Publications by Year: 2025

BACKGROUND: Right heart catheterisation (RHC) with pulmonary capillary wedge pressure (PCWP) assessment is the reference standard for diagnosis of heart failure with preserved ejection fraction (HFpEF), remains however largely underused. Different approaches for non-invasive PCWP calculation have been proposed. However, as left atrial strain (LA Es) and volume index (ESVi) emerge as a key-criteria in HFpEF, we sought to investigate them for PCWP calculation.

METHODS: The derivation population consisted of patients referred to RHC and cardiovascular magnetic resonance (CMR) imaging who were enrolled in a prospective monocentric registry. Patients were classified by RHC according to current guideline recommendations. The external validation population consisted of patients included in the HFpEF-Stress trial who underwent exercise-stress RHC and CMR with follow-up after 4 years for hospitalised cardiovascular events. Performance of strain-derived PCWP was compared to a published LA volume (LAV) and LV mass (LVM) derived method.

RESULTS: The derivation population consisted of n=209 patients, n=123 underwent exercise-stress RHC (n=55 without PH, n=72 pre-capillary, n=27 combined post- and pre-capillary pulmonary hypertension (CpcPH), n=15 isolated post-capillary pulmonary hypertension (IpcPH), n=34 exercise and n=6 unclassified PH). Linear regressions models identified the following formulae for functional PCWPrest 10.304-0.095*Es+0.098*ESVi and functional PCWPstress 24.666-0.251*Es+0.056*ESVi calculation. The validation population consisted of n=74 patients (n=15 without, n=5 pre-capillary, n=8 CpcPH, n=10 IpcPH and n=32 exercise PH with n=4 remaining unclassified). Functional PCWPrest (11.8) and RHC-derived PCWPrest (11mmHg) were statistically similar (p=0.285) and showed moderate correlation (r=0.53, p<0.001). Functional PCWPrest (AUC 0.80) and PCWPstress (AUC 0.85) accurately identified HFpEF patients, were superior to LAV/LVM based PCWP (AUC 0.67, p≤0.002) and showed prognostic implications (HR 1.37 (1.16-1.62) and 1.29 (1.14-1.46), p<0.001).

CONCLUSIONS: Functional PCWP may aide in the identification of post-capillary involvement in PH and HFpEF superiorly compared to morphology-derived PCWP and shows prognostic implications.

INTRODUCTION: Visual Patient Predictive (VPP) is an AI-based extension of the Visual Patient Avatar (VPA) that integrates deep learning models to predict upcoming vital sign deviations and display them as dashed visual elements. By explicitly showing anticipated changes, the system aims to support level 3 situation awareness-the projection of future patient states. This multicentre simulation study will evaluate whether predictive algorithms and visualisations integrated into the VPA (resulting in VPP) improve clinicians' ability to anticipate critical vital sign changes compared with conventional number-based and waveform-based monitoring and examine its effects on decision-making, confidence, workload and user acceptance.

METHODS AND ANALYSIS: This investigator-initiated, randomised, within-subjects crossover, computer-based simulation trial will be conducted at five academic centres in Switzerland, Germany and the United States. Medical professionals from anaesthesiology departments will complete scenario-based prediction tasks using both VPP (as the index test) and conventional monitoring (as the reference standard) in randomised order, with the same participant evaluating both modalities and the identical underlying clinical scenario used in each condition, following video-based training and a learnability test. The primary outcome is recall (true positive rate) of vital sign deviation predictions. Secondary outcomes include average lead time, precision, prediction confidence, number and correctness of proposed interventions, perceived workload (NASA-TLX) and qualitative usability feedback. Quantitative data will be analysed using a logistic generalised linear mixed model with random intercepts for centre and participant, and a random slope for the intervention effect. Qualitative interviews will undergo thematic analysis.

