Initiatives

• Biomedical informatics for clinical decision
  support A vision for the 21st century (2004)
• National Institute of Health Bioengineering
  Consortium and Biomedical Information Science and
  Technology Initiative Consortium (NIH
  BECON/BISTIC) Symposium
• CDS Implementers Workbook (2004, 2005)
• Healthcare Information and Management Systems
  Society
• A Roadmap for National Action on CDS (2006)
• American Medical Informatics Association

7
Some Global Initiatives

• Singapore Innovative Healthcare IT Strategic
  Plans
• Electronic medical records exchange (2003 - )
• Innovation in health care IT (2006 - )
• Hong Kong eHealth Consortium
• IT in private practice report (2007)
• E-Health Forum (2006)
• Asia Pacific Medical Informatics Association
• Decision Support working group to develop action
  plan in 2007
• Australian National Institute of Clinical Studies
  strategic plan 2005-2008
• Electronic decision support systems action
  planning report 2004
• The European Union e-Health initiative i2010

8
Common Issues

• How to avoid re-inventing the wheel?
• How to get all stakeholders healthcare
  institutions, researchers, industries to work
  together?
• How to develop or deploy all the relevant
  technologies and applications for supporting
  important clinical decisions?
• How to demonstrate feasibility with small-scale
  pilot projects that can generalize or port to
  other settings?

9
Changing Views

• Previously - technology centric view
• How to solve a technical problem accurately?
• Now socio-technical view
• How to support a clinicians workflow tasks
  efficiently?
• Emerging patient centric view
• How to manage a patients conditions and
  preferences cost-effectively?

10
Key Idea

• Clinical decision support must
  cost-effectively address
  patients conditions and preferences, clinicians
  workflow, and technical
  challenges

11
Emerging Paradigms

• National and regional infrastructure
• Interoperable standards
• Sharable tools and interfaces
• Regulatory and evaluation protocols
• Broad-based education and training
• Solution based technology architecture
• Open source working alongside proprietary systems
  
• Advanced computational and communication
  technologies
• Clinical, research, and industrial collaboration

12
Open Source in Health Care

• Open health care standards
• HL7, DICOM, etc.
• Working groups, consortiums and active projects
• AMIA and IMIA Open Source Working Groups
• The Linux Mednews, Openclinical
• EMRs, clinical research, trials, imaging, master
  patient index
• Potential business models
• Goulde, M., Brown, E. Open Source Software A
  Primer for Health Care Leaders. Report by
  Forrester Research. Mar 2006.
• Weinberg, B. Opinion Open-source can stretch IT
  health care dollars. Computerworld Software.
  September 26, 2006
• Open Research Collaboration programs and
  conferences
• IBM, US universities - Open Research
  Collaboration Principle, 2006
• The Linux Foundation Healthcare Day, 2006, AMIA
  OS Workshop 2007

13
CDS Technology Continuum
Basic
Advanced
Information Access
Guided Choices
Knowledge- Based Prompting
Understanding Clinical
Practice
Passive Visualization
Passive Choices
Active Messages
Todays Technology
Reference HIMSS 2001
14
New Enabling Technologies

• Hybrid techniques to support analytic tasks
• Data mining, diagnosis, prediction, optimization,
  discrimination
• Modeling and analytic models, machine learning
  techniques
• New modeling, analytic, and learning algorithms
• Probabilistic graphical networks
• Natural language processing
• Image-based reasoning
• Emerging general technological platforms
• Mobile and ubiquitous computing
• Business intelligence systems
• User modeling
• Systematic evaluation approaches
• Technical, legal, ethical issues

15
Opportunities and Promises

• Global mandate to improve quality and reduce cost
• Clinical decision support is a necessity, not a
  myth!
• Emergence of interoperable standards and open
  collaboration models
• Introduce next-generation decision support
  capabilities based on common infrastructure and
  open collaboration
• Development of integrative, evidence-adaptive
  CDSSs
• Mobile and other communication technologies to
  support practice of cost-effective,
  evidence-based medicine
• New modeling, analytic, and learning technologies
  incrementally incorporated to enhance
  effectiveness

