Browse Articles

Predicting immune-related adverse events (irAEs) with large language model (LLM) embeddings.

1611 Background: irAEs are a common side effect of immune checkpoint inhibitor (ICI) therapy. Predicting irAEs from clinical records relies on traditional machine learning methods, which require time-consuming and extensive feature engineering such as removing features with many missing values or imputing values. Using LLM vector embeddings to train classifier models may require less feature engineering and leverage LLM understanding of semantic information. We assessed whether this LLM method could predict development of irAEs 12 weeks after initiating an ICI. Methods: We used patient (pt) data from the Immuno-Oncology Registry, a database of cancer pts from 10 DC-Baltimore-based MedStar Health network hospitals and Hackensack Meridian Health system in New Jersey. The data were divided into training (80%) and testing (20%) datasets. The training dataset was oversampled to balance the irAE class distribution. Tabular data was serialized into natural language, then passed into a LLM with original text, 0-shot, 1-shot, or few-shot prompting. Due to LLM context length limits, pt information was separated into 9 components, and each component embedding was max-pooled before concatenation. The embeddings were used to train 4 classifier models with hyperparameter tuning: XGBoost, support vector machine (SVM), logistic regression (LR), and multilayer perceptron (MLP). Five different sentence transformer or text embedding LLMs were tested. Model classification performance was evaluated by Area Under the Receiver Operating Characteristic curve (AUROC). Results: Of the 1459 pts, 585 (40%) experienced irAE(s). The most common cancers were lung cancer (34.5%) and melanoma (27.7%). ICIs included anti-CTLA-4 (11.0%), anti-PD-1 (58.9%), anti-PD-L1 (4.6%), anti-PD-1 + anti-CTLA-4 (14.0%), anti-PD-1 + chemotherapy (3.1%), anti-PD-1 + tyrosine kinase inhibitor (0.07%), and others (8.4%). The most common irAEs were skin rash (14.5%) and colitis (8.9%). Median AUROC was highest for XGBoost (0.654), followed by LR (0.641), MLP (0.627), and finally SVM (0.596). Median AUROC was highest for 0-shot (0.641), followed by original (0.636), 1-shot (0.619), and few-shot (0.516). Of the LLMs tested, intfloat/e5-base combined with XGBoost and the original prompt demonstrated the strongest performance (AUROC = 0.712, see Table), although for 0-shot and 1-shot LR was a stronger classifier. Conclusions: Using LLM embeddings to train classifier models is feasible for making predictions with tabular medical data, works best with XGBoost, and does not need prompt engineering. This method requires less feature engineering and data processing compared to standard approaches and leverages the extensive pre-training and semantic understanding of LLMs to achieve predictive power. e5-base original 0-shot 1-shot few-shot XGBoost 0.71 0.62 0.67 0.62 SVM 0.56 0.61 0.67 0.65 LR 0.64 0.66 0.69 0.64 MLP 0.63 0.65 0.65 0.59

A phase II study of trastuzumab emtansine (T-DM1) combined with ribociclib in patients with HER2-positive metastatic breast cancer (RibHER).

1041 Background: HER2-targeted therapy has significantly improved outcomes for patients with HER2-positive metastatic breast cancer (MBC). Cyclin D and cyclin-dependent kinases 4/6 (CDK4/6) act as downstream effectors of HER2 signaling, providing a rationale for combining CDK4/6 inhibitors with anti-HER2 therapies. However, combining CDK4/6 inhibitors with anti-HER2 antibody–drug conjugates (ADCs) raises the concerns of hematologic toxicities. This prospective, single-center, single-arm, phase II study evaluated the antitumor activity and safety of trastuzumab emtansine (T-DM1) plus ribociclib in patients with HER2-positive MBC previously treated with trastuzumab and a taxane, and with no prior ADC exposure. Methods: Patients received T-DM1 (3.6 mg/kg IV, day 1 of each 21-day cycle) and ribociclib (400 mg orally, days 8-21). Those with ER-positive disease could receive concomitant endocrine therapy (aromatase inhibitor or fulvestrant) per prior treatment; premenopausal patients also received ovarian function suppression. The primary endpoint was overall response rate (ORR). Secondary endpoints included clinical benefit rate (CBR), progression-free survival (PFS), overall survival (OS), and safety. Results: From August 2020 to October 2024, 12 patients were enrolled. With the subsequent availability of trastuzumab deruxtecan (T-DXd), the study was terminated prematurely due to difficulties in patient enrollment. As of December 1, 2025, the median follow-up was 22.5 months (range: 8.7-25.2). The overall ORR was 41.7% and the CBR was 58.3%. Median PFS was 5.25 months (95%CI: 3.3-NA). In subgroup analyses, ORR (28.6% vs. 60.0%), CBR (57.1% vs. 60.0%), and median PFS (6.0 months, 95% CI: 3.5-NA vs. 4.5 months, 95% CI: 3.0-NA) differed between ER-positive and ER-negative subgroups, respectively. Median OS was not reached. All patients experienced treatment-related adverse events (TRAEs), most commonly thrombocytopenia (75.0%), neutropenia (66.7%), leukopenia (66.7%), anemia (50.0%), fatigue (50.0%), and elevated aspartate aminotransferase (50.0%). Grade ≥3 TRAEs included neutropenia (33.3%), leukopenia (25.0%), and thrombocytopenia (16.7%). Thrombocytopenia led to dose reduction in 5 patients and discontinuation in 2; one patient required platelet transfusion and another experienced grade 3 intracranial hemorrhage attributed to thrombocytopenia. No treatment-related deaths occurred. Conclusions: In this small preliminary study, T-DM1 plus ribociclib achieved an ORR of 41.7% in HER2-positive MBC. No clear PFS benefit was observed regardless of ER status. Overlapping hematologic toxicities, especially thrombocytopenia, were notable and warrant consideration in future combination strategies. Clinical trial information: NCT06481956 .

