Paul-Henry Koop
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Paul-Henry Koop has a total of 34 co-authors. Here are the top co-authors by number of shared publications:
- Carolin Victoria Schneider
Department of Medicine III, University Hospital RWTH Aachen, Aachen, Germany
6 shared publications - Jan Clusmann
Else Kroener Fresenius Center for Digital Health, Technical University Dresden, Dresden, Germany
6 shared publications - Daniel Truhn
Diagnostic and Interventional Radiology, University Hospital Aachen, Aachen, Germany
5 shared publications - Jakob Nikolas Kather
Else Kroener Fresenius Center for Digital Health, Technical University Dresden, Dresden, Germany
5 shared publications - Kai Markus Schneider
Department of Medicine III, University Hospital RWTH Aachen, Aachen, Germany
5 shared publications - Tony Bruns
Department of Internal Medicine III University Hospital RWTH Aachen Aachen Germany
5 shared publications
Latest Publications
- Open Access
- 1 July 2026
- Published by American Association for Cancer Research (AACR)
Data from Machine Learning Predicts Hepatocellular Carcinoma Risk from Routine Clinical Data: A Large Population-Based Multicentric Study
Hepatocellular carcinoma (HCC) is a highly fatal tumor, for which risk stratification is crucial yet remains challenging. In this study, we develop an interpretable machine learning (ML) framework for HCC risk stratification based on routinely collected clinical data. We utilize prospectively collected multimodal data from more than 900,000 individuals and 983 cases of HCC across two population-scale cohorts: the UK Biobank study (development) and the All of Us Research Program (external testing). We assess individual and cumulative contributions of data modalities, including demographics, lifestyle, health records, blood, genomics, and metabolomics. Our final random forest–based models significantly outperform all publicly available state-of-the-art risk scores on both internal and external test sets. We demonstrate robustness across ethnic subgroups, provide comprehensive interpretability, and release all code, model weights, and a web calculator for external validation and agentic integration. Our study presents PRE-Screen-HCC, a robust and interpretable ML framework for HCC risk stratification and early detection.Significance:Using data from population-scale cohorts, we develop and externally validate an ML framework for HCC risk stratification. Models trained on routine clinical data outperform published scores, perform on par with metabolomics and genomics, generalize across subgroups, and remain interpretable.See related commentary by Foda, p. 1252
- 1 July 2026
- Published by American Association for Cancer Research (AACR)
Open AccessData from Machine Learning Predicts Hepatocellular Carcinoma Risk from Routine Clinical Data: A Large Population-Based Multicentric Study
Hepatocellular carcinoma (HCC) is a highly fatal tumor, for which risk stratification is crucial yet remains challenging. In this study, we develop an interpretable machine learning (ML) framework for HCC risk stratification based on routinely collected clinical data. We utilize prospectively collected multimodal data from more than 900,000 individuals and 983 cases of HCC across two population-scale cohorts: the UK Biobank study (development) and the All of Us Research Program (external testing). We assess individual and cumulative contributions of data modalities, including demographics, lifestyle, health records, blood, genomics, and metabolomics. Our final random forest–based models significantly outperform all publicly available state-of-the-art risk scores on both internal and external test sets. We demonstrate robustness across ethnic subgroups, provide comprehensive interpretability, and release all code, model weights, and a web calculator for external validation and agentic integration. Our study presents PRE-Screen-HCC, a robust and interpretable ML framework for HCC risk stratification and early detection.Significance:Using data from population-scale cohorts, we develop and externally validate an ML framework for HCC risk stratification. Models trained on routine clinical data outperform published scores, perform on par with metabolomics and genomics, generalize across subgroups, and remain interpretable.See related commentary by Foda, p. 1252
- Open Access
- 1 July 2026
- Published by American Association for Cancer Research (AACR)
Supplementary Tables ST1 - ST29 from Machine Learning Predicts Hepatocellular Carcinoma Risk from Routine Clinical Data: A Large Population-Based Multicentric Study
