Toward a Reference Architecture for Intelligent Systems in Clinical Care

A Software Architecture for Precision Medicine

Intelligent systems in clinical care leverage the latest innovations in machine learning, real-time data stream mining, visual analytics, natural language processing, ontologies, production rule systems, and cloud computing to provide clinicians with the best knowledge and information at the point of care for effective clinical decision making. In this post, I propose a unified open reference architecture that combines all these technologies into a hybrid cognitive system for clinical decision support. Indeed, truly intelligent systems are capable of reasoning. The goal is not to replace clinicians, but instead to provide them with cognitive support during clinical decision making. Furthermore, Intelligent Personal Assistants (IPAs) such as Apple's Siri, Google's Google Now, and Microsoft's Cortana have raised our expectations on how intelligent systems interact with users through voice and natural language.

In the strict sense of the term, a reference architecture should be abstracted away from concrete technology implementation. However in order to enable a better understanding of the proposed approach, I take liberty in explaining how available open source software can be used to realize the intent of the architecture. There is an urgent need for an open and interoperable architecture which can be deployed across devices and platforms. Unfortunately, this is not the case today with solutions like Apple's HealthKit and ResearchKit.

The specific open source software mentioned in this post can be substituted with other tools which provide similar capabilities. The following diagram is a depiction of the architecture (click to enlarge).

 

Clinical Data Sources

Clinical data sources are represented on the left of the architecture diagram. Examples include electronic medical record systems (EMR) commonly used in routine clinical care, clinical genome databases, genome variant knowledge bases, medical imaging databases, data from medical devices and wearable sensors, and unstructured data sources such as biomedical literature databases. The approach implements the Lambda Architecture enabling both batch and real-time data stream processing and mining.

Predictive Modeling, Real-Time Data Stream Mining, and Big Data Genomics

The back-end provides various tools and frameworks for advanced analytics and decision management. The analytics workbench includes tools for creating predictive models and data streaming mining. The decision management workbench includes a production rule system (providing seamless integration with clinical events and processes) and an ontology editor.

The incoming clinical data likely meet the Big Data criteria of volume, velocity, and variety (this is particularly true for physiological time series from wearable sensors). Therefore, specialized frameworks for large scale cluster computing like Apache Spark are used to analyze and process the data. Statistical computing and Machine Learning tools like R are used here as well. The goal is knowledge and patterns discovery using Machine Learning model builders like Decision Trees, k-Means Clustering, Logistic Regression, Support Vector Machines (SVMs), Bayesian Networks, Neural Networks, and the more recent Deep Learning techniques. The latter hold great promise in applications such as Natural Language Processing (NLP), medical image analysis, and speech recognition.

These Machine Learning algorithms can support diagnosis, prognosis, simulation, anomaly detection, care alerting, and care planning. For example, anomaly detection can be performed at scale using the k-means clustering machine learning algorithm in Apache Spark. In addition, Apache Spark allows the implementation of the Lambda Architecture and can also be used for genome Big Data analysis at scale.

In another post titled How Good is Your Crystal Ball?: Utility, Methodology, and Validity of Clinical Prediction Models, I discuss quantitative measures of performance for clinical prediction models.

Visual Analytics

Visual Analytics tools like D3.js, rCharts, ploty, googleVis, ggplot2, and ggvis can help obtain deep insight for effective understanding, reasoning, and decision making through the visual exploration of massive, complex, and often ambiguous data. Of particular interest is Visual Analytics of real-time data streams like physiological time series. As a multidisciplinary field, Visual Analytics combines several disciplines such as human perception and cognition, interactive graphic design, statistical computing, data mining, spatio-temporal data analysis, and even Art. For example, similar to Minard's map of the Russian Campaign of 1812-1813 (see graphic below), Visual Analytics can help in comparing different interventions and care pathways and their respective clinical outcomes over a certain period of time by displaying causes, variables, comparisons, and explanations.

