# Tree models for assessing covariate-dependent method agreement

**Authors:** Siranush Karapetyan, Achim Zeileis, André Henriksen, Alexander Hapfelmeier  
**arXiv:** [2306.04456](https://arxiv.org/abs/2306.04456)  
**Date:** 2023-06-07  
**Summary page:** https://www.summarizepaper.com/en/arxiv-id/2306.04456v1/

## Key points

- The paper explores the agreement of measurements made by multiple methods in method comparison studies.
- The authors introduce the concept of conditional method agreement and propose a modeling approach to account for covariate-dependent agreement.
- They use recursive partitioning to identify subgroups with heterogeneous agreement based on covariates.
- Three different modeling approaches are considered: CTreeTrafo, DistTree, and MOB.
- The performance of these models is evaluated using simulation studies.
- An application example using real data from accelerometer device measurements is provided.
- A two-sample Bland-Altman test is proposed for hypothesis testing of differences in agreement between subgroups.
- The COAT approach enables exploratory analysis of method agreement considering covariates and facilitates hypothesis testing for group differences.
- The COAT framework is implemented in an R package called coat.

## Summary

The paper titled "Tree models for assessing covariate-dependent method agreement" by Siranush Karapetyan, Achim Zeileis, André Henriksen, and Alexander Hapfelmeier explores the agreement of measurements made by multiple methods in method comparison studies. The commonly used Bland-Altman analysis assumes that the differences between measurements are identically distributed for all observational units and application settings. However, this assumption may not hold true in practice. To address this limitation, the authors introduce the concept of conditional method agreement and propose a modeling approach to account for covariate-dependent agreement. They embed the Bland-Altman analysis within the framework of recursive partitioning to identify subgroups with heterogeneous agreement based on covariates in an exploratory analysis. The authors consider three different modeling approaches: conditional inference trees with a suitable transformation of the modeled differences (CTreeTrafo), distributional regression trees (DistTree), and model-based trees (MOB). They evaluate the performance of these models in terms of type-I error probability, power, and their ability to uncover given subgroups using simulation studies. Additionally, they demonstrate the applicability of their approach through an application example using real data from accelerometer device measurements. They show that their models can accurately detect relevant covariates that may affect agreement as the sample size increases. Furthermore, the authors propose a two-sample Bland-Altman test for exploratory or confirmatory hypothesis testing of differences in agreement between subgroups. This test allows researchers to assess group differences in method agreement. Overall, the authors conclude that their proposed conditional method agreement trees (COAT) enable exploratory analysis of method agreement considering covariates and facilitate hypothesis testing for group differences. The COAT approach is implemented in an R package called coat which is publicly available. In summary, this paper presents a novel modeling approach that addresses limitations in traditional method comparison studies by incorporating covariate-dependent agreement. The proposed COAT framework provides a valuable tool for researchers to analyze and understand method agreement in various application settings.