16
The Ultimate Goal!

• The very concept of a decision support system
  itself will fade away, as intelligent assistants
  that can enhance the judgment of healthcare
  workers blend into the infrastructure of
  healthcare delivery.
• Automated decision support will take place with
  every practitioners routine access to clinical
  data in a manner that is unobtrusive,
  transparent, and tailored to the specific patient
  situation.
• Source Musen et al. Clinical Decision Support
  Systems, in Shortliffe and Cimino, eds.,
  Biomedical Informatics, 3rd ed., Springer, 2006

17
Thank you!

• Contact information
• leongty_at_comp.nus.edu.sg

18
References

• (2004). "Electronic decision support systems
  action planning report." National Institute of
  Clinical Studies. Retrieved January 2007, From
  http//www.nicsl.com.au/asp/index.asp?pagemateria
  ls/materials_subject_articlecid5212id409.
• (2005). "Biomedical Informatics for Clinical
  Decision Support A Vision for he 21st Century."
  NIH BECON/BISTIC Symposium (BB2004) Symposium
  Final Report. Retrieved 2007, From
  http//www.becon.nih.gov/symposia_2004/becon2004_f
  inal_report.pdf.
• (2005). "World Health Organization eHealth
  Resolution." 58th World Health Assembly,
  Resolution 28. From http//www.who.int/gb/ee_wha58
  .html.
• Berlin, A., M. Sorani, et al. (2006). "A
  taxonomic description of computer-based clinical
  decision support systems." J.of Biomedical
  Informatics 39(6) 656-667.
• Goulde, M. and E. Brown. (2006). "Open Source
  Software A Primer for Health Care Leaders.
  Report by Forrester Research." California
  Healthcare Foundation. From http//www.chcf.org/to
  pics/view.cfm?itemID119091.

19
References

• Musen, M. A., Y. Shahar, et al. (2006). Clinical
  Decision-Support Systems. Biomedical Informatics
  Computer Applications in Health Care and
  Biomedicine. Shortliffe Edward H. and James J.
  Cimino, Springer 698-736.
• NEDST (2003). Electronic decision support for
  Australias health sector. Canberra, Australian
  Government Department of Health and Ageing.
  Retrieved January 2007, From http//www.health.gov
  .au/
• Osheroff, J. A., E. A. P. M, et al. (2005).
  Improving Outcomes with Clinical Decision
  Support, HIMSS..From http//www.himss.org
• Osheroff, J. A., J. M. Teich, et al. (2006). "A
  Roadmap for National Action on Clinical Decision
  Support." American Medical Informatics
  Association. Retrieved 27 January, 2007, From
  http//www.amia.org/inside/initiatives/cds/.
• Weinberg, B. (2006). "Opinion Open-source can
  stretch IT health care dollars." Computerworld
  Software. Retrieved 30 January, 2007, From
  http//www.computerworld.com/action/article.do?com
  mandviewArticleBasicarticleId9003597pageNumber
  1
• Leong, T. Y., K. Kaiser and S. Miksch, "Free and
  open source enabling technologies for
  patient-centric, guideline-based clinical
  decision support A survey". Methods of
  information in Medicine, 46, no. Suppl 1 (IMIA
  Yearbook of Medical Informatics) (2007) 74-86.

20
A Case Study
21
Project ResEasy

• Experimental platform to support best practices
  in chronic and acute disease and care management
• Open source workflow management, outcome
  analysis, guideline implementation
• Public-private collaboration initiative pilot
  toward cost-effective health care in Singapore
  and the region
• Asthma and acute respiratory distress syndrome
  management (ARDS)
• Funded by the Infocomm Development Authority
  (IDA) and The Enterprise Challenge (TEC) in
  Singapore
• Partners include public and private hospitals,
  university, and engineering companies

22
Our Trial Framework
Pathology Information System
Clinical Information System
Paper Forms

Participating Site 1
Encrypted Disk
Participating Site 2
Participating Site n
Internal Review Board

Certification Authority
Trusted Third Party
Internal Review Board
Encrypted Disk
Hospital
Cluster-wide EMR system
Health Care group
Pathology Information System
Clinical Information System
Paper Forms

23
Trial Settings and Tasks

• ResEasy Asthma
• Singapore National Asthma Program
• Main pilot site National University Hospital
• Process management and guideline implementation
• Risk factor identification
• Information protection
• ResEasy ARDS
• Gleneagles Hospital ICU
• Process management
• Alert generation
• Clinical guideline
• Video streaming and information protection

