Browse Articles

Burned-out lymph nodes in testicular cancer: A new entity in imaging and management.

e17008 Background: Germ cell tumors are the most common cancer in men between 15 and 40, with a primary testicular tumor in 95% of cases. In rare instances, this primary tumor may spontaneously regress — a phenomenon known as a “burned-out testicular tumor » (BOTT). It mainly involves pure seminomas, showing clinical and histological regression along with a characteristic appearance on scrotal imaging. We have identified a similar phenomenon of spontaneous regression of lymph node metastases. This newly recognized clinical entity should be acknowledged and considered in treatment and follow-up. Methods: We reviewed retrospectively all new diagnoses of metastatic germ cell tumors between January 1, 2020, and August 31, 2025 in two high-volume French centers. Patients with sequential conventional and metabolic imaging and histological evidence confirming the phenomenon of burned-out lymph nodes (BOLNs) were identified. Data on patient characteristics, treatment, serum tumor markers and lymph node size over time were collected. Results: From January 2020 to August 2025, we identified five cases. The median age was 30.5 years (range 19–49 years). All patients were diagnosed with pure seminoma, histologically confirmed either from the primary tumor (4/5) or from lymph node biopsies. The median lymph node size at diagnosis was 18 mm (range 12–35 mm). In four out of five patients, the lymphadenopathies were ipsilateral retroperitoneal nodes relative to the testicular tumor; in one patient, they were mediastinal nodes, with biopsy confirming the presence of seminoma. All patients underwent baseline CT and 18F-FDG PET scans, repeated at least twice, showing partial morphological regression and complete metabolic response in all cases. Only one patient has so far required chemotherapy. The median follow-up duration for the remaining patients was 13.5 months (range 6–31 months). As of August 2025, no patient showed evidence of active disease on imaging. Conclusions: BOTTs are a well-documented phenomenon, but the concept of BOLNs in testicular cancer, with spontaneous regression observed in retroperitoneal or mediastinal lymphadenopathy without systemic chemotherapy has not been previously described, making this the first case series on the subject.It is important to recognize this entity in order to guide the most appropriate treatment strategy and avoid overtreatment. In case of complete lymph node regression and normalization of tumor markers (if initially elevated), close surveillance with thoracoabdominopelvic CT scans and serum marker testing is indicated. In case of partial regression, a biopsy is necessary to determine whether viable tumor cells remain. If the biopsy is inconclusive, a lymph node dissection may be warranted. Management should be tailored based on pathology, imaging evolution, and tumor marker dynamics.

Quantifying the macroeconomic burden of tobacco-attributable cancers in 2023: Global disparities in value-of-lost-welfare (VLW) in 189 countries.

10567 Background: Tobacco use is the leading preventable cause of cancer globally, yet the macroeconomic welfare losses attributable to tobacco-related cancers remain incompletely quantified. We estimated the global economic burden of tobacco-attributable cancers in 2023 using a standardized value-of-lost-welfare (VLW) framework. Methods: Country-level disability-adjusted life years (DALYs) attributable to tobacco for all cancers were obtained from the Global Burden of Disease (GBD) 2023 framework. Gross domestic product (GDP; purchasing power parity, current international dollars) and population data were sourced from the World Bank. VLW was calculated as tobacco-attributable cancer DALYs multiplied by country-specific GDP per capita (PPP). Outcomes included total VLW, per-capita VLW, and VLW as a percentage of national GDP. Results: In 2023, tobacco-attributable cancers resulted in 48.44 million DALYs globally (95% UI: 41.06–57.06 million), corresponding to a global value of lost welfare (VLW) of USD 1.61 trillion. The largest absolute economic welfare losses were observed in China (USD 415.3 billion), India (USD 262.8 billion), and the United States (USD 219.4 billion), followed by Japan (USD 96.1 billion) and Brazil (USD 63.7 billion). Together, these five countries accounted for 65.9% of the global VLW attributable to tobacco-related cancers. When normalized to national economic output, the highest VLW-to-GDP ratios were observed in India (1.12%), China (0.69%), Egypt (0.61%), Bangladesh (0.58%), and Pakistan (0.52%), indicating substantial macroeconomic vulnerability in several low- and middle-income countries. By cancer type, tracheal, bronchus, and lung cancer accounted for 24.6 million DALYs, corresponding to a VLW of USD 0.82 trillion, followed by esophageal cancer (6.1 million DALYs; USD 0.21 trillion), stomach cancer (5.4 million DALYs; USD 0.18 trillion), liver cancer (4.9 million DALYs; USD 0.16 trillion), and oral cavity cancer (3.2 million DALYs; USD 0.11 trillion). Collectively, these five cancer types accounted for 91.4% of total tobacco-attributable cancer VLW worldwide. Conclusions: In 2023 alone, tobacco-attributable cancers resulted in nearly 50 million DALYs and over $1.6 trillion in lost economic welfare worldwide, highlighting tobacco use as a dominant driver of both health loss and macroeconomic harm. Regional distribution of disability-adjusted life years (DALYs) and economic welfare loss (VLW) attributable to tobacco-related cancers in 2023. Region DALYs (millions) VLW (USD trillion) VLW/GDP (%) High-income 11.6 0.75 1.81 Central Europe, Eastern Europe, and Central Asia 4.58 0.17 1.27 Southeast Asia, East Asia, and Oceania 22.87 0.53 1.14 North Africa and Middle East 2.27 0.05 0.57 Latin America and Caribbean 1.72 0.03 0.29 South Asia 4.33 0.04 0.22 Sub-Saharan Africa 0.98 0.00 0.12

Comparison of real-world effectiveness of all approved BRAF/MEK inhibitor combinations in <i>BRAF</i> -mutant melanoma patients.

