Intelligent aggregation and adaptive visualization

Abstract

Delivering the right health information at the right time to the right patient is a crucial objective that a fully effective E-health system should deliver. The IT solutions used should be able to share such information and have autonomous integration amongst themselves. The national audit report conducted on the use of IT in the Maltese health sector highlights the fact that there is a lack of linkage between the IT solutions. Due to the current segregation of the IT solutions that exist in the primary state-run hospital in Malta, this project aims to link the segregated IT solutions together with other external related datasets into one consolidated system. With the use of adaptive visualizations and machine learning algorithms, we hypothesize that information accessibility at the appropriate level of detail and usability on the visualizations should improve. This proof of concept takes the colorectal patient pathway section in the oncology department as a case study. To create one consolidated system, this research uses semantic web technologies to link the local datasets with external linked data such as NHS (National Health Service), BioBank, DrugBank, and Bio2RDF so that more information on the T, N, M stages (location of the cancer in the colorectal area), drug side effects and dosage are presented to the user. To improve knowledge extraction, the developed system uses visualizations that present the data from a patient aggregated statistical point of view down to patient-based information, while also using outlier detection and a rule association algorithm. For better usability, a rule-based adaptive visualization technology is used, so that visualizations adapt to the users' preferences. This research uses entry and exit questionnaires, task lists, Cronbach's Alpha values, and expert analysis as a qualitative evaluation for the adaptive visualization, linked data and machine learning algorithms. Precision and recall are used to evaluate the linkage between drug side effects and dosage and the machine learning algorithms. The Cronbach's Alpha values, which evaluates the current system with the proposed system indicate that the system fares better than the current systems in use at the primary-state hospital system when the participant is able to verbalize his past actions on the system (verbalizability). It received approximately the same values for extracting the correct extra information (serendipity) but received worse ratings on the intuitiveness of the proposed system (gut feeling) and how comfortably a given task is made (effortlessness). From the entry and exit questionnaires conducted on the participants it was concluded that more information was extracted from the proposed IT solution successfully. However, further evaluation should be made to determine if usability can be improved. These results provide motivation for future work.