Felix Nensa

University Hospital Essen
Country flagGermany

Research Interests

ApplyKite TwinLearns as you apply

Make your next move.

Prepare your application

Review drafts before you send them.

Explore related searches

Articles (6)

BACKGROUND Unstructured clinical text remains a major barrier to interoperable data reuse and large-scale secondary analysis in healthcare. Large language models (LLMs) have the potential to automate the extraction of structured clinical information; however, their application is limited by the scarcity of high-quality annotated training data. OBJECTIVE - METHODS We evaluated an LLM–based pipeline for extracting structured clinical information from cancer-related discharge letters and mapping it to representations compatible with Fast Healthcare Interoperability Resources (FHIR). To enable large-scale supervised training, we developed a random sample generator that creates synthetic discharge letters using Qwen3 235B by randomly sampling and aggregating structured FHIR data from 41,175 cancer patients. The resulting synthetic discharge letters (n=75k) were paired with their originating structured data, forming a large-scale dataset for fine-tuning MedGemma 27B. Evaluation was conducted on the synthetic test dataset (n=7,500), real-world discharge letters (n=30) which are evaluated by physicians and a medical student, and a comparative one-shot approach using open-source models (Qwen3, LLaMA, and GPT-OSS). RESULTS The fine-tuned model achieved high extraction performance across multiple clinical entities, including full ICD diagnosis codes (F1 = 0.84), tumor-related information (0.99), laboratory values (0.99), medication names and dosages (0.99), and ATC medication codes (0.94). Extraction of procedure-related information was more challenging but remained reliable, with F1 scores of 0.63 for OPS codes and 0.90 for procedure descriptions. In a one-shot comparison of general-purpose LLMs with the fine-tuned model, the fine-tuned model consistently outperformed general-purpose LLMs in nearly all extraction categories. When applied to real-world discharge letters, performance remained robust, with F1 scores of 78.9% for ICD diagnoses, 86.1% for tumor-related information, 93% for medications, and 61.3% for procedures. CONCLUSIONS These results demonstrate that synthetic text generation from structured clinical data enables effective and scalable training of LLMs for extracting interoperable, multi-entity clinical information from unstructured documentation.

Year:

2026

BACKGROUND The standardized extraction of postoperative complications from unstructured routine clinical documentation remains a major unresolved challenge in digital surgery and health informatics. Although the Clavien–Dindo classification is the established standard for grading postoperative complications, its application in routine clinical documentation is largely implicit and unstructured. OBJECTIVE To assess the capability of open-weight and proprietary large language models (LLMs) to classify postoperative complications according to the Clavien–Dindo system using discharge letters, benchmarked against expert consensus annotations. METHODS We analyzed discharge letters from 650 surgical cases of patients who underwent hepatobiliary surgery between 2010 and 2024. The cohort included Grade I–II complications in 24%, Grade III–IV in 19%, and Grade V (death) in 6% of patients. Representative open-weight (Qwen 3, Llama 3.3, Ministral 3, GPT-OSS) and proprietary (GPT 5.1, Gemini 3 Pro) LLMs were prompted to infer complication grades directly from the discharge letters. Model performance was evaluated against expert assessment using accuracy and a detailed deviation analysis. RESULTS All models were capable of identifying and classifying complications from the unstructured documentation in the discharge letters. On the full 650-case dataset, open-weight models achieved accuracies up to 0.775 for fine-grading prediction and 0.94 for binary classification. On a balanced 50-case subset, proprietary models achieved the highest performance, with accuracies of 0.78 for fine-grading and 0.98 for binary classification. In contrast, open-weight models reached accuracies up to 0.70 in fine-grading and 0.94 for binary grading, with lower computational requirements. An ensemble approach yielded additional gains in classification performance. CONCLUSIONS LLMs can accurately classify postoperative complications from discharge letters, enabling scalable and objective monitoring of surgical outcomes. Their use may reduce manual abstraction workload and promote consistent, data-driven quality assessment in surgical care.