e21525 Background: Currently, three distinct combinations of BRAF and MEK inhibitors (BRAFi/MEKi) are approved for treatment of advanced melanoma patients. However, they have not been compared in a head-to-head trial. Objective of this study was to investigate if selection of BRAF/MEK inhibitors combination impacted progression-free survival (PFS) in real-world setting. Methods: This was a multi-centre, retrospective study, conducted in major melanoma centres in Poland. Only patients treated in routine clinical practice (not clinical trials) were included. Multivariable Cox proportional hazards (PH) regression model was created to assess prognostic impact of the selected variables. Results: A total number of 790 BRAF -mutant melanoma patients received BRAFi/MEKi as first-line systemic therapy in advanced metastatic/unresectable setting. Dabrafenib + trametinib (D+T) was used in 497 (62.9%) patients, encorafenib + binimetinib (E+B) in 146 patients (18.5%) and vemurafenib + cobimetinib (V+C) in 147 patients (18.6%). Table 1 provides baseline clinical characteristics and overall response rates (ORR). Median PFS was 8.11 (95% CI: 7.33 – 9.33) months in patients treated with D+T, 10.35 (7.95 – 12.42) months in patients treated with E+B and 8.71 (7.13 – 11.14) in patients treated with V+C. Median overall survival since the start of treatment was 14.1 (12.6 – 15.6) months, 14.7 (12.6 – 21.7) and 12.2 (10.7 – 16) months respectively. Multivariate Cox model included BRAFi/MEKi, baseline melanoma stage, age and gender as covariates, while lactate dehydrogenase level and ECOG score were used as stratification variables due to failure to meet PH assumption. Melanoma stage and age were the only significant prognostic factors with hazard ratio (HR) of 4.48 (2.34– 8.57; p &lt; 0.001) in M1d patients, 2.41 (1.34 – 4.33; p = 0.003) in M1c patients, 2.08 (1.03 – 4.18; p = 0.04) in M1b patients and 0.74 (0.3 – 1.83; 0.511) in unresectable stage III patients compared to M1a patients as reference and HR 1.01 (1-1.03; p = 0.044) for age (continuous). Importantly, combination of BRAFi/MEKi was not a significant prognostic factor, with HR of 0.86 (0.52 – 1.42; p = 0.552) in E+B patients and 0.90 (0.53 – 1.50; p = 0.676) in V+C patients compared to D+T patients. Conclusions: In this retrospective analysis of real-world patients, type of used BRAFi/MEKi combination was not a significant prognostic factor, suggesting similar clinical activity of these combinations. These results require validation in randomized, prospective studies. Combination D+T E+B V+C Number of patients 497 146 147 TNM: M1c 189 (38%) 52 (35.6%) 50 (34%) TNM: M1d 135 (27.2%) 40 (27.4%) 48 (32.7%) TNM: M1a 65 (13.1%) 23 (15.8%) 21 (14.3%) TNM: M1b 70 (14.1%) 21 (14.4%) 21 (14.3%) TNM: stage III unresectable 33 (6.6%) 9 (6.2%) 7 (4.8%) ECOG: 1+ 359 (72.2%) 112 (76.7%) 98 (66.7%) ECOG: 0 122 (24.5%) 34 (23.3%) 49 (33.3%) ORR 245 (49.3%) 91 (62.3%) 85 (57.8%)

A capacity assessment of pediatric oncology research biorepositories across a regional collaborative group in the Middle East, North Africa, and Asia.

10023 Background: Biobanks advance translational research in pediatric oncology. Understanding the existing biobanking infrastructure is critical for fostering collaboration and improving research capacity in low- and middle-income countries (LMICs). Methods: A previous research capacity survey within institutions of the Pediatric Oncology East and Mediterranean (POEM) Group collaborative network identified 26 out of 69 institutions across 23 countries, with access to clinical tissue specimens for research use. We surveyed these centers regarding sample/data management, patient population, governance, standard operating procedures (SOPs), quality systems, and funding. This study was determined non-human subjects research by the Stanford IRB. Biobanking capacity and participation were evaluated using descriptive statistics (medians, ranges, and frequencies). Results: Fifteen of the 26 (58%) institutions confirmed that they have a research-specific biobank; 10 of 15 (67%) participated. All surveyed biobanks collect cancer samples using broad consent, and have ongoing funding through institutional support, external collaborations or both. Principal investigators lead 20% of the biobanks, the others are institution-based and led. A median of 7 staff are involved in biobanking activities per institution (range, 3–28). Most institutions (80%) rely on Microsoft Excel for sample tracking. A quality management system is implemented in 7 of 10 (70%) of institutions, with most adhering to ISBER best practices. Five of 10 (50%) of the biobanks lack disaster management SOPs. Eight of 10 (80%) biobanks are capable of sharing samples and associated clinical data with other institutions through material/data sharing agreements. A median of 3500 samples (range, 368–6000) are stored yearly with a median of 300 samples (range, 100–1000) released for authorized research. While most biobanks viewed their national legal frameworks as supportive, 2 biobanks (20%) indicated that national regulations constrain biobanking activities. All the respondents indicated a need to improve their national regulation to support ethical biobanking practices – specifically need for improved regulation for cross-border collaborations, data privacy, and intellectual property (60%, 40% and 30%, respectively). Conclusions: Biobanks exist in 21.7% (15 of 69) centers within the POEM group. Biobanks in the region have robust consenting and sample collection procedures. Significant opportunities exist to enhance capacity through implementation of international quality standards, and improved disaster preparedness. Further understanding of the determinants of successful initiation and maintenance of these biobanking efforts in LMICs will guide targeted initiatives to further expand regional biobanking infrastructure and augment research collaborations.

Trastuzumab (Tmab) plus chemotherapy (CT) vs CT alone in the perioperative treatment of HER2-positive gastric cancer: A systematic review and meta-analysis of randomized controlled trials (RCTs).

e16129 Background: HER2 overexpression occurs in 15-20% of gastric cancers (GC) and is associated with poor prognosis. While Tmab combined with CT improves survival HER2-positive GC, its role in resectable disease remains unclear. We conducted a meta-analysis of RCTs to evaluate the efficacy and safety of Tmab plus CT versus CT alone in the perioperative treatment of in HER2-positive GC and gastroesophageal junction (GEJ) adenocarcinomas. Methods: We systematically searched PubMed, Embase, and Cochrane databases for RCTs comparing Tmab plus CT with CT alone in resectable HER2-positive GC. Outcomes included pathologic complete response (pCR), R0 resection rate, progression-free survival (PFS), overall survival (OS), treatment completion, and diarrhea incidence. Data were pooled using a random-effects model. Heterogeneity was assessed via I² statistics. Analyses were performed using R software v4.3.2. Results: Three RCTs (229 patients) met inclusion criteria. Tmab showed a non-significant reduction in death risk (HR 0.74; 95% CI: 0.44–1.25; p = 0.26; I² = 0%) and PFS (HR 0.72; 95% CI: 0.48–1.06; p = 0.09; I² = 0%). No significant differences were found in pCR (RR 0.24; 95% CI: 0.04–1.45; p = 0.12; I² = 37.3%) or R0 resection rates (RR 1.41; 95% CI: 0.89–2.24; p = 0.15; I² = 0%). Similarly, treatment completion (RR 1.04; 95% CI: 0.95–1.13; p = 0.45; I² = 0%) and diarrhea incidence (RR 1.01; 95% CI: 0.93–1.09; p = 0.84; I² = 0%) were comparable between groups. Heterogeneity was low or absent across all the outcomes. Conclusions: In conclusion, the addition of Tmab to perioperative chemotherapy in HER2-positive resectable GC did not yield statistically significant improvements in survival, pathological response, or surgical outcomes, although the regimen was well tolerated. Further large-scale trials are warranted to better define the role of anti-HER2 therapy in the curative-intent setting.

