Multi-agent architecture for the Virtual Health Record

This article marks one of the first moments when my research on the Virtual Healthcare Record makes an explicit transition towards an agent-based architecture. If in previous texts the emphasis was primarily on the longitudinal record and continuity of care, here the idea emerges that the virtual record must be surrounded by a population of agents capable of mediating information exchange, supervising care plans, integrating medical devices, and supporting cooperation among healthcare system actors.

Bibliographic Reference

Authors:
Luca Dan Șerbănați, Andrei Vasilățeanu

Title:
Multi-agent architecture for the Virtual Health Record

Published in:
Proceedings of the Conference on WWW/Internet

Location: Timișoara, Romania
Year: 2010
Type: conference paper

Download article PDF

Article Figure

Thumbnail (click to enlarge)

Figure 1 – Local agents interact with the world

Fig. 1. Local agents interacting with the real world: medical devices, service providers, care plans, and VHCR infrastructure.

Conceptual Contribution

The novelty of the article lies in the fact that the Virtual HealthCare Record is no longer treated merely as a repository or a simple mechanism for integrating clinical documents, but as a central service surrounded by a population of software agents.

The article brings to the forefront several ideas that will become important in subsequent works:

the transition from the longitudinal record to an agent-based architecture;
correlating continuity of care with monitored clinical workflows;
integrating medical devices through adapter agents;
the role of agents in discovery, cooperation, notification, and synthesis;
the connection between VHCR and an open, distributed infrastructure capable of semantic interoperability.

Description

The article starts from the idea that the Electronic Health Record is a central component of an e-health infrastructure, but that, in its classic form, it is not sufficient to truly support continuity of care. Therefore we propose a model called Virtual HealthCare Record – abbreviated in the text as VHCR – understood as a patient-centric platform, with a longitudinal character, capable of integrating data from distributed and heterogeneous sources and interacting with a dynamic, nondeterministic world made up of numerous actors in the care process.

Right from the start, the article very clearly formulates the problems the system must solve. First, it's about integrating information originating from medical documents generated in different places and updating a single, coherent entity – the VHCR itself. Then comes the problem of semantic interoperability, because agents must be able to interpret and translate messages from different standards. Furthermore, there is medical device integration, the longitudinal nature of the record, the supervision of clinical workflows, and finally, the issue of patient awareness and notification.

The VHCR is described in a very interesting way: not as a document repository, but as a healthcare service provider. It keeps the information extracted from clinical documents in a finer model than the documentary model and organizes it in a relational structure, precisely to enable complex queries and a coherent view of the patient's condition. Source documents are not lost, but their information is reinterpreted and integrated within the context of concepts such as health issue, episode of care, and contact.

This organization is essential. Health issue designates the actual health problem, with its life cycle, sometimes very long, as in the case of a chronic condition. Episode of care logically groups the relevant data for an instance of treatment and enables both contextualized access to information and finer control of confidentiality rights. Within the episode, contacts appear, the concrete encounters between patient and service providers, each resulting in documents, prescriptions, referrals, test results, or care plans.

From this point, the article makes the decisive step: around the VHCR, an agent community is built. The first is the VHCR Agent, the only one that knows the internal structure of the record and is responsible for data integration and queries regarding the current condition, history, and ongoing treatments. Then comes the Synthesis Agent, which produces specialized summaries in different formats – for instance CCD, CCR, or proprietary formats – for clinics and external providers.

A distinct group is formed by Device Agents, which represent medical devices in the virtual space. They acquire data from the device and transmit it to the VHCR Agent, and can also send configuration requests to the device. The idea is important because it shows how real-time measurement of vital parameters can be integrated into the longitudinal record.

At the same time, the article introduces collaboration agents for doctors' offices and clinics. Each partner can be associated with an HCP Agent, responsible for receiving signed documents, transmitting them for integration, and routing queries. Thus, the cooperation between heterogeneous applications is shifted to a more flexible logic, where agents mediate the interaction between systems and the virtual record.

Very interesting are the Clinical Workflow Agents. They monitor the execution of care plans associated with an episode and a health issue. A plan can contain observations, procedures, appointments, or drug administrations, and the agent tracks whether the activities marked as mandatory actually take place. If the patient does not confirm, for instance, that they have taken a daily medication, the agent can generate a notification. We can clearly see here one of the directions that will develop later: the transition from simple information storage to orchestrating the care process.

Equally suggestive are the Health Monitoring Agents, reactive agents based on <situation>.<action> rules. They are notified when new measurements appear, compare them with acceptable thresholds, and can generate alarms or notifications to the patient or doctor. The article even provides an example concerning the early detection of a flu, based on temperature and some symptoms declared by the patient. This idea anticipates very well the later interest in context awareness, continuous monitoring, and distributed intervention.

Overlaid on all these are the support agents: a Broker Agent for discovery and communication, an Agent Lifecycle Manager for creating and removing agent instances, and an ontology translation agent for terminological translation between vocabularies, including by calling UMLS. In this way, the article doesn't just offer a few examples of agents, but outlines a true internal ecology of the system.

Technically speaking, the article is still anchored in a classic enterprise architecture: VHCR is implemented component-based, using Enterprise JavaBeans 3.0, and the JADE platform is chosen for agents, along with rules in JESS. The possibility of using WADE for workflows is also discussed, and communication complies with FIPA specifications. This combination is typical for that moment: a robust enterprise core, on top of which a multi-agent infrastructure starts to be layered.

In retrospect, the article has special value precisely through its nature as a transitional text. It does not yet reach the digital health ecosystem in the full sense that the concept will later acquire, but it takes the decisive step from a virtual medical record towards a world of agents, devices, workflows, rules, and distributed cooperation.

Place in the Research Journey

In the economy of my research, this article occupies an essential place. It still belongs to the Virtual Healthcare Record branch, but directly paves the way for subsequent works on workflow-based security, PHR integration, digital health ecosystems, and later, agentification.

In other words, a shift in emphasis occurs here: the longitudinal record is no longer just the object of informational integration, but becomes the center of an infrastructure of agents capable of cooperating, reacting, and supervising clinical processes. This is why the article deserves to be treated as one of the early links of the direction later developed with Andrei Vasilățeanu.

Also evident here is the continuity between the practical experience of designing the VHCR and the subsequent reflection on digital ecosystems and avatars: before arriving at the notion of the Digital Health Ecosystem, an intermediate stage was needed where actors, devices, and processes were already represented and orchestrated by software agents. This article is exactly one of the pieces of that stage.

Luca Dan Șerbănați

Emeritus Professor at Politehnica University of Bucharest

Research, teaching, industry and memoirs