The requirements for semantic infrastructure are defined as they relate to the architecture, use cases, and stakeholders. This section presents functional requirements with tracing up to the use cases and down to the service capabilities specified later in this document. This section is not an exhaustive list of requirements and is expected to evolve as additional requirements are analyzed and defined. In addition, Semantic Infrastructure 2.0 will fully support existing caDSR users, including supporting forms created in caDSR.
This section provides a description of the following requirement categories:
The requirements listed above address one or more use cases in each domain. In addition to the domain specific use-cases, the requirements also address CBIIT internal development and architecture requirements. Specifically, CBIIT has standardized on Service-Oriented Architecture (SOA) as the foundational principle for applications architecture and interoperability. CBIIT has also adopted a formal approach, Enterprise Conformance and Compliance Framework (ECCF), for defining service specifications. These service specifications (as defined in the NCI CBIIT SAIF Implementation Guide address both the requirements for supporting semantic interoperability, and the need to publish formal specifications that can be adopted by external organizations and vendors.
The following sections provide a detailed description of the requirements categories. Where possible the requirements are tied to specific use cases described in the previous section.
Artifact Management
Artifact management includes support for different formats of models (for example, Unified Modeling Language (UML), Web Ontology Language (OWL), or text), both static and dynamic, as well as the ability to manage content and clinical forms. Artifact management primarily deals with managing artifact lifecycle and authoring of artifact metadata.
A service specification is made up of service metadata, artifacts and the metadata supporting these artifacts. Artifact management enables creating a service specification and helps to accomplish the following:
Improve visibility through publication. When the management service can be integrated into the development, testing and production cycle, artifacts become available for review and discussion, as well as reference for supporting development. This helps insure proper understanding of applications and services being developed, and provides a standard and controlled method of access.
Annotate artifacts to expand understanding. To further improve the understanding of artifacts, the management service provides the ability to add annotations to both the parts of an artifact (depending on artifact type) and the artifact as a whole. Adding additional semantic definitions to an artifact allows for the searching and location of elements across artifact type, as well as makes clear the intent of a given artifact.
Support governance. When the management services allows for artifact versioning, along with state representation, artifact elements which require governance can be located and interacted with. This functional aspect of the artifact management provides a change history as well as links to external change control systems.
Types of Artifacts
Static Models
Static models include a variety of models with different representations. Static models include but are not limited to:
- Syntactical and semantic models: XML, OWL, RDF representations
- Information Models: HL7 MIF, UML, 11179 representations
- Meta Models
- HL7 RIM (Reference Information Model)
- BRIDG (Biomedical Research Integrated Domain Group)
- LS-DAM (Life Sciences Domain Access Model)
- Transforms
- Object Management Group (OMG) Ontology Definition Metamodel Tranforms
- Model Constraints
- Object Constraint Language (OCL), Schematron
- Data Types
- ISO 21090 and HL7 R2
- HL7 R1
- Primitives
Behavioral Models
In the context of this paper, behavioral dynamic models capture the behavior of services. Behavior of services provides an unambiguous definition of the service constraints, capabilities, dependencies and interactions. The metadata and grammar required to realize service behavior is called behavioral semantics. Behavioral semantics provide a mechanism for better service discovery and enforcing the constraints at design and runtime.
Dynamic models include but are not limited to:
- HL7 SAIF (Service-Aware Interoperability Framework) behavioral model (which provides a formal model and grammar for service contracts)
- Orchestrations and Workflows
- Business Rules
Content
Content includes all unstructured text and other forms of content that make up a service specification. Examples include storyboards and scope. Content is an integral part of service specification, and content is leveraged across the enterprise for documentation and communications. Content includes:
- Service specification content, primarily unstructured text
- Images and other representations of static content
Forms
Forms include Clinical Data Interchange Standards Consortium (CDISC) Operational Data Model (ODM), HL7 Clinical Document Architecture (CDA) documents, and HL7 Version 3 RIM derived forms. This includes all aspects of the document including the style, definitions and semantics. CDISC and NCI CBIIT require a Distributed, Collaborative Form Template Development Environment and a Distributed Knowledge Repository to capture and manage its metadata. The following are required:
- Form Templates
- Reusable Form Sections
- Form Definitions
Specification Content
The National Cancer Institute has created many specification documents which include extended datatype flavors for the ISO 21090 datatypes as well as the ECCF specifications for the behavioral framework, information framework, and governance framework. The specifications are an integral part of the semantic infrastructure, allowing the user to fully understand and appropriately apply the many artifacts stored in the ECCF registry.