Artificial intelligence and machine learning for predicting immune checkpoint inhibitor–associated cardiotoxicity: A systematic review and narrative synthesis.

e24028 Background: Immune checkpoint inhibitor (ICI) associated cardiotoxicity, including myocarditis and major adverse cardiovascular events (MACE), is uncommon but associated with high mortality. Scalable risk prediction models may enable targeted surveillance and early intervention. Methods: We searched PubMed/MEDLINE, Embase, Web of Science, and Google Scholar from inception through January 2026 for human studies validating artificial intelligence and machine learning (AI/ML) models to predict ICI related cardiac events. We extracted data on demographics, type of cancer, ICI agents, follow-up duration, outcome definitions, model family, and discrimination metrics. Meta-analysis could not be performed due to heterogeneity in endpoints, feature sets, and validation strategies. Results: Six studies (n = 56,721) comprised 5 retrospective cohorts (2 single-center: n = 419 and 4,960; 2 multicenter: n = 615 and 2,258; 1 nationwide claim: n = 48,446) and 1 case series (n = 23). Age (reported in 4/6) ranged from 64.4 years to 72 years; sex (3/6) ranged 24.7–46.4% female and 53.6–75.3% male. Lung cancer was most common (4 studies; 30.6–82.7%, including 59.4% in claims), followed by melanoma/skin (3 studies; 2.0–14.0%) and renal/genitourinary cancers (3 studies; 0.16–22.0%); GI tumors were also frequent (3 studies; 3.6–47.4%). Pembrolizumab was administered in 26.3–48.0%, nivolumab in 30.0–65.9%, atezolizumab in 5.7–17.0% and durvalumab in 2.1–6.9% as reported in 3 studies. Follow-up spanned from 30 days to 299 days. Across cohorts, cardiotoxicity incidence varied markedly by endpoint definition and sampling frame: composite cardiac events occurred in 8.4% of single center cohort (418/4,960), immune-related cardiac adverse events in 23.9% of a multicenter cohort (147/615), and composite of myocarditis/MACE in 11.7% (264/2,258) with myocarditis alone 3.0% (59/2,258). In contrast, nationwide claims-defined myocarditis was rare (0.24%; 117/48,446), while a case series reported myocarditis in 65.2% (15/23), including 4 fatalities. Models included gradient boosting (3 studies), random forests (2), soft-voting two-stage ensemble (1), multimodal fusion deep learning (1), and neural networks (1). AUROC ranged from 0.635–0.851 with lowest in the LightGBM model and highest in the two-stage ensemble. Conclusions: Current AI/ML models demonstrate moderate-to-high discriminatory performance for predicting ICI-associated cardiotoxicity. However, evidence remains limited by retrospective study designs, heterogeneous endpoints, insufficient external validation and calibration reporting. Larger prospective studies incorporating standardized outcome definitions and clinically actionable thresholds are required for real world adoption.

Trial-level association between age-enriched populations and treatment effect in frontline randomized multiple myeloma studies.

e19555 Background: Age at diagnosis has traditionally been considered a host-related factor in MM, influencing treatment tolerability and competing mortality. However, emerging biologic and clinical observations raise the possibility that age-enriched trial populations may differ in disease behavior and treatment responsiveness. We performed a trial-level comparative analysis to evaluate whether median age of enrolled populations is associated with treatment effects across contemporary frontline randomized MM trials. Methods: We systematically assembled a dataset of 18 randomized frontline multiple myeloma clinical trials reporting treatment outcomes and baseline demographic characteristics. Trial-level median age was extracted for each study. The primary analysis used inverse-variance weighted meta-regression to evaluate the association between median age and log(HR) for PFS. A prespecified sensitivity analysis additionally adjusted for baseline control-arm risk using PFS-12 as a proxy. Exploratory analyses evaluated the robustness of findings across selected trial characteristics. Results: Among the 18 identified frontline trials, 14 trials reported sufficient data for inclusion in the primary inverse-variance meta-regression, and 13 trials were included in the sensitivity analysis adjusting for baseline risk. Median age across included trials ranged from approximately 57 to 73 years, reflecting systematic age enrichment across trial populations. In inverse-variance weighted meta-regression, median age demonstrated a modest positive association with log(HR) for PFS, indicating that trials enrolling younger populations tended to show greater relative PFS benefit from experimental therapies compared with trials enrolling older populations. The age association remained directionally consistent after adjustment for baseline risk and in exploratory analyses accounting for trial characteristics, suggesting that differences in control-arm efficacy or selected trial features alone do not fully explain the observed pattern. However, the magnitude of the association was small and substantial heterogeneity was observed across trials. Conclusions: In this trial-level analysis of frontline randomized multiple myeloma studies, age-enriched trial populations demonstrated a modest association with observed PFS treatment effects, with younger-enriched trials tending to show greater benefit from treatment intensification. While exploratory, the consistency of the association across sensitivity analyses suggests that age-related differences in treatment responsiveness may contribute to variability in outcomes across trials. These findings remain hypothesis-generating and do not establish age as an independent prognostic or biologic determinant at the patient level.

Neoadjuvant anti–PD-1 prolgolimab monotherapy in resectable high-risk melanoma: Pathologic response and early event-free survival in a prospective single-arm phase II CRISTINA study.