DCE-MRI for assessment of pathologic complete response after completion of neoadjuvant therapy in triple-negative breast cancer patients.

611 Background: Triple-negative breast cancer (TNBC) is an aggressive molecular subtype that accounts for approximately 15%-20% of all breast cancer diagnoses. We aimed to investigate the utility of presurgical DCE breast MRI as a predictive marker for pathologic complete response (pCR) after neoadjuvant treatment (NAT) in TNBC patients and to compare the predictive value of early versus delayed enhancement for residual tumor detection. Methods: A total of 308 Stage I–III TNBC patients who underwent preoperative DCE-MRI after completion of NAT followed by surgery were enrolled in an IRB-approved prospective clinical trial (NCT02276433). Tumor size was measured using three-dimensional measurements of the index lesion during both the early (1 min) and delayed (6 min) phases of DCE-MRI. Treatment response at surgery (pCR vs. non-pCR) and the pathologic size of residual disease were documented. Correlation between pCR and residual enhancement on DCE-MRI was assessed using McNamar’s test. Spearman’s rank correlation coefficient was used to assess concordance between the longest diameter on MRI and pathology. Differences between longest diameter on DCE-MRI and pathology were compared using the Wilcoxon signed-rank test. Results: Among the 308 TNBC patients, 47% (145/308) achieved pCR following treatment. Residual disease detection on the early phase of DCE-MRI demonstrated higher sensitivity for predicting pCR compared to the delayed phase (79% vs. 69%, p < 0.001); however, it had lower specificity (78% vs. 84%, p = 0.008). Absence of enhancement in both early and delayed phase DCE-MRI predicted pCR with positive predictive values (PPV) of 80% and 83%, respectively. Residual enhancement in both phases predicted non-pCR with negative predictive values (NPV) of 77% and 71%, respectively. Both early and delayed DCE-MRI phases demonstrated a similar moderate positive correlation with pathology (r = 0.64 vs. 0.62). There was no significant difference between the longest diameter measured on early phase DCE-MRI and pathology (p = 0.706), whereas a significant difference was observed for the delayed phase (p < 0.001), which over estimated residual disease. Conclusions: Presurgical DCE-MRI demonstrated strong performance in predicting pCR among TNBC patients following NAT. The early and delayed phases of DCE-MRI may each capture different aspects of tumor characteristics, potentially providing complementary information for prediction. Clinical trial information: NCT02276433 .

Evaluating mobile app performance through sentiment analysis with SimCLR and MobileBERT

Mobile apps have significantly enhanced user engagement and convenience, providing seamless access to services anytime, anywhere. This paper presents a sentiment analysis (SA) model comprising SimCLR (Simple Contrastive Learning of Representations) and MobileBERT to measure the perceived performance of the mobile apps. The reviews of mobile apps are manually labeled as High and Low based on performance parameters (insightfulness, transparency, pertinence, precision, consistency, elaboration, and practical usability insights). These labels are aligned with attitudinal components, i.e., appreciation and judgment from appraisal theory. Appreciation concerns insightfulness, transparency, and pertinence of mobile app features, while judgment pertains to precision, consistency, elaboration, and practical usability. By integrating appraisal theory into this proposed model, a more comprehensive insight into user sentiment is achieved as compared to traditional SA models, which are heavily dependent on a very limited set of sentiment categories, i.e., positive and negative. The accuracy of manual annotation is ensured by the SHAP (SHapley Additive exPlanations) method of Explainable AI (XAI). In this study, SimCLR, a self-supervised learning framework, is employed to improve feature extraction and enhance the model’s ability to generalize across different datasets, while MobileBERT, a lightweight transformer model for sentiment classification, ensures computational efficiency and high performance. The MobileBERT makes the proposed approach scalable, efficient, and highly suitable for real-time SA in resource-limited environments. Knowledge distillation is incorporated to teach a student model that is similar to the teacher model, increasing the robustness of the system. In this work, the Augmentation of data by replacing synonyms in the data using WordNet increases the quantity of the training data to make the model more robust to linguistic variations. The proposed model is optimized using the PCA (Principal Component Analysis) method for dimensionality reduction. The Optuna method is used for hyperparameter optimization, automating the search for the most effective configuration, including learning rate, batch sizes, etc. K-fold stratified cross-validation ensures model robustness. The results demonstrated that the integration of SimCLR, MobileBERT, distillation, data augmentation techniques, and fine-tuning using Optuna provides a highly efficient and accurate SA-based model, with a mean fold accuracy and ROC AUC of 88.63% and 90.91%, respectively. This approach offers a scalable solution for NLP (Natural Language Processing) tasks. The proposed approach will be highly beneficial not only for the developers, yielding a deeper and nuanced understanding of the user feedback to identify specific areas for improvement in app performance, but also to stakeholders like marketers, product managers, and user experience (UX) researchers.

