Transcript Proteus - Introduction
PRO
cesses &
T
ransactions
E
ditable by
US
ers
www.proteme.org
Hemant Shah M.D., M.Surg.
Sr. Research Informatician Clinical Systems Research & Integration Henry Ford Health System Detroit, MI
What Is Proteus?
A Model for constructing Clinical Workflows with integrated Decision Support (AKA Guidelines) The workflows are constructed with entities called
Knowledge Components
or
KCs
KCs are: Modular (like Lego blocks) Executable Editable Reusable
The Clinical Process
The Clinical Process
The Clinical Process
Knowledge Component Repository Internet Resource Cardiovascular System Expert Diabetes Expert
Proteus can be used for…
Creating executable clinical practice guidelines to provide decision-support to Clinicians about patient care Creating process-oriented EMR systems with integrated clinical decision-support Creating kernel of integrated healthcare information systems
Proteus Model Contains…
A specification of an architecture for: KCs Executable Workflows (Guidelines) built with KCs Tools and Systems to handle them A graphical notation system for Workflows (Guidelines) – human & machine readable
Knowledge Component (KC)
A modular building block for Clinical Processes To represent Clinical Actions, Events and Processes A software component with a discrete bit of knowledge Complete in itself Can manage its own internal affairs Can be “connected” with other KCs to work cooperatively with them Contains knowledge about a clinical activity: Actions to be performed Events to look for Data to be collected from the actions and events Interpretation and implications of that data Supplementary information about the activities (e.g. links to websites)
Value of KC
Knowledge Component
Knowledge Component (KC)
Abstraction Lump Tenderness Vomiting Temperature
KC Represents: Clinical Process (e.g. diagnosis of acute abdomen pain) Clinical Transaction, which may be: Clinical Action (e.g. palpation of liver) Clinical Event (e.g. vomiting)
KC may contain data-fields describing the underlying clinical entity
Knowledge Component
KCs can be Nested
To represent composite processes To reduce complexity
KC to Guidelines
KCs can be linked by Activity-links
To represent processes To define Guidelines
Guidelines to EMR
Lump Tenderness present severe Vomiting yes Temperature 102
F
Instantiated (executed) KCs become medical record
Knowledge Component
Abstraction Lump Tenderness Vomiting Temperature
Inference tool
Part of KC, yet separate Just an interface Technology neutral Pluggable Decides Abstraction the KC – The value of Activity within the KC
Pluggable Inference Tool
Test A Test B Test C
Knowledge Component Inference Tool User’s System
Action A Action B
Inference tool reference
Inference Tools
•Algorithm •Decision Tables •Decision Theory •Rule Based System •Neural Network •Fuzzy System •Patient assisted decisions •Human expert (even user) •User Defined •User Specified •Combination of these Inference Tool Network Inference Tool Internet
Proteus Graphical Language
A Quest for a Language for Clinical Knowledge For long Medical Informatics has struggled to develop language(s) to express clinical knowledge.
The Languages are needed to express: Facts about clinical medicine Logic useful in clinical medicine Processes of clinical medicine For
Facts
, we have: GELLO HL7 RIM (indirectly) For
Logic
, we have: Arden Syntax GELLO For
Processes
: ?? Guideline Languages
Quest for Language
None of the Languages are Clinician Friendly
Proteus Graphical Language (PGL)
attempt to fill that void PGL Expresses Actions and Events Processes Logic underlying them Intuitive and Graphical Amenable to tooling Interpretable by an execution engine is an
PGL Basics – Two Knowledge Components Atomic KC Process KC
PGL Basics – Atomic KC
Atomic KC Icon
Represents: Action Event Contains: Data Elements The underlying entity also contains Abstraction Inference tool
Atomic KC Icon
Atomic KC Name Atomic KC Value
Data 1 Value
Data 2 Value
Data n Value
PGL Basics – Process KC
Process KC Icon
Represents Clinical process Contains Nested KCs Activity Links Underlying Entity also contains Abstraction inference tool Action inference tool
Process Icon
Process Name Process Value Atomic KC Process Atomic KC
Nested KCs Activity Links
PGL Basics – Activity Links
Represent the sequence of Triggering of KCs and how they are triggered
Activity Links
Inferential Link Sequential Link Synchronous Link Inferential Stop Link Sequential Stop Link
Protean – Prototype Software Tool
A Demo
System Overview
Knowledge Component Servers “Publish” KCs Naming Server Get KC references Healthcare Delivery Organization Organization (a) Inference tool (a) Access KCs Access Inference tool (b) Organization (b) EMR Human Expert as an “inference tool” Inference Tools Expert Knowledge Managers Independent Clinician Knowledge Users
Administrator
Layers for Unlimited Extensibility
Researcher Physician Accountant
A New Ecosystem
Creates References Inference Tool Guideline Author
Knowledge Repository Actionable Literature Inference Tool
References Inference Tool Uses References Inference Tool Modifies References Inference Tool Physician Providing Care Authors
Medical Literature
Medical Author Creates
Inference Tool
Software Developer Personnel Management Inventory Management Performance Evaluation Projections Other… Administration Codes Bills Misc.
Medical Coder Billing Department Others
Proteus - The Great Integrator
The Guideline (Clinical Process) Attached Processes Other Software Tools Medical Informatics Tools
•AI & DS •Vocabulary Tools •Telemedicine •Data Standards •Multimedia Access •Data Entry •EPR
Question: What is the Core Process for the Healthcare Industry?
Proxy Clinical Process
Question: What is the Core Process for the Healthcare Industry?