9579 Background: In resectable high-risk melanoma, neoadjuvant anti–PD-1 aims to exceed the ~40% MPR benchmark; while combos may deepen response, they add tox, complexity, and cost. Recently reported data for prolgolimab+nurulimab (N = 205) showed pCR 38.5%, near-CR 4.4%, motivating evaluation of prolgolimab mono as a simpler potentially non-inferior approach. Methods: MelPRO-0322 (CRISTINA; NCT06299878) is a prospective single-arm study in resectable stage IIIB–IV melanoma: 3 neoadj doses of prolgolimab followed by radiologic assessment followed by surgery (regional lymphadenectomy or M1a-equivalent metastasectomy). Central pathology used INMC criteria. Interim analysis after full enrollment and pathology assessment. Results: Pt characteristics (N = 82): Median age 61.5 y (range 23.2–87.3); median BMI 28.2 kg/m² (21.0–46.5). Females 47 (57.3%); ECOG 0/1: 56 (68.3%)/26 (31.7%). Primary melanoma: cutaneous or UPO 73 (89.0%), acral 7 (8.5%), mucosal 2 (2.4%). Stage: IIIB 12 (15.8%), IIIC 54 (71.1%), IIID 3 (3.9%), IV 7 (9.2%); stage missing 6 (7.3%). BRAF V600 mut in 43/74 (58.1%). Median neoadj tx duration 28 d (0–108). Median time from start of neoadj tx to surgery 67 d (29–176). LND/resection performed in 66/82 (80.5%). RECIST 1.1 (ITT, N = 82): ORR 23 (28.0%); CR 5 (6.1%), PR 18 (22.0%), SD 27 (32.9%), PD 24 (29.3%); not evaluable 8 (9.8%). Pathologic response (ITT, N = 82): pCR 28 (34.1%), near-pCR 7 (8.5%), MPR 35 (42.7%); pathology not available in 16 (19.5%). Among evaluable (n = 66): pCR 42.4%, MPR 53.0%. Achieving MPR correlated with better radiologic response by RECIST (p &lt; 0.001) and lower nodal tumor burden (fewer positive nodes; p = 0.012). With a median follow-up of 11.8 mo by reverse KM (95% CI 7.5–17.4), EFS in the ITT cohort (N = 82; events = relapse/progression/death) showed 20 events with median EFS 32.2 mo (95% CI not estimable) and RMST 29.8 mo (SE 4.71). In the pathology-evaluable set (n = 66), EFS was longer in pts achieving MPR vs non-MPR/PD (log-rank p = 0.002): median EFS 19.3 mo for non-MPR/PD (n = 31; 10 events) vs not reached for MPR (n = 35; 2 events); RMST 20.3 vs 30.1 mo. In exploratory multivariable Cox models, MPR remained strongly associated with improved EFS (HR 0.043, 95% CI 0.003–0.684; p = 0.026), while higher PLT/LYM ratio (HR 1.029, 95% CI 1.006–1.052; p = 0.015) and male sex (HR 8.74, 95% CI 1.12–68.21; p = 0.039) showed associations with inferior EFS, while time from neoadj start to surgery and receipt of adjuvant therapy were not associated with EFS. Conclusions: NST prolgolimab monotherapy achieved MPR rates above the historical benchmark with encouraging early EFS. MPR—especially pCR—was strongly associated with improved EFS, with no progressions observed among pCR pts. In exploratory analyses, neither time from neoadj start to surgery nor receipt of adjuvant therapy appeared to materially affect EFS. Clinical trial information: NCT06299878 .

Burden and site-specific spectrum of second primary malignancies after adult acute myeloid leukemia: A population-based SEER standardized incidence ratio study with latency-stratified risk profiling (2000–2022).

11070 Background: As survival after acute myeloid leukemia (AML) improves, second primary malignancies (SPMs) represent an increasingly relevant survivorship outcome. We quantified overall and site-specific SPM risk following adult AML and characterized latency-dependent patterns and demographic heterogeneity. Methods: We performed a population-based cohort analysis using SEER 17 registries (2000–2022). Adults (≥18 years) with AML were followed for incident SPMs occurring ≥2 months after AML diagnosis. Risk was summarized using standardized incidence ratios (SIRs; observed/expected) with 95% confidence intervals (CIs), overall and by cancer site, and stratified by latency (2–11, 12–59, 60–119, ≥120 months), age, sex, and race. Results: Overall, 751 SPMs were observed versus 563.87 expected (SIR 1.33; 95% CI 1.24–1.43). Solid tumors overall were not significantly increased (SIR 1.03; 0.94–1.12), but select sites showed excess risk: oral cavity/pharynx (SIR 2.35; 1.64–3.27), skin excluding basal/squamous (SIR 1.42; 1.04–1.90), kidney (SIR 1.88; 1.31–2.60), and eye/orbit (SIR 4.32; 1.18–11.06), while breast cancer was reduced (SIR 0.65; 0.47–0.86). Hematologic SPMs were markedly elevated (lymphatic/hematopoietic SIR 2.48; 2.06–2.96), including leukemia overall (SIR 4.42; 3.45–5.59) and acute lymphoblastic leukemia (SIR 21.51; 11.76–36.08). All-site risk remained elevated across latency windows (SIR range 1.30–1.38), while solid-tumor risk increased after 5 years (SIR 1.27 at 60–119 months and ≥120 months). Risk was highest in ages 0–59 (SIR 1.91), and remained elevated in males (1.37) and females (1.27). Conclusions: Adult AML survivors experience a sustained excess risk of SPMs, driven primarily by secondary hematologic malignancies with additional site-specific solid-tumor excesses emerging in longer-term follow-up, supporting risk-adapted survivorship surveillance. Site-specific risk of second primary malignancies after adult AML (SEER 17, 2000–2022). Site Observed Expected O/E CI Lower CI Upper All Sites 751 563.87 1.33# 1.24 1.43 All Solid Tumors 506 491.44 1.03 0.94 1.12 All Lymphatic and Hematopoietic Diseases 123 49.66 2.48# 2.06 2.96 Lymphoma 42 24.97 1.68# 1.21 2.27 Leukemia 70 15.83 4.42# 3.45 5.59 Acute Lymphocytic Leukemia 14 0.65 21.51# 11.76 36.08 Acute Myeloid Leukemia 13 4.71 2.76# 1.47 4.72 Chronic Myeloid Leukemia 22 2.11 10.40# 6.52 15.75

Multi-modal liquid biopsy integrating cfDNA fragmentomics and serum biomarkers for the differential diagnosis of ovarian tumors.