Supplementary Table 1 lists HCC case counts per UKB assessment center used in the study. Supplementary Table 2 lists the assessment-visit variables extracted and harmonized in UKB. Supplementary Table 3 reports PheWAS results used to derive disease-level features in UKB. Supplementary Table 4 lists the diagnosis-based inclusion criteria defining the PAR cohort in UKB. Supplementary Table 5 lists all single ICD code features used across UKB and AOU. Supplementary Table 6 lists grouped ICD (phecode/category) features used across UKB and AOU. Supplementary Table 7 maps self-reported conditions to ICD-based features in UKB. Supplementary Table 8 summarizes at-risk subset definitions and corresponding HCC/control sample sizes in UKB. Supplementary Table 9 provides the UKB ICD code summary representations. Supplementary Table 10 lists blood count and biochemistry features used for modeling in UKB and AOU. Supplementary Table 11 reports UKB blood count/biochemistry summary statistics stratified by HCC status and sex. Supplementary Table 12 lists genomic features (variants, coding, and availability) used in UKB modeling. Supplementary Table 13 lists metabolomics features used in UKB modeling. Supplementary Table 14 reports UKB metabolomics summary statistics stratified by HCC status. Supplementary Table 15 details the center-based train/test splits applied in UKB. Supplementary Table 16 provides UKB–AOU variable mappings and conversion summaries (min/max/median/mean and unit conversion where applicable). Supplementary Table 17 reports UKB threshold-independent performance metrics (AUROC, AUPRC) for all models and benchmarks. Supplementary Table 18 reports UKB threshold-dependent performance metrics across predefined operating points. Supplementary Table 19 reports UKB DeLong test results for AUROC comparisons between models/benchmarks. Supplementary Table 20 reports feature importances for Models A–E and TOP75–TOP15 in UKB. Supplementary Table 21 provides confusion matrices for selected thresholds in UKB and AOU (All and PAR where applicable). Supplementary Table 22 provides the AOU ICD code summary representations. Supplementary Table 23 reports AOU blood count/biochemistry summary statistics stratified by HCC status and sex. Supplementary Table 24 reports AOU threshold-independent performance metrics (AUROC, AUPRC) for all models and benchmarks. Supplementary Table 25 reports AOU threshold-dependent performance metrics across predefined operating points. Supplementary Table 26 reports AOU DeLong test results for AUROC comparisons between models/benchmarks. Supplementary Table 27 compares predicted risk distributions and related summary statistics across UKB and AOU. Supplementary Table 28 summarizes missingness (NA) patterns across features in AOU. Supplementary Table 29 provides an overview of all trained models, including modalities, feature sets, and evaluation cohorts (UKB/AOU).
- 1 July 2026
- Published by American Association for Cancer Research (AACR)
Open AccessSupplementary Tables ST1 - ST29 from Machine Learning Predicts Hepatocellular Carcinoma Risk from Routine Clinical Data: A Large Population-Based Multicentric Study
Supplementary Table 1 lists HCC case counts per UKB assessment center used in the study. Supplementary Table 2 lists the assessment-visit variables extracted and harmonized in UKB. Supplementary Table 3 reports PheWAS results used to derive disease-level features in UKB. Supplementary Table 4 lists the diagnosis-based inclusion criteria defining the PAR cohort in UKB. Supplementary Table 5 lists all single ICD code features used across UKB and AOU. Supplementary Table 6 lists grouped ICD (phecode/category) features used across UKB and AOU. Supplementary Table 7 maps self-reported conditions to ICD-based features in UKB. Supplementary Table 8 summarizes at-risk subset definitions and corresponding HCC/control sample sizes in UKB. Supplementary Table 9 provides the UKB ICD code summary representations. Supplementary Table 10 lists blood count and biochemistry features used for modeling in UKB and AOU. Supplementary Table 11 reports UKB blood count/biochemistry summary statistics stratified by HCC status and sex. Supplementary Table 12 lists genomic features (variants, coding, and availability) used in UKB modeling. Supplementary Table 13 lists metabolomics features used in UKB modeling. Supplementary Table 14 reports UKB metabolomics summary statistics stratified by HCC status. Supplementary Table 15 details the center-based train/test splits applied in UKB. Supplementary Table 16 provides UKB–AOU variable mappings and conversion summaries (min/max/median/mean and unit conversion where applicable). Supplementary Table 17 reports UKB threshold-independent performance metrics (AUROC, AUPRC) for all models and benchmarks. Supplementary Table 18 reports UKB threshold-dependent performance metrics across predefined operating points. Supplementary Table 19 reports UKB DeLong test results for AUROC comparisons between models/benchmarks. Supplementary Table 20 reports feature importances for Models A–E and TOP75–TOP15 in UKB. Supplementary Table 21 provides confusion matrices for selected