ETHICS AND DISSEMINATION: The leading ethics committee (Zurich, Switzerland; BASEC-Req-2023-00465) reviewed and approved the study protocol. Ethics committees at the other participating centres have obtained their respective approvals or waivers. Bonn: 2025-144-BO, Boston: 2025P000501, Heidelberg: S-376/2025, Munich: 2025-357 W-CB. As this simulation study involves only healthcare professionals performing prediction tasks based on simulated vital sign scenarios-without collection of patient data or any medically relevant personal data-it does not constitute human subjects research under applicable regulations. Study results will be disseminated through peer-reviewed publications and presentations at scientific conferences.

BACKGROUND: Chimeric antigen receptor (CAR) T-cell therapies are highly efficacious for several different hematologic cancers. However, for most CAR T targets it is observed that low surface antigen density on tumors can significantly reduce therapeutic efficacy. In this study, we explore this dynamic in the context of CD72, a surface antigen we recently found as a promising target for refractory B-cell cancers, but for which CD72 low antigen density can lead to therapeutic resistance in preclinical models.

METHODS: Primary samples were accessed via institutional review board-approved protocols. Affinity-matured and humanized nanobody clones were previously described in Temple et al. (2023). CAR T cells were generated via lentiviral transduction. In vitro cytotoxicity assays were performed using luciferase-labeled cell lines. In vivo studies were performed using cell line-derived or patient-derived xenografts implanted in NOD scid gamma mice.

RESULTS: We first confirmed ubiquitous CD72 expression across a range of primary B-cell non-Hodgkin lymphomas. We further found that after resistance to CD19-directed therapies, across both B-cell acute lymphoblastic leukemia (B-ALL) models and primary tumor samples, surface CD72 expression was largely preserved while CD22 expression was significantly diminished. Affinity maturation of a nanobody targeting CD72, when incorporated into CAR T cells, led to more effective elimination in vitro of isogenic models of CD72 low-expressing tumors. These results suggested that nanobody-based CAR T cells (nanoCARs) may exhibit a similar relationship between binder affinity, antigen expression, and efficacy as previously demonstrated only for single chain variable fragment-based CAR T cells. Surprisingly, however, this significantly improved in vitro efficacy only translated to modest in vivo survival benefit. As a parallel strategy to enhance CAR T function, we found that the small molecule bryostatin could also significantly increase CD72 surface antigen density on B-cell malignancy models. Structural modeling and biochemical analysis identified critical residues improving CD72 antigen recognition of our lead affinity-matured nanobody.

CONCLUSIONS: Together, these findings support affinity-matured CD72 nanoCARs as a potential immunotherapy product for CD19-refractory B-cell cancers. Our results also suggest that for B-ALL in particular, CD72 may be a preferable second-line immunotherapy target over CD22.

Liver Imaging Reporting and Data System (LI-RADS) category 3 (LR-3) observations remain indeterminate and often result in repeated follow-up or biopsy. Prostate-specific membrane antigen (PSMA) is overexpressed in hepatocellular carcinoma (HCC) neovasculature and may serve as a useful imaging biomarker. This study aimed to evaluate whether [68Ga]Ga-PSMA-11 PET/MRI improved characterization of LR-3 observations in patients with cirrhosis compared with MRI alone. Methods: In this prospective study, conducted between March 2022 and June 2024, 19 patients with cirrhosis and 54 LR-3 observations identified on prior MRI underwent [68Ga]Ga-PSMA-11 PET/MRI. An observation was classified as HCC if it demonstrated focal 68Ga-PSMA uptake greater than background liver combined with at least 1 LI-RADS major or ancillary feature. The reference standard was histopathology or a follow-up MRI within 12 mo. Diagnostic metrics were calculated. Univariable logistic regression and decision tree analysis were performed to identify imaging predictors of malignancy. Results: Of the 54 LR-3 observations, 13 (24%) were confirmed as HCC and 41 (76%) as benign. [68Ga]Ga-PSMA-11 PET/MRI correctly identified 12 of 13 HCCs (sensitivity, 92%; 95% CI, 66.7-99.6) and 39 of 41 benign observations (specificity, 95%; 95% CI, 81.9-99.3). Overall diagnostic accuracy was 94%, with a positive predictive value of 86% and negative predictive value of 97%. Diagnostic performance was significantly better than MRI alone (McNemar test, P < 0.001). [68Ga]Ga-PSMA-11 uptake was the only significant imaging predictor of malignancy on univariable analysis (odds ratio, 5.7; P = 0.017). Decision tree analysis identified [68Ga]Ga-PSMA-11 uptake, observation size, and hepatobiliary phase hypointensity as principal discriminators. Conclusion: [68Ga]Ga-PSMA-11 PET/MRI demonstrates high diagnostic accuracy in differentiating malignant from benign LR-3 liver observations in patients with cirrhosis. This technique may reduce unnecessary follow-up imaging and biopsy. These results support further validation of [68Ga]Ga-PSMA-11 PET/MRI as a promising imaging approach for indeterminate liver observations.