5573 Background: Ovarian cancer (OC) remains the leading cause of mortality among gynecological malignancies, primarily due to delayed diagnosis. Current diagnostic modalities, including imaging and serum biomarkers, exhibit suboptimal specificity, leading to unnecessary surgical interventions for benign adnexal masses. This study explores the clinical utility of cell-free DNA (cfDNA) fragmentomics, a high-dimensional genomic feature reflecting chromatin accessibility, in distinguishing malignant from benign ovarian tumors. Methods: This multi-center study enrolled 595 participants across three cohorts: a training set (n=236) and an internal validation set (n=172) from Yunnan Cancer Hospital, and an independent external validation set (n=136) from Beijing Obstetrics and Gynecology Hospital. Additionally, a prediction cohort included borderline ovarian tumors (BOT, n=51). Using whole-genome sequencing (WGS), we characterized cfDNA via four fragmentomic dimensions: copy number variations (CNVs), fragment size profiles (FSPs), nucleosome footprints (NFs), and genomic element perturbations (ARTEMIS). An ensemble model was trained to classify malignant ovarian cancer versus benign disease, and a generalized linear model was constructed by integrating fragmentomic scores with CA125 and HE4 levels to further improve the performance. Results: The cfDNA-only model demonstrated robust performance with Area Under the Curve (AUC) values of 0.937, 0.930, and 0.932 in the training, internal, and external cohorts, respectively. At 90% sensitivity in training (internal: 88.4%; external: 89.2%), specificities remained stable at 77.12%, 77.91%, and 78.87%. The model exhibited exceptional sensitivity for High-Grade Serous Ovarian Cancer (HGSOC) (&gt;90%) with sensitivity increased with stages; early-stage (I–II) sensitivities were maintained ≥77%. Integrating the cfDNA fragmentomic score with CA125 and HE4 significantly enhanced AUCs to 0.961 (training), 0.964 (internal) and 0.964 (external). While maintaining high sensitivity (~90%), the combined model substantially increased specificity to 90.1% (training). Among 37 surgically treated stage I–III patients, higher pre-operative predictive scores were significantly associated with shorter disease-free survival (DFS) (p &lt; 0.05), independent of age, stage, and histology. In BOT patients, 39.2% were classified as higher malignant-risk, consistent with an intermediate phenotype. Conclusions: The integration of cfDNA fragmentomics with traditional biomarkers provides a highly sensitive and specific non-invasive tool for the differential diagnosis of ovarian masses. This multi-modal approach significantly outperforms current clinical standards, offering a promising strategy for early detection and clinical risk stratification in ovarian cancer management.

Canadian Cancer Trials Group (CCTG) study PR21 (PLUDO): Results of crossover treatment from a randomized trial of ¹⁷⁷ Lu-PSMA-617 (LuP) vs docetaxel (DOC) in patients with metastatic castration-resistant prostate cancer (mCRPC).

5019 Background: The PLUDO trial randomized patients with mCRPC to receive either LuP or DOC, with cross-over permitted at radiographic progression (RP). At the primary analysis, there was no significant difference in the primary endpoint of 1 st line radiographic progression free survival (rPFS) (HR 1.01, 90% CI: 0.77, 1.31). However, overall survival (OS) was in favor of patients randomized to receive docetaxel first (HR 1.64, 95% CI: 1.14, 2.35) (KN Chi, et al. ESMO Congress, 2025). To provide insights into potential reasons for the survival difference, we report the prespecified secondary objective of rPFS after cross-over therapy. Methods: Multi-centre open-label randomized phase II trial. 199 patients with chemotherapy-naïve, PSMA-PET positive mCRPC progressing after ARPI therapy were randomized 1:1 to receive either LU-P 7.4 GBq IV q6 weeks or DOC 75 mg/m2 IV q3 weeks, with cross-over permitted at progression. rPFS2 was measured from randomization to RP or death after cross-over therapy, “2 nd line rPFS” was from time of start of cross-over therapy to RP or death, OS was defined from time of initial randomization to death, and “2 nd line OS” was from time of cross-over therapy to death. Results: 159 patients had an rPFS event on 1 st line therapy (LuP: n = 79, DOC: n = 80) including 24 deaths (LuP: n = 16, DOC: n = 8). At the time of 1 st rPFS, there were no substantive differences between arms for grade 3-4 adverse events, but patients on LuP reported better quality of life (FACT-P). Of the 135 patients alive, 104 received cross-over therapy (LuP → DOC: 42, DOC → LuP: 62). There were no clinically relevant differences in baseline characteristics of patients who had cross-over therapy between treatment arms. OS was worse for patients that did not cross-over compared to those who did cross-over (LuP arm: HR 3.18 (95% CI 1.93, 7.42); DOC arm: HR 4.73 (95% CI 1.96, 11.38)). For the 104 patients who received both lines of therapy, there was no differences in efficacy outcomes for 2 nd line rPFS, rPFS2, OS, or 2 nd line OS (TABLE). PSA decline ≥ 50% was higher with cross-over Doc than LuP (69% vs 40%, P = 0.004). There were 9 grade 3 adverse events related to cross-over therapy in each arm. Conclusions: In the PLUDO study, there was no difference in first-line or cross-over rPFS between LuP and DOC. In patients who received cross-over therapy, this analysis shows no OS difference, suggesting that imbalance in cross-over from LuP to Doc likely impacted the OS difference in the ITT population and emphasizes the importance of both treatments on efficacy outcomes. Clinical trial information: NCT04663997 . LuP → DOC(median, months)n = 42 DOC → LuP(median, months)n = 62 HR(DOC → LuP/LuP → DOC) 2 nd line rPFS 4.8 5.4 0.86 (90% CI: 0.58, 1.26) rPFS2 18.5 15.8 0.91 (90% CI: 0.61, 1.35) 2 nd line OS 8.1 8.3 0.94 (95% CI: 0.53-1.64) OS 23.2 20.0 0.88 (95% CI: 0.50, 1.53)

XNW27011, a Claudin 18.2(CLDN18.2) targeted antibody-drug conjugate (ADC), in patients with CLDN18.2-positive gastric/gastroesophageal junction adenocarcinoma (G/GEJA): A phase 2 study.

