Julius Keyl
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Julius Keyl has a total of 314 co-authors. Here are the top co-authors by number of shared publications:
- Jens Kleesiek
Technical University Dortmund
10 shared publications - Philipp Keyl
Institute of Pathology, Ludwig-Maximilians-University Munich, Munich, Germany
10 shared publications Stefan Kasper
University Hospital Essen
9 shared publications- Martin Schuler
University Hospital Essen
8 shared publications Jan Egger
Affiliation Institute for Artificial Intelligence in Medicine, University Hospital Essen, Essen, Germany
7 shared publications- Sylvia Hartmann
Institute of Pathology, University Hospital Essen, Essen, Germany
7 shared publications
Latest Publications
- Open Access
- 12 November 2026
- Published by Springer Nature in Nature Cancer
A single-cell lung cancer atlas reveals cellular and molecular drivers of metastasis and clinical outcome
Non-small-cell lung cancer (NSCLC) remains a leading cause of cancer-related mortality, characterized by extensive cellular and molecular heterogeneity. Here, to capture this complexity, we present LuCaS+, a multicenter single-cell RNA sequencing atlas comprising over 2.5 million nuclei from 454 clinically annotated patients with NSCLC with long-term follow-up. LuCaS+ consists of a discovery dataset (Berlin and Cologne, 364 patients) and an independent validation dataset (Munich, 90 patients). Leveraging this dataset, we identified tumor subpopulations with high metastatic potential associated with oncogenic signaling, immune evasion and epithelial–mesenchymal transition. In early-stage adenocarcinoma (lung adenocarcinoma), these cells showed strong prognostic value. Integrating tumor and microenvironmental cell profiles improved survival prediction across all stages, identifying distinct tumor–immune–stromal interaction patterns. These included monocyte- and fibroblast-driven inflammation (lung adenocarcinoma), VEGF–TGFβ-mediated angiogenic feedback loops (lung squamous cell carcinoma) and immune-permissive niches promoting antitumor immunity. LuCaS+ offers a clinically anchored single-cell resource to enable personalized, biology-driven therapies in NSCLC.
- 12 November 2026
- Published by Springer Nature in Nature Cancer
Open AccessA single-cell lung cancer atlas reveals cellular and molecular drivers of metastasis and clinical outcome
Non-small-cell lung cancer (NSCLC) remains a leading cause of cancer-related mortality, characterized by extensive cellular and molecular heterogeneity. Here, to capture this complexity, we present LuCaS+, a multicenter single-cell RNA sequencing atlas comprising over 2.5 million nuclei from 454 clinically annotated patients with NSCLC with long-term follow-up. LuCaS+ consists of a discovery dataset (Berlin and Cologne, 364 patients) and an independent validation dataset (Munich, 90 patients). Leveraging this dataset, we identified tumor subpopulations with high metastatic potential associated with oncogenic signaling, immune evasion and epithelial–mesenchymal transition. In early-stage adenocarcinoma (lung adenocarcinoma), these cells showed strong prognostic value. Integrating tumor and microenvironmental cell profiles improved survival prediction across all stages, identifying distinct tumor–immune–stromal interaction patterns. These included monocyte- and fibroblast-driven inflammation (lung adenocarcinoma), VEGF–TGFβ-mediated angiogenic feedback loops (lung squamous cell carcinoma) and immune-permissive niches promoting antitumor immunity. LuCaS+ offers a clinically anchored single-cell resource to enable personalized, biology-driven therapies in NSCLC.
