Zu den Inhalten springen

Institute of Medical Biometry and Statistics (IMBI)

Selected project overview

Distributed Computing Under Data Protection Constraints

Individual patient data fall into the category of the highly sensitive data that enjoy special protection. Therefore data sharing for research purposes is prohibited even within joint projects.

To make a joint analysis of the data possible nonetheless, for example using the data from several different university medical centers, we are working on specialized statistical modelling methodology, which complies with data protection regulations by only sharing aggregated data. We are working within the framework of the projects MIRACUM and GESA and are building on the platform DataSHIELD.

Responsible person: Harald Binder

Participation: Daniela Zöller, Stefan Lenz

MIRACUM - Medical Informatics in Research And Care in University Medicine

The MIRACUM Consortium, an association of initially 8 university medical centers (Erlangen, Frankfurt, Freiburg, Gießen, Mainz, Magdeburg, Mannheim and Marburg) receives its funding from the Federal Ministry of Education and Research (BMBF) within the scope of the Medical Informatics Initiative Germany. The aim of the MIRACUM Consortium is to build a research infrastructure that allows distributed data from treatment and research to be analyzed jointly. This will be demonstrated by the example of three so called „Use Cases“.

Use Case 2 („From Data to Knowledge”), under the direction of Harald Binder, focuses on the identification and of patterns and subgroups in patient data. It consists of two parts. The first part looks at clinical and lab data of asthma/COPD patients, while the second part examines clinical and genetic data (gene methylation) of patients with brain tumors.

Person responsible: Harald Binder

Participation: Stefan Lenz, Kiana Farhadyar

SoftMeta – Software for Meta-Analysis

The goal of this project is the implementation and distribution of statistical methods for meta-ananalyses in the freely available software R. Starting point of the project are the R-packages meta, metasens, netmeta and diagmeta, which are available on GitHub. You can find further information on our project page at ResearchGate and the corresponding website to our course book "Meta-Analysis with R."

Person responsible: Guido Schwarzer

Participation: Gerta Rücker

Modelling of ROC curves in meta-analyses of diagnostic test accuracy studies and network meta-analysis

This DFG (German Research Foundation) project comprises two current research areas in the field of evidence synthesis in medicine: the meta-analysis of diagnostic test accuracy studies and network meta-analysis. We developed a new approach to the meta-analysis of diagnostic test accuracy studies that allows the pooling of entire ROC curves and additionally implemented a corresponding new R package called diagmeta. In the area of network meta-analysis we continuously add new methods to our R package netmeta, f.e. to separate the individual effects in combination therapies. The final objective is to combine both areas - network meta-analysis of diagnostic test accuracy studies.

Person responsible: Gerta Rücker

Participation: Guido Schwarzer, Susanne Steinhauser (diagnostics)

DynaMORE - Dynamic Modeling of Resilience

The EU (HORIZON 2020) project DynaMORE uses mathematical modeling for the advancement of mental health and well-being.

As one of twelve partners of the consortium, we integrate multimodal and high dimensional data (e.g. fMRI and behavioral data, as well as ecological monitoring of smartphones and wearables) in the models and contribute technology for dynamic predictions. In both areas we compare deep learning with traditional approaches.

Person responsible: Harald Binder

Participation: Göran Köber

GESA - Gender-Sensitive Analyses of Psychological Health

Cross-sectional studies developed many leads to differences in the prevalence of mental health problems in men and women. Sex- and gender-sensitive analyses, as they will be conducted in the GESA project, can contribute to a clarification of the causes of the observed differences.

The project combines three large-scale cohorts (SHIP / GHS / KORA), whose data allow a longitudinal approach. The aim is to develop leads to causal relationships between predictors and to develop effective models.

Person responsible: Harald Binder

Participation: Daniela Zöller

MeInBio - BioInMe: Exploration of Spatio-Temporal Dynamics of Gene Regulation Using High-Throughput and High-Resolution Methods

The experimental design of single-cell RNA sequencing (scRNA-seq) studies, particularly determining the sample size, is a crucial step in testing biological hypotheses, because the right sample size guarantees sufficient statistical power.

We examine the distribution hypotheses of single-cell gene expression data and use generative models from the field of deep learning to uncover complex structures in the data. These structures are being used to determine the sample size as well as for the experimental design of the analysis of differentially expressed cells, to discover groupings of cells and to describe continuous cell development.

Person responsible: Harald Binder

Participation: Martin Treppner

AgeGain - Transfer of Cognitive Training Gains in Cognitively Healthy Aging: Mechanisms and Modulators

AgeGain is a multicenter randomised controlled study, which investigates the basis of transfer learning in order to identify effective measures to conserve the mental capacities of the elderly. Cooperation partners for this project are the University Hospital Cologne, the German Sport University Cologne, as well as the University Medical Center Mainz and Rostock. The participants are people in the age range between 60 and 85 without any severe physical or mental impairments. Part of them receives a specialised cognitive training and another part receives an additional fitness training. Data regarding physical and social activities are being gathered, brain scans (fMRI and PET) performed and tests checking the mental and physical abilities.

Our task in this taking care of the statistical aspects, especially the unique challenges that come with the analysis of such a multimoal and high-dimensional data set.

Person responsible: Harald Binder

Participation: Stefan Lenz


Prof. Dr. Harald Binder

  • Machine Learning (esp. Deep Learning)
  • Integration of Molecular and Clinical Data