Answer: The Clinical Process
Hierarchy for Variability within Standards
Disease Population Medical Setup/Skills
Top - Level Guideline
Top Level
Level A Guideline
Level A
Level B Guideline
Level B
Suspected Condition Positive
Clinical Test
Doubtful Negative
Special Test A Special Test B
Positive
Action A
Negative
Action B
Suspected Condition Positive
Clinical Test
Doubtful Negative
Special Test
Positive
Action A
Negative
Action B
Suspected Condition Positive
Clinical Test
Doubtful
Special Test
Negative
New Test
Positive
Action A Action B
Negative
thank you
Physician’s Burden
Burden of Knowledge Medical Knowledge is evolving since 5000 Years Medical Knowledge quadruples during life time of a doctor 2 million medical articles are published every year Not just new facts are added, old ones also change Human genome knowledge is making this even more pronounced Burden of Information 2 Million pieces of information required for normal clinical performance Very fluid information, frequently uncertain 1/3 rd of doctors’ time is spent in organizing information Doctors need to carry in their heads
“a constantly expanding and reinterpreted database”
1/3 rd the cost of Hospitals is for communication
Physician’s Burden
Every year 44,000 to 188,000 patients die of medical errors Many more harmed Patients are becoming more knowledgeable Payment for Performance (P4P) is around the corner
Physicians’ Response
Less than 25% of doctors in USA use computerized health record systems Few reasons: Different tools for different tasks Hard coded applications, not flexible enough to deal with variations in diseases, organizations etc.
Computerized systems require data entry but do not give much back Lack of collaborative decision making
Poor use of Computers by Doctors: Impact A chain is as strong as its weakest link Physician is the weak link
Proteus Model – UML Class Diagram
Proteus Model – UML Class Diagram – Proteus Guideline Component Class
Execution and Inference
C Process KC 1 is executed Process KC 2 is executed Transaction KC A is executed Process KC 2 ’s abstraction is changed Process KC 1 ’s abstraction is changed Guideline’s abstraction is changed Process KC 2 ’s inference tool decides next action. The Cycle is repeated
Protean – A Software Environment for Proteus Guidelines
And Other Ancillary Tools
A Complex Executed Process
Features of Protean
Loading and Display of Guideline Execution Inferencing and Decision Support What actions to perform What events to look for Interpretation based on the data Supplementary information Data Entry Support EMR
Features of Protean
Editing Creating new elements Deleting Modifying existing elements Reuse Changing the Inference tool Changing the inference tool behavior UMLS Knowledge Source Server access to associate an entity with a UMLS term Extensibility – JIT feature
Future Actions
Web-services based repository of KCs UMLS based data dictionary Full EMR features HL7 compatibility ISO/IEC 11179 integration Multiple Interfaces Associated processes capability New graphical engine Enhancements in existing tools
Future Actions
Create inference tools of new kinds Image processing based Calculation based Remote human inferencing Adapt inference tools of diverse kinds Arden Syntax based MLMs Artificial Neural Network based tools Temporal Inferencing
Comparison with Other Approaches
Other models – Modularity Missing PRO
forma
Attributes of the generic task Attribute Description Name Unique identifier of task Caption Descriptive title of task Description Textual description of task Goal Purpose of task Pre-conditions Conditions necessary before a task may be started Trigger conditions Conditions which will initiate a task Post-conditions Conditions true on task completion
Inaccurate lines of separation •Workflow entities are part of the clinical activity entities.
•Inferencing elements are part of clinical activity elements
Other models – why we cant have modularity GLIF Inaccurate lines of separation •Programmatic (inferencing) entities are represented as steps just like clinical activity entities.
Other models – why we cant have modularity GLIF Inaccurate lines of separation •next_step is an attribute in each step which is directive for the next action.
Other models – why we cant have modularity Eon Inaccurate lines of separation •Sequential Step attribute has followed_by within the Guideline entities.
Object oriented design principles
Encapsulation and information hiding
Abstraction
Independent development of entities Different inference technologies Pluggable inference tools Ease of editability Reusability Abstraction of Contained Activities/Observations Semantic containment in contrast with just ‘physical’ or temporal containment
Proteus Others
Accessing the UMLS Knowledge Source Server from Protean to Map KC to a Concept
Other inferencing approaches and deficiencies because of that
Migration or change in data structure
Data and inference tool bundled together
Migration
Proteus based system
Data and inference tool bundled together
Non Proteus system
Reuse
Reuse as in protégé approach. The reusable elements are available to create a tool or service. Once they are created for a context, they become dedicated to it.
Reuse
Reuse as in KC approach 3 types of reuse for KCs.
1.
Reuse after accessing and then modifying 2.
3.
Reuse with different data by remapping Using in a context different than original
The Clinical Process – The Core of Healthcare Associated Process (e.g., Billing)
Data Decision Action Action Data Decision
Core (clinical) Process Almost everything in healthcare can be mapped to the elements of the Clinical Process Proves that clinical process is the core Gives unlimited extensibility
Action Action
Two Critical Assumptions
Physicians will use Proteus based solution Physicians will share their KCs
Why would a Physician use Proteus?
Reduced Errors Lowered litigation risks Lowered professional liability Insurance costs Personal or Local standards for team and subordinates More Compliance with Best Practices Higher reimbursements based on Payment for Performance (P4P) Staying current with the latest medical advances Ability to acquire expertise of others Control over application – Freedom from Developers Communicating unambiguous workflows and clinical logic with colleagues, insurance providers & payers Because it is
Fun !!
Why would Physicians Share KCs?
For the reasons they publish in journals To show to the world their impressive work To make a difference to others’ work To build name recognition in certain areas of expertise To earn
Conclusion In the Healthcare Systems Puzzle Proteus is the Missing Piece