Artifact Management Functions
Artifact lifecycle management and metadata requirements include the ability to:
- Manage lifecycle, governance and versioning of the models, content and forms
- Establish relationships and dependencies between models, content and forms
- Determine provenance, jurisdiction, authority and intellectual property
- Create representation and views of the information, realized through the appropriate transforms
- Provide access control and other security constraints
- Create annotations for better discovery and searching of artifacts
- Develop usage scenarios and context for the information
- Provide terminology and value set binding
- Provide rules and algorithms for the use of the artifacts in a particular service
The artifacts are bound to the services via the service metadata. The service metadata combined with the artifacts and supporting metadata provide a comprehensive service specification.
The artifact management requirements listed above are derived from the following use cases:
Electronic Health Records: The caBIG® Clinical Information Suite project has adopted ECCF for specifications and Clinical Document Architecture (CDA) documents for interoperability. Project requirements include the need for an infrastructure for managing all the artifacts generated during specification process, including HL7 models and documents. The project also intends to publish these artifacts for the community and vendors. The infrastructure must support better discovery, making all the relevant information available in the right context.
Office of the National Coordinator and other external EHR adopters: ONC has adopted the Continuity of Care Document (CCD) and Continuity of Care Record (CCR) for meaningful use. All national EHR implementations are expected to support forms and the semantics of these forms play a critical role in interoperability. The semantic infrastructure must provide a mechanism to create, store and manage these forms.
Clinical Trials: Clinical trials use forms to capture clinical information, and the semantics captured by these forms are critical for interoperability and reporting. The semantic infrastructure must provide a mechanism to manage the lifecycle of these forms.
Service Discovery and Governance
Service discovery and governance allows service developers to specify rich metadata about services. This enables better discovery, and governance of services. Service discovery and governance help to accomplish the following:
Promote service reuse: The use of well defined service metadata promotes better discovery and reuse of services during design and run time. Service metadata includes information about service interactions and dependencies. It also includes a classification scheme for organizing services based on business objectives, domain, and usage. It links services to all the supporting artifacts in the specification and provides a placeholder for conformance statements. This enables better reuse across the enterprise and eliminates redundancy.
Establish service policies: Service policies help establish constraints on the service specifications and mandate an approach. Policies can be specified around governance, access control and other design and runtime constraints.
Provide governance: This includes predefined templates, workflows, and governance policies for governing the service lifecycle as well as an approval and review process for service specifications and the ability to promote services through the stages of the service lifecycle.
Enable better discovery: Complex search offers a natural and user-friendly way to find services by progressively refining search results using a variety of criteria including attributes, artifacts, classification, usage scenarios, and dependencies. This includes runtime contract discovery, a powerful query mechanism that allows either the service orchestrator or a program to find the services that best fit the requirements of a given process. This increases both runtime and design time flexibility by enabling selection of services based on computable metadata.
Service Discovery Functions
Service discovery functions include the ability to:
- Identify the service endpoint for analysis
- Identify the service directory endpoint for analysis
- Extract the service interface
- Annotate the service interface providing undiscovered features or behaviors
- Manage lifecycle, governance and versioning of the service interfaces
The requirements listed above are derived from the following use cases:
Electronic Health Records: The caBIG® Clinical Information Suite project is developing service specifications and lacks the infrastructure to govern these services. Vendors and external implementations are expected to leverage the caBIG® Clinical Information Suite service specifications and there is currently no infrastructure that allows easy discovery and consumption of this information.
