Implementing a Machine Learning Screening Tool for Malnutrition: Insights From Qualitative Research Applicable to...

Area of Science:

• Clinical Informatics
• Health Services Research
• Artificial Intelligence in Healthcare

Background:

• Machine learning (ML)-based clinical decision support systems (CDSS) face challenges in usability, interpretability, and effectiveness.
• Evaluating ML-based CDSS implementation is crucial for clinician acceptance and high-quality healthcare.
• Malnutrition is a prevalent, underdiagnosed condition in hospitals with significant adverse outcomes.

Purpose of the Study:

• To evaluate the implementation of the Malnutrition Universal Screening Tool (MUST)-Plus, an ML tool for predicting high-risk malnutrition patients.
• To identify optimal implementation practices for ML-based CDSS in clinical settings.

Main Methods:

• A qualitative postimplementation evaluation was conducted.
• In-depth interviews were performed with registered dietitians (RDs) using the MUST-Plus tool.
• Emergent themes were analyzed using the nonadoption, abandonment, scale-up, spread, and sustainability (NASSS) framework.

Main Results:

• Enhancements improved MUST-Plus accuracy and usability; the tool identified previously unseen patients.
• Perceived usefulness was highest at the original site, with accuracy varying by respondent and site.
• Integration into workflows and electronic health records occurred, with a desire for a single automated screener.

Conclusions:

• Continuous stakeholder involvement is vital for buy-in, especially with staff turnover.
• Qualitative research can uncover ML tool biases, promoting health equity.
• Collaboration between developers and clinicians can optimize CDSS acceptability and use.