This thesis focuses on theory, methods and technology for developing clinical decision support system (CDSS) for personalized learning and skill development, as is outlined in this figure.
Knowledge representation and transparent methods for reasoning with uncertain and incomplete information constitute the foundation for the CDSS. In addition, methods and technology for knowledge acquisition and management that can improve efficiency and quality are required for building and managing the CDSS.
Apart from assisting physicians to make diagnostic decisions, an important purpose of a CDSS is to educate novice physicians and support their process of developing towards experts. Therefore, evidence-based medicine (EBM) and clinical practice guidelines need to be disseminated through the CDSS in a transparent way to allow clinicians to learn and develop their skills. Moreover, the design of the interactive support for the diagnostic reasoning should meet this requirement.
To address these challenges, three complementary aspects are focused on in the thesis:
1) development of support for the diagnostic reasoning process through an argument-based multi-agent dialogue system;
2) end-user driven development of CDSSs using Activity-Centered Knowledge and Interaction Tailored to Users (ACKTUS), in particular for the dementia domain;
and 3) evaluation of the dementia CDSS motivated by the need for continuing medical education and support for diagnostic reasoning.
The thesis is summarized in the following description of the three aspects.
1) Development of support for the diagnostic reasoning process through an argument-based dialogue system (Paper I-II):
- Research Questions: Novice physicians expressed the wish for a CDSS supporting rapid assessment in an earlier evaluation study of DMSS (Dementia Management Support System). This lead to the research question 1) can the novice physician be allowed to begin with a hypothesis, as in causal reasoning, as opposed to diagnostic reasoning, and still be provided full support for the dementia assessment as in diagnostic reasoning? Additional research questions include how to make a diagnostic decision and educate the end users in the situation when 2) users and data is decentralized in different places/organisations, and 3) each subset of data contains incomplete and uncertain information?
- Objectives: Support a rapid diagnostic assessment beginning with supporting causal reasoning, and support distributed and cooperative decision making, when data are uncertain, inconsistent and incomplete.
- Approach: The approach is to develop argument-based inquiry dialogues, initiated by the user through the selection of a hypothetical diagnosis, and generated by a multi-agent system. The formal inquiry dialogues are based on a combination of possibilistic logic and argumentation theory. To implement and conduct experiments, the inquiry dialogue systems approach presented by Black and Hunter was evaluated, and further developed to become possible to implement in a CDSS. The information models of ACKTUS were used in the implementation of the dialogue system, and the system was integrated in DMSS-web version (DMSS-W) for demonstration purposes.
- Contributions: A multi-agent system was developed, which was based on possibilistic logic, formal argumentation theory, and a developed version of Black and Hunter’s inquiry dialogue systems. Possibilistic logic was used for capturing uncertain information, formal argumentation was used for dealing with inconsistent knowledge in a distributed environment, and inquiry dialogues were used for allowing the user to follow how the assessment process is unfolding for education purposes. Moreover, the combination of these formal theories in the dialogue system allowed the transparent generation of a set of potentially conflicting arguments in favor or against the initially suggested hypothesis. This was expected to provide the users the opportunity to develop skills, even if starting with a hypothesis that is not well-founded, and support their decision making. This approach was implemented and applied in the dementia domain, which makes this contribution into one of the very few examples of implementation in the field of formal argumentation.
The formal results of this work also constituted the theoretical basis for the knowledge representation and reasoning outcomes of the other two aspects of the thesis.
2) ACKTUS as tool for end-user development and knowledge engineering of decision support systems (Paper III-IV), and in particular DMSS-W for the dementia domain (Paper IV):
- Research Questions: i) How can domain experts hands-on model and manage the knowledge in a CDSS?
ii) How can domain experts design the interaction and model the support for the clinical reasoning process with the purpose to facilitate knowledge development in the user? iii) How can a content management system (CMS) facilitate adaptation to different organisations, users and development of knowledge?
- Objective: Develop instruments for facilitating domain experts’ knowledge acquisition and management, and for building new CDSSs.
- Approach: ACKTUS is a platform developed for knowledge engineering and designing knowledge-based systems, designed for allowing end users manage the development. As part of this research, ACKTUS was further extended and used in the development of a CDSS for dementia.
- Contributions: An improved architecture for building CDSSs was proposed and developed. In this new architecture two modules were added that distinguishes the approach from traditional development of CDSS, and a reasoning engine was developed: 1) a CMS built on semantic web technology to achieve modularity, reusability, customization, and the possibility to allow medical experts to model the medical knowledge as well as structuring the information that builds up the design of the user interface; 2) a graphical user interface (GUI) generator that automatically generates the user interface whenever the user logs in, so that the interface is synchronized with updates of the KB; and 3) a reasoning engine that handles uncertainty, and generates hypothetical diagnoses and their level of certainty following clinical practice guidelines. The engine implemented the possibilistic framework presented in Paper II.
A CDSS implemented using this architecture can be further adapted to different situations and education purposes, without significantly modifying the software structure.
Since the domain experts can directly modify the knowledge base, and the content and the information structure automatically displayed through the CDSS user interface, it reduces the need of knowledge engineers and iterations due to misunderstandings, and reduces time for development and maintenance. The CDSS can be tailored to different organizations, and integrate different methods for supporting diagnostic reasoning.
DMSS-W was developed using the proposed architecture, and is currently used by a medical association in Japan.
