Authors (including presenting author) :
Mok JHM(1), Chan JKY(1), Fung VH(1), Kong WT(2), Chung APM(2), Cheng MCY(2), Yeung SWC(3)
Affiliation :
(1) Health Informatics, (2) Business Systems, (3) Clinical Departmental Systems, Information Technology and Health Informatics Division, Hong Kong Hospital Authority
Introduction :
To explore AI can be an option for the clinical decision support system to perform complicated rules without being explicitly programmed.
Objectives :
We collected a set of HBV DNA lab data for one month in HA. A total of ~7000 lab results data were retrieved, in which 3800 data were labelled as positive and the others were labelled as negative, for the training and validation of machine learning.
Methodology :
An automated machine learning tool (DataRobot) was applied in this study. The best algorithm recommended by the DataRobot was the “Gradient Boosted Greedy Tree Classifier”. We found the accuracy of the machine learning results was nearly prefect - True Positive Rate (Sensitivity) was 1, True Negative Rate (Specificity) was 0.9991, and the AUC was 0.9999. For traditional programming in developing an expert system for clinical decision support, complicated rules and programming are always needed. In this study, we demonstrated that using AI might be an alternative to provide computers to learn and to perform complicated rules without being explicitly programmed.
Result & Outcome :
I DISAGREE to send the abstract to Continuous Quality Improvement Initiatives System (CQIs) for sharing after HA Convention.