Citations

Retrieval-augmented generation (RAG) is rapidly emerging as a transformative paradigm for large language models (LLMs), especially in high-stakes domains like oncology that demand precision, factual grounding, and up-to-date knowledge. By pairing LLMs with external knowledge repositories, RAG systems explicitly ground model outputs in relevant retrieved documents, helping to reduce hallucinations and ensure responses reflect current evidence. In oncology, where clinical knowledge evolves continually with new research and drug approvals, RAG offers a way to integrate the latest data (e.g., trial results, guidelines, genomic databases) into decision-making. This review synthesizes the technical foundations of RAG, including its architecture and key components, and examines current applications in oncology such as clinical decision support, patient education, radiology reporting, pathology analysis, and genomics-driven precision medicine. We highlight recent studies that demonstrate RAG’s potential—for instance, improving treatment recommendations by incorporating genetic profiles and literature, and enhancing diagnostic accuracy by integrating guidelines. We also discuss emerging developments like multimodal RAG (combining text with imaging or other data), ensemble model approaches, and new explainability tools that trace model outputs to sources. Finally, we critically analyze the limitations and challenges of deploying RAG in healthcare, including computational costs, retrieval errors, noise or conflicts in retrieved information, and ethical and regulatory considerations. While RAG-based systems show promise in augmenting oncologists’ expertise with timely knowledge, careful implementation, high-quality curation of knowledge bases, and human oversight will be crucial for safe and effective adoption in clinical practice.

Abstract Background The IEC exposure index (EI), deviation index (DI), and target exposure index (EI T ), represent critical standardized metrics for the evaluation of exposure and quality in radiographic imaging. Purpose This work develops and validates a systematic procedure to estimate the EI T for eight of the most common radiography imaging protocols utilizing automatic exposure control (AEC) from measurements acquired under reference conditions. Methods A model was developed to define the relationship between a systems AEC logic, and an estimation of the EI under flat‐field conditions (EIFFC). Separately, clinical data and acquisition protocol information for resultant EI during patient studies were also collected for the eight protocols studied: Chest posteroanterior (PA), Chest lateral, Abdomen anteroposterior (AP), Pelvis AP, L‐Spine AP, C‐Spine AP, T‐Spine AP, and Ribs AP. Data were collected from 41 x‐ray units spanning seven institutions. For each protocol on each unit the EIFFC was computed based on the acquisition protocol, as well as median EI from clinical exams to produce a scaling factor (SF). Kruskal–Wallis statistical tests were used to compare SF's between vendors and AEC cell configurations. Results SFs for eight radiographic imaging protocols have been produced per vendor and per AEC cell selection. A workflow has been established for end‐users to follow to apply these SFs to flat‐field measurements taken at their own locations to establish local EI T . Conclusions The study results show that in seven of the eight imaging protocols, the SFs for most units included in the study report SFs within 1 DI (± 25%) of their respective final vendor reported SF (218/228). The ribs protocol is the exception to this finding (n = 26). SFs have high utility for establishing EI T values on individual x‐ray units and normalizing EI value distributions for quality assurance purposes.

Background Failure of conventional imaging to detect pancreatic ductal adenocarcinoma (PDA) at its visually occult pre-diagnostic stage is a primary barrier to improving its otherwise poor rate of survival. Objective To develop and validate the Radiomics-based Early Detection MODel (REDMOD), an AI framework to identify subvisual radiomic signatures of pre-diagnostic PDA on standard-of-care CT. Designs REDMOD was trained on a multi-institutional cohort (n=969; 156 pre-diagnostic, 813 control) and tested on an independent set (n=493; 63 pre-diagnostic, 430 control), simulating a low prevalence (~1:6) early detection paradigm. The fully automated framework couples AI-driven segmentation with a heterogeneous ensemble architecture trained on a 40-feature radiomic signature derived from Synthetic Minority Over-sampling Technique (SMOTE)-balanced data. A tunable Youden Index-optimised classification threshold enables performance calibration without retraining. Validation included direct comparison with radiologists, longitudinal test–retest analysis and external specificity validation across two independent cohorts (n=539 and n=80). Results On an independent test set (n=493), REDMOD identified occult PDA (AUC 0.82; 73.0% sensitivity) at a median 475-day lead time. This represented nearly twofold higher sensitivity than radiologists (38.9%; p<0.001), which grew to nearly threefold (68.0% vs 23.0%) at >24 months lead time. REDMOD showed strong longitudinal stability (90–92% concordance) and generalisable specificity across multi-institutional (81.3%; n=539) and public (87.5%; n=80) datasets. Mechanistic analyses confirmed predictive power derived principally from multi-scale wavelet-filtered textural features (90% of selected signature), which outperformed unfiltered features (AUC 0.82 vs 0.74; p=0.007) in capturing subvisual architectural disruptions. Conclusions REDMOD is an automated, mechanistically grounded, longitudinally stable, externally validated AI that surpasses radiologists for PDA detection at its visually occult pre-diagnostic stage. These attributes position it for prospective validation in high-risk cohorts, a necessary step towards shifting the paradigm from late-stage symptomatic diagnosis to proactive pre-clinical interception.