3036 Background: The prognosis of G/GEJA remains poor, especially in late line patients (pts). mPFS for G/GEJA pts who received ≥ 2 prior lines of standard therapy is 1.6 - 2.6 months (mos). XNW27011 is a novel antibody-drug conjugate (ADC) targeting CLDN18.2, a promising therapeutic target that is aberrantly expressed in patients with G/GEJA. Methods: This phase 2 study aims to evaluate the efficacy and safety of XNW27011. Adult pt with advanced solid tumors who has failed standard therapy or is intolerable to available standard therapy/without available standard therapy, and with CLDN18.2 expression in ≥5% tumor cells (TC) with IHC staining ≥2+ (IHC≥2+, TC≥5%, EPR19202) were enrolled, including G/GEJA pts. All patients had measurable disease (RECIST v1.1) and ECOG 0-1. Patients received XNW27011 at 2.4-4.8 mg/kg Q3W. Endpoints included ORR for efficacy, TRAE events for safety and PK parameters. Results: As of Dec 29, 2025, a total of 86 eligible G/GEJA pts were enrolled at 2.4-4.8 mg/kg. IHC≥2+, TC≥20% was chosen as the CLDN18.2 expression criteria for further development of XNW27011. Here, we report the results in G/GEJA pts with CLDN18.2 expression of IHC≥2+, TC≥20%. Safety: 71 enrolled pts had CLDN18.2 expression of IHC≥2+, TC≥20%. The median age was 57.0 years, 87.3% of the pts had received ≥2 lines of systemic therapy, 85.9% of the pts had received immune checkpoint inhibitors. The common TRAEs (≥20%) included anemia, nausea, white blood cell count decreased, decreased appetite, neutrophil count decreased, vomiting, weight loss, hypoalbuminemia, asthenia, platelet count decreased, lymphocyte count decreased and hypokalemia. Efficacy: At 3.0 mg/kg, 26 patients were evaluable for efficacy, with 23 and 3 pts had received ≥2 and 1 prior lines of systemic therapy, respectively. The median follow-up(mFU) was 11.3 mos, cORR and cDCR were 65.4% and 84.6%, respectively. mPFS was 5.7 mos and mOS was 11.7 mos. Among the 23 pts who had ≥2 prior lines of therapy, the mPFS was 6.8 mos and mOS was 11.9 mos, with the mFU of 11.5 mos. Notably, one patient previously treated with CLDN18.2-targeted therapy still demonstrated durable clinical benefit, with PFS of 11.3 mos and still remains on treatment as of data cutoff date. Pharmacokinetics: At doses from 0.6 - 6.0 mg/kg in pts with solid tumors including G/GEJA, XNW27011 exposure (C max and AUC 0-∞ ) increased in an approximately dose proportional manner with a half-life of 5~7 days. Across all dose levels, circulating blood payload concentrations were low with 3.3% ADA positive rate. Conclusions: XNW27011 demonstrated promising anti-tumor activity and a manageable safety profile in CLDN18.2-positive G/GEJA. A pivotal phase 3 study evaluating XNW27011 as a ≥3 line therapy in G/GEJA pts is currently ongoing. Clinical trial information: NCT06792435 .

CCTG BR.38: Consolidative use of radiotherapy to block (CURB2) oligoprogression in patients with metastatic non–small-cell lung cancer—A randomized phase III trial.

TPS8124 Background: Immune checkpoint inhibitor (ICI)–based regimens are standard first-line therapy for metastatic non–small-cell lung cancer (NSCLC) without actionable driver mutations; however, most patients experience disease progression within 12 months. A subset develop oligoprogression, defined by progression at a limited number of metastatic sites with otherwise controlled disease. Oligoprogression represents a therapeutic window in which local ablative therapy could be used to prolong disease control. In the prior phase II CURB trial, SBRT was shown to significantly prolong progression-free survival (PFS) in patients with oligoprogressive NSCLC. Methods: CCTG-BR38 is an international, multicenter, open-label, randomized phase III trial conducted through the NCI National Clinical Trials Network. Adults with stage IV NSCLC receiving first-line ICI ± chemotherapy who develop ≤5 extracranial oligoprogressive lesions are randomized 1:1 to: (1) switch to second-line standard systemic therapy or (2) SBRT (30 Gy in 3 fractions) to all oligoprogressive sites followed by continuation of the same first-line systemic therapy. Randomization is stratified by type of first-line systemic therapy (ICI alone vs ICI + chemotherapy), number of oligoprogressive lesions (1–2 vs 3–5), and ECOG performance status (0-1 vs 2). Patients with treated, stable brain metastases are eligible. Primary Endpoints: Dual primary endpoints are PFS and overall survival (OS). Secondary endpoints include safety (CTCAE v5.0), patient-reported adverse events (PRO-CTCAE), quality of life (EORTC QLQ-C30/LC13), and cost-effectiveness (EQ-5D-5L; Canadian sites). Optional blood and tissue collection for exploratory biomarker analyses are planned. Statistical Considerations: A total of 320 patients (160 per arm) will be enrolled. The sample size is based on detecting a HR of 0.7 for OS (an improvement in median OS from 10 to 14.3 months) with 80% power using a 1-sided 2% level test (the overall 1-sided 2.5% type I error will be assigned 2% to OS and 0.5% to PFS). Interim analyses for futility and efficacy are planned. Current Status: The trial was centrally activated on April 3, 2025. Accrual is ongoing across Canadian and US NCTN sites. Clinical trial information: NCT06686771 .

Cancer survival for rural and urban populations during the opioid epidemic: A SEER analysis.

e23065 Background: The opioid epidemic has disproportionately affected rural communities, with associated increases in overdose deaths, non-overdose medical mortality and healthcare costs. Few studies, however, have evaluated changes in cancer related mortality in the rural communities most affected by the opioid epidemic. Methods: We analyzed cause-specific five-year survival data for prostate, breast, colorectal, pancreatic, lung, and liver cancers diagnosed from 2000-2016 using the SEER Database (22 registries covering 48% of the US population). Rural and urban residence was defined using USDA Rural-Urban Continuum Codes. Ordinary least squares regression assessed temporal survival trends, and analysis of covariance compared the statistical significance in annual change in survival between rural and urban groups. Results: We created a cohort of 2.96 M people with cancer including 5 of the most commonly diagnosed cancers between 2000 and 2016. The cohort included 93% urban and 7% rural people with cancer and was 78% white and 22% non-white. There were significant differences in demographics between urban and rural patients in all cancer types. Five-year survival improved over time for all cancers except prostate; however, gains were consistently inferior for rural compared to urban populations. Rural-urban differences in survival were present at year 2000 and widened by 2016 for breast (1.8% to 2.9%), colorectal (2.1% to 3.0%), pancreatic (0.2% to 4.2%), lung (2.5% to 8.7%) and liver cancer (2.1% to 14.0%). We compared the rate of change in cancer mortality between rural and urban populations and found significant differences in all of cancers (p&lt;0.05) except for prostate cancer (p=0.57). Conclusions: From 2000 to 2016, rural-urban disparities in cancer survival widened for five of six major cancers. Over the same period, national opiate overdose rates increased in rural populations and then exceeded urban populations. Consistent with previous studies showing increases in non-overdose related mortality in populations with higher opioid exposure, our study confirms widening disparities between urban and rural populations across a broad range of cancers. These findings warrant additional efforts to understand the root causes of widening disparities for rural patients with cancer. Cause-specific 5-year survival difference (urban - rural) and difference in ordinary linear regression slopes. Urban Rural Difference in year 2000 Urban Rural Difference in year 2016 Change from 2000 to 2016 Average Yearly Change in Difference (slope of difference) Prostate 2.5% 2.6% 0.10% 0.04 Breast 1.8% 2.9% 1.10% 0.12 * Colorectal 2.1% 3.0% 0.90% 0.15 * Pancreas 0.2% 4.2% 4.00% 0.19 * Lung 2.5% 8.7% 6.20% 0.38 * Liver 2.1% 14.0% 11.90% 0.48 * *p&lt;0.05 determined by ANCOVA assessing the significance of the difference in slopes between the urban and rural groups.

