Tree models for covariate-dependent method agreement with repeated measurements in clinical research

Siranush Karapetyan, Achim Zeileis, Moritz Flick, Bernd Saugel, Alexander Hapfelmeier · 2026

Read Tree models for covariate-dependent method agreement with repeated measurements in clinical research by Siranush Karapetyan, Achim Zeileis, Moritz Flick, Bernd Saugel, Alexander Hapfelmeier (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2603-29539.txt.

Introduction

Karapetyan and co-authors extend agreement trees to repeated measures in clinic studies. Bland-Altman analysis compares two ways of measuring the same subjects. It assumes the two agree the same for all patients. Their trees let agreement shift with patient traits. The mean and spread of gaps between methods split into subject-level parts. That split covers paired and unpaired repeat designs. Simulations held error rates at the set level. A heart output device study showed traits can move agreement.

Catalog record

CREATORSiranush Karapetyan, Achim Zeileis, Moritz Flick, Bernd Saugel, Alexander Hapfelmeier
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE42734
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2603.29539

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Tree models for covariate-dependent method agreement with repeated measurements in clinical research Siranush Karapetyan, Achim Zeileis, Moritz Flick, Bernd Saugel, Alexander Hapfelmeier Abstract Background: In clinical research, the Bland-Altman analysis is commonly used to assess agreement of metric measurements made by two or more techniques, devices or methods. The approach can also deal with repeated measurements per subject or observational unit. However, a strong and implicit assumption is that agreement of methods is homogeneous across subjects or any other internal or external factors. Objective: To extend the previously introduced multivariable modeling approach for conditional method agreement with single measurements per subject to the frequent case of repeated measurements. Methods: Appropriate regression trees, called conditional method agreement trees (COAT), are generalized to capture the dependence of the parameters of the Bland-Altman analysis on covariates. These parameters, namely the expectation and variance of the differences between the methods are decomposed into subject-specific components to detect heterogeneity while accounting for repeated measurements. Whilst the theoretical, asymptotic properties of tree models are known, a supportive simulation study was carried out to assess the performance of COAT in finite samples. A comparison of devices measuring cardiac output serves as an application example. Results: The simulation study showed that COAT is applicable to the two relevant cases of paired and unpaired repeated measurements. It controlled the type-I error at the nominal level and could detect covariate-dependent method agreement with increasing sample size. The Adjusted Rand Index, a measure of concordance between the estimated and true subgroups, reached very high values of up to 0.9 0.9 , close to the maximum of 1 1 . The analysis of cardiac output showed that patients’ characteristics and the size of measurements may have influence on the agreement between measuring devices, with implications for their use in patient care. Conclusion: COAT can explicitly define subgroups of heterogeneous method agreement in dependence of covariates with appropriate statistical significance testing in the case of repeated measurements. It is implemented in the free and open-source R package coat . ∗ Corresponding author. E-mail address: [email protected] Keywords: Bland-Altman analysis, method comparison study, recursive partitioning, subgroup analysis, replicates, conditional inference. 1 Introduction In clinical medicine, it is crucial to assess whether measurements made by new technologies are equivalent or at least clinically comparable to those made by established and accurate measurement methods. For example, cardiac output (CO) is a complex hemodynamic variable that is difficult to measure because established measurement methods are invasive, can cause complications, require training, and are costly. With the development of new technologies, clinicians have the option to use alternative methods to measure CO that potentially alleviate these problems. However, before such new technologies can be implemented, their measurement performance must be assessed in method comparison studies to ensure optimal patient care and safety ( 4 ; 8 ; 13 ; 23 ) . A systematic review of method comparison studies revealed that 85\% of studies used Bland-Altman (BA) analysis to evaluate method agreement ( 25 ) , underscoring the fact that BA analysis is an established standard ( 2 ; 3 ; 4 ; 5 ) . It essentially estimates two quantities. The so-called ‘bias’, which measures the average difference of measurements made by two methods, and the ’Limits of Agreement’ (LoA), which are a 95\% prediction interval, describing a respective range of the expected difference between the methods when used to make a single measurement of an unknown value. The LoA are derived from estimates of the bias and the standard deviation of the differences of measurements. In our research, we question the universal and implicit assumption of BA analysis, that the agreement of methods is constant across all subjects (or any other kind of observational units), independent of internal and external factors. We therefore allow for differences of measurements to be influenced by patient characteristics, pre-existing conditions, experimental settings, and so on. For example, the agreement of CO measurements between invasive and non-invasive methods may relate to physics and other characteristics of the patients, such as age, sex, body mass index (BMI) or comorbidities. In an earlier work, we already introduced the concept of conditional method agreement and developed multivariable tree models for the case of single measurements per subject to test and model covariate-dependent method agreement ( 20 ) . In the present work, we extend the concept and develop a new multivariable tree model for the very common but more complex case of repeated measurements per subject. Conditional inference trees ( 18 ) with a suitable transformation of the observed measurements are used to fulfill the purpose. Repeated measurements are accounted for by subject-specific components of the bias, the between-subject variance and the within-subject variance. The two sources of variance have already been defined by Bland and Altman, such that the new approach closely resembles the established BA analysis ( 4 ; 5 ) . The proposed approach is referred to as conditional method agreement trees (COAT), and its relevance for medical research is demonstrated by application to data from repeated CO measurements made with different methods ( 21 ) . In addition, the ability of COAT to control the type-I error probability at a nominal level, the power to detect associations, and the ability to accurately define subgroups is investigated in simulation studies. 2 Methods Method agreement is characterized by two quantities, the bias and the LoA, which quantify the expected difference of the methods’ measurements on average or for the majority of measurements, respectively. The LoA are a 95% prediction interval of the differences between single measurements. The bias and LoA are obtained from a BA analysis, as we estimate the expectation \mathbb{E}(Y) and the variance \mathrm{Var}(Y) of a random variable Y=A-B . Y represents the differences between measurements A and B made by two distinct methods. Respective estimates, which are the parameters of a BA analysis, are given by the mean and the standard deviation of the observed differences y=a-b (see Section 2.1 for computational details). In the context of repeated measurements, the methods are used to generate m_{Ai} and m_{Bi} measurements or realizations of A and B for the i -th subject ( i\in\{1,\ldots,n\} ). These numbers of measurements can be equal or unequal, as outlined below for the specific cases of paired and unpaired measurements. In our work, we allow the agreement of methods and therefore the parameters of the BA analysis to be dependent on covariates X . Consequently, we define conditional method agreement by estimating the conditional expectation (or bias) \mathbb{E}(Y|X) and conditional variance \mathrm{Var}(Y|X) . It is important to note that we use the term ‘subjects’ to refer to an observational unit, although the more general term ‘item’ is also commonly used. However, this purely notational difference has no impact on the approach presented here. The following Section 2.1 describes the original BA analysis for estimating the basic parameters of the BA analysis, i.e., the bias \mathbb{E}(Y) and variance \mathrm{Var}(Y) of the differences Y in the case of repeated measurements. The novel and corresponding modeling of the conditional BA parameters \mathbb{E}(Y|X) and \mathrm{Var}(Y|X) is based on these calculations and is proposed in Section 2.2 with appropriate cross-references to the original approach. 2.1 The Bland-Altman analysis In the case

Read the complete text