Multi-agent systems in health care An example in

Multi-agent systems in health care. An example in palliative care. Antonio Moreno Multi-Agent Systems Group (Gru. SMA) Research Group on Artificial Intelligence Computer Science and Maths Dep. University Rovira i Virgili (URV) Tarragona, Spain Czech Technical University, Prague May 31 st 2005

Outline of the talk l Introduction – – Information and Communication Technologies Intelligent agents and Multi-Agent Systems l l Pallia. Sys project – l l MAS applied in Health Care Use of ICT and MAS to help to manage the care of palliative patients Research and development challenges on the use of agents in HC Some final thoughts

Information and Communication Technologies l End of 20 th century: enormous development of information technologies – – l Mobile phones Personal and portable computers Personal Digital Assistants (PDAs) Internet Information Society – Easy, flexible and cheap access to information

Computer Science: Intelligent Agents l Definition by Dr Michael Wooldridge: – – An intelligent agent is a computational process that is able to perform tasks autonomously and that communicates with other agents in order to solve problems through cooperation, coordination and negotiation. These agents work in a complex and dynamic environment and interact with it in real time to achieve their goals.

ICT and MAS l l Recent trend: join the intelligent performance of multi-agent systems with the flexible access to information through new technologies. Future scenario: ambient intelligence, in which ubiquitous agents communicate wirelessly to provide intelligent services to users. – In particular, Am. I@Medicine

Basic agent properties (I) l l Reactivity: awareness of the environment Autonomy: control over its own actions Proactivity: anticipation to user’s requests Reasoning/planning (AI): basis of intelligent behaviour

Basic agent properties (II) l l Learning: improvement of its performance Communication: exchange of information with other agents; implies standardization of languages and protocols; allows cooperation, negotiation, . . .

Agent classification l l l Collaborative: group of agents that cooperate in the joint solution of a problem Interface: collaboration with a user to solve a task Internet: manage the search and manipulation of information through Internet Mobile: physical movement through different machines Hybrid: combination of the previous types

Collaborative Agents - Motivation l l l To solve problems that are too complex for a single agent To be able to solve inherently distributed problems To be able to interconnect already existing systems (agentification) Multi-Agent Systems

Multi-agent systems l A multi-agent system is a set of autonomous agents that can communicate (exchange information) and thus negotiate and cooperate in the joint solution of a distributed problem.

Domains of application of MAS l l l Distributed knowledge Joint effort of a set of autonomous entities Problem decomposable in subproblems – Possibly inter-dependent

Health Care problems l Distributed knowledge – l Coordinated effort – l E. g. receptionist, general and specialised doctors, nurses, tests personnel, . . . Complex problems – l E. g. different units of a hospital E. g. organ transplant management Great amount of information – E. g. medical information in Internet

MAS applied in Health Care l Summary of main motivations – – Use of spatially distributed knowledge Coordination of tasks among several autonomous entities Complex problems decomposable in subproblems Personalised information to doctors and patients Example: national organ transplant coordination (Cortés – CARREL, Moreno et al. - URV, Calisti –Switzerland)

Growing interest l l AI in Medicine special issue (2003) Specialised workshops at AA 00, ECAI 02, ECAI 04. – l Int. Workshop on Health Care Applications of Intelligent Agents – February 2003 – l l AI-Communications special issues (2003, 2005) Book on Whitestein Series on Agent Technology (2003) Forthcoming workshop at IJCAI 05, Edinburgh. Agent. Cities WG on HC applications => Agent. Link III TFG on HC applications

Fields of application l l l l Patient scheduling Patient monitoring Agent-based decision support systems Information agents in Internet Community care, care of old/disabled people Access to medical information Management of distributed processes

Outline of the talk l Introduction – – Information and Communication Technologies Intelligent agents and Multi-Agent Systems l l Pallia. Sys project – l l MAS applied in Health Care Use of ICT and MAS to help to manage the care of palliative patients Research and development challenges on the use of agents in HC Some final thoughts

Pallia. Sys Project l l l Integration of Information Technologies and Multi-Agent Systems to improve the care given to palliative patients. Spanish research project, 2004 -05. Work conducted between the Research Group on Artificial Intelligence at URV and the Palliative Care Unit of the Hospital de la Santa Creu i Sant Pau of Barcelona.