ABSTRACT Aims To evaluate the real‐world associations between semaglutide 2.4 mg and cardiometabolic comorbidities, biomarkers and cardiovascular risk among adults with overweight or obesity. Methods This retrospective cohort study used US claims data and laboratory measurements from the Komodo Research Database (2016–2024). Adults with obesity or with overweight and ≥ 1 obesity‐related comorbidity were included. Patients on semaglutide 2.4 mg for ≥ 12 months were compared with non‐users (including users of other obesity medications) using propensity score weighting with residual imbalances adjusted in regression model. Outcomes included time to onset of cardiometabolic comorbidities, change in cardiometabolic biomarkers and changes in 10‐year ASCVD risk score at 12 months. Results 25 783 semaglutide 2.4 mg users and 8 948 913 non‐users were identified. Semaglutide 2.4 mg was associated with significantly lower risks for incident prediabetes or type 2 diabetes, type 2 diabetes only, hypertension, ASCVD, heart failure and CVD (HRs 0.45–0.75; all p < 0.05), as well as greater improvements in BMI (−4.1 kg/m 2 ), SBP (−6.2 mmHg), HbA1c (−0.4%), LDL cholesterol (−7.7 mg/dL), triglycerides (−28.6 mg/dL) and HDL cholesterol (+1.9 mg/dL; all p < 0.001). At 12 months, semaglutide 2.4 mg users ( N = 161) demonstrated a greater reduction in ASCVD risk score (−0.7% vs. +0.3%; between‐group difference −1.0%, p < 0.001), with a decrease in the proportion at intermediate‐high ASCVD risk (14.9%–9.3% vs. 17.7%–18.7%; OR 0.3 [95% CI: 0.1, 0.6], p < 0.001). Conclusions Semaglutide 2.4 mg was associated with lower risk of multiple cardiometabolic comorbidities, favourable changes in biomarkers, and lower estimated ASCVD risk in this population.