Adsorptive elimination of ofloxacin antibiotic from aqueous solution using green synthesized iron nanoparticles

Composition Restoration Enables Recycling of Mixed‐Cation, Mixed‐Halide Perovskites for Solar Cells

ABSTRACT The rapid industrial emergence of perovskite photovoltaics (PV) highlights their potential to complement silicon PV in meeting the growing global solar demand. As deployment scales, closed‐loop recycling of perovskite PV will be beneficial to conserve critical resources and mitigate environmental risks associated with lead. However, mixed‐cation, mixed‐halide perovskites—typical in record‐efficiency devices—undergo systematic composition drift during device fabrication. Consequently, material recovered from end‐of‐life modules inherits these deviations, degrading cell performance if reused without adjustment. To overcome this fundamental bottleneck in circular manufacturing, we developed a comprehensive quantification framework to audit and restore perovskite composition. By combining nuclear magnetic resonance (NMR), inductively coupled plasma‐optical emission spectroscopy (ICP‐OES), and ion chromatography (IC), we obtained full compositional fingerprints of the hybrid perovskite recovered from processed solar‐cell stacks, allowing us to resolve their altered composition and restore the material to match the original precursor formulation. Composition restoration effectively closed the performance gap, yielding recycled perovskite cells with efficiencies comparable to pristine devices. A cost analysis demonstrates this approach can achieve a 69.1% cost reduction, while preserving supply‐constrained elements like Cs and I. These results demonstrate a practical, compositionally informed pathway for the sustainable, closed‐loop manufacturing of complex perovskite absorbers.

High‐Performance Flexible Pyroelectric Energy Harvesting System Enabled by Light‐Driven Thermomechanical Coupling in Liquid Crystal Elastomer

ABSTRACT Liquid crystal elastomers (LCEs) with reversible thermal actuation are promising platforms for multifunctional flexible electronics. Herein, we present a PVDF/LM‐LCE (PVDF, polyvinylidene fluoride; LM, liquid metal) composite in which PVDF is polymerized in situ within the LCE matrix to achieve seamless mechanical coupling and efficient stress transfer. LM nanodroplets enhance mechanical robustness, charge transport, and photothermal conversion, enabling LCEs to serve as photothermally driven transducers that amplify the piezoelectric and pyroelectric outputs in these flexible systems. The optimized composite achieves a pyroelectric coefficient of −4.81 nC·cm − 2 ·K − 1 , 1.8 times higher than conventional PVDF films. Furthermore, the composite device powers two LEDs and digital sensors using low‐grade photothermal fluctuations. This LCE‐based light‐driven thermomechanical‐to‐electrical conversion strategy offers a generalizable pathway for high‐performance, flexible pyroelectric and piezoelectric energy‐harvesting materials.

Respiratory sound-based AI screening of asthma and COPD via multi-feature fusion and CatBoost classification

Structure-guided discovery of non-catechol dopamine D1 receptor ligands with biased agonism and antagonism

Exploring pancreatic cancer risk in rheumatoid arthritis and role of rheumatoid factor test: A retrospective cohort study using NIH All of Us database.

e16496 Background: Pancreatic cancer (PC) has rising mortality, emphasizing the need for improved risk stratification. Rheumatoid arthritis (RA) is a chronic autoimmune disease with a known risk of hematologic malignancies and pancreatitis, but its association with PC is less well defined. Methods: A retrospective cohort study was done using NIH All of Us program. RA cases were identified using ICD codes, and the index date was the first documented RA diagnosis. Each case of RA was matched 1:1 to non-RA control on birth year, sex at birth and calendar year of index date. The primary outcome was incidence of PC occurring after RA index date. Cox proportional hazards models estimated hazard ratios (HRs) and 95% confidence intervals (CIs). Lag analyses at 2 and 5 years after RA diagnosis were conducted to mitigate potential reverse causation and surveillance bias. In a secondary analysis, we assessed the relevance of Rheumatoid Factor (RF) testing among PC patients. Patients with PC diagnosis with ICD codes were identified and RF test was extracted. To reduce confounding from malignancy-associated immune activation, only RF testing performed at least one year prior to PC diagnosis was included. Multivariable logistic regression was performed adjusting for demographic covariates. Survival outcomes were analyzed using adjusted Cox proportional hazards for regression. Results: A total of 11,375 patients with RA were matched to 11,375 non-RA controls with comparable follow-up (median 2,350 vs 2,141 days). During follow-up,195 patients with RA had PC incidence compared with 3 among the control group. RA diagnosis was strongly associated with PC risk (HR of 55.8 , 95% CI 17.8-174.5 , p < 0.001). Further lag analysis was done in 2 years and 5 years which were consistent with high HR (2-year lag: HR 30.2, 95% CI 9.6–95.4; 5-year lag: HR 17.0, 95% CI 5.3–54.6;both p< 0.001).Secondary analysis identified 1203 PC patients and 140 of those patients had RF level checked. Among the 140 PC patients with RF testing, 114 (81.4%) were RF-positive. and 26 (18.6%) were RF-negative. The overall cohort had a median age of 73.5 years (IQR 65.5–79.8) and a mean age of 73.2 ± 10.3 years. In multivariable logistic regression, age was the only independent factor associated with RF positivity, with a 75% increase in odds per 10-year age increase (OR 1.75; p = 0.011). Female sex was not independently associated with RF positivity (OR 1.27; p = 0.61). In adjusted Cox regression analysis, RF positivity was not associated with overall survival (HR 1.15, 95% CI 0.42–3.15; p = 0.79). Conclusions: This study showed that RA is an independent risk factor for PC, with risk persisting after lag analyses. However, antecedent RF testing and positivity may not have significant utility for screening or prognosis of PC. Further longitudinal studies are required to better understand the causal relationship and screening utility.

Efficacy and safety of neoadjuvant disitamab vedotin (DV) in combination with toripalimab (Tor) and concurrent/sequential chemotherapy (chemo) in patients with HR-negative, HER2-low breast cancer: A randomized multicenter phase 2 study.