Palliative care l l Palliative patients are in a very advanced stage of a fatal disease. The aim of their care is to ease their pain. These patients may be located in hospitals (Palliative Care Units-PCU, or other units of the hospital), specialised hospice centres or at their own homes.

Aims of the Pallia. Sys project l l To improve the process of collecting information from the palliative patients. To improve the access to this information by patients and doctors. To monitor the state of the patients. To apply intelligent data analysis techniques on the data of the PCU.

Information Technologies Multi-Agent System WAP Server Simul. Data Anal. Web Server Web interface PCU Database PCU Head Patient DB Wrapper Patient PALLIASYS Architecture Alarm management Doctor Web interface

Information collection (I) l l l Patients have to send periodically non-technical information relative to their health state. Fill in a form with 10 items to be valued [0 -10] In the current prototype forms can be sent – – – through a web page, or with a mobile phone via WAP (simulated). Other communication means (PDAs, e-mails, SMS messages) have not (yet) been implemented; a study of their potential usefulness is being carried out with a questionnaire given to patients.

Information collection (II – future extensions) l l We could associate an agent to each bed in the PCU, that would periodically send information about the patient status. A doctor might also send information to the system when he is performing a home visit, through an agent running on a mobile phone or a PDA via GPRS. – – We have already been available to implement agents in Nokia ngage mobile phones using the JADE-LEAP environment, and they can communicate succesfully with agents running on a standard PC via GPRS. A MSc-Final Year Project on tourism information using this kind of agents will be presented in June 2005.

Information access l All the data of the palliative patients is stored in a central Data Base at the PCU of the hospital. – l Personal information, family data, auto-evaluations, health record Patients and doctors may make queries on the stored information. – – Patient queries are made directly on the DB (via web or WAPsimulated interface). Doctor queries are made through agent communication (the Doctor Agent requesting the information from the DB Wrapper).

Data Base at the PCU / Security l l There is an agent that controls the access to the Data Base (the Data. Base Wrapper). The whole system includes security mechanisms to protect the privacy of the medical data. – – User authentication (private-public keys) Encrypted messages (SSL) Access through login/password Permissions associated to user types

Information Technologies Multi-Agent System WAP Server Simul. Data Anal. Web Server Web interface PCU Database PCU Head Patient DB Wrapper Patient PALLIASYS Present State Alarm management Doctor Web interface

Patient agents l l There is a patient agent associated to each palliative patient. It has to continuously monitor the status of the patient, and send alarms to the doctor associated to the patient if something goes wrong.

Doctor agents l l A doctor agent is an agent associated to each doctor of the PCU, which would be running in the doctor’s desktop computer. It provides a graphical interface to help: – – – Request information about his patients. Define alarm situations. Receive alarm signals from patient agents.

Classes of alarms l General alarms – l They are defined by the PCU head (through his Doctor Agent), and they have to be applied to all the patients of the unit. Doctor-specific alarms – A doctor can define personal alarms, and he can assign them l l to a single patient, or to all his patients.

Patient auto-evaluation l l l There are 10 differents aspects in patient’s auto -evaluation forms (weakness, pain, anxiety, hunger, etc). Each of the aspects has to be evaluated by the patient with an integer number from 0 to 10. Each patient has to send an auto-evaluation form every 2 -3 weeks.

Alarm types (I) l Alarms defined on a single auto-evaluation – – – (Weakness >7) and (Pain > 8) : extreme_weakness (Hunger < 3) and extreme_weakness: dangerous_weakness Extreme_weakness => patients 1, 3 and 4 Dangerous_weakness => patients 2, 3 and 7. They can be combined with and/or/not operators. Basic alarms can be used to define more complex alarms.

Alarm types (II) l Alarms defined on a sequence of auto-evaluations – – – (Last 2 evaluations a, b) Weaknessb-Weaknessa > 2 : fast_weakness_increase (Last 4 autoevaluations a, b, c, d) Paind-Paina > 3: extreme_pain_increase (Evaluations received in the last 3 weeks) Increase of pain degree > 4 These types of alarms may be defined on the last n evaluations or on the evaluations received in a certain amount of time. The use of Boolean operators and the definition of complex alarm situations is also allowed.