thresholds in UKB and AOU (All and PAR where applicable). Supplementary Table 22 provides the AOU ICD code summary representations. Supplementary Table 23 reports AOU blood count/biochemistry summary statistics stratified by HCC status and sex. Supplementary Table 24 reports AOU threshold-independent performance metrics (AUROC, AUPRC) for all models and benchmarks. Supplementary Table 25 reports AOU threshold-dependent performance metrics across predefined operating points. Supplementary Table 26 reports AOU DeLong test results for AUROC comparisons between models/benchmarks. Supplementary Table 27 compares predicted risk distributions and related summary statistics across UKB and AOU. Supplementary Table 28 summarizes missingness (NA) patterns across features in AOU. Supplementary Table 29 provides an overview of all trained models, including modalities, feature sets, and evaluation cohorts (UKB/AOU).
- Open Access
- 1 July 2026
- Published by American Association for Cancer Research (AACR)
Supplementary Figures S1 - S12 from Machine Learning Predicts Hepatocellular Carcinoma Risk from Routine Clinical Data: A Large Population-Based Multicentric Study
Supplementary Figure S1 shows participant inclusion and exclusion flowcharts for UKB and AOU, detailing data processing and cohort assembly steps. Supplementary Figure S2 shows age distributions at first HCC diagnosis in UKB and AOU and associations between metabolic biomarkers and HCC risk in PAR. Supplementary Figure S3 shows additional model performance metrics in UKB, including ROC curves, AUC comparisons across estimators, benchmark risk scores, and prediction score distributions. Supplementary Figure S4 shows the impact of missing-data thresholds on cohort size, disease prevalence, and model performance in UKB. Supplementary Figure S5 shows comparative performance of clinical models and AFP in the UKB proteomics subcohort using ROC and PRC analyses. Supplementary Figure S6 shows feature importance rankings for Models C and E across cohorts, including top-ranked features and aggregated feature group contributions. Supplementary Figure S7 shows calibration performance of the TOP15 model in the UKB All cohort before and after Platt scaling. Supplementary Figure S8 shows calibration performance of the TOP15 model in the UKB PAR cohort before and after Platt scaling. Supplementary Figure S9 shows additional external validation metrics in AOU, including prediction score distributions and confusion matrices across thresholds and sex strata. Supplementary Figure S10 shows the relationship between recall and NNS across thresholds in UKB and AOU for All and PAR cohorts. Supplementary Figure S11 shows calibration performance of the TOP15 model in the AOU All cohort before and after Platt scaling. Supplementary Figure S12 shows calibration performance of the TOP 15 model in the AOU PAR cohort before and after Platt scaling.
- 1 July 2026
- Published by American Association for Cancer Research (AACR)
Open AccessSupplementary Figures S1 - S12 from Machine Learning Predicts Hepatocellular Carcinoma Risk from Routine Clinical Data: A Large Population-Based Multicentric Study
Supplementary Figure S1 shows participant inclusion and exclusion flowcharts for UKB and AOU, detailing data processing and cohort assembly steps. Supplementary Figure S2 shows age distributions at first HCC diagnosis in UKB and AOU and associations between metabolic biomarkers and HCC risk in PAR. Supplementary Figure S3 shows additional model performance metrics in UKB, including ROC curves, AUC comparisons across estimators, benchmark risk scores, and prediction score distributions. Supplementary Figure S4 shows the impact of missing-data thresholds on cohort size, disease prevalence, and model performance in UKB. Supplementary Figure S5 shows comparative performance of clinical models and AFP in the UKB proteomics subcohort using ROC and PRC analyses. Supplementary Figure S6 shows feature importance rankings for Models C and E across cohorts, including top-ranked features and aggregated feature group contributions. Supplementary Figure S7 shows calibration performance of the TOP15 model in the UKB All cohort before and after Platt scaling. Supplementary Figure S8 shows calibration performance of the TOP15 model in the UKB PAR cohort before and after Platt scaling. Supplementary Figure S9 shows additional external validation metrics in AOU, including prediction score distributions and confusion matrices across thresholds and sex strata. Supplementary Figure S10 shows the relationship between recall and NNS across thresholds in UKB and AOU for All and PAR cohorts. Supplementary Figure S11 shows calibration performance of the TOP15 model in the AOU All cohort before and after Platt scaling. Supplementary Figure S12 shows calibration performance of the TOP 15 model in the AOU PAR cohort before and after Platt scaling.
