Artificial Intelligence-Powered Clinical Decision Support Systems in Internal Medicine Reviewed

Faced with diagnostic diversity and uncertainties in the field of internal medicine, the potential uses and impacts of artificial intelligence and machine learning-based clinical decision support systems were addressed in a comprehensive study.

Introduction and Background

Internal medicine is known for having a broad diagnostic spectrum and decisions made under uncertainty. Computerized clinical decision support systems introduced to reduce cognitive load have provided modest benefits with traditional rule-based approaches, but have struggled to adapt to complex data flows.

Artificial Intelligence and Data Flow

Artificial intelligence, machine learning, and deep learning technologies offer a qualitatively different approach by enabling direct learning from electronic health record data, physiological waveforms, and unstructured text.

Model Classes

Three main model classes stand out in the literature: supervised machine learning applied to tabular electronic health records for continuous risk prediction, deep learning applied to raw physiological waveforms such as 12-lead ECGs, and large language models applied to unstructured clinical text.

Translational Gap

The maturity of foundational models has outpaced their clinical evaluations. Most published artificial intelligence studies rely on retrospective discrimination metrics based on historical data where the output was not seen by clinicians.

Expectations and Solution

Prospective deployment, where the model output is presented to the clinician within a live clinical workflow, is of great importance for understanding actual changes in clinician behavior and patient outcomes, and this systematic review focuses on this issue.