- Open Access
- 11 September 2026
- Published by Wiley in European Journal of Haematology
Transplantation Strategies in Treatment of Primary Plasma Cell Leukemia in Germany—A Multi‐Center Analysis
Objectives: Primary plasma cell leukemia (pPCL) is a rare disease with poor outcomes. As data on randomized phase‐III trials are lacking, treatment recommendations are predominately based on retrospective analyses incorporating time periods with outdated therapies. Thus, the optimal consolidation strategy remains to be defined. Methods: We conducted a retrospective, multi‐center analysis with 49 transplant‐eligible patients diagnosed with pPCL between 2011 and 2025 focusing on different stem cell transplantation (SCT) strategies as consolidation treatment in first remission: single autologous (auto) or single allogeneic (allo) SCT versus patients with tandem cell therapy strategies (auto‐auto, allo‐auto, auto‐chimeric antigen receptor therapy). Results: The median follow‐up after first SCT was 31 months. The cumulative incidence of initiating next treatment was lower after tandem cell therapy compared to single SCT (at 12 months: 42% vs. 57%, p = 0.006). Moreover, in a multivariable analysis incorporating cytogenetic risk and response to induction therapy, tandem transplantation strategies tended to be the only independent predictor for prolonged time to next treatment (TTNT, p = 0.077). Noteworthy, differences in TTNT between patients with a single SCT and patients with tandem cell therapy did not translate into an improved overall survival, possibly due to potent salvage therapies. Conclusions: Our data support recent recommendations advising treating patients with tandem transplantation regimes.
- 11 September 2026
- Published by Wiley in European Journal of Haematology
Open AccessTransplantation Strategies in Treatment of Primary Plasma Cell Leukemia in Germany—A Multi‐Center Analysis
Objectives: Primary plasma cell leukemia (pPCL) is a rare disease with poor outcomes. As data on randomized phase‐III trials are lacking, treatment recommendations are predominately based on retrospective analyses incorporating time periods with outdated therapies. Thus, the optimal consolidation strategy remains to be defined. Methods: We conducted a retrospective, multi‐center analysis with 49 transplant‐eligible patients diagnosed with pPCL between 2011 and 2025 focusing on different stem cell transplantation (SCT) strategies as consolidation treatment in first remission: single autologous (auto) or single allogeneic (allo) SCT versus patients with tandem cell therapy strategies (auto‐auto, allo‐auto, auto‐chimeric antigen receptor therapy). Results: The median follow‐up after first SCT was 31 months. The cumulative incidence of initiating next treatment was lower after tandem cell therapy compared to single SCT (at 12 months: 42% vs. 57%, p = 0.006). Moreover, in a multivariable analysis incorporating cytogenetic risk and response to induction therapy, tandem transplantation strategies tended to be the only independent predictor for prolonged time to next treatment (TTNT, p = 0.077). Noteworthy, differences in TTNT between patients with a single SCT and patients with tandem cell therapy did not translate into an improved overall survival, possibly due to potent salvage therapies. Conclusions: Our data support recent recommendations advising treating patients with tandem transplantation regimes.
- Open Access
An explainable transformer model learning from entire treatment timelines for pan-cancer risk profiling across healthcare systems
Cancer outcomes vary widely between individual patients, each accumulating an irregular record of treatments, diagnoses, measurements, and complications. Current prognostic models reduce this complexity into a single snapshot, focus on narrow clinical settings, and rarely generalize across hospitals. Here we introduce Chronicle, an explainable transformer that learns from entire patient trajectories, predicts diverse clinical outcomes throughout the disease course while capturing both short- and long-term temporal dependencies. Trained on 53.7 million longitudinal data points from 51,711 patients spanning 67 cancer types, Chronicle operates natively on irregular data without imputation and jointly predicts eight endpoints within a flexible framework adaptable to additional outcomes. Chronicle outperformed cross-sectional models for overall survival prediction (C-index 0.84 vs 0.76-0.79), stratified patients more accurately than established prognostic systems, including TNM stage, and predicted seven adverse event and transfusion endpoints (AUC 0.80-0.92). Applied without retraining to 69,341 patients in Germany, Switzerland, and the United States, Chronicle generalized across healthcare systems and improved further with local fine-tuning. Integrated explainability traced each risk update to patient-specific clinical factors, revealing distinct temporal persistence of prognostic information, with relevance half-lives ranging from weeks for therapies to nearly one year for baseline characteristics. These findings demonstrate that learning from hospital-wide patient trajectories enables interpretable and continuously updated predictions, providing a scalable framework to support individualized treatment decisions.