Abstract Aims Chronic kidney disease (CKD) is common in patients with heart failure (HF) with preserved ejection fraction (HFpEF), and its presence is associated with more severe echocardiographic abnormalities and poorer outcomes through unclear mechanisms. CKD-associated endothelial dysfunction, inflammation, and activation of profibrotic pathways could worsen pulmonary vascular disease (PVD) in HFpEF, but such relationships have not been explored. Methods Consecutively evaluated patients with HFpEF undergoing invasive haemodynamic cardiopulmonary exercise testing were stratified by baseline kidney function to characterize potential differences in pulmonary vascular loading, haemodynamics, cardiopulmonary reserve, and outcomes based upon the presence and severity of kidney dysfunction. Results Of 925 patients with HFpEF, 319 (34.5%) had estimated glomerular filtration rate (eGFR) &lt; 60 ml/min/1.73 m2. Patients with more severe kidney dysfunction were older and more likely to have diabetes, atrial fibrillation, and hypertension. At rest, reduced kidney function was associated with lower haemoglobin, higher biventricular filling pressures, and lower cardiac output, but the severity of kidney dysfunction was most conspicuously associated with worsening PVD, evidenced by increasing pulmonary arterial (PA) pressures due to marked increases in pulmonary vascular resistance (PVR) as kidney dysfunction progressed (median [IQR] 1.3 [0.86, 1.88] to 2.8 [2.0, 3.8] WU from eGFR ≥ 90 to eGFR &lt; 30, P &lt; .001), along with progressively lower PA compliance (PAC, 4.7 [3.6, 5.8] to 2.2 [1.8, 3.4] ml/mmHg from eGFR ≥ 90 to eGFR &lt; 30, P &lt; .001). With exercise, differences in the severity of PVD became even more striking, with more severe elevation in PA pressures and PVR, and lower PA compliance, leading to lower transmural left-sided distending pressures, with no difference in exertional PA wedge pressure. Patients with worse kidney dysfunction displayed the most dramatic cardiac output limitations with exercise, which along with more severe anaemia and markedly reduced O2 delivery, caused marked impairment in aerobic capacity. Worsening kidney function was associated with a striking gradient of increased risk of a composite of all-cause death and HF hospitalization [ranging from HR 2.38 (95% CI 1.01–5.59) for eGFR 60–90 to HR 16.77 (95% CI 6.65–42.28) for eGFR &lt; 30 (vs. reference eGFR ≥ 90)]. Conclusions Kidney dysfunction in patients with HFpEF is characterized by more severe PVD at rest and with exercise, leading to reduced cardiac output reserve, impaired exercise capacity, and increased risk for adverse events. Further study is warranted to better understand causal relationships between CKD and PVD, and determine whether novel therapies to improve kidney function might target these impairments to improve clinical status.

Abstract Solitary bone plasmacytoma (SBP) and SBP with minimal marrow involvement (SBPmm) are treated similarly with definitive radiation therapy (RT) followed by observation. However, differences in the natural disease course after RT between the two, are not well described. We performed a multicenter study assessing outcomes after RT in 141 patients with SBP and 102 with SBPmm diagnosed between January 2010 and December 2024. About 92% of patients with SBPmm progressed to multiple myeloma (MM) vs 76% of those with SBP (P = .01). Among patients with SBP, 21% progressed with a second solitary bone lesion vs 4% with SBPmm (P&lt; .01). Patients with longer time to progression (TTP) after initial RT could achieve durable responses with RT to a second solitary plasmacytoma in the absence of other myeloma-defining events (MDE). Multifocal lytic bone lesions were the most common MDE, occurring in 89% and 81% of patients with SBPmm and SBP, respectively. Patients with SBPmm had a median TTP of 15 months (95% confidence interval [CI], 14-21) compared to 43 months (95% CI, 34-55) for those with SBP (P&lt; .0001); 67% with SBPmm progressed within 2 years of diagnosis. Del(17p) and 1q+ (gain or amplification) were associated with shorter TTP to MM in SBPmm, with 2-year progression rates of 100% and 82%, respectively. We report significant differences in risk and modes of progression between SBP and SBPmm. Patients with SBPmm are at particularly high risk for early progression to MM, with most developing bone disease as the MDE. These findings support prospective studies incorporating frontline systemic therapy for SBPmm.

ABSTRACT Immediate early genes (IEGs) encode master transcription factors, including EGR1 , FOS , JUN , and MYC genes that drive robust transcription in the G1 phase of cell cycle. Our previous studies indicate that topoisomerase IIβ (TOP2B) critically regulates IEG transcription and that the activities of TOP2B are dynamically controlled through post-translational modifications by BRCA1-BARD1 and ERKs. Here, we show that two E2 enzymes, UBCH5b and UBC13/MMS2, differentiate the effects and functions of TOP2B ubiquitination by BRCA1-BARD1. Comprehensive transcriptomics, proteomics, and biochemical and molecular cellular analyses revealed a close relationship between BARD1 and key mitogen-activated protein kinase pathway genes and identified activated ERK2 as a novel kinase that phosphorylates BARD1 at S391, a previously reported mitotic phosphorylation site, whose genetic mutation has been linked to tumorigenesis. Mechanistically, the catalytic activity of ERK2 stimulates TOP2B ubiquitination mediated by BRCA1-BARD1 in complex with UBCH5b and UBC13/MMS2, which controls the binding and function of TOP2B and BARD1 for transcriptional activation at representative IEGs. Taken together, our data propose that there is a functional regulatory circuit involving TOP2B, BARD1, and ERK2, three key transcriptional activators for IEG transcription, in which the gene association and catalytic activity of TOP2B are regulated through E2-differentiated ubiquitination by BRCA1-BARD1 and the phosphorylation of BARD1 by ERK2 for productive transcription.