CBIIT Projects: CBIIT has adopted SOA. Service lifecycle management and governance are industry best practices for all organizations adopting SOA. Better service discovery and reuse improves productivity, avoids redundancy and makes it easier for the CBIIT enterprise architecture governance team to manage NCI's enterprise services portfolio.
Life Sciences: Service discovery based on a rich metadata and semantics of the underlying data play a critical role in developing research pipelines. Research pipelines are developed by connecting data and analytical services together to achieve a research objective.
Other national initiatives: All EHR vendors and national initiatives rely on a services paradigm for integration and interoperability. A standardized services metamodel makes it easier for participating organizations to discover and reuse services.
caGRID 2.0 Platform: The caGRID 2.0 Platform provides a runtime registry for service discovery. This service registry relies on a small subset of information for discovery. The semantic infrastructure provides a mechanism to leverage rich service and artifact metadata to extend this capability.
Clinical Data Forms Definition and Modeling
Clinical Data Forms are the primary channel for capturing information in the healthcare and clinical domain. Forms also play a key role in information exchange and are critical to supporting interoperability in healthcare.
A form differs from a document, in that a document is used to capture information, while a form defines skip patterns, validation rules, and other aspects required to capture or render information for a document.
A document in this context is specifically a clinical document which represents information about a clinical activity. The document contains the specific information gained during that clinical activity and supports the broader definitions of a document. Documents can be transformed into human readable forms, and be transferred or transmitted electronically for use across different systems.
Clinical data forms definition and modeling help to accomplish the following:
Define data entry forms using robust data representation. Ultimately the data that is captured on a form is used in many ways, but that data must provide a high level of meaningful use to insure the consumer knows how the data was captured and what context it represents. In this way even a simple question on a form may result in a much more complex representation in the data. As an example, a Yes or No question on a form may result in a codified representation of an observation.
Reuse contextual representation. Since a given form may collect data for a context that might be common to many forms, being able to reuse these elements in a way that insures contextual consistency is a must. Forms created with the form definition tool must retrieve from well defined metadata sources that provide common contexts, default values, and coded representations including value set binding.
Reuse form elements. When defining a form element which is bound to a specific contextual representation, it should be easy to reuse that element with minimal reconfiguration.
Provide governance support. Forms and the supporting schemas need to be versioned as well as support the governance workflows. This insures that documentation follows a consistent and planed use.
Clinical Data Form Functions
The functions of clinical data forms include the ability to:
- Define model objects for reuse
- Define form templates
- Bind value set to data element
- Provide default form delivery
- Provide form data transformation
- Manage lifecycle, governance and version of forms and document schemas
Based on the use cases the key forms requirements include:
- Tools and services for defining form templates
- Ability to leverage models and reusable segments for defining these forms
- Ability to bind terminology in the form of value sets to form controls
- User friendly tools that hide the complexity of the underlying semantics
The requirements listed above are derived from the following use cases:
- Electronic Health Records
- ONC and Other external EHR adopters
- Clinical Trials
Decision Support and Reasoning
One of the primary reasons for having structured data is to provide the ability to automate decision support and reasoning across information models, data types, and the terminology associated with the attributes of each data type. For the ECCF registry to provide maximal value to end users, it is necessary to support common decision support functions across the enterprise and to extend that through services to the end users. In effect the semantic infrastructure must provide the tools to support Decision Support solutions:
Identify sources of valued information. Using the semantic metadata as a source, reasoning systems need to be able to identify the sources of information which are key to a given decision support solution. The services, models, and annotations provide definitions which can identify candidate sources for integration.
Common representations and transformations. To make decision support services viable, it is necessary that information be consistent and provide the ability to transform data for use in various tools and reasoning solutions.