BACKGROUND: Estimators of biological age, such as epigenetic clocks, are promising biomarkers for neurological disorders where the risk significantly increases with age, such as Parkinson's disease (PD). The purpose of this study was to prospectively investigate whether epigenetic age acceleration can predict PD risk, age at PD onset and time to phenoconversion.

METHODS: We conducted a prospective, nested case-control study within the Nurses' Health Study, including participants who provided two blood samples before being diagnosed with PD. DNA methylation profiles were obtained from 75 individuals who developed PD, 79 individuals who developed prodromal features suggestive of PD and 154 age-matched controls. We estimated epigenetic age acceleration using six different epigenetic clocks (Horvath, Hannum, PhenoAge, GrimAge, DunedinPACE and the cortical epigenetic clock) and assessed their associations with PD risk, age at PD onset and time to PD onset.

RESULTS: Epigenetic age acceleration was not consistently associated with a higher PD risk, using estimates of biological ageing neither in the first sample (collected a median of 19 years before PD onset) nor in the second sample (collected a median of 8 years before PD onset). These results remained similar in multivariable models adjusted for smoking status, physical activity, body mass index, caffeine intake, alcohol intake and Mediterranean diet score. Furthermore, epigenetic age acceleration was not associated with earlier age at PD onset or time to PD phenoconversion.

CONCLUSIONS: In our study, epigenetic clock-based biomarkers do not reliably predict PD risk, age at PD onset or time to PD phenoconversion.

Major depressive disorder (MDD) is an etiologically diverse psychiatric disease with heterogeneous manifestations, making it difficult to diagnose with conventional assessment standards. In addition, the obvious incompatibility of the standard survey-based tests for human MDD and the behavioral assays for depressive-like phenotypes in mice makes clear the requirement for a non-invasive method for quantifying the expression of depressive-like state in naturalistic contexts. Here, we introduce a self-supervised machine learning platform, CLOSER (Contrastive Learning-based Observer-free analysis of Spontaneous behavior for Ethogram Representation), to monitor the spontaneous behavior in a depressive disease model with enhanced precision, reliability, and efficiency. This framework incorporates 3D pose skeleton data and kinematic features in a unique data augmentation strategy to characterize semantic behavioral syllables with a high-quality feature space. Using CLOSER, we uncovered distinct motion profiles in chronically stressed mice across both sexes and different disease stages. Furthermore, we quantified the drug-specific recovery of psychomotor symptoms, highlighting CLOSER's discriminative power for identifying drug efficacy. In offering an artificial intelligence (AI)-driven decoding of exploratory behaviors, CLOSER proposes the standardization of depressive-like phenotype assessment in mouse models, thereby bridging preclinical and clinical diagnostics for psychiatric drug discovery.

OBJECTIVE: External saphenous vein graft stenting has been shown to reduce intimal hyperplasia, lumen irregularities, and flow disturbances after coronary artery bypass grafting (CABG). The objective of this study is to evaluate the effect of saphenous vein graft external stenting on clinical outcomes up to 5 years.

