Minimal algorithms for knowledge representation in clinical decision support systems research: a theoretical and empirical analysis

Abstract

Clinical decision support systems (CDSS) figures out as one of the most promising technologies for data-centered and AI-prompted healthcare. Its current developments are mainly guided by two disparate mindsets, namely a machine learning-centered framework and a classical rule-based framework. These respective approaches presents contrastive pros and cons. In the present study we provide an analysis showing that these two mindsets are actually related to each other, and straightforward algorithms are feasible by combining current standards for machine learning and classic decision tables algorithms. A theoretical analysis are provided, as well a computational implementation (in python). A real case scenario on radiological immaging exam prescription is used to ilustrate the successfully application of our results. Future work on benchmarking the proposed algorithms embodied in a fully operational clinical decision support system could extend our findings towards daily used systems.

Exploring the KISS principle (Keep It Simple) for decision support systems

“…We assume that all users have an email account, which can be added to (or removed from) the receiver’s list of updates, to be eventually emailed when modifications are found in the protocols repository. The information from these documents (protocols) is manually tabulated by a team of specialist professionals, forming a database that will be used to parameterize a decision table algorithm (see (11)), which will serve as a knowledge representation model. Recommendations are obtained through user interaction with the decision table algorithm through an interface algorithm, in which basic and clinical patient data are provided by the user (usually, a healthcare professional), who will receive the recommendation of action to be taken.…”

“…We assume that all users have an email account, which can be added to (or removed from) the receiver's list of updates, to be eventually emailed when modifications are found in the protocols repository. The information from these documents (protocols) is manually tabulated by a team of 3 specialist professionals, forming a database that will be used to parameterize a decision table algorithm (see (11)), which will serve as a knowledge representation model.…”