598 Background: For patients (pts) with triple-negative early breast cancer (BC), recommended neoadjuvant regimens include chemo with or without immunotherapy, which achieve a modest rate of pathological complete response (pCR). Emerging data suggest that anti-HER2 antibody-drug conjugates (ADCs) have antitumor activity in HER2-low advanced BC. Notably, DV (anti-HER2 ADC) has shown encouraging efficacy both as monotherapy and in combination with Tor (PD-1 inhibitor) in pts with HER2-low advanced disease. We evaluated the efficacy and safety of neoadjuvant DV-containing regimens in pts with hormone receptor (HR)-negative HER2-low BC in a randomized phase 2 trial. Here, we report results from the pts receiving DV plus Tor combined with concurrent/sequential chemo. Methods: Eligible pts were aged ≥18 years with untreated histologically confirmed HR-negative and HER2-low (centrally confirmed as IHC 1+, or IHC 2+/ISH-) BC (T1cN1-2M0 or T2-3N0-2M0 as per AJCC 8th edition) who were planned to undergo curative-intent breast cancer surgery. Pts randomized to group A received DV (2.0 mg/kg Q2W) + Tor (3.0 mg/kg Q2W) + carboplatin (AUC 3 Q2W or AUC 1.5 QW) for 18 weeks. Pts randomized to group B received DV + Tor for 12 weeks followed by epirubicin (90 mg/m 2 Q3W) + cyclophosphamide (600 mg/m 2 Q3W) + Tor for additional 12 weeks. Stratification factors were PD-L1 status (positive or negative) and clinical staging (stage II or III). After neoadjuvant treatment, surgery was performed. Enrollment in group B was closed after 28 pts had been randomized to this group, while enrollment in group A continued. The primary endpoint was total pCR (tpCR, defined as ypT0/TisN0) rate; secondary endpoints included breast pCR (bpCR, defined as ypT0/Tis) rate, objective response rate (ORR), and safety. Results: 40 pts were enrolled in group A (median age: 46.0 years; clinical stage III: 37.5%; clinical N+: 82.5%; IHC 1+: 72.5%; PD-L1 CPS < 10: 37.5%) and 28 in group B (54.0; 42.9%; 82.1%; 67.9%; 32.1%). The tpCR was achieved in 25 pts (62.5% [95% CI: 45.8-77.3]) in group A and 14 pts (50.0% [31.9-71.3]) in group B. The bpCR rate was 65.0% (95% CI: 48.3-79.4) and 53.6% (35.3-74.5) for groups A and B, respectively. The corresponding ORR was 92.5% (95% CI: 79.6-98.4) and 75.0% (55.1-89.3), respectively. Grade 3-4 treatment-related adverse events (TRAEs) occurred in 29 (72.5%) of pts in group A and 23 (82.1%) in group B. Serious TRAEs occurred in 6 (15.0%) and 7 (25.0%) of pts, respectively. Immune-related adverse events occurred in 3 (7.5%) of pts in group A and 7 (25.0%) in group B. No grade 5 TRAEs occurred. Conclusions: Both treatment regimens showed manageable safety profiles; DV + Tor + carboplatin showed numerically better efficacy in pts with previously untreated, HR-negative, HER2-low, early BC. Clinical trial information: NCT06227117 .

Risk assessment models for venous thromboembolism in patients with multiple myeloma: A systematic review and meta-analysis.

e24168 Background: Thrombosis prevention is a cornerstone of supportive care in multiple myeloma (MM), as venous thromboembolism (VTE) drives morbidity, treatment interruptions, mortality, and reduced quality of life. The heightened VTE risk in MM arises from patient-, disease-, and treatment-related factors, alongside an increased bleeding risk. Several risk assessment models (RAMs) have been developed to guide thromboprophylaxis, but validation studies show variable performance and heterogeneous designs, limiting generalizability. We conducted a systematic review and meta-analysis to evaluate the predictive performance of available VTE RAMs in adults with MM. Methods: A systematic literature search of Cochrane, Embase, PubMed, and Web of Science was conducted from inception through September 2025. Eligible studies were randomized controlled trials (RCT) or observational studies that externally validated VTE RAMs in adults with MM. Abstract screening, full-text review, and data extraction were performed in duplicate. C-statistics for model discrimination and observed-to-expected (O:E) ratio for model calibration were pooled by random-effects meta-analysis. Heterogeneity was measured with the I 2 statistic. Results: After removal of duplicates, 1,419 records were screened for eligibility, of which 26 were included in the final analysis. Most studies were retrospective cohorts, except for 1 RCT and 2 prospective cohorts. IMPEDE was validated in 21 studies, SAVED in 15 studies, PRISM in 7 studies, and IMWG in 4 studies. Overall, 13,312 individuals were included in this review. Median age ranged from 60 to 75.5 years; proportion of males from 41.5% to 98%; overweight/obesity from 13% to 62%; use of immunomodulatory drugs from 7.3% to 100%; aspirin use from 10.5% to 95.4%; and anticoagulant use from 3.4% to 59.7%. IMPEDE demonstrated the highest pooled discrimination (c-statistic 0.65, 95% CI 0.61–0.69, I 2 = 25%), followed by SAVED (0.60, 95% CI 0.53–0.66, I 2 = 73%), PRISM (0.59, 95% CI 0.47–0.69, I 2 = 69%), and IMWG (0.58, 95% CI 0.50–0.65, I 2 = 0%). Calibration was assessable for IMPEDE and SAVED and was acceptable on average, with pooled O:E ratios of 1.12 (95% CI 0.73–1.72, I 2 = 88%) and 1.06 (95% CI 0.62–1.84, I 2 = 94%), respectively, although CIs were wide. In direct comparisons from seven studies, there was no statistically significant difference in discrimination between IMPEDE and SAVED (pooled c-statistic difference 0.04, 95% CI −0.02 to 0.10, I 2 = 0%). Conclusions: In adults with MM, existing VTE RAMs show only modest discriminative ability, with limited accuracy in distinguishing patients at higher versus lower thrombotic risk, and perform inconsistently across patient populations. Future work should focus on contemporary external validation, improved calibration, as well as refinement or updating of existing models to enhance predictive performance.