METHODS: Outcomes for patients who received external vein graft stenting in an international, real-world cohort were compared in a propensity matched analysis with patients from the Arterial Revascularization Trial (ISRCTN46552265). All eligible patients required an internal mammary artery graft to the left anterior descending coronary artery, received at least one vein graft, and survived to discharge. The primary end point was major adverse cardiovascular and cerebrovascular events at 1 year after surgery, consisting of all-cause mortality, myocardial infarction, repeat revascularization, and cerebrovascular accident. Secondary end points included 5-year major adverse cardiovascular and cerebrovascular events with and without stroke and annualized target vessel revascularization.

RESULTS: In total, 789 treated and 2205 control patients were included. At 1 year after CABG, the weighted hazard ratio comparing outcomes between treated and control patients was 0.60 (90% confidence interval, 0.38-0.94, P = .03). The benefits associated with external stenting for the composite outcome persisted through 5 years' post-CABG (hazard ratio, 0.70; 95% confidence interval, 0.51-0.98, P = .04). Annual target vessel revascularization rates in vein grafts were significantly lower in the venous external support cohort at 2 to 5 years after surgery (P = .009-.03).

CONCLUSIONS: The current study demonstrates that external vein graft stenting is associated with a significantly lower risk of experiencing adverse clinical outcomes up to 5 years after surgery compared with standard of care.

Immunological non-response (INR) to antiretroviral therapy (ART) is a critical concern for PLHIV, characterized by inadequate CD4+ T-cell recovery despite virological suppression. This retrospective study analyzed medical records of virologically suppressed adult PLHIV on ART (2004-2024) at two hospitals in Surabaya, Indonesia, using four operational categories to identify clinical and demographic determinants of INR. Patients were classified as immunological responders (IRs) or non-responders (INRs) based on four definitions: INR1 (CD4+ gain < 100 cells/mm3), INR2 (CD4+ < 350 cells/mm3), INR3 (meeting of either criterion), and INR4 (meeting of both criteria). Of 464 patients, 382 were analyzed. Baseline CD4+ < 200 cells/mm3 strongly predicted INR2 (aOR = 5.60, 95% CI: 2.95-10.62) and INR3 (aOR = 4.46, 95% CI: 2.39-8.29), while anal sexual transmission was protective against INR2 (aOR = 0.42, 95% CI: 0.19-0.92) and INR3 (aOR = 0.41, 95% CI: 0.19-0.89). By month 12, IR groups had over 350 CD4+ cells/mm3, with faster recovery slopes in months 0-6 (IR: >20 vs. INR: <10 cells/mm3/month). INR1 and INR4 had flat or negative slopes at 12-24 months, while IR groups had positive slopes. Baseline CD4+ was the strongest INR predictor, suggesting the value of early ART and individualized care for Indonesian PLHIV.

Dose prediction plays a key role in knowledge-based planning (KBP) by automatically generating patient-specific dose distribution. Recent advances in deep learning-based dose prediction methods necessitates collaboration among data contributors for improved performance. Federated learning (FL) has emerged as a solution, enabling medical centers to jointly train deep-learning models without compromising patient data privacy. We developed the FedKBP framework to evaluate the performances of centralized, federated, and individual (i.e. separated) training of dose prediction model on the 340 plans from OpenKBP dataset. To simulate FL and individual training, we divided the data into 8 training sites. To evaluate the effect of inter-site data variation on model training, we implemented two types of case distributions: 1) Independent and identically distributed (IID), where the training and validating cases were evenly divided among the 8 sites, and 2) non-IID, where some sites have more cases than others. The results show FL consistently outperforms individual training on both model optimization speed and out-of-sample testing scores, highlighting the advantage of FL over individual training. Under IID data division, FL shows comparable performance to centralized training, underscoring FL as a promising alternative to traditional pooled-data training. Under non-IID division, larger sites outperformed smaller sites by up to 19% on testing scores, confirming the need of collaboration among data owners to achieve better prediction accuracy. Meanwhile, non-IID FL showed reduced performance as compared to IID FL, posing the need for more sophisticated FL method beyond mere model averaging to handle data variation among participating sites.