Production Rule System, Ontology Reasoning, and NLP

The architecture also includes a production rule engine and an ontology editor (Drools and Protégé respectively). This is done in order to leverage existing clinical domain knowledge available from clinical practice guidelines (CPGs) and biomedical ontologies like SNOMED CT.  This approach complements machine learning algorithms' probabilistic approach to clinical decision making under uncertainty. The production rule system can translate CPGs into executable rules which are fully integrated with clinical processes (workflows) and events. The ontologies can provide automated reasoning capabilities for decision support.

NLP includes capabilities such as:

• Text classification, text clustering, document and passage retrieval, text summarization, and more advanced clinical question answering (CQA) capabilities which can be useful for satisfying clinicians' information needs at the point of care; and
• Named entity recognition (NER) for extracting concepts from clinical notes.

The data tier supports the efficient storage of large amounts of time series data and is implemented with tools like Cassandra and HBase. The system can run in the cloud, for example using the Amazon Elastic Compute Cloud (EC2). For real-time processing of distributed data streams, cloud-based solutions like Amazon Kinesis and Lambda can be used.

 

Clinical Decision Services

The clinical decision services provide intelligence at the point of care typically using deployed predictive models, clinical rules, text mining outputs, and ontology reasoners. For example, Machine Learning algorithms can be exported in predictive markup language (PMML) format for run-time scoring based on the clinical data of individual patients, enabling what is referred to as Personalized Medicine. Clinical decision services include:

• Diagnosis and prognosis
• Simulation
• Anomaly detection 
• Data visualization
• Information retrieval (e.g., clinical question answering)
• Alerts and reminders
• Support for care planning processes.

The clinical decision services can be deployed in the cloud as well. Other clinical systems can consume these services through a SOAP or REST-based web service interface (using the HL7 vMR and DSS specifications for interoperability) and single sign-on (SSO) standards like SAML2 and OpenID Connect.

Intelligent Personal Assistants (IPAs)

Clinical decision services can also be delivered to patients and clinicians through IPAs. IPAs can accept inputs in the form of voice, images, and user's context and respond in natural language. IPAs are also expanding to wearable technologies such as smart watches and glasses. The precision of speech recognition, natural language processing, and computer vision is improving rapidly with the adoption of Deep Learning techniques and tools. Accelerated hardware technologies like GPUs and FPGAs are improving the performance and reducing the cost of deploying these systems at scale.

Hexagonal, Reactive, and Secure Architecture

Intelligent Health IT systems are not just capable of discovering knowledge and patterns in data. They are also scalable, resilient, responsive, and secure. To achieve these objectives, several architectural patterns have emerged during the last few years:

• Domain Driven Design (DDD) puts the emphasis on the core domain and domain logic and recommends a layered architecture (typically user interface, application, domain, and infrastructure) with each layer having well defined responsibilities and interfaces for interacting with other layers. Models exist within "bounded contexts". These "bounded contexts" communicate with each other typically through messaging and web services using HL7 standards for interoperability.


• The Hexagonal Architecture defines "ports and adapters" as a way to design, develop, and test an application in a way that is independent of the various clients, devices, transport protocols (HTTP, REST, SOAP, MQTT, etc.), and even databases that could be used to consume its services in the future. This is particularly important in the era of the Internet of Things in healthcare.


• Microservices consist in decomposing large monolithic applications into smaller services following good old principles of service-oriented design and single responsibility to achieve modularity, maintainability, scalability, and ease of deployment (for example, using Docker).


• CQRS/ES: Command Query Responsibility Segregation (CQRS) and Event Sourcing (ES) are two architectural patterns which consist in the use of event-driven messaging and an Event Store for separating commands (write-side) from queries (read-side) relying on the principle of Eventual Consistency. CQRS/ES can be implemented in combination with microservices to deliver new capabilities such as temporal queries, behavioral analysis, complex audit logs, and real-time notifications and alerts.