Open AccessAn explainable transformer model learning from entire treatment timelines for pan-cancer risk profiling across healthcare systems
Cancer outcomes vary widely between individual patients, each accumulating an irregular record of treatments, diagnoses, measurements, and complications. Current prognostic models reduce this complexity into a single snapshot, focus on narrow clinical settings, and rarely generalize across hospitals. Here we introduce Chronicle, an explainable transformer that learns from entire patient trajectories, predicts diverse clinical outcomes throughout the disease course while capturing both short- and long-term temporal dependencies. Trained on 53.7 million longitudinal data points from 51,711 patients spanning 67 cancer types, Chronicle operates natively on irregular data without imputation and jointly predicts eight endpoints within a flexible framework adaptable to additional outcomes. Chronicle outperformed cross-sectional models for overall survival prediction (C-index 0.84 vs 0.76-0.79), stratified patients more accurately than established prognostic systems, including TNM stage, and predicted seven adverse event and transfusion endpoints (AUC 0.80-0.92). Applied without retraining to 69,341 patients in Germany, Switzerland, and the United States, Chronicle generalized across healthcare systems and improved further with local fine-tuning. Integrated explainability traced each risk update to patient-specific clinical factors, revealing distinct temporal persistence of prognostic information, with relevance half-lives ranging from weeks for therapies to nearly one year for baseline characteristics. These findings demonstrate that learning from hospital-wide patient trajectories enables interpretable and continuously updated predictions, providing a scalable framework to support individualized treatment decisions.
- Open Access
- 8 July 2026
- Published by Public Library of Science (PLoS) in PLOS Digital Health
- Vol. 5 (7) , e0001546
Large language models enable prognostic stratification of cancer patients using real-world clinical notes
In medical documentation, vast amounts of unstructured text are generated that are still underutilized in current prognostic models. We investigate the potential of self-hosted large language models (LLM) to extract clinically meaningful, patient-specific information from routine clinical notes for personalized risk stratification in cancer care. We collected real-world medical notes from 2,708 non-small cell lung cancer (NSCLC) patients and 814 colon cancer patients documented before treatment at a large comprehensive cancer center. LLMs extracted key prognostic indicators, including comorbidities, metastatic sites, and qualitative descriptors of patient condition, in a zero-shot manner without prior task-specific training. Integrating these LLM-derived features into machine learning models significantly improved the prediction of overall survival compared to TNM staging alone (C-Index: NSCLC, 0.72 vs 0.64; colon cancer, 0.70 vs 0.59), and surpassed models using text embeddings. Based on the LLM-informed risk scores, patients were stratified into four distinct risk groups, enabling reclassification of 61.4% of NSCLC and 68.3% of colon cancer patients. Analysis of model drivers revealed that LLM-derived factors, such as the physical condition, substantially modulated the prognostic impact of TNM stage. These findings highlight the potential of self-hosted LLM to derive prognostically relevant information from unstructured clinical documentation and support clinical decision-making.
- 8 July 2026
- Published by Public Library of Science (PLoS) in PLOS Digital Health
- Vol. 5 (7) , e0001546
Open AccessLarge language models enable prognostic stratification of cancer patients using real-world clinical notes
In medical documentation, vast amounts of unstructured text are generated that are still underutilized in current prognostic models. We investigate the potential of self-hosted large language models (LLM) to extract clinically meaningful, patient-specific information from routine clinical notes for personalized risk stratification in cancer care. We collected real-world medical notes from 2,708 non-small cell lung cancer (NSCLC) patients and 814 colon cancer patients documented before treatment at a large comprehensive cancer center. LLMs extracted key prognostic indicators, including comorbidities, metastatic sites, and qualitative descriptors of patient condition, in a zero-shot manner without prior task-specific training. Integrating these LLM-derived features into machine learning models significantly improved the prediction of overall survival compared to TNM staging alone (C-Index: NSCLC, 0.72 vs 0.64; colon cancer, 0.70 vs 0.59), and surpassed models using text embeddings. Based on the LLM-informed risk scores, patients were stratified into four distinct risk groups, enabling reclassification of 61.4% of NSCLC and 68.3% of colon cancer patients. Analysis of model drivers revealed that LLM-derived factors, such as the physical condition, substantially modulated the prognostic impact of TNM stage. These findings highlight the potential of self-hosted LLM to derive prognostically relevant information from unstructured clinical documentation and support clinical decision-making.
