Conditional Method Agreement Trees (COAT)

• Applicant: Professor Dr. Alexander Hapfelmeier
• Subject Area: Medical Informatics and Medical Bioinformatics ; Cardiology, Angiology
• Term: since 2021
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 447467169

Method comparison studies are performed to assess the agreement of measurement methods. They find application in any domain but originate from medical research where (gold-)standard methods to obtain a patient’s clinical measurements of interest can be costly, time-consuming, complex, invasive, stressful and risky. Therefore methods have been redeveloped to save costs, time and to lower the physical burden for patients, thereby retaining accuracy.The most commonly used analysis method has been proposed by Martin Bland and Douglas Altman (i.e. the Bland-Altman plot) more than three decades ago. Since then methodological work adhered to a one-fits-all paradigm while, quite contrary, subgroup analyses become more and more important in the ages of stratified and personalized medicine. Thinking of increasingly complex and highly developed methods (involving computer science) and experimental settings, it is questionable whether the one-fits-all paradigm is true. Sensors, rating systems, imaging analysis etc. might just not be equally accurate in any situation and for any patient due to internal and external influences. The proposed research provides an innovative solution, called ‘conditional method agreement trees’ (COAT), by introducing the concept of ‘conditional agreement’ and respective multivariable modeling. It exploits the fact that agreement can be modeled through properly specified linear mixed-effects models. In combination with the recently developed mixed-effects model trees, it is possible to adapt model-based recursive partitioning to define subgroups with different agreement of measurement methods. The high relevance of the research question has been confirmed by my collaborators consisting of leading scientist in the fields of medical, statistical and informatics research.

• DFG Programme: Research Grants