• Functional Programming: Functional Programming languages like Scala have several benefits that are particularly important for applying Machine Learning algorithms on large data sets. Like functions in mathematics, functions in Scala have no side effects. This provides referential transparency. Machine Learning algorithms are in fact based on Linear Algebra and Calculus. Scala supports high-order functions as well. Variables are immutable witch greatly simplifies concurrency. For all those reasons, Machine Learning libraries like Apache Mahout have embraced Scala, moving away from the Java MapReduce paradigm.


• Reactive Architecture: The Reactive Manifesto makes the case for a new breed of applications called "Reactive Applications". According to the manifesto, the Reactive Application architecture allows developers to build "systems that are event-driven, scalable, resilient, and responsive."  Leading frameworks that support Reactive Programming include Akka and RxJava. The latter is a library for composing asynchronous and event-based programs using observable sequences. RxJava is a Java port (with a Scala adaptor) of the original Rx (Reactive Extensions) for .NET created by Erik Meijer.
  
  Based on the Actor Model and built in Scala, Akka is a framework for building highly concurrent, asynchronous, distributed, and fault tolerant event-driven applications on the JVM. Akka offers location transparency, fault tolerance, asynchronous message passing, and a non-deterministic share-nothing architecture. Akka Cluster provides a fault-tolerant decentralized peer-to-peer based cluster membership service with no single point of failure or single point of bottleneck.
  
  Also built with Scala, Apache Kafka is a scalable message broker which provides high-throughput, fault-tolerance, built-in partitioning, and replication  for processing real-time data streams. In the reference architecture, the ingestion layer is implemented with Akka and Apache Kafka.

• Web Application Security: special attention is given to security across all layers, notably the proper implementation of authentication, authorization, encryption, and audit logging. The implementation of security is also driven by deep knowledge of application security patterns, threat modeling, and enforcing security best practices (e.g., OWASP Top Ten and CWE/SANS Top 25 Most Dangerous Software Errors) as part of the continuous delivery process.

An Interface that Works across Devices and Platforms

The front-end uses a Mobile First approach and a Single Page Application (SPA) architecture with Javascript-based frameworks like AngularJS to create very responsive user experiences. It also allows us to bring the following software engineering best practices to the front-end:

• Dependency Injection
• Test-Driven Development (Jasmine, Karma, PhantomJS)
• Package Management (Bower or npm)
• Build system and Continuous Integration (Grunt or Gulp.js)
• Static Code Analysis (JSLint and JSHint), and 
• End-to-End Testing (Protractor). 

For mobile devices, Apache Cordova can be used to access native functions when desired. The main goal is to provide a user interface that works across devices and platforms such as iOS, Android, and Windows Phone.

Interoperability

Interoperability will always be a key requirement in clinical systems. Interoperability is needed between all players in the healthcare ecosystem including providers, payers, labs, knowledge artifact developers, quality measure developers, and public health agencies like the CDC. These standards exist today and are implementation-ready. However, only health IT buyers have the leverage to demand interoperability from their vendors.

Standards related to clinical decision support (CDS) include:

• The HL7 Fast Healthcare Interoperability Resources (FHIR)
• The HL7 virtual Medical Record (vMR)
• The HL7 Decision Support Services (DSS) specification
• The HL7 CDS Knowledge Artifact specification
• The DMG Predictive Model Markup Language (PMML) specification.

Overcoming Barriers to Adoption

In a previous post, I discussed a practical approach to addressing challenges to the adoption of clinical decision support (CDS) systems.

Navigating in Scala Land

In a previous post titled Toward Polyglot Programming on the Java Virtual Machine (JVM), I described my preliminary  exploration of the other languages and frameworks on the JVM including Groovy, Gradle, Grails, Scala, Akka, Clojure, and the Play Framework. I made the switch from Maven to Gradle, a Groovy-based build language that combines the best of Ant and Maven. I was seduced by the scaffolding capabilities of Grails, but decided to make the jump to Scala and functional programming. So I have been navigating in Scala land recently. In this post, I described my journey.