3) Evaluation of DMSS-W built using ACKTUS, motivated by the need for continuing medical education and support for diagnostic reasoning (Paper V-VI):
- Research Questions: DMSS-W integrates two methods for supporting reasoning about diagnoses in a patient case. Can different reasoning patterns be detected based on log data collected when DMSS-W is used that could motivate the development of personalized reasoning support? Can novices and experts be distinguished by how they use the system, and how could support be tailored to these different categories of users? Can learning and skill development be detected?
- Objective: Explore the possibility to detect reasoning patterns to measure level of knowledge, learning and skill development, which could function as instrument for automated evaluation of a CDSS.
- Approach: At the first stage, four physicians were using the system and their use of DMSS-W was evaluated in the first case study. Then, 29 physicians had access to DMSS-W during a period of two years. Click events were logged and categorized. The data was analysed semi-automatically to find patterns of use. A second case study was conducted where two frequent users (one expert and one novice) were further analysed for the purpose to find differences between the particular novice and expert, which could indicate learning and different needs for support. Another purpose with the case studies was to evaluate a methodology for detecting patterns.
- Contributions: A methodology was proposed to detect reasoning patterns, which was evaluated in the case studies. A typical reasoning pattern was detected where the novice was seen to develop an expert-style forward reasoning eventually, indicating that learning took place when the novice was using the CDSS. The results serve as the starting point for a continued study, where more people will use the system.
My PhD thesis can be found here.
Publications: (the following articles will be included in the thesis)
- Paper I:
Chunli Yan, Helena Lindgren.
Hypothesis-Driven Agent Dialogues for Dementia Assessment.
In Proceedings of VIII Workshop on Agents Applied in Health Care (A2HC), Murcia, Spain, pp. 13-24, 2013.
[pdf].
- Paper II:
Chunli Yan, Helena Lindgren, Juan Carlos Nieves.
A Dialogue-Based Approach for Dealing with Uncertain and Conflicting Information in Medical Diagnosis.
Accepted by the journal Autonomous Agents and Multi-Agent Systems
[pdf].
- Paper III:
Helena Lindgren, Chunli Yan.
ACKTUS - A Platform for Developing Personalized Support Systems in the Health Domain.
In Proceedings of the 5th International Conference on Digital Health 2015, Florence, Italy pp. 135-142, 2015.
[pdf].
- Paper IV:
Chunli Yan, Helena Lindgren.
A Generic Approach for Data Management and End-User Development of Clinical Decision Support Systems.
Technical Report / UMINF 18.08, ISSN 0348-0542
[pdf].
- Paper V:
Helena Lindgren, Chunli Yan.
Detecting Learning and Reasoning Patterns in a CDSS for Dementia Investigation.
Studies in Health Technology and Informatics 210: 739-742, 2015
[pdf].
- Paper VI:
Chunli Yan, Helena Lindgren.
Diagnostic Reasoning Guided by a Decision-Support System: a Case Study.
In Proceedings of the ACM European Conference on Cognitive Ergonomics (ECCE-17), Umeå, pp. 25-30, 2017.
[pdf].
Other Publications:
-
Helena Lindgren, Ming-Hsin Lu, Yeji Hong, Chunli Yan.
Applying the Zone of Proximal Development When Evaluating Clinical Decision Support Systems: a Case Study.
Studies in Health Technology and Informatics, vol 247, pp. 131-135, 2018.
[pdf].
-
Jayalakshmi Baskar, Chunli Yan, and Helena Lindgren.
Instrument-Oriented Approach to Detecting and Representing Human Activity for Supporting Executive Functions and Learning.
In Proceedings of the ACM European Conference on Cognitive Ergonomics (ECCE-17), Umeå, 2017, pp. 105-112.
[pdf].
- Helena Lindgren, Jayalakshmi Baskar, Esteban Guerrero, Juan Carlos Nieves, Ingeborg Nilsson, and Chunli Yan.
Computer-Supported Assessment for Tailoring Assistive Technology.
In Proceedings of the 6th International Conference on Digital Health Conference, pp. 1-10, 2016.
[pdf].
-
Chunli Yan, Juan Carlos Nieves, Helena Lindgren.
A Multi-Agent System for Nested Inquiry Dialogues.
In Proceedings of International Conference on Practical Applications of Agents and Multi-Agent Systems,
Springer, pp. 303-314, 2014.
[pdf].
-
Jayalakshmi Baskar, Helena Lindgren, Chunli Yan.
User's Control of Personalised Intelligent Environments Supporting Health.
In Proceedings of the 9th International Conference on Intelligent Environments, Athens, Greece, IEEE Computer Society Press, pp. 270-273, 2013.
[pdf].
-
Helena Lindgren, Farahnaz Yekeh, Jayalakshmi Baskar and Chunli Yan.
Agent-Supported Assessment for Personalized Ambient Assisted Living.
In Proceedings of Workshop on Agents Applied in Health Care(A2HC), Valencia, Spain, pp. 141-150, 2012.
[pdf].
-
Jayalakshmi Baskar, Helena Lindgren, Dipak Surie, Chunli Yan, and Farahnaz Yekeh.
Personalisation and User Models for Support in Daily Living.
In Proceedings of the 27th Annual Workshop of the Swedish Artificial Intelligence Society (SAIS), Örebro, Sweden, pp. 7-15, 2012.
[pdf].
-
Helena Lindgren, Patrik J Winnberg, and Chunli Yan.
Collaborative Development of Knowledge-Based Support Systems: a Case Study.
Studies in Health Technology and Informatics, vol 180, pp. 1111-1113, 2012.
[pdf].
Chunli Yan
chunli@cs.umu.se
