Citations

Measurable residual disease (MRD) has established prognostic significance in patients with myeloid disorders; however, there is limited data regarding its prognostic and predictive value and the optimal technique for measurement in the setting of post allogeneic hematopoietic cell transplantation (allo-HCT). A multi-gene next generation sequencing (NGS) panel can overcome the limitations of conventional techniques for measurement of MRD and can shed light onto the longitudinal evolution and genomic complexity of patients undergoing allo-HCT for acute myeloid leukemia (AML) and myelodysplastic syndrome (MDS). In this study, patients with AML or MDS who underwent their first allo-HCT and completed NGS testing at day +100 and were in a state of morphological remission were identified from the transplant registry of Mayo Clinic in Florida. The primary aim was to evaluate the prognostic impact of somatic variants detected via NGS at day +100 post allo-HCT on long-term outcomes. In total, 105 patients who underwent allo-HCT for AML and MDS and completed NGS testing on day +100 were included in the study population. Seventeen patients were found to have detectable variants, and 88 patients were without a detectable variant. At a median follow up of 1.4 years (0.1–5.5), presence of any variant at day +100 was associated with inferior 4-year overall survival (OS) (32% vs. 69%, P < 0.001), decreased progression free survival (PFS) (34% vs. 61%, P < 0.001), and a higher incidence of relapse (64% vs. 23%, P < 0.001). In univariate models, a lower Karnofsky performance score and presence of variants post allo-HCT remained significantly associated with inferior OS, higher relapse incidence, and a decreased PFS. The presence of variants at day +100 post allo-HCT remained significant in multivariate models for relapse ( P = 0.002) but not for OS ( P = 0.06) and PFS ( P = 0.08). These results indicate that the presence of detectable somatic variants at day +100 is associated with poor long-term outcomes and are predictive of a higher incidence of relapse.

Abstract Background Cardiomegaly is a clinically significant incidental finding on chest computed tomography (CT) associated with heart failure, arrhythmias, and sudden cardiac death. Qualitative radiologist assessment is variable, and automated AI tools may enable objective opportunistic cardiac volumetry. Purpose To evaluate whether AI-enabled total cardiac volume (TCVAI) derived from non-ECG-gated, non-contrast chest CT can identify cardiomegaly as defined by echocardiography. Materials and Methods This retrospective study included 307 consecutive patients (median age, 67 years; 56% male) who underwent non-contrast chest CT at a single center on 7 scanner types (4 vendors) and clinically indicated echocardiography within 31 days. A commercially available AI tool (AI-Rad Companion, Siemens Healthineers) automatically quantified TCVAI, indexed to body surface area (TCVAI/BSA). Echocardiography reports were reviewed for chamber dilation and left ventricular hypertrophy (LVH), collectively defined as cardiomegaly. Associations between TCVAI/BSA and echocardiographic findings were assessed using correlation, ordinal regression, and receiver operating characteristic (ROC). Interscan repeatability was evaluated in 248 patients with 544 repeat CT examinations. Prespecified sex-specific thresholds were tested in a temporally independent validation cohort of 50 patients. Results Median TCVAI was higher in patients with cardiomegaly than those without (1061.9 vs 798.4 mL; P &lt; .001). TCVAI/BSA was associated with chamber dilation and LVH severity on univariate analysis and remained associated in multivariable ordinal models, except for right ventricular dilation. Discriminatory performance was fair to good, with area under the curve (AUC) 0.81 (95% CI, 0.75-0.87) in men and 0.77 (95% CI, 0.69-0.85) in women. Interscan repeatability was excellent (intraclass correlation coefficient [ICC]: 0.93). In independent validation, performance ranged from sensitivity 89.3%/specificity 27.3% at a high-sensitivity threshold to sensitivity 28.6%/specificity 100% at a high-specificity threshold. Conclusion AI-derived cardiac volume from routine chest CT shows fair to good performance for identifying echocardiography-defined cardiomegaly with high measurement repeatability, supporting a potential role for automated cardiac volumetry as an objective, opportunistic biomarker.

Neuromodulation is becoming a common treatment strategy for patients with drug-resistant epilepsy who are not candidates for resective surgery. Responsive neurostimulation can effectively reduce seizure burden, however it remains unclear which neural signals are most important for triggered stimulation. Furthermore, the high-dimensional parameter space remains systematically underexamined. As more patients become candidates for neurostimulation, clinical teams will need to decide on the optimal site and stimulation program for therapeutic neuromodulation. In this review, we highlight how recent insight into the biological rhythms of epilepsy may pair with personalization protocols in the emerging field of neuropsychiatric neuromodulation. Finally, we discuss future directions for epilepsy neuromodulation, specifically how machine learning can offer clinical teams actionable feedback on neuromodulation efficacy.