Year:

2026

Objective To address the challenges of data heterogeneity and manual feature engineering in clinical predictive modeling, we introduce FHIR-Former, an open-source framework integrating Fast Healthcare Interoperability Resources (FHIR) with large language models (LLMs) to automate and standardize clinical prediction tasks. Materials and Methods FHIR-Former dynamically processes structured (eg, lab results, medications) and unstructured (eg, clinical notes) data from FHIR resources. The pipeline supports multiple classification tasks, including 30-day readmission, imaging study prediction, and ICD code classification. Leveraging open-source LLMs (GeBERTa), we trained models on 1.1 million data points across ten FHIR resources using retrospective inpatient data (2018-2024). Hyperparameters were optimized via Bayesian methods, and outputs were mapped to FHIR RiskAssessment resources for interoperability. Results FHIR-Former achieved an F1-score of 70.7% and accuracy of 72.9% for 30-day readmission, 51.8% F1-score (88.1% accuracy) for mortality prediction, and 61% macro F1-score for imaging study classification. The ICD code prediction model attained 94% accuracy. Performance demonstrated promising performance for readmission and showed scalability across tasks without manual feature engineering. Discussion FHIR-Former eliminates institution-specific preprocessing by adapting to diverse FHIR implementations, enabling seamless integration of multimodal data. Its configurable architecture outperformed prior frameworks reliant on static inputs or limited to unstructured text. Real-time risk scores embedded in FHIR servers enhance clinical workflows without disrupting existing practices. Conclusion By harmonizing FHIR standardization with LLM flexibility, FHIR-Former advances scalable, interoperable predictive modeling in healthcare. The open-source framework facilitates automation, improves resource allocation, and supports personalized decision-making, bridging gaps between AI innovation and clinical practice.

Year:

2025

BACKGROUND Fast Healthcare Interoperability Resources (FHIR) is a widely used standard for storing and exchanging health care data. At the same time, image-based artificial intelligence (AI) models for quantifying relevant body structures and organs from routine computed tomography (CT)/magnetic resonance imaging scans have emerged. The missing link, simultaneously a needed step in advancing personalized medicine, is the incorporation of measurements delivered by AI models into an interoperable and standardized format. Incorporating image-based measurements and biomarkers into FHIR profiles can standardize data exchange, enabling timely, personalized treatment decisions and improving the precision and efficiency of patient care. OBJECTIVE This study aims to present the synergistic incorporation of CT-derived body organ and composition measurements with FHIR, delineating an initial paradigm for storing image-based biomarkers. METHODS This study integrated the results of the Body and Organ Analysis (BOA) model into FHIR profiles to enhance the interoperability of image-based biomarkers in radiology. The BOA model was selected as an exemplary AI model due to its ability to provide detailed body composition and organ measurements from CT scans. The FHIR profiles were developed based on 2 primary observation types: Body Composition Analysis (BCA Observation) for quantitative body composition metrics and Body Structure Observation for organ measurements. These profiles were structured to interoperate with a specially designed Diagnostic Report profile, which references the associated Imaging Study, ensuring a standardized linkage between image data and derived biomarkers. To ensure interoperability, all labels were mapped to SNOMED CT (Systematized Nomenclature of Medicine – Clinical Terms) or RadLex terminologies using specific value sets. The profiles were developed using FHIR Shorthand (FSH) and SUSHI, enabling efficient definition and implementation guide generation, ensuring consistency and maintainability. RESULTS In this study, 4 BOA profiles, namely, Body Composition Analysis Observation, Body Structure Volume Observation, Diagnostic Report, and Imaging Study, have been presented. These FHIR profiles, which cover 104 anatomical landmarks, 8 body regions, and 8 tissues, enable the interoperable usage of the results of AI segmentation models, providing a direct link between image studies, series, and measurements. CONCLUSIONS The BOA profiles provide a foundational framework for integrating AI-derived imaging biomarkers into FHIR, bridging the gap between advanced imaging analytics and standardized health care data exchange. By enabling structured, interoperable representation of body composition and organ measurements, these profiles facilitate seamless integration into clinical and research workflows, supporting improved data accessibility and interoperability. Their adaptability allows for extension to other imaging modalities and AI models, fostering a more standardized and scalable approach to using imaging biomarkers in precision medicine. This work represents a step toward enhancing the integration of AI-driven insights into digital health ecosystems, ultimately contributing to more data-driven, personalized, and efficient patient care.

