The utilization of machine learning (ML) has expanded significantly within laboratory medicine. Existing literature indicates substantial potential for clinical applications. However, numerous parties have highlighted potential drawbacks, especially when the intricacies of development and validation pipelines lack meticulous control.
This talk illustrates recommendations from a guidance document by a working group of the International Federation for Clinical Chemistry and Laboratory Medicine [1]. It presents consensus recommendations for optimal practices aiming to enhance the quality of developed and published ML models intended for deployment in clinical laboratories. These practices encompass all stages of model development, ranging from problem formulation to predictive implementation. While it is impractical to comprehensively address every potential pitfall in ML workflows, this talk encapsulates best practices aimed at avoiding the most common and potentially hazardous errors in this evolving field. Another major focus of this talk is on the role of the specialist in laboratory medicine in ensuring that the ML application can be relied upon for patient treatment.
Literature:
[1] Master, Stephen R., et al. "Machine Learning in Laboratory Medicine: Recommendations of the IFCC Working Group." Clinical Chemistry (2023): hvad055.