I am still learning, but I can tell you that I never had so much fun learning a new programming language. It probably has something to do with the use of pure mathematical functions in Scala. I did spend a year studying pure mathematics at the University of Abomey-Calavi after graduating from high school. My first exposure to functional programming was with XSLT and XQuery and I very much enjoyed programming without side effects when using those languages. XSLT 3.0 is a fully-fledged functional programming language with support for functions as first class values and high-order functions. XQuery 3.0 is a typed functional language for processing and querying XML data.


Scala is a complex and ambitious language as it supports both object-oriented and functional programming. When I started to learn Scala, I took the wrong directions several times and had to make several U-turns. So the following steps have been effective for me:

• The Coursera class Functional Programming Principles in Scala taught by Martin Odersky (the designer of Scala) is a good place to start. It explains the motivations behind Scala and emphasizes its mathematical and functional nature without trying to map pre-existing knowledge (of Java or Python, or any other language) to Scala. This is a refreshing approach because Scala is a different language although it can interoperate with Java. A good companion to this course is the book Programming in Scala: A Comprehensive Step-by-Step Guide, 2nd Edition by Martin Odersky, Lex Spoon and Bill Venners.


• If you're a Java programmer moving to Scala, then the book Scala for the Impatient by Cay S. Horstmann would be a good reference. 


• If you're interested in building a web application in Scala, then I would recommend the book Play for Scala by Peter Hilton, Erik Bakker and Francisco Canedo. Scala and Play come with their own ecosystem of tools. This includes a Scala-based build system called sbt (simple build tool), testing tools (like Spec2 and ScalaTest), IDEs (Eclipse-based Scala IDE and IntelliJ), database drivers (like ReactiveMongo and Slick), and authentication/authorization (SecureSocial and Deadbolt). Play has first-class support for JSON and REST, supports asynchronous responses (based on the concepts of "Future" and "Promise"), reactive programming with Akka, caching, iteratees (for processing large streams of data), and real-time push-based technologies like WebSockets and Server-Sent Events.


• A this point, if you decide to dive into the deep waters of Scala, you might want to consider learning Reactive Programming and purely functional data structures. 


• There is a second Scala course at Coursera titled Principles of Reactive Programming taught by Martin Odersky, Erik Meijer, and Roland Kuhn. The Reactive Manifesto makes the case for a new breed of applications called "Reactive Applications". According to the manifesto, the Reactive Application architecture allows developers to build "systems that are event-driven, scalable, resilient, and responsive."  In Scala land, there are currently two leading frameworks that support Reactive Programming: Akka and RxJava. The latter is a library for composing asynchronous and event-based programs using observable sequences. RxJava is a Java port (with a Scala adaptor) of the original Rx (Reactive Extensions) for .NET created by Erik Meijer. Based on the Actor Model, Akka is a framework for building highly concurrent, asynchronous, distributed, and fault tolerant event-driven applications on the JVM (it supports both Java and Scala).


• For purely functional data structures, there is a Scala-based library called Scalaz. The book Functional Programming in Scala by Paul Chiusano and Rúnar Bjarnason is a good resource for exploring Scalaz.


• Readers of this blog probably know that I am a proponent of Domain Driven Design (DDD) in building complex software systems. So I have been investigated how DDD principles can be implemented with a functional and reactive approach. Vaughn Vernon recently presented a podcast on Reactive DDD with Scala and Akka. In a post titled Functional Patterns in Domain Modeling - Anemic Models and Compositional Domain Behaviors, Debasish Ghosh provides an interesting perspective on the subject of anemic domain models in DDD done within a functional programming language as opposed to an object-oriented one.


