630
P.-Y. S. Hsueh et al.
Besides, the framework could be developed with different smart learning principles, ranging from supervised learning to unsupervised learning and reinforcement
learning. The data sources could be structured (e.g., electronic health records),
unstructured (e.g., clinical notes), or semi-structured (e.g., medical images). The
knowledge sources could be clinical guidelines (e.g., NICE guideline for type 2
diabetes management [106]), well-known risk models (e.g., Framingham Cardiovascular Disease 10-year risk model), and medical concept graphs (e.g., a knowledge
graph of medical concepts for estimating feature correlations). The applications of
such frameworks widely range from the determination of clinical trial eligibility
[107] to risk assessment and treatment recommendations [56].
In retrospect of the MoCaB case study, quite a few lessons about the best practice
and requirements for future implementation have been learned. For example, at
the beginning of the study, the plan was to implement a fully rule-based system.
However, because only a small portion of the experts’ knowledge is explicitly
expressible, this plan had to be adapted by finding an appropriate assessment to
extract the implicit knowledge for the domain model. Yet this task has been proven
to be time-consuming, since it involves an iterative process incorporating extensive
cooperation between domain experts in healthcare and experts in computer science.
This workload should be regarded when planning to develop such systems in the
future.
Furthermore, the integration of the target users plays an important role for the
development of an ITS system for care management. The choice of the system’s
target user group and recruiting of participants should account for not only their
information and social need, but also the burden on their time commitment.
Our results indicate that AI-based systems can contribute to adaptive education of
caregivers for care management. While this approach necessitates intensive efforts
in terms of participatory system design and evaluation, its potential benefits may
prove rewarding.
In this chapter, we have gathered requirements for evolving a smart learning
framework to incorporate N-of-1 actionable insights and knowledge learning symbiosis in healthcare, aiming to generate real-world evidence in practice. In particular,
this chapter focuses on promising applications and smart learning methods that help
combine data with a priori knowledge for decision support. This chapter aims to
share with leading practitioners and researchers in the field about the requirements
for the next-generation, data-knowledge fusion systems, especially in areas where a
large amount of evidence has been accumulated in knowledge bases.
We provided an overview for the principles and methodologies of the framework
for better interpretability. To achieve this, there exist new challenges that would
require breakthroughs in both continuous learning analytics and implementation
science. While the former uncovers relationships between patient-specific characteristics, treatment trajectory and patient response to interventions, the latter provides
guidance to embed actionable insights into care flows in practice for sense-making
and decision support.
In the future, the smart learning framework for mobile/IOT e-health is expected
to further help patients in chronic disease management scenarios, or monitor
P.-Y. S. Hsueh et al.
Besides, the framework could be developed with different smart learning principles, ranging from supervised learning to unsupervised learning and reinforcement
learning. The data sources could be structured (e.g., electronic health records),
unstructured (e.g., clinical notes), or semi-structured (e.g., medical images). The
knowledge sources could be clinical guidelines (e.g., NICE guideline for type 2
diabetes management [106]), well-known risk models (e.g., Framingham Cardiovascular Disease 10-year risk model), and medical concept graphs (e.g., a knowledge
graph of medical concepts for estimating feature correlations). The applications of
such frameworks widely range from the determination of clinical trial eligibility
[107] to risk assessment and treatment recommendations [56].
In retrospect of the MoCaB case study, quite a few lessons about the best practice
and requirements for future implementation have been learned. For example, at
the beginning of the study, the plan was to implement a fully rule-based system.
However, because only a small portion of the experts’ knowledge is explicitly
expressible, this plan had to be adapted by finding an appropriate assessment to
extract the implicit knowledge for the domain model. Yet this task has been proven
to be time-consuming, since it involves an iterative process incorporating extensive
cooperation between domain experts in healthcare and experts in computer science.
This workload should be regarded when planning to develop such systems in the
future.
Furthermore, the integration of the target users plays an important role for the
development of an ITS system for care management. The choice of the system’s
target user group and recruiting of participants should account for not only their
information and social need, but also the burden on their time commitment.
Our results indicate that AI-based systems can contribute to adaptive education of
caregivers for care management. While this approach necessitates intensive efforts
in terms of participatory system design and evaluation, its potential benefits may
prove rewarding.
In this chapter, we have gathered requirements for evolving a smart learning
framework to incorporate N-of-1 actionable insights and knowledge learning symbiosis in healthcare, aiming to generate real-world evidence in practice. In particular,
this chapter focuses on promising applications and smart learning methods that help
combine data with a priori knowledge for decision support. This chapter aims to
share with leading practitioners and researchers in the field about the requirements
for the next-generation, data-knowledge fusion systems, especially in areas where a
large amount of evidence has been accumulated in knowledge bases.
We provided an overview for the principles and methodologies of the framework
for better interpretability. To achieve this, there exist new challenges that would
require breakthroughs in both continuous learning analytics and implementation
science. While the former uncovers relationships between patient-specific characteristics, treatment trajectory and patient response to interventions, the latter provides
guidance to embed actionable insights into care flows in practice for sense-making
and decision support.
In the future, the smart learning framework for mobile/IOT e-health is expected
to further help patients in chronic disease management scenarios, or monitor