- Open Access
Explainable AI Predicts Hematoxicity from Cancer Treatment Using Multimodal Real-World Data
Adverse drug effects remain a major barrier to safe and effective cancer therapy, underscoring the need for tools that predict treatment-related toxicities. We analyzed multimodal real-world data from 14,596 cancer patients across 38 cancer entities, encompassing 330 clinical, tumor, and imaging characteristics, along with 89 anticancer agents. Hematological adverse events (HAE), defined by nadirs of hemoglobin, leukocyte, neutrophil, and platelet values within two months of treatment initiation, were highly prevalent (87.7%; 33.1% severe). We developed Toxix , an explainable artificial intelligence (xAI) framework modeling interactions between patient characteristics and drug combinations. Toxix achieved strong predictive performance for severe toxicities (median AUROC 0.85 for anemia; >0.76 for leukopenia, neutropenia, and thrombocytopenia) and was validated in an external cohort of 2,768 patients with non-small cell lung cancer. Model explainability enabled systematic characterization of drug-patient interactions underlying HAEs. Toxix provides a real-world informed framework for personalized and toxicity-aware cancer therapy planning.
Open AccessExplainable AI Predicts Hematoxicity from Cancer Treatment Using Multimodal Real-World Data
Adverse drug effects remain a major barrier to safe and effective cancer therapy, underscoring the need for tools that predict treatment-related toxicities. We analyzed multimodal real-world data from 14,596 cancer patients across 38 cancer entities, encompassing 330 clinical, tumor, and imaging characteristics, along with 89 anticancer agents. Hematological adverse events (HAE), defined by nadirs of hemoglobin, leukocyte, neutrophil, and platelet values within two months of treatment initiation, were highly prevalent (87.7%; 33.1% severe). We developed Toxix , an explainable artificial intelligence (xAI) framework modeling interactions between patient characteristics and drug combinations. Toxix achieved strong predictive performance for severe toxicities (median AUROC 0.85 for anemia; >0.76 for leukopenia, neutropenia, and thrombocytopenia) and was validated in an external cohort of 2,768 patients with non-small cell lung cancer. Model explainability enabled systematic characterization of drug-patient interactions underlying HAEs. Toxix provides a real-world informed framework for personalized and toxicity-aware cancer therapy planning.
- 5 January 2026
- Published by American Association for Cancer Research (AACR) in Clinical Cancer Research
- Vol. 32 (5) , 938-946
Deep Learning–Derived Sarcopenia Marker Predicts Benefit from Anti-EGFR Therapy in Patients with RAS Wild-type Metastatic Colorectal Cancer
Purpose: The benefit of treatment intensification in metastatic colorectal cancer (mCRC) may be influenced by host-related factors that are not accounted for in clinical trials or standard care. We investigated the prognostic and predictive value of the muscle/bone ratio (MBR), a sarcopenia marker automatically derived from computed tomography (CT) images, in patients with mCRC from the prospective PanaMa study and a real-world validation cohort. Experimental Design: PanaMa (AIO KRK 0212; NCT01991873) randomized patients with RAS wild-type mCRC, following induction therapy, to maintenance therapy with fluorouracil and folinic acid (FU/FA) with or without panitumumab (Pmab). MBR was automatically calculated from baseline CT images using a validated deep learning model, and patients were stratified by MBR tertiles. Associations with progression-free survival (PFS) and overall survival (OS) were studied using Kaplan–Meier and Cox regression analyses. A retrospective real-world cohort of patients with mCRC treated with cetuximab was used for validation. Results: Premaintenance CT images were available for 189 of 248 randomized patients (76.2%) from PanaMa. In patients receiving FU/FA + Pmab, high MBR was associated with longer PFS [HR, 0.43; 95% confidence interval (CI), 0.25–0.73; P = 0.002] and OS (HR, 0.41; 95% CI, 0.21–0.77; P = 0.006), whereas no association was observed in patients receiving FU/FA alone. Pmab provided a PFS benefit only in patients with high MBR (HR, 0.42; 95% CI, 0.24–0.73; P = 0.002). The association of high MBR with superior PFS (P = 0.002) and OS (P < 0.001) was confirmed in the real-world cohort. Conclusions: The benefit of anti-EGFR therapy in mCRC is confined to patients with a high MBR. Automated sarcopenia assessment holds promise for personalized treatment intensification in mCRC.