24
Asthma Workflow Management
Retrieve todays records from PC to PDA
Go to clinic with PDA, update old records/create
new records
Integrate updated records on PDA with database on
PC
workflow
Information flow
25
Asthma Portable Electronic Forms
26
Asthma Action Plan and Asthma Control Test (ACT)
27
Outcome Analysis in Asthma

• Objectives
• Outcome prediction
• Control indication
• Cost-effectiveness analysis
• Resource planning
• Outcome measures
• Asthma under control (control indicators)
• Unscheduled physician visits for nebulization
• Emergency department visits
• Hospitalization

28
ARDS Management System
29
(No Transcript)
30
NIH BECON/BISTIC Symposium

• Theme
• Biomedical informatics for clinical decision
  support A vision for the 21st century
• Establish
• Scientific vision of the future where healthcare
  information technologies may be more fully
  deployed in the clinical workflow to deliver
  efficiency and outcomes
• Areas addressed
• Heterogeneous data collection methods
• Data management (databases and digital libraries)
• Enabling technologies (modeling, software tools,
  techniques)
• Translational informatics

Source BECON 2004 Final Report (2/05)
31
Meeting recommendations

• Establish clinical data collection strategies
• Harmonize data acquisition across biosensors
• Support development and evaluation of
  translational informatics tools
• Adopt software engineering approaches
• Provide mechanisms and regulatory approval of
  software tools
• Foster public private partnerships
• Implement demonstration projects

Source BECON 2004 Final Report (2/05)
32
Some Technical Recommendations

• Software Tools for Modeling, Data Analysis, Data
  Integration, and Workflow
• Support training and development of new curricula
  to facilitate adoption and best practice
  application of information technology in clinical
  research and care.
• Previous directions have focused on data vs.
  tools vs. research vs. practice, rather than a
  more holistic approach.
• Biologists/bioengineers/computer scientists and
  doctors need to function as true peers.
• Focus has been on very large scale data, but
  ultimate impact is measured in terms of clinical
  decisions.
• Adopt a solution architectures approach
  (problem-driven view of entire complex of
  tools/data /computation needed to solve similar
  type of problems, including validation
  requirements)

Source BECON 2004 Final Report (2/05)
33
CDS Implementers Workbook

• Main topics
• Identifying Stakeholders and Goals
• Cataloging Available Information Systems
• Selecting and Specifying CDS Interventions
• Specifying and Validating the Details, and
  Building the Interventions
• Putting Interventions into Action
• Measuring Results and Refining the Program
• Standards Pertinent to CDS
• Medico-legal Considerations with CDS
• Pilot Site Selection
• Additional Statistics and Reports for Evaluating
  Alerts
• Source
• Osheroff, J. A., Pifer, E. A., Teich, J. M.,
  Sittig, D. F., and Jenders, R. A. Improving
  Outcomes with Clinical Decision Support, HIMSS,
  2005.

34
Roadmap for National Action on CDS

• Enhanced health and health care through CDS
  Three Pillars
• Best knowledge available when needed
• Represent clinical knowledge and CDS
  interventions in standardized formats
• Collect, organize, and distribute clinical
  knowledge and CDS interventions
• High adoption and effective use
• Address policy/legal/financial barriers and
  create additional support enablers
• Improve clinical adoption and usage of CDS
  interventions
• Continuous improvement of knowledge and CDS
  methods
• Assess and refine the national experience with
  CDS
• Advance care-guiding knowledge
• Source
• Osheroff, J. A., Pifer, E. A., Teich, J. M.,
  Middleton, B. F., Steen, E. B., Wright, A., and
  Detmer, D. E. A Roadmap for National Action on
  Clinical Decision Support, AMIA, June 13, 2006.

35
EU e-Health Initiative

• Part of the eEurope strategy toward better
  access, quality and effectiveness of care
• Sets out roadmap for greater use of technologies,
  new services and systems, toward objective of a
  European e-Health Area
• Identifies practical steps to facilitate
  communication through
• Developing interoperable electronic health
  records, standard patient identifiers and health
  cards, and high speed Internet access
• Continuing collaboration with National Competence
  Centers
• Consulting relevant stakeholders through a public
  consultation as well as meetings and workshops
• Calls on member states to develop national and
  regional e-Health strategies

Reference European Union Information Society
e-Health website
36
From Research to Practice