- Open Access
- 26 March 2026
- Published by American Association for Cancer Research (AACR) in Cancer Discovery
- Vol. 16 (7) , 1304-1322
Machine Learning Predicts Hepatocellular Carcinoma Risk from Routine Clinical Data: A Large Population-Based Multicentric Study
Hepatocellular carcinoma (HCC) is a highly fatal tumor, for which risk stratification is crucial yet remains challenging. In this study, we develop an interpretable machine learning (ML) framework for HCC risk stratification based on routinely collected clinical data. We utilize prospectively collected multimodal data from more than 900,000 individuals and 983 cases of HCC across two population-scale cohorts: the UK Biobank study (development) and the All of Us Research Program (external testing). We assess individual and cumulative contributions of data modalities, including demographics, lifestyle, health records, blood, genomics, and metabolomics. Our final random forest–based models significantly outperform all publicly available state-of-the-art risk scores on both internal and external test sets. We demonstrate robustness across ethnic subgroups, provide comprehensive interpretability, and release all code, model weights, and a web calculator for external validation and agentic integration. Our study presents PRE-Screen-HCC, a robust and interpretable ML framework for HCC risk stratification and early detection. Significance: Using data from population-scale cohorts, we develop and externally validate an ML framework for HCC risk stratification. Models trained on routine clinical data outperform published scores, perform on par with metabolomics and genomics, generalize across subgroups, and remain interpretable. See related commentary by Foda, p. 1252
- 26 March 2026
- Published by American Association for Cancer Research (AACR) in Cancer Discovery
- Vol. 16 (7) , 1304-1322
Open AccessMachine Learning Predicts Hepatocellular Carcinoma Risk from Routine Clinical Data: A Large Population-Based Multicentric Study
Hepatocellular carcinoma (HCC) is a highly fatal tumor, for which risk stratification is crucial yet remains challenging. In this study, we develop an interpretable machine learning (ML) framework for HCC risk stratification based on routinely collected clinical data. We utilize prospectively collected multimodal data from more than 900,000 individuals and 983 cases of HCC across two population-scale cohorts: the UK Biobank study (development) and the All of Us Research Program (external testing). We assess individual and cumulative contributions of data modalities, including demographics, lifestyle, health records, blood, genomics, and metabolomics. Our final random forest–based models significantly outperform all publicly available state-of-the-art risk scores on both internal and external test sets. We demonstrate robustness across ethnic subgroups, provide comprehensive interpretability, and release all code, model weights, and a web calculator for external validation and agentic integration. Our study presents PRE-Screen-HCC, a robust and interpretable ML framework for HCC risk stratification and early detection. Significance: Using data from population-scale cohorts, we develop and externally validate an ML framework for HCC risk stratification. Models trained on routine clinical data outperform published scores, perform on par with metabolomics and genomics, generalize across subgroups, and remain interpretable. See related commentary by Foda, p. 1252
- 22 May 2025
- Published by Massachusetts Medical Society in NEJM AI
- Vol. 2 (6)
Incidental Prompt Injections on Vision–Language Models in Real-Life Histopathology
- 22 May 2025
- Published by Massachusetts Medical Society in NEJM AI
- Vol. 2 (6)
Incidental Prompt Injections on Vision–Language Models in Real-Life Histopathology
- Open Access
- 25 April 2024
- Published by Wiley in Liver International
- Vol. 44 (8) , 2038-2053
Comorbidities, mortality and metabolic profile in individuals with primary biliary cholangitis—A Phenome‐Wide‐Association‐Study