Evaluation of a novel diagnostic kit using the semi-dry dot-blot method with an automated reader for detecting breast cancer metastases in sentinel lymph nodes: A combined analysis of two multicenter prospective studies in Japan.

586 Background: The semi-dry dot-blot (SDB) method is a diagnostic procedure for detecting lymph node (LN) metastases using an anti-cytokeratin (CK) antibody, based on the premise that epithelial components such as CK protein are not present in normal LNs. Thus, the presence of CK protein in the lavage fluid of sectioned LNs indicates metastasis. Because this method does not involve loss of LN tissue, it can be used in parallel with conventional histopathological diagnosis. We developed a novel diagnostic kit utilizing the SDB method and a newly developed anti-CK19 antibody to detect macrometastases over 2.0 mm in diameter. Two multicenter prospective studies were conducted to evaluate the diagnostic performance of this kit, and we herein report a combined analysis of the two studies. Methods: A total of 825 sentinel LNs were dissected between January and December 2021, and 909 LNs between August 2023 and November 2024, at seven institutions in Japan from 880 patients with breast cancer. Patients who received neoadjuvant chemotherapy or endocrine therapy were excluded. LNs were sliced at 2-mm intervals and washed with phosphate-buffered saline. Cells in the lavage fluid were centrifuged and lysed, and the extracted proteins were applied to the SDB kit. Absorbance was measured using an automated reader to determine the presence of metastases. The washed LNs underwent blinded intraoperative and permanent histopathological examination. Diagnostic performance of the SDB kit was compared with permanent histology, with primary endpoints being sensitivity, specificity, and overall concordance for detecting macrometastases. Results: Of the 1,734 LNs analyzed, 199 were diagnosed as macrometastases and 1,535 as non-macrometastases, including 64 micrometastases measuring more than 0.2 mm and not up to 2.0 mm in diameter, based on permanent histopathology. Using a CK absorbance cutoff of 11.9, the SDB kit correctly identified 185 of the 199 macrometastatic LNs and 1,496 of the 1,535 non-macrometastatic LNs. This corresponded to a sensitivity of 93.0%, specificity of 97.5%, and overall concordance of 96.9%. Additionally, the kit enabled diagnosis within approximately 20 minutes, at a cost of under $30 per test and less than $3,000 for the reader device. Conclusions: The novel SDB kit with an automated reader demonstrated high accuracy, rapid turnaround, and cost-effectiveness in detecting breast cancer LN metastases, while preserving LN tissue. It proved particularly effective in identifying macrometastases. Commercial availability of the kit and reader is planned in the near future.

Clinical outcomes of enfortumab vedotin (EV) or pembrolizumab (P) interruption in metastatic urothelial carcinoma (mUC).

e16592 Background: EV–P is a standard first-line therapy for mUC. In routine practice, one agent is frequently discontinued due to toxicity with continuation of the alternate agent as monotherapy (mono). The clinical impact remains unclear. Methods: We performed a retrospective analysis of patients (pts) with mUC on EV–P. Pts were stratified by interruption of one agent with continuation of the alternate agent as mono versus uninterrupted EV-P. Overall survival (OS) and progression-free survival (PFS) were analyzed using multivariable Cox models adjusted for age, first line EV–P, synchronous metastases, extranodal disease, treatment duration, and immune-related adverse events (irAEs). Same covariates were used for 1:1 propensity score matching. Baseline characteristics were compared using Wilcoxon rank-sum and Chi-squared tests. Results: Among 159 pts, 36 (23%) interrupted treatment and continued mono, while 123 (77%) received uninterrupted EV-P combination. Baseline characteristics were comparable, including age (74 vs 72 years, p=0.35), first-line use (58.3% vs 56.9%, p=0.88), and extranodal metastases (38.9% vs 36.6%, p=0.80). Median follow-up was 9.5 months (IQR 5–18). irAEs were more frequent in the interruption group (72.2% vs 29.5%, p<0.001). Median treatment duration was longer with interruption (8.6 vs 3.8 months, p<0.001). Among the interrupted cohort, 53% continued P and 47% continued EV. In propensity-matched analysis, interruption was associated with improved PFS (HR 0.32, 95% CI 0.25–0.70, p=0.0026) and OS (HR 0.24, 95% CI 0.09–0.66, p=0.03). Among the interrupted cohort, continuation of P had higher PFS (HR 0.14, p=0.003) and OS (HR 0.28, p=0.01) than continuation of EV monotherapy. Conclusions: In this cohort, discontinuation of one EV–P component with continuation of the other agent as mono was associated with improved survival versus uninterrupted combination. Toxicity-guided treatment de-escalation may achieve durable disease control in selected pts. Patient characteristics and treatment outcomes. Variable Interrupted EV or P (N=36) No interruption in EV-P (N=123) P value Baseline demographic characteristics Age at Tx initiation 74.00 (65.75, 80.25) 72.00 (64.00, 77.50) 0.350 Sex 0.623 - Female 25.0% (9) 21.1% (26) - Male 75.0% (27) 78.9% (97) Duration of treatment 8.60 (5.33, 14.98) 3.83 (1.60, 8.32) <0.001 M stage Tx initiation 0.379 No distant metastases (M0) 16.7% (6) 23.6% (29) Metastatic disease present (M1) 83.3% (30) 76.4% (94) Treatment characteristics and discontinuation patterns Adverse events from ICI 72.2% (26) 29.5% (36) <0.001 Pembro discontinuation due to adverse event 51.4% (18) _ Pembro discontinued due to progression 11.4% (4) 27.9% (34) 0.075 Adverse events from EV 72.2%(26) 41.8% (51) 0.001 EV discontinuation due to adverse event 42.9% (15) _ EV discontinued due to progression 34.3% (12) 29.5% (36) 0.739

Temporal evolution of prostate cancer incidence and incidence-based mortality from 2000 to 2022: A retrospective analysis by race and age at diagnosis based on USPSTF screening guidelines.

