Machine Learning in Laboratory Medicine: Recommendations of the IFCC Working Group

Area of Science:

• Clinical Laboratory Science
• Medical Informatics
• Artificial Intelligence in Medicine

Background:

• Machine learning (ML) shows significant potential in laboratory medicine for predictive tasks.
• However, uncontrolled development and validation pipelines pose risks for ML applications.
• Potential pitfalls in ML workflows require careful consideration for reliable clinical use.

Purpose of the Study:

• To provide consensus recommendations for best practices in laboratory medicine ML.
• To address challenges and mitigate pitfalls in applying ML in this domain.
• To improve the quality and reproducibility of ML models for clinical laboratories.

Main Methods:

• A working group of the International Federation for Clinical Chemistry and Laboratory Medicine was convened.
• The group developed a guidance document based on consensus.
• Recommendations cover all phases of ML model development and implementation.

Main Results:

• The manuscript presents consensus recommendations for best practices in ML for laboratory medicine.
• These guidelines aim to improve the quality of developed and published ML models.
• The practices span from problem formulation to predictive implementation.

Conclusions:

• Adherence to these best practices will enhance ML quality and reproducibility in laboratory medicine.
• Implementation ensures valid and reproducible ML models for clinical laboratory questions.
• These guidelines address common and critical errors in the emerging field of laboratory medicine ML.