Abstract Individuals who have serum elevations of anti-cyclic citrullinated protein (anti-CCP) antibodies are at risk for developing rheumatoid arthritis (RA), yet immunologic factors that lead to a transition from pre- to clinical RA remain unclear. Here, we used materials from anti-CCP antibody-positive individuals enrolled in a clinical trial that evaluated the efficacy of hydroxychloroquine to prevent clinical RA, and performed multi-modal single-cell profiling (transcriptome, surface proteins, T/B-cell receptor sequencing, and chromatin accessibility) on samples obtained at baseline and at RA onset in those who developed clinical RA (Converters) or follow-up point in matched Nonconverters. At both baseline and follow-up, Converters had expansions of peripheral helper T (Tph) cells and CD8 + T cells expressing GZMK and GZMB , along with elevated potentially autoreactive T-cell receptors in CD4 + T cells compared to Nonconverters. Induction of age-associated B cell signatures was observed in B cells of Converters prior to RA onset. Epigenetic profiling further identified chromatin accessibility changes in Converters over time, particularly within myeloid and NK cells. Lastly, predictive modeling using baseline immune features, including Tph cells, GZMK + XCL1 + CD8 + , and GZMB + CD57 + CD8 + T cells, together with clinical features such as anti-CCP3 levels, RF-positivity, and HLA shared epitope status, stratified RA risk and predicted time to onset. These findings define immune endotypes in pre-RA that could serve as targets for future preventive interventions and be used to stratify the risk of developing clinical RA in anti-CCP antibody-positive individuals.

Abstract Background Integrating advanced artificial intelligence (AI) into clinical decision-support often requires the sharing of sensitive patient data with external services, raising privacy concerns. Homomorphic encryption (HE) allows computing directly on encrypted data, without revealing the underlying patient information. Objectives To develop a large language model (LLM)-assisted diagnosis framework while preserving patient privacy in the clinical text analysis, by leveraging HE and using rare disease (RD) diagnosis as a representative application. To demonstrate HE does not hinder the system performance. Materials and Methods Texts from patient histories and a RD knowledge base were embedded by LLMs into vectors, then encrypted using HE to obscure private information while retaining the semantic nuances. Diagnostic recommendations were generated by computing and ranking the similarities between the patient history and RD vectors in the encrypted space. The system was evaluated using 50 synthetic case reports (5 RDs, each with 10 reports). Results Applying HE did protect private information from reverse-embedding attacks. HE imposed little disruption to the diagnostic accuracy, with normalized discounted cumulative gains (nDCG) of 0.6108 ± 0.3412 (encrypted) versus 0.6083 ± 0.3415 (unencrypted). The accuracy and computational performance were tunable and consistent, as demonstrated across five different LLMs. Discussion Our privacy-preserving framework opens tremendous opportunities toward hosting and serving powerful AI solutions across institution boundaries, which would remove the need for local deidentification and incentivize users to access secure external decision-support services. Conclusions Integrating HE with LLM retrieval can promote the dissemination of nonredundant, high-capacity AI services by preserving both privacy and accuracy.

566 Background: FGFR inhibitors (FGFRi) have an established role in treating FGFR2-altered (alt) cholangiocarcinoma (CCA). Tinengotinib, a novel FGFRi, has shown promising activity overcoming acquired resistance to prior FGFRi. We present biomarker analyses evaluating tinengotinib’s ability to overcome prior FGFRi resistance and to elucidate mechanism of resistance to tinengotinib. Methods: Eligible patients (pts) with advanced/metastatic CCA who had received ≥ 1 prior systemic chemotherapy were treated with tinengotinib 10 mg daily across 4 Cohorts: A1: FGFR2 fusion (fus) with primary progressive disease (PD) on prior FGFRi; A2: FGFR2 fus with PD after prior response to FGFRi; B: FGFR alt without FGFR2 fus; C: FGFR wild-type. Efficacy was assessed by RECIST v1.1. All pts underwent liquid biopsy for circulating tumor DNA (ctDNA) at baseline; additional timepoints include cycle 3 day 1 (C3D1), time of PD, and end of treatment (EOT). Samples were analyzed using FoundationOne Liquid panel. Results: 27 out of 55 enrolled pts had biomarker samples collected at baseline and C3D1. Among them, pts in A1 (n=10, 107 mutations (mut)), B (n=8, 90 mut) and C (n=5, 74 mut) showed a significant decrease in short variant allele frequencies (VAF) from baseline to C3D1 by a median relative VAF reduction of 100% (p<0.0001) for each of the three cohorts. Although a median relative VAF reduction of 33.3% was seen in A2 (n=4, 15 mut), this was not significant (p=0.6093) likely due to low baseline mutation burden. A ≥50% reduction in ctDNA was associated with improved progression free survival (p<0.0014). Additionally, 100% of median relative reductions in FGFR2 kinase domain (KD) mut were seen in A1 and B (p<0.0001, 95% CI 100-100% and p=0.0156, 95% CI 83.6-100%, respectively), with 92% and 57% clearance of initial FGFR2 KD mut by C3D1. In a subgroup of 10 pts with biomarker sampling at baseline and EOT, 3 pts in A1 with acquired resistance showed emergence of novel FGFR2 mut, loss of FGFR-sensitive mut and new mut in alternate pathways (e.g., AKT, EGFR, RAS/MAPK) while retaining the original FGFR2 fus. 1 pt from B developed resistance due to new FGFR2 rearrangements and FGFR2 mut. 4 FGFR wild-type pts from C developed resistance involving mut in TP53, KRAS, and other oncogenic drivers. Conclusions: These findings provide genomic evidence that tinengotinib may overcome resistance to prior FGFRi treatment in FGFR2 fus-positive CCA, as reflected by significantly reductions in ctDNA VAFs and frequent clearance of FGFR2 KD mutations. The emergence of resistance alterations including de novo FGFR2 mutations and bypass pathways activations indicates acquired resistance that may limit long-term benefit. These results support further investigation with expanded biomarker sampling to better characterize resistance evolution and guide therapy optimization in an ongoing Phase 3 trial. Clinical trial information: NCT04919642 .