- 5 January 2026
- Published by American Association for Cancer Research (AACR) in Clinical Cancer Research
- Vol. 32 (5) , 938-946
Deep Learning–Derived Sarcopenia Marker Predicts Benefit from Anti-EGFR Therapy in Patients with RAS Wild-type Metastatic Colorectal Cancer
Purpose: The benefit of treatment intensification in metastatic colorectal cancer (mCRC) may be influenced by host-related factors that are not accounted for in clinical trials or standard care. We investigated the prognostic and predictive value of the muscle/bone ratio (MBR), a sarcopenia marker automatically derived from computed tomography (CT) images, in patients with mCRC from the prospective PanaMa study and a real-world validation cohort. Experimental Design: PanaMa (AIO KRK 0212; NCT01991873) randomized patients with RAS wild-type mCRC, following induction therapy, to maintenance therapy with fluorouracil and folinic acid (FU/FA) with or without panitumumab (Pmab). MBR was automatically calculated from baseline CT images using a validated deep learning model, and patients were stratified by MBR tertiles. Associations with progression-free survival (PFS) and overall survival (OS) were studied using Kaplan–Meier and Cox regression analyses. A retrospective real-world cohort of patients with mCRC treated with cetuximab was used for validation. Results: Premaintenance CT images were available for 189 of 248 randomized patients (76.2%) from PanaMa. In patients receiving FU/FA + Pmab, high MBR was associated with longer PFS [HR, 0.43; 95% confidence interval (CI), 0.25–0.73; P = 0.002] and OS (HR, 0.41; 95% CI, 0.21–0.77; P = 0.006), whereas no association was observed in patients receiving FU/FA alone. Pmab provided a PFS benefit only in patients with high MBR (HR, 0.42; 95% CI, 0.24–0.73; P = 0.002). The association of high MBR with superior PFS (P = 0.002) and OS (P < 0.001) was confirmed in the real-world cohort. Conclusions: The benefit of anti-EGFR therapy in mCRC is confined to patients with a high MBR. Automated sarcopenia assessment holds promise for personalized treatment intensification in mCRC.
