Covariate adjustment for multivariate outcomes

In healthcare, the approval of new treatments relies on systematically collected data on the effectiveness and safety of medical interventions (e.g., drugs). For this systematic data collection, so- called clinical trials provide a methodologically sound and rigorous framework that is well-accepted also by regulatory agencies. Typically, for quantifying the efficacy of an intervention in a clinical trial, one single measure is defined as the primary endpoint, that is, the measurement which is of primary importance for assessing efficacy (e.g., for a dietary intervention, the primary endpoint might be weight loss within a certain period). However, this approach is often insufficient to capture the full range of benefits from an intervention, to jointly evaluate benefit-risk, or to analyze objective measurements and patient-reported outcomes (e.g., quality of life, pain) simultaneously. Statistical methods for analyzing multiple endpoints (or outcomes) have been developed, but are limited in several ways, for example regarding the number and types of outcomes that can be combined. Recently, the Generalized Pairwise Comparisons (GPC) method has been suggested, which addresses these concerns and has been successfully used in clinical applications, even leading to drug approvals. The disadvantage of GPC is that it lacks covariate adjustment: In other words, potential imbalances (e.g., regarding age, sex, disease characteristics) between groups of patients in a clinical trial which are treated with different interventions cannot be accounted for. This problem may be solved by so-called Probabilistic Index Models (PIM), which however cannot deal with multivariate outcomes, nor with missing values. So, summing up, both approaches GPC and PIM have their advantages and drawbacks. Hence, the objective of this project is to bring the advantages from these two worlds together by extending the PIM methodology to being capable of handling multivariate outcomes and missing data. On the application-oriented side, these novel methodological developments will yield considerable added value for conducting clinical trials in patients with multi-faceted diseases, in particular so-called rare diseases (i.e., affecting less than 1 in 2,000 persons in the population). The advantages of PIMs should be investigated by comparing them to alternative methods for covariate adjustment (e.g. joint models and semiparametric ANCOVA). To facilitate the application in clinical practice, all newly developed methods will be implemented in the open-source statistical software R.