5124 Background: Prostate-specific antigen (PSA) screening guidelines have evolved substantially since 2000, influencing detection patterns and outcomes. Despite widespread implementation, significant racial disparities persist in prostate cancer incidence and mortality. Understanding temporal trends across guideline eras, stratified by race and age, is critical for optimizing screening strategies and addressing health equity. Methods: Using SEER*Stat, we extracted data from the Surveillance, Epidemiology, and End Results (SEER) database (2000–2022). We employed Jointpoint regression to analyze the secular trend in incidence and incidence-based mortality (IBM), which we classified and arranged by race, age at diagnosis, and year of diagnosis. Annual percentage change (APCs) were calculated using weighted least square method. Results: A total of 1,259,662 patients met the inclusion criteria for our study. These included diagnosed cases of prostate cancer from 2000-2022. The total age-adjusted incidence rate was 60.7 per 100,000. The APC from 2000-2007 (before 2008 guidelines) was -0.1 (p < 0.05), signifying a cumulative annual decrease. The APC was -2.6 (p < 0.05) from 2008-2011, 0.1 (p < 0.05) from 2012-2017, and 1.5 (p < 0.05) from 2018-2022. Based on race, the APC from 2000-2022 was -2.1 for white, -1.7 for blacks, -2.2 for American Indian/Alaska native (AIAN) and -2.7 for asian and pacific islander (API). Based on age at diagnosis, the APC was -1.9 from 55-64 years, -1.7 from 65-74 years, -2.0 from 75- 84 years, and -2.2 at 85 years and above. In terms of incidence based mortality, rate increased from 2.4 (2000) to 37.7 (2022) in white, with APC of 6.3% (95% CI 5.0–7.7), in black, rate increased from 4.8 (2000) to 60.5 (2022), with APC of 5.2% (95% CI 4.0– 6.4), in API, rate increased from 1.8 (2000) to 17.7 (2022) while in AIAN, rate increased from 2.4 (2000) to 21.5 (2022), with APC of 5.0% (95% CI 3.4–6.6). Conclusions: Between 2000 and 2022, prostate cancer incidence demonstrated distinct temporal patterns aligned with changes in PSA screening recommendations, with a marked decline from 2008 to 2011, minimal change from 2012 to 2017, and an increase from 2018 to 2022. Across the same period, incidence-based mortality (IBM) increased steadily in all racial groups, despite long-term declines in overall incidence. Incidence-based mortality rates were consistently higher in black patients compared with white patients throughout the study period. These findings demonstrate a difference between incidence and incidence-based mortality over time and highlight persistent racial differences in prostate cancer outcomes across multiple screening eras.

SCALE-2: A phase 2 trial evaluating neoadjuvant short-course radiotherapy combined with chemotherapy and toripalimab in resectable locally advanced esophageal squamous cell carcinoma.

4092 Background: The SCALE-1 Phase Ib trial demonstrated that short-course neoadjuvant radiotherapy combined with toripalimab and chemotherapy, followed by esophagectomy, exhibited manageable toxicity and promising efficacy in patients (pts) with resectable locally advanced ESCC (RLaESCC). This Phase II trial aims to further valuate efficacy and safety of this novel short-course regimen in a larger cohort. Methods: RLaESCC pts with clinical stages cT3-4aN0M0/cT1-4aN+M0 received neoadjuvant paclitaxel (135 mg/m 2 ), carboplatin (AUC=5), and toripalimab (240 mg) every 3 weeks for two cycles. Short-course neoadjuvant radiotherapy (30 Gy/12f; 5 days per week) was administered between the two doses of neoadjuvant immune-chemotherapy. Compared to SCALE-1, high-risk nodal regions were selectively included in the irradiation volume. Weekly weight changes were used to assess preoperative waiting time. Esophagectomies were scheduled 8-10 weeks after completing neoadjuvant treatment. The primary endpoint was pathological complete response (pCR) rate, with secondary endpoints included disease free survival (DFS), overall survival (OS) and safety. Results: From April 28, 2024 to February 26, 2025, 63 pts were enrolled (intention-to-treat population, ITT), with stage distribution:I (3.2%), II (14.3%), III (63.5%) and IVA (19%). Following the exclusion of 3 pts who refused surgery, pathological assessment of the surgical specimens revealed that 56.7% (34/60) pts achieved pCR (ypT0N0 or ypTisN0) and 91.7% (55/60) achieved MPR. The median number of resected lymph nodes was 19 (range: 6–45). At a median follow-up of 23.9 months (cutoff date: Jan 19, 2026), median OS and DFS were not reached. In the ITT population, the 2-year OS and DFS rates were 90.4% (95% Cl: 82.4-98.4%) and 81.5% (95% Cl: 70.5-92.5%), respectively. Postoperative complications occurred in 45% (27) pts, with 18.3%(11) experiencing Grade ≥3 events. The most frequent Grade 3 and above complications were anastomotic leakage (n=4) (with or without pleural effusion or hemothorax) and airway mucus obstruction (n=4). Moreover, one case of postoperative delirium was observed (Grade IV), and one patient died of septic shock secondary to an anastomotic leak. All 63 pts completed neoadjuvant radiotherapy. However, 2 pts discontinued the second dose of toripalimab, and 16 pts required dose reductions during the second cycle of neoadjuvant chemotherapy. The most frequent grade 3/4 adverse events were neutropenia (n=27, 42.9%) and leukopenia (n=21, 33.3%). Notably, no grade 3 esophagitis or pneumonitis occurred. Conclusions: The SCALE regimen demonstrated potent antitumor activity, achieving 56.7% pCR and 91.7% MPR. Favorable 2-year OS and DFS further highlight its potential to improve long-term outcomes for RLaESCC pts. Clinical trial information: NCT05424432 .