ABSTRACT Current laboratory testing for Lyme disease (LD) relies on serology. We evaluated the performance of standard two-tiered testing (STTT) and modified two-tiered testing (MTTT) algorithms using samples obtained from well-characterized patients with early LD in the U.S. East Coast and Upper Midwest. Participants with signs and symptoms of early LD (cases) and controls were enrolled by Lyme Disease Biobank. We compared the performance of four FDA-cleared STTT or MTTT algorithms using serum samples from 251 participants (107 cases, 69 with a convalescent draw; 144 endemic controls). At the initial blood draw, algorithm sensitivity ranged from 22% to 36%, with specificity ranging from 98% to 100%. MTTT algorithms showed higher sensitivity compared with STTT algorithms ( P ≤ 0.05). One STTT algorithm was less sensitive than the other ( P = 0.035), and there was no significant difference in sensitivity between MTTT algorithms. There was also discordance between algorithms; only 22 of the 45 samples classified as laboratory confirmed by Lyme Disease Biobank testing were positive using all algorithms evaluated. Likelihood of positive two-tiered serology among cases with a suspected erythema migrans (EM) skin lesion increased with longer lesion duration and/or when presenting with >1 constitutional symptom. Participants with suspected EM without constitutional symptoms were unlikely to test positive by any algorithm evaluated. Testing convalescent samples did not improve LD detection, and seroconversion was rare. While MTTT confirmed more early LD cases than STTT, all two-tiered algorithms evaluated were insensitive in this population. Novel diagnostics that improve laboratory confirmation for early LD are urgently needed. IMPORTANCE This study confirms that two-tiered serologic testing algorithms are insensitive in early Lyme disease, particularly within the first 2 weeks after symptom onset or when erythema migrans is not accompanied by constitutional symptoms. It also demonstrates that seroconversion is rare after antibiotic treatment. These results highlight the need for novel diagnostics for early Lyme disease that do not rely on serologic testing.

Abstract Alzheimer’s disease (AD) emerges from multi-scale interactions between molecular pathology and disruptions in large-scale brain network dynamics. Understanding how these processes co-evolve and relate to disease stages is essential for advancing complex systems models of aging and AD, and for developing system-informed interventions. However, progress has been limited by a lack of large-scale longitudinal data. To address this, we examined the longitudinal relationship between subsystems of the default mode network (DMN) (posterior DMN, ventral DMN, anterior dorsal DMN) using task-free functional MRI (fMRI) and amyloid positron emission tomography (PET) imaging in a large longitudinal cohort spanning the clinico-biological spectrum of AD (n = 1,451; 2,763 time points) using mixed-effect models. We also assessed whether patterns of DMN connectivity predicted conversion to amyloid positivity, mild cognitive impairment (MCI), and dementia using Cox proportional hazards models. Our findings reveal a dynamic interplay between amyloid accumulation and connectivity within and between DMN subsystems, with both hyper- and hypoconnectivity emerging across DMN subsystems in association with increasing amyloid burden. Importantly, survival models showed that DMN connectivity patterns predicted conversion to critical stages of the disease, including not only conversion to MCI and dementia, but also conversion to amyloid positivity in otherwise clinically unimpaired individuals who were amyloid negative at baseline. These associations were independent of age, APOE4 status, sex, education, and in-scanner motion. These results support a model in which breakdowns in tightly regulated feedback loops governing DMN physiology represent a core systems-level pathophysiology of AD. Notably, this functional dyshomeostasis precedes detectable amyloidosis on imaging. Future studies should focus on the development of robust biomarkers of brain function that can be applied at the individual level, which could in turn help support the development of therapeutic approaches targeting system-level pathophysiology.