Prompt engineering and readability of large language model–generated patient education materials in hematology and oncology.

e13668 Background: Patients with cancer must understand complex diagnostic and treatment concepts to make informed healthcare decisions, which disadvantages individuals with limited health literacy. To address this, the Agency for Healthcare Research and Quality and the National Institutes of Health recommend that patient-facing materials be written at a fifth- to sixth-grade reading level. However, most oncology education materials exceed this threshold. Large language models (LLMs) are increasingly used to generate patient-facing educational content and, with appropriate prompt engineering, may be able to produce materials at recommended readability levels across hematologic and solid malignancies. Methods: We generated patient-facing educational explanations for five malignancies: acute myeloid leukemia, diffuse large B-cell lymphoma, breast cancer, colorectal cancer, and metastatic lung cancer. Three LLMs were queried using predefined prompt engineering techniques with varying instructional constraint, including general prompts, explicit grade-level–targeted prompts, meta-generated prompts, and teach-back prompts. Readability of English-language outputs was assessed using the Simple Measure of Gobbledygook (SMOG) and Flesch–Kincaid Grade Level (FKGL) via a Python-based text analysis pipeline. The primary analytic goal was to evaluate whether prompt structure shifted outputs toward or below the recommended sixth-grade reading level. Differences across prompt types were evaluated using one-way ANOVA with post-hoc Tukey HSD testing. Results: Prompt type significantly influenced readability for both SMOG (F = 121.2, p &lt; 0.001) and FKGL (F = 81.2, p &lt; 0.001). General prompts produced the least readable outputs. Grade-level–targeted, meta-generated, and teach-back prompts significantly improved readability compared with general prompts (p &lt; 0.001). There were no significant differences in SMOG or FKGL between grade-level–targeted, meta, and teach-back prompts. Persona-style prompts produced higher reading grade levels than constrained and meta prompts. No prompt consistently achieved a sixth-grade reading level by SMOG, though some models reached this threshold by FKGL. Conclusions: Prompt structure meaningfully affects the readability of LLM-generated oncology patient education materials. Explicitly constrained and meta-generated prompts improve readability, whereas persona-based prompting may increase textual complexity. These findings offer practical guidance for clinicians using LLMs to support accessible, patient-centered cancer education.

Cross-platform concordance between tumor-informed and tumor-naive ctDNA MRD assays in early-stage NSCLC: A paired-sample pilot comparison of Signatera and Guardant Reveal.

e20082 Background: Circulating tumor DNA (ctDNA)–based minimal residual disease (MRD) assays are increasingly integrated into postoperative risk assessment for early-stage non–small cell lung cancer (NSCLC). Despite growing clinical adoption, real-world data directly comparing tumor-informed and tumor-naïve MRD platforms remain limited. To inform the design of a planned prospective multi-platform MRD study, we performed a pilot paired-sample comparison of two widely used technologies: Natera Signatera (tumor-informed) and Guardant Reveal (tumor-naïve). Methods: Eleven patients with NSCLC underwent same-day paired blood draws, with MRD testing conducted in parallel using Signatera and Guardant Reveal. Binary MRD calls were evaluated for concordance using overall agreement, positive and negative percent agreement (PPA, NPA), Cohen’s kappa (κ), and Gwet’s AC1. Exact binomial 95% confidence intervals (CIs) were calculated for agreement metrics. Results: Among 11 paired assessments, 8 (72.7%) were concordant. The 2×2 distribution included: both positive (n=3), both negative (n=5), Signatera-positive/Reveal-negative (n=3), and no Reveal-positive/Signatera-negative cases. Overall agreement was 72.7% (95% CI, 39.0–94.0). PPA was 50.0% (95% CI, 11.8–88.2) and NPA was 100% (95% CI, 47.8–100). Agreement was moderate by both κ (0.48) and AC1 (0.47). All discordant results favored Signatera as the positive caller. Conclusions: In this pilot evaluation, Signatera and Guardant Reveal demonstrated moderate cross-platform agreement, with high concordance among negative results but reduced alignment for positive MRD detection. The directionality of discordance driven solely by Signatera-positive/Reveal-negative results highlights fundamental technological and biological differences between tumor-informed and tumor-naïve MRD approaches. Larger, standardized prospective studies are warranted to determine comparative accuracy, define clinical interchangeability, and guide optimal MRD assay selection in early-stage NSCLC.

A single-institution retrospective analysis of the distinct differences between spontaneous malignant transformation and treatment-associated transformation in <i>IDH</i> -mutant gliomas.