Pretreatment CT-based radiomics and machine learning models for predicting treatment response in lung cancer: A diagnostic test accuracy meta-analysis.

e20023 Background: Lung cancer remains one of the most commonly diagnosed malignancies and the leading cause of cancer-related mortality worldwide. CT-based noninvasive predictive biomarkers, including radiomics and machine learning models, may aid in predicting treatment response and guiding therapy selection. However, heterogeneous evidence underscores the need for pooled analyses to define their clinical utility. Methods: This PRISMA-compliant meta-analysis was prospectively registered in PROSPERO. We systematically searched PubMed, Embase, Cochrane CENTRAL, and ClinicalTrials.gov from inception to January 2026 to identify studies on radiomics signatures and machine learning models for predicting treatment response (TR) and pathological response (PR), reporting poolable AUCs. Logit-transformed AUCs were used for meta-analysis, and pooled estimates were calculated using a generic inverse-variance model with REML tau estimation in R (version 2025.05.0+496; 26 Posit), with a two-sided significance threshold of p < 0.05. Results: The pooled logit AUC for radiomics signatures predicting treatment response was 1.85 [1.02–2.68]; I² = 77.0%, with combined radiomics–clinical models showing slightly improved performance (1.87 [1.02–2.71]; I² = 0%). Machine learning models demonstrated a pooled AUC of 1.42 [0.61–2.23]; I² = 94%, with the highest-performing models reaching 2.70 [0.53–4.87]. In chemotherapy-alone cohorts, the pooled AUC was 2.02 [0.78–3.26], whereas chemoimmunotherapy cohorts exhibited lower performance (0.42 [0.25–0.59]; I² = 0%, P = 0.026). Validation cohorts achieved an AUC of 1.01 [0.27–1.75]; I² = 71%, with multicenter studies reporting 0.85 [0.44–1.26] and single-center studies 1.41 [0.69–2.12]. For pathological response, radiomics signatures achieved a pooled logit AUC of 1.27 [0.92–1.63]; I² = 69.4%, with multicenter studies performing better (1.66 [1.42–1.89]; I² = 0%) and validation cohorts reaching 1.01 [0.77–1.26]; I² = 0%. Machine learning models achieved a pooled AUC of 1.40 [1.13–1.67]; I² = 42.3% in training cohorts and 0.94 [0.69–1.18]; I² = 0% in validation cohorts. Subgroup analyses did not reveal statistically significant differences. Conclusions: Radiomics signatures and high-performing machine learning models show strong predictive value for treatment response, particularly when combined with clinical features. However, their predictive performance for pathological response is moderate and inconsistent, highlighting the need for further multicenter validation before routine clinical use.

Genomic and clinical characterization of estrogen receptor–positive/HER2-negative early breast cancer: Results from the NextGIM study.

e12590 Background: Estrogen Receptor–Positive (ER+)/HER2-Negative early breast cancer (BC) is biologically heterogeneous: while most patients experience a favourable prognosis, 15–30% still relapse despite standard adjuvant therapy. As therapeutic options expand, reliable biomarkers are needed to guide escalation and de-escalation strategies. Genomic signatures and low HER2 expression have emerged as potential tools for refining prognosis and predicting treatment benefit, although their roles beyond specific therapeutic contexts remain uncertain. Methods: NextGIM is a multicenter retrospective translational study integrating clinicopathologic and molecular data from patients with ER+/HER2-negative early BC enrolled in the GIM2, GIM4, and GIM10 trials. Reassessment of tumor-infiltrating lymphocytes (TILs) and HER2 expression was performed, and associations with invasive disease-free survival (iDFS) and overall survival (OS) were evaluated. Comprehensive gene expression profiling was performed using the nCounter BC 360 panel platform. Results: Overall, 272 patients were included: median age was 60 years (IQR 53–67), and 203 (74.6%) underwent surgery before 2011. Most tumors were T1 (163, 59.9%), and nearly half of patients had N1 disease (128, 47.1%). At diagnosis, HER2 reporting reflected trial-era pathology, with 106 (39.0%) tumors classified as HER2-negative without a documented immunohistochemistry (IHC) score, 77 (28.3%) as HER2-zero, 52 (19.1%) as HER2 1+, and 37 (13.6%) as HER2 2+ without ISH amplification. Among the 225 samples already reassessed, 126 (56.0%) were classified as HER2-null, 37 (16.4%) as HER2-ultralow, 43 (19.1%) as HER2 1+, 18 (8.0%) as HER2 2+, and 1 (0.4%) as HER2 3+. Most HER2-null BC were diagnosed before 2011 (76.2%), whereas HER2-ultralow BC displayed a more even distribution (54.1% before vs. 45.9% after 2011; p = 0.050). TILs ≤1% were observed in 55.6% of HER2-null BC, 67.6% of HER2-ultralow cases, 34.9% of HER2 1+ tumors, and 27.8% of HER2 2+ BC. After a median follow-up of 20.3 years (IQR 10.1–21.8), iDFS and OS did not differ across HER2 subgroups defined by reassessment (iDFS: p = 0.385; OS: p = 0.792). Preliminary gene expression data from 123 samples showed a progressive increase in ERBB2 expression with rising HER2 scores, accompanied by only modest variation in proliferation-related genes (e.g., AURKA, FGFR4, MKI67, CCNA2, CDK1, CDKN3). Conclusions: The predominance of HER2-null cases among older specimens suggests a potential contribution of pre-analytical and storage-related factors, which may limit the reliability of HER2 IHC reassessment in archival tissue. Repeat testing on a recent specimen should be considered when HER2 status is re-evaluated to inform treatment decisions.