Objectives: To identify factors associated with short-term mortality among patients receiving microaxial flow pump (mAFP) therapy for acute myocardial infarction-related cardiogenic shock (AMI-CS). Data Sources: We searched four databases (MEDLINE, Embase, CENTRAL, and Scopus) from January 1, 2004, to January 1, 2025. Study Selection: We selected English-language studies that included adults with AMI-CS receiving mAFP and evaluated factors associated with short-term mortality. We excluded patients receiving concurrent venoarterial extracorporeal membrane oxygenation, as well as studies that solely included patients presenting with out-of-hospital cardiac arrest. Data Extraction: Two authors performed citation screening and data extraction. For each factor evaluated in at least two studies, we performed meta-analyses of adjusted odds ratios (aORs) using a random-effects model. Risk of bias was evaluated using the Quality in Prognosis Studies tool, and the certainty of evidence was evaluated using Grading of Recommendations, Assessment, Development, and Evaluations methodology. Data Synthesis: Our primary analysis included 18 studies, encompassing 20,617 patients. Median short-term mortality across studies was 50.7% (interquartile range 38.4–55.3%). Factors associated with short-term mortality based on high-certainty evidence included: increased age (aOR, 1.04 per year [95% CI, 1.03–1.05 per year] or ≥ 65 yr (aOR, 2.42 yr [95% CI, 0.77–7.64 yr]), female sex (aOR, 1.26 [95% CI, 1.09–1.45]), higher body mass index (aOR, 1.05 per point [95% CI, 1.04–1.07 per point]), higher heart rate (aOR, 1.02 per beats/min [95% CI, 1.01–1.02 per beats/min]), higher serum creatinine (aOR, 1.35 per mg/dL [95% CI, 1.08–1.70 per mg/dL]), mechanical ventilation (aOR, 2.53 [95% CI, 1.82–3.53]), vasopressors (aOR, 1.52 [95% CI, 1.11–2.08] for any vasopressors and aOR, 1.37 [95% CI, 1.18–1.58] per each vasopressor), presentation with ST-elevation myocardial infarction (aOR, 1.59 [95% CI, 1.11–2.26]), cardiac arrest (aOR, 2.85 [95% CI, 2.22–3.64]), and hypoxic-ischemic brain injury (aOR, 5.36 [95% CI, 3.03–9.47]). Conclusions: We identified several prognostic factors associated with short-term mortality in AMI-CS patients receiving mAFP support. This work may help inform clinicians, patients, and families regarding utilization of mAFP in AMI-CS.