• From research to practice
• Targeted e-Health research funding of 1000
  million
• Emergence of new e-Health industry with potential
  to be the third largest industry in the health
  sector with a turnover of 11 billion
• By 2010 it is expected to account for 5 of the
  total health budget of the European Union's
  Member States.
• Frost Sullivan market research report (Aug
  2006)
• Clinical decision support is a nascent market
  with strong growth
• Clinical Decision Support Systems markets in
  Europe earned revenue of 238.5 million in 2005
  and estimates this to reach 430.7 million in
  2012
• Development of robust CDSSs key to increase
  adoption in Europe
• Website http//www.healthcare.frost.com

Reference European Union Information Society
e-Health website
37
Hong Kong eHealth Consortium

• Response to call in 58th WHA eHealth resolution
• A public-private partnership created after the
  SARS and H5N1 outbreaks in 2003
• Key initiatives
• Data sharing and standardization
• Education and capacity building
• eHealth Forum
• Lays foundation for future sharing of healthcare
  information
• Envisions and defines road map for the future
• Enhances communication between public private
  sectors.

38
Australian National Institute of Clinical Studies
strategic plan

• The national electronic decision support
  taskforce report (NEDST, 2003)
• Electronic decision support systems action
  planning report 2004
• Presented to Australian Health Information
  Council's Electronic Decision Support Steering
  Committee, the national body overseeing the
  implementation of the EDSS Taskforces'
  recommendations.

39
Literature-Based Evidence

• Randomized trials, systematic reviews, guidelines
  
• Constitutes only small fraction of research
  literature
• Study design and reporting problems abundant
• Electronic resources mostly not
  machine-interpretable
• The Cochrane Library, Best Evidence, Clinical
  Evidence, etc.
• Emerging machine-interpretable knowledge bases
• The Trial Bank, genomic information databases,
  etc.
• Need advanced free-text understanding techniques

40
Practice-Based Evidence

• Local databases and data warehouses from
• registries and repositories, health information
  systems, electronic medical records, laboratory
  systems, etc.
• Complements and supplements general,
  literature-based evidence
• Required for risk and outcome analysis and
  practice guideline development
• Improve process and intervention designs

41
Research-Based Evidence

• Experimental data and results generated through
  specific design and analysis
• Can be sliced and diced into various formats
  and categories for further processing
• Complements and supplements practice-based
  evidence
• Required for risk and outcome analysis and
  practice guideline development
• Improve process and intervention designs

42
Human-Directed Evidence

• Policy makers or clinicians objectives
• Patients preferences and concerns through
• direct interactions
• feedback from health-related resources, e.g.,
  websites, surveys, etc.
• Increase health care quality through
• Facilitating communication
• Fostering shared decision making
• Personalized care plan
• Improving clinical outcomes

43
Executive Information Systems

• Target users
• Health policy makers, quality assurance managers,
  hospital administrators, medical directors,
  department chiefs, etc.
• Functions
• Integrate information from different sources
• Keep track of internal and external changes
• Identify and monitor resource utilization
• Support risk analysis and risk management
• Objectives
• Achieve strategic vision and mission
• Gain high level perspective on
• key performance indicators
• trends in organization

44
Monitoring and Control Systems

• Target users
• Clinicians, pharmacists, administrators
• Functions
• Selectively monitor clinical data continuously
• Test data against predefined criteria to send
  alerts
• Objectives
• Detect and prevent adverse events
• Alarming laboratory results
• Drug contraindications
• Critical care monitoring

45
Risk or Outcome Prediction Systems

• Target users
• Clinicians, surgery or treatment planning teams,
  health policy makers, quality assurance managers,
  hospital administrators
• Functions
• Perform classification and prediction of outcome
  or risk with respected to specific outcome
  measures, e.g., length of stay, death,
  complications, based on data collected in a
  population
• Derive outcome predictors, staging scores or risk
  stratification indices
• Support risk analysis and risk management at the
  bedside and in policy planning
• Objectives
• Facilitate decision making in routine and complex
  situations
• Serve as educational and communication tools

46
Clinical Diagnostic Treatment Systems

• Target users
• Clinicians, patients, students
• Functions
• Recommend diagnosis and treatment planning
• Detect adverse or specific events
• Critique care management plans
• Objectives
• Facilitate decision making in routine and complex
  situations
• Provide reference and confirmation information
• Support scenario analyses for better insights
• Serve as educational and communication tools

