An Introduction to Machine Learning for the Pediatric Hospitalist.

Area of Science:

• Clinical Informatics
• Biostatistics
• Artificial Intelligence in Medicine

Background:

• Machine learning (ML) models are prevalent in clinical research for predicting patient outcomes.
• Clinicians often lack the necessary training to critically evaluate these ML studies.
• A gap exists in accessible education on ML for healthcare professionals without computational backgrounds.

Purpose of the Study:

• To provide a conceptual introduction to machine learning for pediatric hospitalists.
• To explain supervised learning, classification, and regression in a clinical context.
• To equip readers with knowledge for critical appraisal of ML research.

Main Methods:

• Focus on supervised learning, a common ML application in medicine.
• Explanation of core tasks: classification and regression.
• Discussion of intuitive models (e.g., decision trees) and advanced methods (e.g., ensembles).

Main Results:

• Highlights key concepts for critical appraisal: overfitting, leakage, interpretability, and data bias.
• Emphasizes the importance of model validation and differentiating prediction from causation.
• Illustrates principles through deconstruction of a published pediatric ML study.

Conclusions:

• Pediatric hospitalists can gain foundational knowledge to understand and evaluate ML research.
• Informed consumership of ML studies is crucial for augmenting clinical judgment.
• Empowering clinicians to engage with the ML ecosystem ensures responsible technology integration.