Skip to main content Skip to page footer

LeMeDaRT: Lean Medical Data - the Right data at the right Time

Aims & Objectives

There is a scarcity or often even absence to have patient reported outcomes data available in a structured fashion. Where data is digitally recorded they often reside in well-protected silos. Several processes still rely on manual and paper-based processes. Additionally, patients are plagued with repeating even basics such as giving their medical history from provider to provider.

LeMeDaRT aims to address these issues by building a ”liquid data” concept to link and digitally support the patient journey from population health over prevention to interaction of rural care including complex interventions. The core goal is to provide the right relevant data in each instance during the patient journey to leverage patient benefit and provider efficiency. In practice, these goals are realized for three different main Use Cases: 

UC 1: Prehabilitation and enhanced postsurgical surveillance in complex abdominal cancer surgery

UC 2: Prevention and early intervention for non-alcoholic fatty liver disease (NAFLD)

UC 3: Infection surveillance in respiratory viral disease

All three use cases share a common goal: The development of a digital hub to support practitioners during the decision-making and facilitating as well as improving the patient care with utilizing the correct data. More detailed information about the modelling tasks for each use case (from the perspective of EMCL) is given below.
The entirety of the project is composed of multiple interdisciplinary groups of scientists, physicians and computer experts from Medizinische Fakultät Mannheim der Uni Heidelberg and Universität Heidelberg.

Role of the EMCL

The EMCL is part of the Cluster-Group 1 which is responsible for the Data Integration & Technology as well as Data Science. EMCL focuses on developing AI methods and intelligent approaches for the three Use Cases of the LeMeDaRT. More specifically, the following tasks are being addressed by the EMCL for each Use Case respectively:

  • UC 1 Modelling Goal: Predict the compliance and the benefit of the prehabilitation program for each individual patient.
    • Develop AI methods (namely Deep Neural Networks & Random Forests) for the compliance modelling. More specifically, the prediction should be whether the patient is likely to complete and adhere to the individually designed prehabilitation training program based on the psychological, nutritional and physiological attributes. Consequently, these patients are treated then as high risk and low risk candidates. 
    • Predict the causal individual treatment effect of the training program for the patient. For this, the EMCL (together with the MIISM Group of Prof. Dr. Jürgen Hesser in Mannheim) published a review paper [1] bench-marking and improving numerous Deep Neural Networks for potential outcome modelling for treatment effect estimation.
  • UC 2 Modelling Goal: Develop a referral system for early detection of NAFLD with a fibrosis stage prediction based on biomarkers.
    • Implement an intelligent rule-based algorithm to refer patients with increased biomarker values from the general practitioner to the clinic.
    • Develop and test Random Forest AI models such as XGBoost to infer the fibrosis stage of the liver solely from biomarkers (i.e. without the need of a fibroscan or biopsy).
      Include shapely value analysis to provide explainable/transparent AI.
  • UC 3: Modelling Goal: Implement a diagnostic predictive tool for infectious diseases given the symptom progression. 
    • Model the probabilities of several infectious diseases in a time-serious fashion and update the probabilities given the symptom progression.
      In particular, the EMCL focuses on utilizing Bayesian Neural Networks with prior distribution based on the symptom observations.
    • Refer patients to practitioner in case that the suspected infectious disease requires medical attention.
    • Derive a “heat map” of the current diagnosed infectious diseases for the project region.

Research topics

  • Uncertainty quantification of neural networks and Bayesian Models
  • Machine learning based patient classification
  • Data assimilation for a standardized clinical data format
  • Causal Inference and patient classification using neural networks
  • Translation of the innovative developments to clinical practice


People from the EMCL


For a detailed description of the partners see BMBF/LeMeDaRT - "Partner im digitalen Fortschrittshub LeMeDaRT".
Furthermore the EMCL closely works with the MIISM group in Mannheim.


  • Marcus Buchwald



  • Sirazitdinov, Andrei; Buchwald, Marcus; Hesser, Jürgen; Heuveline, Vincent
    (2022). Review of Deep Learning Methods for Individual Treatment Effect Es-
    timation with Automatic Hyperparameter Optimization. TechRxiv. Preprint.