• For me, Big Data is real, not just a buzzword. I believe in analyzing humongous amounts of data to find hidden patterns and obtain insight for solving complex problems. Dean Wampler called copious data, the killer app for functional programming. Scalding by Twitter can be used for writing MapReduce jobs in Scala. Apache Spark which is written in Scala can run Machine Learning programs up to 100x faster than Hadoop MapReduce in memory. A talk titled Why Spark is the Next Top Compute Model by Dean Wampler explains why Spark has emerged as the most likely replacement for MapReduce in Hadoop applications.

How Not to Build A Big Ball of Mud, Part 2

In a previous post entitled How not to build a big  ball of mud, I described the complexity of modern software systems and the challenges faced today by software developers and architects. Domain Driven Design (DDD) is a proven pattern language that can foster a disciplined approach to software development. DDD was first introduced by Eric Evans nine years ago in a seminal book entitled: Domain-Driven Design: Tackling Complexity in the Heart of Software. Over the last 9 years, a community of practice has emerged around DDD and many lessons have been learned in applying DDD to real world complex software development projects. During that time, software complexity has also increased significantly. Changes in the field of software development during the last few years include:

• The proliferation of client devices which requires a Responsive Web Design (RWD) approach. RWD is made possible by open web standards like HTML5, CSS3, and Javascript which have displaced proprietary user interface technologies like Flex and Silverlight. RWD frameworks like Twitter Boostrap and Javascript Libraries like JQuery have become very popular with developers. The demands put on Javscript on the client side have created the need for Javascript MVC frameworks like AngularJS and EmberJS.


• The importance of the user experience in a competitive online marketplace. Performing usability testing early in the software development life cycle (using wireframes or mockups) to test design ideas and obtain early feedback from future users is extremely valuable for creating the right solution. Metrics such as the System Usability Scale (SUS) can be used to assess the results of usability testing.


• The prevalence of REST, JSON, OAuth2, and Web APIs for achieving web scale.


• The emergence of Polyglot Persistence or the use of different persistence mechanisms such as relational, document, and graph databases within the same application. Developers are discovering that modeling data for NoSQL databases has many benefits, but also its own peculiarities.


• The demands for quality and faster time-to-market have led to new techniques like test automation and continuous delivery.

The open source community has responded to these challenges by creating many frameworks which are supposed to facilitate the work of developing software. Software developers spend a considerable amount of time researching, learning, and integrating these various frameworks to build a system. Some of these frameworks can indeed be very helpful when used properly. However, DDD puts a big emphasis on understanding the domain. Here is what I learned from applying DDD over the last few years:

• DDD is a significant intellectual investment, but with a potential for big rewards. To be successful in applying DDD, one must take the time to understand and digest the underlying principles, from the building blocks (entities, aggregates, value objects, modules, domain events, services, repositories, and factories) to the strategic aspects of applying DDD. For example, understanding the difference between an aggregate, a value object, and an entity is essential. Learning the right approach to designing aggregates is also very important as this can significantly impact transactions and performance. I highly recommend reading the recently published Implementing Domain Driven Design by Vaughn Vernon. The book provides a contemporary approach to applying DDD. For example, it covers important topics in applying DDD to modern software systems such as:  sub-domains, domain events, event stores and event sourcing, rules for aggregate design, transactions, eventual consistency, REST, NoSQL, and enterprise application integration with concrete examples.


• Proper application layering (user interface, application, domain, and infrastructure), understanding the responsibility of each layer (for example, an anemic domain model and a fat application layer are anti-pattern), and coding to interfaces. DDD is object-oriented (OO) design done right. The SOLID Principles of OO design are still applicable.


• Determine if DDD is right for your project. Most of my work during the last few years has been in the healthcare domain. The HL7 CCDA and the Virtual Medical Record (vMR) define an information model for Electronic Healthcare Records (EHR) and Clinical Decision Support (CDS) systems respectively. Interoperability is an important and challenging issue in healthcare. DDD concepts such as "Strategic Design", "Context Map", "Bounded Context", and "Published Language" are very helpful in addressing and navigating this type of complexity.