Abstract TP53 and PPM1D are key regulators of DNA damage response and repair, and somatic mutations in these genes often co-occur in hematopoietic cells, expanding under genotoxic stress. Unlike TP53 mutations, where mechanisms of progression are defined, pathways underlying clonal fitness and transformation in PPM1D mutant cells remain unclear. In collaboration with 5 academic institutions, we analyzed the clinical and molecular landscape of 337 patients with clonal hematopoiesis (CH) and clonal cytopenia of undetermined significance (CCUS) across 4 genotypes: PPM1Dmt/TP53wt (n = 170 [50%]), PPM1Dmt/TP53mt (n = 25 [7%]), TP53mt/PPM1Dwt (n = 17 [5%]), and TP53wt/PPM1Dwt (n = 125 [38%]). All PPM1D variants were truncating, located in exon 6 of the gene, with a median variant allele frequency (VAF) of 6% (range, 0.3%-64%). The PPM1Dmt/TP53mt genotype was most frequently encountered in therapy-related CH/CCUS (t-CH/t-CCUS; 80%, 66.5%, 76.5%, and 19%; P ≤ .001) and had a shorter time interval to detection from last genotoxic exposure (6.2, 5.9, 11.25, and 24.5 months; P ≤ .001) compared with PPM1Dmt/TP53wt, TP53mt/PPM1Dwt, and TP53wt/PPM1Dwt genotypes, respectively. Acknowledging the short follow-up duration, rates of malignant transformation were lower in the PPM1Dmt/TP53wt (2%) and PPM1Dmt/TP53mt (4%) groups compared with PPM1Dwt/TP53wt (12%) and PPM1Dwt/TP53mt (17%) groups (P ≤ .001), respectively. In summary, PPM1D mutations are frequently observed in t-CH/t-CCUS, with low median VAFs, and are associated with low rates of progression, even when comutated with TP53.

Abstract Background Self‐administered and remote digital cognitive assessments provide a scalable method for screening and longitudinal cognitive monitoring. Mayo Test Drive (MTD) is one such assessment with previously demonstrated usability, reliability, validity and multi‐device compatibility. MTD includes a computer‐adaptive word list memory test (Stricker Learning Span) and a measure of processing speed/executive functioning (Symbols Test) that combine to provide an MTD Composite. This study aimed to develop a conditional normative model predicting longitudinal cognitive change for MTD, and to validate the model in two independent samples of cognitively unimpaired (CU) and mild cognitive impairment (MCI)/dementia (DEM) participants. Method 1355 CU participants at initial MTD assessment with 2‐4 consecutive MTD assessments every 7.5 months comprised the development dataset. Linear mixed effects models were used to construct a conditional normative model for MTD assessments 2‐4. The baseline model was chosen a priori , consisting of age, sex, education, baseline score, number of assessments, and time. Device type, race, and baseline score by time interactions were also considered. Variables were only added if a >1% increment in r‐squared was observed. The models were externally validated in 113 CU and 51 MCI/DEM patients. Following model development, the data were updated and additional follow‐up MTD assessments were added for many participants, including the addition of additional CU validation data at test 3 ( N = 126) and 4 ( N = 279). T‐tests compared observed and predicted differences and residual z‐scores between development and validation at each follow‐up assessment. Result Table 1 presents demographic variables. No additional variables increased the r‐squared by >1%. The final models are presented in Table 2. Age, baseline score, sex, and education were significant. Residual z‐scores were not significantly different between CU development and CU validation cohorts, indicating observed scores were in line with model predicted scores. Residual z‐scores were significantly worse among MCI/DEM compared to CU development in 11/12 comparisons (Figure 1), indicating observed scores were below model predicted expectations and implying sensitivity to longitudinal cognitive decline. Conclusion These longitudinal normative models can be used to identify individuals with follow‐up scores outside of expected predicted ranges, signifying abnormal longitudinal cognitive decline.

Abstract Background Neuropathologic evaluation is essential for diagnosing and treating dementia, but definitive diagnoses is only available after post‐mortem brain autopsy. Positron emission tomography (PET) offers a potential in vivo proxy for predicting neuropathology. This study proposes a semi‐supervised learning (SSL) model using a single FDG‐PET image to predict underlying neuropathology. Method The SSL model leverages a self‐organized graph structure based on a patient similarity network (PSN) created from 1,495 FDG‐PET images across the aging and neurodegenerative spectrum. Each patient is represented by a node, with edges indicating image similarity derived from principal component analysis features and cosine similarity weights. The PSN structure connects 20 nearest neighbors. Neuropathologic labels were available for 204 autopsied patients. The Poisson Learning Method was employed to propagate labels from these annotated nodes to unlabeled nodes, guided by the PSN structure. FDG‐PET images were collected from clinical and research participants at the Mayo Clinic and processed using an MRI‐free pipeline. Neuropathologic labels were assigned based on widely accepted diagnostic criteria. Leave‐one‐out cross‐validation (LOOCV) was used to assess model performance. Results The confusion matrix demonstrated high accuracy in predicting Alzheimer's disease, frontotemporal lobar degeneration with tau inclusions, and Lewy body disease pathology. A Sankey diagram and label distribution analysis revealed strong correlations between clinical and predicted neuropathologic labels. Conclusion This study highlights the potential of a graph‐based SSL pipeline to predict neuropathological changes using FDG‐PET imaging data, despite limited pathology labels. The findings underscore the value of SSL in bridging the gap between abundant neuroimaging data and scarce neuropathologic labels, contributing to improved dementia diagnosis and understanding.