Skip to main content Skip to page footer
EMCL logo EMCL logo

LeMeDaRT 2: Lean Medical Data in the digital Hub

Aims & Objectives


LeMeDaRT 2 represents the follow-up project of LeMeDaRT & unites six regional DigiHubs to accelerate the digital transformation of German health care. The overarching aim is to build a secure, standards‑based data ecosystem that links research, clinical care and prevention for high‑risk surgical patients.  

1. Method‑level goal – Define, implement and validate common data models (FHIR‑profiles, MII‑core dataset) and interoperable interfaces (TI/KIM, ePA).  
2. Content‑level goal – Deliver AI‑driven, patient‑centred pathways that span pre‑habilitation, intra‑operative management, early rehabilitation and post‑acute follow‑up.  

Specific objectives:  


* Integrate ambulatory and inpatient sources into a single, pseudonymised record‑linkage service.  
* Build and continuously improve predictive models for risk stratification, complication forecasting and individual‑treatment‑effect (ITE) estimation.  
* Embed real‑time decision support into clinical workflows as FHIR‑based Clinical Decision Assistants (CDA).  
* Demonstrate measurable impact: ≥15 % reduction in postoperative complications, ≥1‑day shorter hospital stay, and ≥80 % clinician acceptance of AI tools.  

Use Cases

Use CaseScopeAI Deliverable
InMeD – Intelligent Medical Data for Pre‑habilitationExtends the existing multimodal pre‑habilitation app (nutrition, psychology, movement) to a sector‑wide platform for high‑risk patients scheduled for major surgery.
  • Gradient‑Boosting risk classifier (frailty, renal & cardiac scores).
  • GNN‑based ITE model quantifying benefit of intensive physiotherapy vs. standard care.
  • Survival models with Conformal Prediction delivering calibrated 95 % prediction intervals for intra‑ and postoperative events.
  • Real‑time CDA that alerts OR/ICU staff of elevated bleeding or infection risk. 
Cross‑Hub Patient Journey – End‑to‑End Digital Care PathProvides a seamless, data‑driven patient journey from pre‑habilitation through operation, early rehab, and home‑based aftercare, linking primary, secondary and tertiary providers as well as regional practice‑ and rehab‑networks
  • ML‑based risk stratification at each phase (Frailty, organ‑specific risk).
  • Deep‑Learning analysis of intra‑operative video streams (e.g., laparoscopic image quality).
  • Time‑to‑event models predicting mobilisation success and readmission.
  • Anomaly‑detection on post‑discharge ePA entries for early relapse detection.
  • Federated‑learning framework that updates models from real‑world data across all hubs while keeping patient data on‑site

 

Role of the EMCL

  1. AI Modelling (WP 3 Lead) – Designing the end‑to‑end AI stack, selects model families (GBM, GNN, survival nets, Conformal Prediction) and packages them as reusable Docker/Kubernetes micro‑services.  
  2. Compliance Tasks – Documenting validation studies, runs Explainable‑AI (SHAP/LIME) analyses, and secures CE‑marking/MDR conformity for every AI component.  
  3. Federated‑Learning Tasks – Orchestrating decentralized model training, aggregates updates without moving patient‑level data, and monitors distribution‑shift to maintain performance.  

Research Topics

  1. Standardised Data Modelling – FHIR‑profiles aligned with the MII‑core dataset for cross‑sector exchange.  
  2. Predictive Risk Stratification – Gradient‑Boosting classifiers for frailty, renal and cardiac risk pre‑surgery.  
  3. Individual Treatment‑Effect (ITE) Estimation – Graph‑Neural‑Network linking patient‑diagnosis‑treatment‑outcome.  
  4. Survival & Complication Forecasting – DeepSurv/Cox‑PH with Conformal Prediction for calibrated risk intervals.  
  5. Explainable & Trustworthy AI – SHAP/LIME visualisations, model‑drift detection, audit trails for MDR compliance.  

Through these tightly coupled aims, responsibilities, use cases and research topics, LeMeDaRT 2 will deliver a reusable, AI‑enhanced care continuum that improves outcomes for high‑risk surgical patients and establishes a scalable blueprint for nationwide, data‑driven health‑care delivery.

Funding

People from the EMCL

Partners

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.

Contact

Links

Publications

  • 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.
    doi.org/10.36227/techrxiv.20448768.v2