Support for classification. The system provides for data classification, discovering new knowledge about key elements. This classification process is based on description logic and business rules which process the semantic structures of artifacts. Classification information should be added to the pool of knowledge about given structures and related information
Support for expert system rule processing and choreography. Using systems such as the OWL classifiers (Pellet, Fact++, Hermit), rule based expert systems (Jess, Drools), and work with RDF (Resource Description Framework) choreography languages (SPIN), the decision support system should be able to applied in a choreographed layered fashion. Key to this process is a choreography engine which matches data with rules and a reasoning environment. Because of the complexity of the reasoning requirements, the OWL 2 specification is required in order to support the Semantic Infrastructure 2.0 requirements.
Integration with service registries. Since the artifact metadata provides definitions of data, the service registry provides the data access needed to process information. If a given artifact is a service, the decision support system determines the necessary definitions to integrate a service into decision support for the gathering of data.
Decision Support Functions
Decision support functions include the ability to:
- Query artifact metadata to locate useful artifacts for decision support
- Query service metadata to locate services matching artifacts and metadata definitions
- Create a decision support definition
- Create a decision support session
- Provide scheduling and access information to choreographer
- Select rules and rule system environment
- Execute reasoning systems against gathered data providing classification and additional data
The requirements listed above are derived from the following use cases:
- Electronic Health Records
- Clinical Trials
Conformance Testing
Services specifications developed by NCI and the community have to be testable to ensure that the implementation conforms to the specification. Conformance testing leverages the artifact and service registries along with predefined reasoning systems to validate that an implementation adequately addresses the requirements stated in the service specification. An example of service requirement is the ability to specify a response time in the specification (design time) and validate that this response time is adequate for an implementation of the service. Additional test points include but are not limited to binding to specific terminologies and domain models.
Conformance testing allows both CBIIT and other HL7 SAIF adopters to validate specifications as follows:
Analyze a given artifact for its stated ECCF purpose. Determine if a given artifact satisfies the requirements of the ECCF artifact that it declares itself to be. This analysis should look at such things as datatypes matching the appropriate level (abstract data types in a Platform Specific Model (PSM)).
Analyze a given artifact to verify traceability. Determine if a given artifact provides correct traceability from level to level. The analysis should look at naming conventions and stereotypes to determine correctness along with promotion of data types from different levels of abstraction.
Analysis of accessibility and interoperability. Used to determine if a given service matches its proposed service specification. Also determine if an artifact or specification is complete as it relates to data binding and value set binding.
Conformance Testing Functions include the ability to:
- Analyze an artifact for ECCF Conformance and traceability
- Produce a non-conformance statement
- Interact with governance systems
The requirements listed above are derived from the following use cases:
CBIIT's adoption of ECCF: ECCF requires all specification developers to make conformance statements; the conformance testing framework leverages these conformance statements to generate validation tests.
Other national initiatives: Other national organizations like NIST are adopting a similar approach to conformance testing.
caGrid 2.0 Platform and Terminology Integration
The semantic infrastructure has to support seamless integration with the caGrid 2.0 platform. The following are some high-level platform and terminology requirements that are either supported or addressed by the semantic infrastructure.
Service Generation
Service generation is the ability to generate services from user defined service metadata. The semantic infrastructure provides this metadata and the platform leverages this metadata for service generation. The constraints and policies specified in the semantic infrastructure are inherited by the platform and are enforced as runtime policies.
Additional platform specific and runtime information is provided by the developer at the time of service generation.
Service Discovery and Utilization
This group of requirements focuses on enabling developers of composite services and applications to discover, compose, and invoke services. This includes the discovery of published services based on service metadata and the generation of client APIs in multiple languages to provide cross-platform access to existing services.
Discovery includes service discovery, data discovery, and policy discovery. Service discovery allows primary users as well as secondary users to locate a service specification and instances based on attributes in the service metadata (for example, via a search for specific microarray analysis services). Data discovery enables secondary users to find the types of data available in the ecosystem as well as summary-level information about available data sets. Policy discovery allows application developers to find and retrieve policies on services.