- 1 January 2026
- Published by American Society of Clinical Oncology (ASCO) in JCO Clinical Cancer Informatics
- Vol. 10 (1) , e2500254
Automated Tumor International Classification of Diseases Coding of Real-World Pathology Reports Using Self-Hosted Large Language Models
PURPOSE: Manual coding of pathology reports with International Classification of Diseases for Oncology (ICD-O)-3 codes is time-consuming, error-prone, and resource-intensive for health care institutions. To evaluate the performance of multiple state-of-the-art large language models (LLMs) in extracting ICD-O-3 topography and morphology codes from real-world pathology reports and assess their potential for clinical implementation, this study compares the performance of state-of-the-art open-source models in multiple evaluation setups. METHODS: We analyzed 21,364 pathology reports from 10,823 patients documented between 2013 and 2025 at a large German hospital. Five LLMs were evaluated: Llama-3.3-70B-Instruct, DeepSeek-R1-Distill-Llama (8B and 70B variants), Qwen3-235B-A22B, and Gemma-3-12B-it. All models were deployed on secured private information technology hospital infrastructure. Three different prompts were developed for topography extraction (with and without anatomic context) and morphology extraction. Performance was evaluated using exact code matches and first three-position matches. RESULTS: For exact ICD-O topography code prediction, Qwen3-235B-A22B achieved the highest performance (microaverage F1: 71.6%), whereas Llama-3.3-70B-Instruct performed best at predicting the first three characters (micro-average F1: 84.6%). For morphology codes, DeepSeek-R1-Distill-Llama-70B outperformed other models (exact microaverage F1: 34.7%; first three characters’ microaverage F1: 77.8%). Large disparities between micro- and macroaverage F1-scores indicated poor generalization to rare conditions. CONCLUSION: Although LLMs demonstrate promising capabilities as support systems for expert-guided pathology coding, their performance is not yet sufficient for fully automated, unsupervised use in routine clinical workflows. LLMs showed poor performance on rare conditions, heavy dependence on contextual information, and substantially lower scores for morphology versus topography classification.
- 1 January 2026
- Published by American Society of Clinical Oncology (ASCO) in JCO Clinical Cancer Informatics
- Vol. 10 (1) , e2500254
Automated Tumor International Classification of Diseases Coding of Real-World Pathology Reports Using Self-Hosted Large Language Models
PURPOSE: Manual coding of pathology reports with International Classification of Diseases for Oncology (ICD-O)-3 codes is time-consuming, error-prone, and resource-intensive for health care institutions. To evaluate the performance of multiple state-of-the-art large language models (LLMs) in extracting ICD-O-3 topography and morphology codes from real-world pathology reports and assess their potential for clinical implementation, this study compares the performance of state-of-the-art open-source models in multiple evaluation setups. METHODS: We analyzed 21,364 pathology reports from 10,823 patients documented between 2013 and 2025 at a large German hospital. Five LLMs were evaluated: Llama-3.3-70B-Instruct, DeepSeek-R1-Distill-Llama (8B and 70B variants), Qwen3-235B-A22B, and Gemma-3-12B-it. All models were deployed on secured private information technology hospital infrastructure. Three different prompts were developed for topography extraction (with and without anatomic context) and morphology extraction. Performance was evaluated using exact code matches and first three-position matches. RESULTS: For exact ICD-O topography code prediction, Qwen3-235B-A22B achieved the highest performance (microaverage F1: 71.6%), whereas Llama-3.3-70B-Instruct performed best at predicting the first three characters (micro-average F1: 84.6%). For morphology codes, DeepSeek-R1-Distill-Llama-70B outperformed other models (exact microaverage F1: 34.7%; first three characters’ microaverage F1: 77.8%). Large disparities between micro- and macroaverage F1-scores indicated poor generalization to rare conditions. CONCLUSION: Although LLMs demonstrate promising capabilities as support systems for expert-guided pathology coding, their performance is not yet sufficient for fully automated, unsupervised use in routine clinical workflows. LLMs showed poor performance on rare conditions, heavy dependence on contextual information, and substantially lower scores for morphology versus topography classification.
- Open Access
- 20 November 2025
- Published by Elsevier in Clinical Surgical Oncology
- Vol. 4 (4)
Automated CT-based sarcopenia assessment for risk stratification of patients undergoing colorectal cancer resection
- 20 November 2025
- Published by Elsevier in Clinical Surgical Oncology
- Vol. 4 (4)
Open AccessAutomated CT-based sarcopenia assessment for risk stratification of patients undergoing colorectal cancer resection
- Open Access
- 5 November 2025
- Published by Elsevier in JHEP Reports
- Vol. 8 (1)
Clinical practice and implications of biomarker testing in biliary tract cancer: An observational study
- 5 November 2025
- Published by Elsevier in JHEP Reports
- Vol. 8 (1)
Open AccessClinical practice and implications of biomarker testing in biliary tract cancer: An observational study