Data-driven web-based intelligent decision support system for infection management at point of care

Abstract

Infectious diseases are caused by the invasion of pathogenic microorganisms such as bacteria, viruses or fungi and are one of the leading causes of mortality worldwide. In the last years, there has been a significant increase in the ability of these microorganisms to resist antimicrobials which were previously effective. This phenomenon, denoted as antimicrobial resistance (AMR), has become a noticeable obstacle to treat infections in health care with misuse and overuse of antimicrobials as one of the leading drivers.

This thesis presents a novel clinical decision support system for infection management to provide personalized, accurate and effective diagnostics at point of care. The proposed system, which has been denoted as EPiC IMPOC, incorporates two main decision support engines: case-based reasoning to facilitate vital sign collection, patient monitoring and further inspection of past similar cases and probabilistic inference to provide stepwise guidance within the infection management pathway followed by clinicians. In addition, a number of local AMR statistics are automatically computed from susceptibility test data to promote education and awareness among clinicians. This decision support system has been implemented as a web-based platform that can be accessed at the point of care from computers or hand-held devices.

The design and implementation of the system has been performed incrementally. As such, the server has been divided into discrete and reusable modules. Firstly, the AMR statistics have been computed and compared with the existing literature to better describe the scope of the problem. After this, the case-based reasoning methodology was included to inform physicians of previous past cases to assist in antimicrobial therapy selection. In order to evaluate the validity and usability of the system, a pilot study was conducted in the ICU. The system advocated the same results as those suggested by infection specialists in 84% of the cases. Furthermore, participants highlighted the utility of the case-based reasoning engine to promote knowledge transfer among health care professionals and the benefits of having access to real-time patient data at the point of care. The system usability score was 68.5 which is above average. On the other hand, participants suggested that the provision of more specific support would be beneficial. For this reason, the probabilistic inference module was included into the system to provide the likelihood of positive culture (AUCROC over 0.90) and the most plausible sites of infection (AUCROC within the range 0.82-0.96). The translational utility of the generated predictive models was assessed retrospectively considering also scenarios with missing data and imbalanced classes.

All these elements combined result in a state-of-the-art clinical decision support system which assists physicians on multiple areas within infection management to facilitate the provision of evidence-based and personalized medicine.