The platform will use the semantic infrastructure service metadata to address all the service discovery requirements. The semantic infrastructure relies on metadata about services and artifacts.
Link to use case satisfied from caGrid 2.0 Roadmap: As institutions share de-identified glioblastoma data sets, they are available to others via data discovery. The treatment recommendation service used by the oncologist is able to discover these new data sets and their corresponding information models, and include that data for subsequent use in recommendation of treatment.
Link to use case satisfied from caGrid 2.0 Roadmap: all of the data management and access services in the use case are utilized by application developers to build the user interfaces that the clinicians use during the course of patient care.
Service Orchestration and Choreography
Service orchestration and choreography allows both application developers and non-developers to discover service "building blocks" that can be composed dynamically to provide business capabilities. Special cases include the orchestration of multiple services for a distributed query, or for a transactional workflow. Service orchestration and choreography will leverage static and behavioral semantics from the Semantic Infrastructure 2.0.
The semantic infrastructure provides the behavioral semantics required for dynamic composibility of services or generation of distributed queries. This includes runtime contract discovery and negotiation to determine composibility of services based on service capabilities and constraints.
Another use case is dynamic retrieval and enforcement of the policies that are in effect for a service interaction in the areas of logging, validations, data transformation, or routing. This information can be used either during the design of the orchestration or during the execution of the defined flow.
Link to use case satisfied from caGrid 2.0 Roadmap: Federated query over The Cancer Genome Atlas (TCGA) data and other data sets is performed using a service orchestration.
Policy and Rules Management
Policy and Rules Management allow non-developer secondary users to create policies and rules and apply them to services. The scope of policies includes, but is not limited to, definition and configuration of business processing policy and related rules, compliance policies, quality of service policies, and security policies. Some key functional requirements for managing policies include capabilities to author policies and store policies, and to approve and validate policies and execute policies at runtime.
The semantic infrastructure will provide a mechanism to specify policies, including business processing policies and related rules, compliance policies, and quality of service policies. Tools and services for creating security specific policies will be provided by the caGrid 2.0 platform and will be used by the semantic infrastructure. All other policies specified in the semantic infrastructure will be enforced by the platform at runtime.
Link to use case satisfied from caGrid 2.0 Roadmap: Each institution has different data sharing needs, access control needs, and business rules for processing that are defined and customized. For example, policy at the pathologist's institution may state that the patient is scheduled for a visit when the review is complete.
Event Processing and Notifications
Event Processing and Notifications enables monitoring of services in the ecosystem and provides for asynchronous updates by services, effectively allowing a loose coordination of services that both provide and respond to conditions (possibly defined in business rules).
The semantic infrastructure will provide a placeholder to specify events and triggering conditions for data and services. The platform monitors these events at runtime and acts on these events.
Link to use case satisfied from caGrid 2.0 Roadmap: As patient care proceeds, the system notifies the designated clinicians that data (for example, images) are ready for review. Similarly, when notifications are received, event processing logic allows the appropriate parties to assign clinicians for care. In order to facilitate better treatment (a learning healthcare system), as new de-identified glioblastoma data is made available, notifications are sent that could indicate a recommended change in the treatment plan.
Data Representation and Information Models
This set of requirements includes providing an application developer with the ability to define application-specific attributes (for example, defined using ISO 21090 healthcare datatypes) and an information model that defines the relationships between these attributes and other attributes in the broader ecosystem. In particular, the last requirement suggests linked datasets, where application developers can connect data in disparate repositories as if the repositories are part of a larger federated data ecosystem. Additional requirements include the ability to publish and discover information models. Support is needed for forms data and common clinical document standards, such as HL7 CDA. To support the use of binary data throughout the system, the binary data must be typed and semantically annotated.
All information models, their representation and binding to datatypes and terminologies will be managed by the semantic infrastructure. The ability to publish and discover information models will be supported by the semantic infrastructure, and the platform will leverage these capabilities.