e14094 Background: Malignant transformation (MT) of IDH mutant gliomas is a poorly characterized process by which a lower grade glioma evolves into a higher grade glioma. We stratified patients based on histology and treatment to examine the prevalence of MT in IDH -mutated patients, the prognostic implications of early and late MT, and the differences between spontaneous, treatment associated, and IDH mutant inhibitor associated MT. Methods: We identified 935 IDH mutant patients seen at UCLA from 1998 to 2025 who provided IRB consent. pMT was defined by a second resection with pathology indicating a worsened grade (2 to 3, 2 to 4, 3 to 4). To increase sample size, new contrast enhancement (CEnew) on MRI scans indicative of true progression (TP) was considered iMT. CEnew identified as pseudoprogression (PsP) was not included as MT. Patients were then stratified by pre-MT treatment (none/spontaneous, radiation and/or chemotherapy, or IDH mutant inhibitor only). We defined early MT as shorter than one standard deviation &lt; median ttMT, and late MT as longer than one standard deviation &lt; median ttMT. Time to MT (ttMT), overall survival (OS), and residual survival (censored patients excluded) were analyzed by Kaplan-Meier and Cox multivariate analyses. Results: In our cohort, 328/935 had MT (pMT + iMT): 88 spontaneous, 234 post chemotherapy and/or radiation, and 6 post IDH mutant inhibitor. 111 were early transformers and 26 were late transformers. Early transformers had shorter median ttMT (early=2.841, late=28.66 years, p&lt;0.0001) as expected, and shorter median OS (early=9.784, late=undefined years, p&lt;0.0001). There was no difference in residual survival after MT. Grade 2 astrocytomas had a median time to transformation of 8.55 years while grade 2 oligodendrogliomas had a median time to MT of 16.42 years. There was no difference between time to iMT versus pMT. Older age predicted shorter ttMT (HR 1.022, p=0.0003) and worse OS (HR 1.026, p&lt;0.0001) while higher KPS predicted increased OS (HR 0.983, p=0.0033). Using grade 3 astrocytoma as reference, lower grade tumors had significantly lower hazard ratios of ttMT (G2O HR 0.05438, p=0.0003; G2A HR 1.118, p=0.4740). Relative to RT/Chemo associated MT, spontaneous (HR 1.407, p=0.0131) and IDH mutant inhibitor MT (HR 2.705, p=0.0202) were associated with shorter ttMT. However, among MT patients only (censored patients excluded), only spontaneous MT was associated with a better OS (HR 0.521, p=0.0013). Residual survival was higher in G2Os (HR 0.4817, p=0.0018), but no differences were observed based on treatment received before MT. Conclusions: Treatment type, along with tumor grade and age at diagnosis, emerged as clinically relevant predictors of transformation risk and survival. Using this clinical cohort, future studies will focus on defining molecular features associated with MT.

Real-world patterns of aspirin and anticoagulant prophylaxis at IMiD-associated VTE onset in multiple myeloma: A phase-specific analysis.

e19534 Background: Immunomodulatory drugs (IMiDs) increase venous thromboembolism (VTE) risk in multiple myeloma (MM). Guidelines endorse risk-adapted prophylaxis with aspirin (ASA) or anticoagulation (AC), but real-world data on IMiD-associated VTE and prophylaxis use across treatment phases are limited. Methods: We conducted a retrospective single-center cohort study of adults with MM treated with IMiD-based regimens who developed image-confirmed VTE between 2020-2025. Clinical and treatment data, including ECOG, IMWG/R-ISS stage, IMPEDE-VTE score, line of therapy, VTE prophylaxis at the time of VTE (ASA, AC, or none), and time from diagnosis and IMiD initiation to VTE, were collected and summarized descriptively. Results: Of the 110 patients screened, 61 met the inclusion criteria and were included in the final analysis. Median age was 72 years, median BMI 30.6 kg/m², and 85% of patients had ECOG 0–1. IMWG/R-ISS stage was I/II/III in 27%/56%/17%. Median IMPEDE-VTE score at VTE was 4; 10% were high-risk (IMPEDE ≥8). At the time of VTE, 39 patients (64%) were on ASA, 7 (11%) on AC, and 15 (25%) were not on prophylaxis. Patients without prophylaxis had the highest IMPEDE scores (median 7 vs 4 with ASA/AC). Median time from MM diagnosis to VTE was 18 months, with events distributed throughout the disease course. Median time from IMiD initiation to VTE was 8 months; ~50% of events occurred within 12 months, and ~25% occurring &gt;24 months. At the time of VTE, 26 (43%) were on induction and 27 (44%) on maintenance therapy, indicating clustering of VTE in these phases. Median time from IMiD initiation to VTE varied by prophylaxis, occurring earliest with AC (3.5 months), later with ASA (13.0 months), and at 4.0 months without prophylaxis. Two-thirds of patients on AC developed VTE within 6 months of IMiD start, whereas patients on ASA showed both early events and a late cluster beyond 24 months. Patients without prophylaxis had high IMPEDE scores, and VTE occurrence did not appear to be associated with timing of disease. VTE recurrence occurred in 5 patients (8%). Conclusions: In this real-world, VTE-enriched cohort of MM patients treated with IMiD therapy, thrombotic events clustered during induction and maintenance, with higher occurrence among those receiving ASA or no prophylaxis. While incidence and comparative effectiveness cannot be determined from this VTE-only dataset, the frequent VTE events observed in patients with elevated IMPEDE-VTE scores who were not anticoagulated highlight opportunities to improve risk stratification and prophylaxis selection. These findings support a phase-conscious approach to VTE prevention, emphasizing early anticoagulant prophylaxis in selected patients and cautious de-escalation as treatment- and disease-related risks diminish.

Gallstone disease classification using SLOA-optimized CatBoost classifier with explainable AI

Gallstones are small stones that form in the gallbladder. Around 80% of individuals with gallstones do not present any symptoms. Despite the high accuracy of image-based machine learning (ML) models in the detection of gallstones, less research has been carried out regarding tabular data. This paper has discussed a publicly available tabular dataset in order to study some predictive models to determine the presence of gallstones. We have employed the Catboost (CB) classifier model and the Sea Lion Optimization Algorithm (SLOA) in our study. In this project, the primary methods that are explored include CatBoost with cross-validation and CB optimized using the SLOA with cross-validation. The CB model techniques using 5-fold cross-validation have attained a mean accuracy of 79.58%, a mean F1-score of 79.01%, a mean precision of 80.91%, and a mean recall of 77.36% using a total of 38 features. In particular, fold-1 has attained an accuracy of 86.46%, F1-score 85.39%, precision of 88.37%, and recall of 82.61% among the 5-fold cross-validation of the CB model. In addition, the SLOA_CB model of 5-fold cross-validation has achieved a mean accuracy of 80.42%, a mean F1-score of 79.94%, a mean precision of 81.97%, and a mean recall of 77.76%, using the 19 selected features. In this case, fold-4 has attained an accuracy of 87.50%, F1-score, precision, and recall of 87.23%, which implies that the classification performance is balanced. Lastly, SHAP, LIME, and DiCE have been applied to the model explainability, and the most influential features in all cases, regarding the prediction of gallstone disease, are C-Reactive Protein (CRP) and Vitamin D.

Nanotechnology-based novel approaches for the treatment of tuberculosis