Year:

2024

Automated medical image analysis systems often require large amounts of training data with high quality labels, which are difficult and time consuming to generate. This paper introduces Radiology Object in COntext version 2 (ROCOv2), a multimodal dataset consisting of radiological images and associated medical concepts and captions extracted from the PMC Open Access subset. It is an updated version of the ROCO dataset published in 2018, and adds 35,705 new images added to PMC since 2018. It further provides manually curated concepts for imaging modalities with additional anatomical and directional concepts for X-rays. The dataset consists of 79,789 images and has been used, with minor modifications, in the concept detection and caption prediction tasks of ImageCLEFmedical Caption 2023. The dataset is suitable for training image annotation models based on image-caption pairs, or for multi-label image classification using Unified Medical Language System (UMLS) concepts provided with each image. In addition, it can serve for pre-training of medical domain models, and evaluation of deep learning models for multi-task learning.

Year:

2024

Platelet demand management (PDM) is a resource-consuming task for physicians and transfusion managers of large hospitals. Inpatient numbers and institutional standards play significant roles in PDM. However, reliance on these factors alone commonly results in platelet shortages. Using data from multiple sources, we developed, validated, tested, and implemented a patient-specific approach to support PDM that uses a deep learning–based risk score to forecast platelet transfusions for each hospitalized patient in the next 24 hours. The models were developed using retrospective electronic health record data of 34 809 patients treated between 2017 and 2022. Static and time-dependent features included demographics, diagnoses, procedures, blood counts, past transfusions, hematotoxic medications, and hospitalization duration. Using an expanding window approach, we created a training and live-prediction pipeline with a 30-day input and 24-hour forecast. Hyperparameter tuning determined the best validation area under the precision-recall curve (AUC-PR) score for long short-term memory deep learning models, which were then tested on independent data sets from the same hospital. The model tailored for hematology and oncology patients exhibited the best performance (AUC-PR, 0.84; area under the receiver operating characteristic curve [ROC-AUC], 0.98), followed by a multispecialty model covering all other patients (AUC-PR, 0.73). The model specific to cardiothoracic surgery had the lowest performance (AUC-PR, 0.42), likely because of unexpected intrasurgery bleedings. To our knowledge, this is the first deep learning–based platelet transfusion predictor enabling individualized 24-hour risk assessments at high AUC-PR. Implemented as a decision-support system, deep-learning forecasts might improve patient care by detecting platelet demand earlier and preventing critical transfusion shortages.

Year:

2023

Collaborators (8)

Alba Garcia Seco de Herrera

Distinguished Researcher

Universidad Nacional de Educación a Distancia

SPAIN
View details

Henning Müller

Professor

HES-SO

SWITZERLAND
View details

Sophia Marie-Therese Schmitz

Universität Duisburg-Essen

GERMANY
View details

Christoph M. Friedrich

Head of Department

Technische Universität Dortmund

GERMANY
View details

Elisabeth Livingstone

University Hospital Essen

GERMANY
View details

Jens Kleesiek

-

GERMANY
View details

Sylvia Thun

Professor at Berlin Institute of Health

Berlin Institute of Health at Charité - Universitätsmedizin Berlin

GERMANY
View details

Hans-Ulrich Prokosch

friedrich-alexander-universität erlangen-nürnberg (fau)

GERMANY
View details