(1) Background: Mortality is a major outcome in research on chronic obstructive pulmonary disease (COPD) with various predictors described. However, the dynamic courses of important predictors over time are disregarded. This study evaluates if longitudinal assessment of predictors provides additional information on the mortality risk in COPD when compared with a cross-sectional analysis.; (2) In a longitudinal, prospective, non-interventional cohort study including mild to very severe COPD patients, mortality and its various possible predictors were annually assessed up to seven years.; (3) Results: 297 patients were analysed. Mean (SD) age was 62.5 (7.6) years and 66% males. Mean (SD) FEV1 was 48.8 (21.4)%. A total of 105 events (35.4%) happened with a median (95% CI) survival time of 8.2 (7.2/NA) years. No evidence for a difference between the raw variable and the variable history on the predictive value for all tested variables over each visit was found. There was no evidence for changing effect estimates (coefficients) across the study visits due to the longitudinal assessment; (4) Conclusions: We found no evidence that predictors of mortality in COPD are time dependent. This implies that cross-sectional measured predictors show robust effect estimates over time and multiple assessments seem not to change the predictive value of the measure.

Handling incomplete data is a common challenge in quantitative research, and multiple imputation (MI) has become a widely adopted solution. However, traditional MI methods can struggle with complex data patterns, potentially leading to biased estimates. To address this, algorithms like random forests, classification and regression trees, and tree-based boosting techniques have been integrated into common imputation tools, improving the preservation of complex data patterns. Despite these advancements, tree-based methods typically require large sample sizes, limiting their efficiency in smaller datasets. Classical regression techniques, including generalized additive models for location, scale, and shape (GAMLSS), have also been applied to MI. While GAMLSS effectively model non-linearities, they require manual specification of interactions and become computationally intensive with many predictors. This paper explores the use of multivariate adaptive regression splines (MARS) within a fully conditional specification MI framework. MARS offers a balance between flexibility and computational speed, efficiently handling continuous variables without requiring the large datasets needed by tree-based models. A MARS-based MI algorithm was implemented and evaluated through a comprehensive simulation study, demonstrating that MARS provides promising results, with less bias and better coverage, particularly in scenarios involving complex data patterns.

Handling missing data is a central challenge in quantitative research, particularly when datasets exhibit complex dependency structures, such as nonlinear relationships and interactions. Multiple imputation (MI) via fully conditional specification (FCS), as implemented in the MICE R package, is widely used but relies on user-specified models that may fail to capture complex dependency structures, especially in high-dimensional settings, or on more sophisticated algorithms that are considered data-hungry. This paper investigates the performance of TabPFN, a transformer-based, pretrained foundation model developed for tabular prediction tasks, for MI. TabPFN is pretrained on millions of synthetic datasets and approximates posterior predictive distributions without dataset-specific retraining, offering a compelling solution for imputing complex missing data in small to moderately sized samples. We conduct a simulation study focusing on univariate missingness in a continuous outcome with complete predictors, comparing TabPFN with standard MI methods. Performance is evaluated using bias, standard error, and coverage of the marginal mean estimand across a range of data-generating and missingness mechanisms. Our results show that TabPFN yields competitive or superior performance relative to Classification and Regression Trees and Predictive Mean Matching. These findings highlight TabPFN as a promising tool for missing data imputation, with particular relevance to health research.

Handling missing data is a central challenge in quantitative research, particularly when datasets exhibit complex dependency structures, such as nonlinear relationships and interactions. Multiple imputation (MI) via fully conditional specification (FCS), as implemented in the mice R package, is widely used but relies on user-specified models that may fail to capture complex dependency structures, especially in high-dimensional settings, or on more sophisticated algorithms that are considered data-hungry. This paper investigates the performance of TabPFN, a transformer-based, pretrained foundation model developed for tabular prediction tasks, for MI. TabPFN is pretrained on millions of synthetic datasets and approximates posterior predictive distributions without dataset-specific retraining, offering a compelling solution for imputing complex missing data in small to moderately sized samples. We conduct a simulation study focusing on univariate missingness in a continuous outcome, comparing TabPFN with standard MI methods. Performance is evaluated using bias, standard error, and coverage of the marginal mean estimand across a range of data-generating - and missingness mechanisms. Our results show that TabPFN yields competitive or superior performance relative to classification and regression trees and predictive mean matching. These findings highlight TabPFN as a promising new tool for missing data imputation, with particular relevance to health research.