Link to use case satisfied from caGrid 2.0 Roadmap: The pathology, radiology and other data have various data formats which must be described, and the information model for the patient record must link between these various datatypes. The complete information model includes semantic links between datasets to build a comprehensive electronic medical record. Annotations on data are defined and included in the information model.
Data Management
Data management includes linking of disparate data sets and updates of data across the ecosystem. Data updates may include updates to multiple data sources, necessitating the need for transactions.
Linkages between the different disparate data sets will be managed by the semantic infrastructure. Data updates that trigger transactions are captured by the platform and are propagated upstream to the semantic infrastructure. An example would be the platform monitoring events to identify changes to data.
Link to use case satisfied from caGrid 2.0 Roadmap: the patient has an electronic medical record that spans multiple institutions. The clinical workup data (for example, genomics and proteomics data) is linked to the clinical care record; similarly pathology and radiology findings must be attached to the patient's electronic medical record.
Data Exploration and Query
The wealth of data must be accessible, resulting in the need for exploration of available datasets. This includes the ability to view seamlessly across independent data sets, allowing a secondary user to integrate data from multiple sources. In addition, the query capability must support sophisticated queries such as temporal queries and spatial queries.
The semantic infrastructure will provide metadata for discovery of these datasets. Complex temporal and spatial queries will be informed by the metadata but will be formulated and executed by the platform.
In order to also discover dataset contents exposed on the grid, the ECCF registry must have linkages from dataset metadata to the metadata about the data they contain. This is distinct from the metadata about the dataset (the owner, creation time, table structure of fields and attributes) and instead describes the type of data contents of the dataset so that a user can retrieve portions of a dataset of some type.
Link to use case satisfied from caGrid 2.0 Roadmap: The oncologist must be able to quickly find glioblastoma data sets, indicating the fields that he is interested in comparing from his clinical data in order to find similar disease conditions and associated treatment plans. Temporal queries allow clinicians to identify changes in patient condition and treatment over time.
Provenance
Provenance encompasses the origin and traceability of data throughout an ecosystem. This is a clear requirement directly from the use case in order to ensure that all steps of patient care and research are clearly linked via the patient record.
The semantic infrastructure will provide data provenance support.
Link to use case satisfied from caGrid 2.0 Roadmap: The origin of data is tied to the data creator, allowing the oncologist performing the match against TCGA data and other datasets to include and exclude data sets based on their origin.
Data Semantics
In a diverse information environment, semantics must be used to clearly indicate the meaning of data. This requirement is expected to be addressed by the semantic infrastructure, although there will be a touchpoint between caGrid 2.0 and Semantic Infrastructure 2.0 to annotate data with semantics. Integration with the semantic Infrastructure will enable reasoning, semantic query, data mediation (for example, ad hoc data transformation) and other powerful capabilities.
Data semantics are captured in the semantic infrastructure and the platform will leverage the semantic infrastructure interfaces for reasoning and analysis.
Link to use case satisfied from caGrid 2.0 Roadmap: The oncologist accesses the TCGA database to search for de-identified glioblastoma tumor data that is similar to the patient data exported from the hospital medical record. During this search, the semantics of the data fields are leveraged to indicate matches between TCGA data fields and the hospital medical record data fields.
External Data Repositories
There are numerous data repositories on the web today. These data repositories contain essential information that must be accessible to services in the ecosystem. As a result, caGrid 2.0 must provide capabilities to integrate these external repositories into the grid with the assumption that the remote service cannot be changed.
The semantic infrastructure will support integration with other metadata repositories, allowing the platform to leverage the semantic infrastructure for federated metadata discovery and analysis. The federated data query capabilities will be implemented by the platform.
Link to use case satisfied from caGrid 2.0 Roadmap: The oncologist searches both TCGA glioblastoma data as well as de-identified data that has been added by care providers around the country. The additional data sets are external data repositories.
Other Functional Requirements
These requirements are provided on a child page for convenience, pending division of this section into multiple pages for ease of navigation.