47
Protocol-Based Decision Systems

• Target Users
• Clinicians, patients, administrators
• Functions
• Create, maintain, and access to disease
  management and best practice guidelines from
  different information sources
• Transform often-ignored guidelines to dynamic
  programs for
• real-time patient-specific management advice
• automated recommendations, reminders, alerts, and
  adjustment of device settings
• Support outcomes analysis and outcomes management
• Objectives
• Promote systematic record keeping
• Support rational decision making
• Improve clinician acceptance
• Improve quality and reduce cost of care

48
Rule-Based Techniques

• Knowledge structured as a set of rules
• If A1,A2,A3 then B1,B2 else C1
• Forward reasoning or data-driven reasoning
• If patients serum potassium level is below 3.0
  then assert hypokalemia
• If hypokalemia, then send report to hospital
  staff
• Backward reasoning or goal-driven reasoning
• If fever and runny nose then flu
• If temperature is higher then 36.9C, then fever
• Assert runny nose

49
Model-Based Techniques

• Semantic networks or frames as knowledge
  representations for diseases and processes
• A set of concepts with a set of attributes
• Concept disease
• Name pneumonia
• ICD code 481
• Body part affected lung
• Standard treatment antibiotic
• Inheritance and other inferences to derive
  conclusions from the concept hierarchies

50
Case-Based Techniques

• Diagnosis or prediction based on similarity to
  previous cases and classifications
• Previous cases of patients with common cold
• C1, C2, C3
• Each with slightly different symptoms and
  recommended treatments
• New case D1
• With some symptoms common to C1 and C2
• With some new symptoms unseen before
• Can D1 be classified as common cold?
• If so, can the previous treatments be used?
• If not, what to do with D1?

51
Neural Network Techniques

• Pattern recognition and analysis of underlying
  disease dynamics
• look for patterns in training sets of data
• learn the patterns
• develop the ability to classify new patterns

52
Business Intelligence Systems

• Major functionalities
• Reporting
• Online analytic analysis (OLAP)
• Dashboards
• Data integration
• Data mining
• Technology categories
• Enterprise BI systems (EBIS)
• Query and reporting tools
• Advanced BI tools OLAP/statistical and
  data-mining tools
• BI platforms

53
Probabilistic Network Systems

• Bayesian networks
• Annotated directed acyclic graphs
• Model partial causality structures with
  incomplete or probabilistic information
• Depict and facilitate communication on
  human-oriented qualitative structures
• Problem characteristics
• Diagnosis or classification
• Causal interpretation or prediction
• Multiple input multiple output

54
Example

• Whether or not a person has a History of smoking
  (H) has a direct influence both on
• whether or not he has Bronchitis (B) and
• whether or not he has Lung cancer (L)
• Presence or absence of each of B and L has direct
  influence on
• whether or not he experiences Fatigue (F)
• Presence or absence of L has a direct influence
  on
• whether or not a Chest X-ray (C) is positive
• Assume B, L, H, F, and C are binary random
  variables

55
Queries

• Given that a patient has history of smoking and
  has a positive chest X-ray
• What are the conditional probabilities of the
  person having
• bronchitis, i.e., P(bh,c)?
• lung cancer, i.e., P(lh,c)?
• Direct calculation

56
A Bayesian Network
57
Reasoning with Bayesian Network
58
Local Probabilistic Models

• Subjective probabilities
• Subjective assessment of uncertainty in terms of
  probability
• How to make such judgments?
• What do such judgments imply?
• Theoretical probability models
• Modeling uncertainty with theoretical probability
  distributions
• Theoretical probability distributions have
  well-defined characteristics and useful
  statistics
• Learning probabilities from data
• Construct structure and probability distributions
  from data
• Fit theoretical probability models with data
• Model relationships with data

59
Learning in Bayesian Networks
A, B, C T, T, T T, ?, F T, F, F
?, F, ? F, T, T F, ?, F
P(A)
P(B)
P(A)
P(B)
? ?
? ?
0.4 0.6
0.8 0.2
A
B
A
B
Learner
C
C
A B
P(CA,B)
P(CA,B)
A B
a b 0.1 0.9 a b 0.3 0.7 a
b 0.9 0.1 a b 0.6 0.4
a b ? ? a b ? ? a
b ? ? a b ? ?
Each ? (or ? ) denotes a parameter
60
Application Areas

• Executive information systems
• Monitoring and control systems
• Risk and outcome analysis systems
• Clinical diagnostic and treatment systems
• Protocol-based decision systems