RulesFest 2011 - Dr Emory Fry: A Knowledge Management Architecture for Healthcare

Dr Emory Fry delivered a presention on the design and implementation of a Knowledge Management repository supporting the immediate and long term care support for medical patients. (As a note, Paul Haley must be happier that by now a number of talks have used the term “knowledge” rather than the term rules. That is a good trend.)

Dr Emory Fry at RulesFest 2011

Healthcare, as a domain, is one of the most difficult to tackle in Decision Management:

• lots of unstructured and distributed data
• regulations
• complex orchestration
• individualization requirements
• education imperatives
• resource allocation
• and let’s not forget that we are dealing with live (hopefully) humans
• etc

which present a significant challenge to formalization and automation. Dr Fry introduced the term “behavorial entropy” to describe this key challenge.

The requirements for the knowledge management architecture include:

• real-time analysis capabilities
• share knowledge modules across systems and organizations
• handle distributed data
• able to handle decision support for populations at large as well as individuals

The outcome of the work by Dr Fry and his team is a relatively sophisticated architecture including a significant number of components. Some of these are the following:

• Common Access Layer, based on HL7
  Access to legacy data, hand-mapped into the canonical data model
• Event service
  Real-time service managing both internal and external events converted to the common representation so that they can be transparently consumed by the decision support systems
• Decision Support Agent
  Leveraging FIPA and Drools (yes, Mark), agents react to events by extracting the relevant rules from the repository, executing and triggering the appropriate results. Human case management is also included, triggered through events.
  The architecture includes both stateless interactions and long running stateful interactions and includes serialization / deserialization of engine state.
• Session Virtual Medical Record
  Composition of facts into a “virtual” medical record then fed into the decision support agent.
• Action Agent or Manager
  Provides binding to actions from the decisions – for example unified communications management
• Predictive model service
  Leveraging the formalization of predictive models into PMML imported into the inference engines
• Optimization service
  Essentially resource allocation and planning
• Presentation services

This is a full blown decision management application such as those that I have seen deployed a few times. In particular, the architecture shows a good maturity in terms of leveraging the right components for the right problem. Combining events, agents, rules, analytics and presentation on top of a normalized data and event model provides a solid base to build agile decision-centric applications on.

Dr Fry insisted, with reason, on the importance of providing the right user interface to the business user. This is, of course, something everybody agrees upon, but something that not many invest a lot of cycles on. A key aspect of the user interface and the workflows involved is the immediate access to intelligent unified communication management, and the clear and transparent handling of both delegation and transfer of responsibility.

His user interface also adds a semantic wiki – based on Wikipedia information – providing educational and informational content, and essentially forming the context for the human or automated decisions. An additional component is problem specific calculators and simulators that allow the users to perform various explorations directly within the user interface and within the context set by the tool.

Dr Fry hit on the issue that rules editing frameworks provided by the vendors do not allow his business users to manage their knowledge, or the orchestration of processes that they involve. For that purpose, a custom built drag-and-drop orchestration builder was developed that business users should be able to use.

This was a good presentation – emphasizing the specifics of the Healthcare industry. A couple of aspects I did not see addressed but that add even more complexity:

• Unstructured data – a large part of the data in Healthcare is unstructured or semi-structured. Converting them to HL7 compliant document is not a small task, and the sources of lots of challenges.
• Even with structured data, values are largely fuzzy. Values such as “from time to time”, “hurts a little bit around here”, “forever”, “never”, etc…, are often part of the captured information and normalizing it is another enormous challenges. Full standardization efforts have been launched on this.

Interesting talk.