Background and Aims: Primary biliary cholangitis (PBC) is a chronic, immune‐mediated liver disease that can lead to fibrosis and cirrhosis. In this cohort study, we aimed to investigate morbidity and mortality in conjunction with metabolomic changes of PBC in a UK population‐based cohort. Methods: 454 participants with PBC and 908 propensity score (age, sex, BMI, ethnicity) matched controls without liver disease were included in the study. A subset of participants with PBC and controls were analysed for their metabolomic profile. Further, PBC‐associated comorbidities were investigated by PheWAS analysis. Lastly, we assessed causes of death in individuals with PBC using a Fine and Grey competing‐risks regression model. Results: Compared to the control group, various pathways associated with the metabolism of amino acids, lipids, and liver biochemistry were significantly enriched in individuals with PBC. We found reduced levels of S‐HDL‐cholesterol and Glycoprotein Acetyls in individuals with PBC as well as an association with diseases of the circulatory system. Notably, PBC individuals had a higher prevalence of digestive diseases, autoimmune diseases, cardiovascular diseases, anaemias, mental disorders, and urinary tract infections compared to the control group. Strikingly, the overall mortality was almost three times higher in the PBC group compared to the control group, with diseases of the digestive system accounting for a significant elevation of the death rate. A subsequent analysis, enhanced by propensity score matching that included the APRI score, demonstrated that the observed morbidity could not be exclusively attributed to advanced hepatic disease. Conclusions: Our study provides a detailed perspective on the morbidity of individuals with PBC. The exploration of potential effects of disease state on morbidity suggest that early detection and early treatment of PBC could enhance patient prognosis and prevent the onset of comorbid diseases. Finally, the metabolomic alterations could represent a link between the pathophysiological processes underlying PBC development, progression, and associated morbidity.
- 25 April 2024
- Published by Wiley in Liver International
- Vol. 44 (8) , 2038-2053
Open AccessComorbidities, mortality and metabolic profile in individuals with primary biliary cholangitis—A Phenome‐Wide‐Association‐Study
Background and Aims: Primary biliary cholangitis (PBC) is a chronic, immune‐mediated liver disease that can lead to fibrosis and cirrhosis. In this cohort study, we aimed to investigate morbidity and mortality in conjunction with metabolomic changes of PBC in a UK population‐based cohort. Methods: 454 participants with PBC and 908 propensity score (age, sex, BMI, ethnicity) matched controls without liver disease were included in the study. A subset of participants with PBC and controls were analysed for their metabolomic profile. Further, PBC‐associated comorbidities were investigated by PheWAS analysis. Lastly, we assessed causes of death in individuals with PBC using a Fine and Grey competing‐risks regression model. Results: Compared to the control group, various pathways associated with the metabolism of amino acids, lipids, and liver biochemistry were significantly enriched in individuals with PBC. We found reduced levels of S‐HDL‐cholesterol and Glycoprotein Acetyls in individuals with PBC as well as an association with diseases of the circulatory system. Notably, PBC individuals had a higher prevalence of digestive diseases, autoimmune diseases, cardiovascular diseases, anaemias, mental disorders, and urinary tract infections compared to the control group. Strikingly, the overall mortality was almost three times higher in the PBC group compared to the control group, with diseases of the digestive system accounting for a significant elevation of the death rate. A subsequent analysis, enhanced by propensity score matching that included the APRI score, demonstrated that the observed morbidity could not be exclusively attributed to advanced hepatic disease. Conclusions: Our study provides a detailed perspective on the morbidity of individuals with PBC. The exploration of potential effects of disease state on morbidity suggest that early detection and early treatment of PBC could enhance patient prognosis and prevent the onset of comorbid diseases. Finally, the metabolomic alterations could represent a link between the pathophysiological processes underlying PBC development, progression, and associated morbidity.