Three-month versus six-month duration of postoperative adjuvant therapy after R0 resection following conversion therapy in hepatocellular carcinoma patients achieving MPR or pCR: A prospective randomized study.

4121 Background: Conversion therapy allows a subset of patients with initially unresectable hepatocellular carcinoma (uHCC) to achieve a deep pathological response, enabling curative resection. However, the optimal duration of postoperative adjuvant therapy remains unclear. This prospective randomized study compared recurrence-free survival (RFS) and overall survival (OS) between 3-month and 6-month durations of postoperative adjuvant therapy after R0 resection in patients achieving major pathological response (MPR; ≥90% tumor necrosis) or pathological complete response (pCR). Methods: This randomized controlled trial (ChiCTR2100051789) enrolled patients with uHCC who achieved a MPR or pCR following conversion therapy. Eligibility criteria included pathological confirmation of MPR or pCR after R0 resection. Eligible patients were randomized 1:1 to receive 3 months (Group 3M) or 6 months (Group 6M) of postoperative adjuvant therapy starting 1 month after surgery. The primary endpoint was RFS; secondary endpoints included OS and recurrence patterns. Results: Between January 15, 2022, and April 20, 2025, 58 patients were randomized (29 per group). Pathological evaluation confirmed MPR in 21 patients and pCR in 37 patients. After a median follow-up of 33.05 months, recurrence occurred in 8 patients. The 1-, 2-, and 3-year RFS rates were 89.3%, 89.3%, and 78.8% in Group 3M and 88.9%, 88.9%, and 88.9% in Group 6M (P = 0.592). Corresponding OS rates were 100%, 96.4%, and 96.4% versus 100%, 91.8%, and 91.8%, respectively (P = 0.451). Conclusions: In patients with uHCC who achieved a deep pathological response (MPR or pCR) after conversion therapy, a shorter duration of adjuvant therapy may be sufficient, with a 3-month course providing survival outcomes comparable to those of a 6-month course. Clinical trial information: ChiCTR2100051789.

Tumor delivery and exposure of aldoxorubicin compared with doxorubicin: Integrated clinical and preclinical analysis.

e15134 Background: Doxorubicin is widely used across solid tumors and sarcomas but is limited by dose-dependent cardiotoxicity and suboptimal tumor delivery. Aldoxorubicin is an albumin-binding prodrug of doxorubicin designed to enhance tumor uptake through albumin transport pathways and to release active doxorubicin in the acidic tumor microenvironment and in tumor cells. Prior studies have independently demonstrated tumor accumulation of aldoxorubicin or its released doxorubicin. We conducted the first integrated analysis comparing tumor-site exposure of aldoxorubicin with doxorubicin using clinical intratumor pharmacokinetics and nonclinical biodistribution and efficacy data. Methods: Tumor and paired skin biopsies were obtained 24–48 hours after dosing from patients with Kaposi’s sarcoma treated with aldoxorubicin (Study ALDOX-P2-KS-01; NCT02029430). Total drug (aldoxorubicin + doxorubicin) was quantified and normalized by administered dose and doxorubicin-equivalent dose. Published Kaposi’s sarcoma intratumor doxorubicin concentrations (Northfelt 1996) served as clinical comparators. Because measured concentrations could not distinguish prodrug from released doxorubicin, two bounding potency scenarios were evaluated: (1) “All Aldox,” assuming all tumor drug was aldoxorubicin and applying a 10-fold lower in-vitro potency (Kratz 2002 and 2007); and (2) “All Dox,” assuming complete conversion to native doxorubicin. Mouse tumor xenograft data (Kratz 2007) comparing aldoxorubicin and doxorubicin were extracted, including tumor AUC 0-24 following 14 C-labeled dosing and tumor growth inhibition at equivalent doses. Exposure and potency-normalized metrics were compared across datasets. Results: Aldoxorubicin was detectable in all human tumor lesions, with linear dose-dependent increases in intratumor concentrations. Dose-normalized tumor exposure was substantially greater for aldoxorubicin compared with published doxorubicin values. Under the All Aldox scenario, potency-normalized exposure was comparable to doxorubicin; under the All Dox scenario, tumor exposure was up to an order of magnitude higher. In mouse xenografts, aldoxorubicin produced higher tumor AUC 0-24 than doxorubicin (92.2 vs 54.0 %ID·h/g) and demonstrated comparable antitumor activity at equivalent exposures and superior activity at higher tolerated doses. Conclusions: Across integrated human and mouse datasets, aldoxorubicin achieved consistently higher tumor exposure than doxorubicin when normalized for dose, with potency-adjusted analyses supporting at least comparable active exposure. Together, these findings support aldoxorubicin as a tumor-targeted anthracycline capable of delivering greater active drug levels to tumors with improved tolerability, offering the potential for enhanced therapeutic index compared with doxorubicin.