Alarm management l l Alarms are defined by doctors through their Doctor Agents. When an alarm is defined, it is automatically sent to the corresponding Patient Agent (or set of agents). When a new auto-evaluation is stored on the DB, the associated Patient Agent gets a signal, and then it checks all the alarms associated to that patient. If any alarm situation is detected, a message is sent to the Doctor Agent that defined it with an explanation of why the alarm has been activated.

Information Technologies Multi-Agent System WAP Server Simul. Data Anal. Web Server Web interface PCU Database PCU Head Patient DB Wrapper Patient PALLIASYS Present State Alarm management Doctor Web interface

Data Analyser: main tasks l l To apply Data Mining and Machine Learning techniques to analyse the information of the DB. To provide general statistics on the data, which are useful to the PCU head to fill in the annual report.

Available medical data l Input data: sequence of treatment episodes – – – Patient location (home, PCU, socio-sanitary centre) Length of stay (days) Medication received by the patient Medical tests and procedures made on the patient General patient health status

Intelligent Data Analysis l l Generation of patient circuits (circuit graph) Automatic detection of patient states – l l Clustering techniques, unsupervised learning Generation of models of patient evolution (state graph) Generation of decision structures (decision trees, set of rules). – Possibility of making predictions on future states and anticipate and prevent undesired situations.

Circuit graph Movement of each patient among different locations

Training sets for each location

State graph

Numerical analysis of the flow of palliative patients

Palliative patients evolution flowchart

Conclusion - Main ideas l l l Information technologies and Intelligent agents may be used to build useful systems in the Health Care domain. The Pallia. Sys project is an example of the use of those tools. Most of the ideas underlying this project might also be applied in elderly care or home care. – – – Use of Information Technologies Automated patient monitoring Intelligent data analysis

Work to be carried out in Pallia. Sys l l l Explore the use of new ICTs to be used by home patients (SMS messages, e-mail). Implement and test the algorithms of data analysis. Test the final prototype at the PCU of the hospital.

Outline of the talk l Introduction – – Information and Communication Technologies Intelligent agents and Multi-Agent Systems l l Pallia. Sys project – l l MAS applied in Health Care Use of ICT and MAS to help to manage the care of palliative patients Research and development challenges on the use of agents in HC Some final thoughts

Some research topics on the use of MAS in Health Care l l Communication standards Medical ontologies Security mechanisms Implementation of agents in mobile devices – l Personalised access to information – l PDAs, mobile phones Less social and professional reluctance to adopt agent technology Legal issues

General research topics on MAS l l l l Service description, discovery, composition Standard agent communication languages and protocols Negotiation, coordination, cooperation techniques Agent-Oriented Software Engineering Trust Human-agent interaction Integration with legacy software. . .

Outline of the talk l Introduction – – Information and Communication Technologies Intelligent agents and Multi-Agent Systems l l Pallia. Sys project – l l MAS applied in Health Care Use of ICT and MAS to help to manage the care of palliative patients Research and development challenges on the use of agents in HC Some final thoughts

Some general thoughts (I) l It is difficult to work with doctors – – l Very busy, unaware of technical details, change requirements… However, they may end up being happy with a rather simple system (e. g. a well-organised DB, statistics for annual report) It is difficult to sell “agents” to hospital computer units – – Understanding, maintenance, … Information systems are hospital-wide, centralised

Some general thoughts (II) l l Security is a matter of degree … Sometimes “real life” technical issues make it unsuitable to use agents – l Use of previous prototypes or programming languages The frontier between “agents” and “non-agents” seems to be difficult to define.

Multi-agent systems in health care. An example in palliative care. Antonio Moreno Multi-Agent Systems Group, Research Group on Artificial Intelligence Computer Science and Maths Dep. University Rovira i Virgili (URV) Tarragona, Spain http: //grusma. etse. urv. es antonio. moreno@urv. net