• As I mentioned earlier, DDD puts a big emphasis on understanding the domain. Developers applying DDD should be prepared to dedicate a considerable amount of time to learning about the domain, for example by collaborating and carefully listening to domain experts and by reading as much as they can about the domain. This is also the key to creating a rich domain model with behavior (as opposed to an anemic one). I found that simply reading industry standards and regulations is a great way to understand a domain. So understanding the domain is not just the responsibility of the Business Analyst. The code is the expression of the domain, so the coder needs to understand the domain in order to express it with code.


• Some developers blame popular frameworks for encouraging anemic domain models. I found that a lack of understanding of the domain and its business rules is a major contributing factor to anemia in the domain model. A rule engine like Drools can help externalize these business rules in the form of declarative rules that can be maintained by domain experts through a DSL, spreadsheet, or web-based user interface.


• There are opportunities in using recent ideas like Event Sourcing and the Command Query Responsibility Segregation (CQRS). These opportunities include: scalability, true audit trails, data mining, temporal queries, application integration. However, being pragmatic can help avoid unnecessary complexity.


• I recommend exploring tools that are specifically designed to support a DDD or Model-Driven Development (MDD) approach. Apache Isis, Roma Meta Framework, Tynamo, and Naked Objects are examples of such tools. These tools can automatically generate all the layers of an application based on the specification of a domain model. By doing so, these tools allow you to really focus your time and attention on exploring and understanding the domain as opposed to framework and infrastructure concerns. For architects, these tools can serve as design pattern automation, constraining the development process to conform to DDD principles and patterns. I believe this is part of a larger trend in automating software development which also includes the essential practice of test automation. We software developers like to automate the job of other people. However, many tasks that we perform (including coding itself) are still very manual. Aspect-Oriented Programming (AOP) (using AspectJ for example) can also be used to enable this type of design pattern automation through compile-time weaving.


• Check my previous post for 20 techniques for achieving software excellence.

Software Architecture Documentation in Agile Projects

One misconception that I often hear in Agile circles is that there is no need for software architecture documentation in Agile because "code is self-documenting". The emphasis in agile is not to eliminate the need for design and documentation, but to avoid Big Up Front Design (BDUF). Design and architecture documentation are still important in Agile. However, you only need just enough design and documentation to start coding. In other words, don't over-document.

As you code and refactor, some of the software architecture documentation will become quickly obsolete and should be discarded. Use tools such as Maven, SchemaSpy, Doxygen, and UmlGraph to auto-generate up-to-date documentation from your source code. A wiki is also a good tool for publishing and sharing architecture documentation. For consistency, I recommend using a template for documenting the architecture.

Provide the documentation only if it is really needed and used by stakeholders. So, don't try to document everything. You do need to document the following:

• Design decisions and their rationale
  
• Design patterns and development frameworks used
  
• The architecture viewpoints and quality attributes that cannot be easily gleaned from the code alone.
  

Far too often, software architecture documentation only covers the code view. This is not enough. Stakeholders are not limited to developers, but also include end users, testers, the operational staff, compliance auditors, etc. When writing software architecture documentation, I first identify all stakeholders and their concerns. To ensure that I provide a 360-degree view of the architecture, I develop the architecture documentation based on the viewpoints and perspectives described by Nick Rozanski and Eoin Woods in their book "Software Systems Architecture: Working With Stakeholders Using Viewpoints and Perspectives" (Addison Wesley, April 2005)

The following are the Architecture Viewpoints:

• Functional
  
• Information
  
• Concurrency
  
• Development
  
• Deployment
  
• Operational
  

And here are the Architecture Perspectives:

• Security
  
• Performance and Scalability
  
• Availability and Resilience
  
• Evolution
  
• Accessibility
  
• Development Resource
  
• Internationalization
  
• Location
  
• Regulation
  
• Usability
  

These viewpoints and perspectives can be described using different notations such as UML (using stereotypes and profiles like SoaML for service oriented architecture), Business Process Modeling Notation (BPMN), and Domain Specific Languages (DSLs).