arXiv subjects

Ricarda Graf

Publications and source records attributed to Ricarda Graf.

3 recordsLinked to original sources

arXiv2025

Comparison of the Cox proportional hazards model and Random Survival Forest algorithm for predicting patient-specific survival probabilities in clinical trial data

The Cox proportional hazards model is often used to analyze data from Randomized Controlled Trials (RCT) with time-to-event outcomes. Random survival forest (RSF) is a machine-learning algorithm known for its high predictive performance. We conduct a comprehensive neutral comparison study to compare the performance of Cox regression and RSF in various simulation scenarios based on two reference datasets from RCTs. The motivation is to identify settings in which one method is preferable over the other when comparing different aspects of performance using measures according to the TRIPOD (Transparent Reporting of a multivariable prediction model for Individual Prognosis or Diagnosis) recommendations. Our results show that conclusions solely based on the C index, a performance measure that has been predominantly used in previous studies comparing predictive accuracy of the Cox-PH and RSF model based on real-world observational time-to-event data and that has been criticized by methodologists, may not be generalizable to other aspects of predictive performance. We found that measures of overall performance may generally give more reasonable results, and that the standard log-rank splitting rule used for the RSF may be outperformed by alternative splitting rules, in particular in nonproportional hazards settings. In our simulations, performance of the RSF suffers less in data with treatment-covariate interactions compared to data where these are absent. Performance of the Cox-PH model is affected by the violation of the proportional hazards assumption.

stat.ML↗

arXiv2023

Descriptive Discriminant Analysis of Multivariate Repeated Measures Data: A Use Case

Psychological research often focuses on examining group differences in a set of numeric variables for which normality is doubtful. Longitudinal studies enable the investigation of developmental trends. For instance, a recent study (Voormolen et al (2020), https://doi.org/10.3390/jcm9051525) examined the relation of complicated and uncomplicated mild traumatic brain injury (mTBI) with multidimensional outcomes measured at three- and six-months after mTBI. The data were analyzed using robust repeated measures multivariate analysis of variance (MANOVA), resulting in significant differences between groups and across time points, then followed up by univariate ANOVAs per variable as is typically done. However, this approach ignores the multivariate aspect of the original analyses. We propose descriptive discriminant analysis (DDA) as an alternative, which is a robust multivariate technique recommended for examining significant MANOVA results and has not yet been applied to multivariate repeated measures data. We provide a tutorial with annotated R code demonstrating its application to these empirical data.

stat.AP↗

arXiv2023

Linear classification methods for multivariate repeated measures data -- a simulation study

Researchers in the behavioral and social sciences use linear discriminant analysis (LDA) for predictions of group membership (classification) and for identifying the variables most relevant to group separation among a set of continuous correlated variables (description). \\ In these and other disciplines, longitudinal data are often collected which provide additional temporal information. Linear classification methods for repeated measures data are more sensitive to actual group differences by taking the complex correlations between time points and variables into account, but are rarely discussed in the literature. Moreover, psychometric data rarely fulfill the multivariate normality assumption.\\ In this paper, we compare existing linear classification algorithms for nonnormally distributed multivariate repeated measures data in a simulation study based on psychological questionnaire data comprising Likert scales. The results show that in data without any specific assumed structure and larger sample sizes, the robust alternatives to standard repeated measures LDA may not be needed. To our knowledge, this is one of the few studies discussing repeated measures classification techniques, and the first one comparing multiple alternatives among each other.

stat.ME↗