Clinical problem selection for machine learning-based clinical decision support in the intensive care unit:...

Area of Science:

• Clinical Informatics
• Artificial Intelligence in Medicine
• Healthcare Systems Engineering

Background:

• Machine learning (ML)-based clinical decision support (CDS) offers significant potential for enhancing medical decision-making and patient outcomes in intensive care units (ICUs).
• A substantial gap exists between the development of ML models and their successful implementation at the patient bedside, hindering the realization of these potential benefits.
• While various factors contributing to this implementation gap are known, the critical phase of problem selection within the ML pipeline remains a significant challenge.

Purpose of the Study:

• To address the lack of a practical framework for guiding multidisciplinary teams in the crucial ML problem selection phase.
• To propose specific, actionable questions to facilitate meaningful evaluation of candidate ML-CDS problems.
• To introduce a tool that helps determine the readiness of a problem-CDS pair for impactful bedside deployment.

Main Methods:

• Leveraging the Information Value Chain Theory and empirical evidence to inform the development of guiding questions for problem selection.
• Focusing questions on the core aspects of complexity and actionability of potential ML-CDS applications.
• Operationalizing these questions into a practical Complexity-Actionability Problem Evaluation (CAPE) checklist for ML teams.

Main Results:

• The proposed questions and CAPE checklist provide a structured method for evaluating candidate ML-CDS problems.
• The CAPE checklist assists ML teams in assessing whether a problem-CDS pair is likely to achieve significant bedside impact or requires refinement.
• This systematic approach aims to improve the selection of ML problems with higher potential for successful clinical integration.

Conclusions:

• Effective problem selection is a critical, yet often overlooked, determinant of success for ML-based CDS in ICUs.
• The CAPE checklist offers a practical framework to guide ML teams in choosing problems that are both complex enough to warrant ML and actionable at the bedside.
• Optimizing the execution of care alongside CDS deployment is essential for maximizing the value of ML technology and achieving scalable improvements in patient outcomes.