Rigorous validation of machine learning in laboratory medicine: guidance toward quality improvement

Area of Science:

• Laboratory medicine
• Artificial intelligence
• Machine learning

Background:

• Artificial intelligence (AI) and machine learning (ML) offer transformative potential for predictive modeling in laboratory medicine.
• Current ML studies often lack essential quality assurance, impacting the reliability of models.
• Factors like analytical variations and pre-analytical errors can affect ML model performance in clinical settings.

Purpose of the Study:

• To highlight the necessity for quality improvement and validation of ML classification models in laboratory medicine.
• To provide researchers, reviewers, and editors with key concepts for developing robust ML models.
• To adapt existing predictive modeling guidelines for ML applications in pathology and laboratory medicine.

Main Methods:

• Overview of ML principles and current guideline recommendations.
• Discussion of advantages and pitfalls of applied ML in laboratory medicine.
• Drawing parallels between clinical assay validation and ML model validation within regulatory frameworks.

Main Results:

• Identification of critical factors affecting ML model robustness, including analytical and physiological variables.
• Emphasis on the importance of classification performance metrics, model explainability, and data quality.
• Recommendations for enhancing journal submission requirements for ML-based studies.

Conclusions:

• Proper validation and quality assurance are essential for the reliable application of ML in laboratory medicine.
• Adapting existing guidelines and focusing on data quality, explainability, and performance metrics will strengthen ML model development.
• The principles discussed are applicable beyond laboratory medicine to broader biomedical science research.