Citations

PURPOSE Large language models (LLMs) have demonstrated remarkable versatility in oncology applications, such as cancer staging and survival analysis. Despite their potential, ethical concerns such as data privacy breaches, bias in training data, lack of transparency, and risks associated with erroneous outputs pose significant challenges to their adoption in high-stakes oncology settings. Therefore, we aim to explore the ethical challenges associated with LLM-based applications in oncology and evaluate emerging techniques designed to address these issues. METHODS Following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses framework, a systematic review was conducted to evaluate publications related to ethical issues of LLMs in oncology across eight academic databases (eg, PubMed, Web of Science, and Embase) between January 1, 2019, and December 31, 2024. RESULTS The search retrieved 4,319 published articles, of which 65 publications were preserved and included in our analysis. We identified six prevalent ethical challenges in oncology, including trust, equity, privacy, transparency, nonmaleficence, and accountability. We then evaluated emerging technical solutions to mitigate ethical challenges and summarized evaluation metrics used to assess these solutions' effectiveness. CONCLUSION This review provides actionable recommendations for responsibly deploying LLMs in oncology, ensuring adherence to ethical guidelines, and fostering improved patient outcomes. By bridging technical and clinical perspectives, this review offers a foundational framework for advancing ethical artificial intelligence applications in oncology and highlights areas for future research.

Background Birt–Hogg–Dubé syndrome (BHD) is a rare autosomal dominant disease characterised by the development of lung cysts, skin fibrofolliculomas and an increased risk for renal cancer. The progression of cystic lung disease with ageing in BHD has not been well characterised. Objectives To assess the severity of lung involvement in patients with BHD aged 50 years or older. Methods A retrospective computer-assisted search was performed to identify patients with BHD seen at Mayo Clinic in Rochester, Minnesota and the First Affiliated Hospital of University of Science and Technology of China from January 2000 to May 2023. Demographic, clinical, pulmonary function and radiological parameters were extracted and analysed. Results Among 82 patients with BHD aged 50 years or older, 49 (59%) were female and the median age was 57 years (range, 50–79) at baseline. Pneumothorax occurred in 46 patients (56%). Lung cysts were observed on CT in 77 patients (94%) and were bilateral in nearly all cases. The number of cysts was≥10 for most patients (84%) and the median size of the largest cyst was 31 mm (range, 4–160 mm). Median (range) % predicted values of forced vital capacity, forced expiratory volume in 1 s and diffusion capacity of the lung for carbon monoxide were 98 (62–147), 92 (56–145) and 85 (61–133), respectively, for those with data available. During follow-up (median, 41 months; range, 3–259), 35% of patients manifested worsening in pulmonary function or cystic lung disease. Conclusion Lung involvement in older patients with BHD manifests varying severity of cystic lung disease with relatively mild functional impairment, if any, and indolent progression.

Background: Lung cancer ranks as the leading cause of cancer-related mortality worldwide. The complexity of tumor delineation, crucial for radiation therapy, requires expertise often unavailable in resource-limited settings. Artificial Intelligence (AI), particularly with advancements in deep learning (DL) and natural language processing (NLP), offers potential solutions yet is challenged by high false positive rates. Purpose: The Oncology Contouring Copilot (OCC) system is developed to leverage oncologist expertise for precise tumor contouring using textual descriptions, aiming to increase the efficiency of oncological workflows by combining the strengths of AI with human oversight. Methods: Our OCC system initially identifies nodule candidates from CT scans. Employing Language Vision Models (LVMs) like GPT-4V, OCC then effectively reduces false positives with clinical descriptive texts, merging textual and visual data to automate tumor delineation, designed to elevate the quality of oncology care by incorporating knowledge from experienced domain experts. Results: The deployment of the OCC system resulted in a 35.0% reduction in the false discovery rate, a 72.4% decrease in false positives per scan, and an F1-score of 0.652 across our dataset for unbiased evaluation. Conclusions: OCC represents a significant advance in oncology care, particularly through the use of the latest LVMs, improving contouring results by (1) streamlining oncology treatment workflows by optimizing tumor delineation and reducing manual processes; (2) offering a scalable and intuitive framework to reduce false positives in radiotherapy planning using LVMs; (3) introducing novel medical language vision prompt techniques to minimize LVM hallucinations with ablation study; and (4) conducting a comparative analysis of LVMs, highlighting their potential in addressing medical language vision challenges.

Abstract The roles of nucleoplasm-residing nucleoporins (NUP) in solid tumors, including prostate cancer, remain unknown. In this study, we reveal the clinical significance and mechanistic role of the off-pore NUP, soluble POM121 (sPOM121), as a crucial transcriptional regulator that enhances the aggressiveness of metastatic prostate cancer. Using orthogonal methodologies in human samples, sPOM121 was identified as the predominantly expressed nucleoplasmic NUP in prostate cancer. Unbiased proteomic and epigenomic studies demonstrate that sPOM121, through its C-terminus, interacts with the chromatin remodeler SMARCA5 at gene promoter sites and localizes at nuclear condensates, reprogramming gene expression. Indeed, sPOM121 regulates a distinct oncogenic gene network, including β-catenin, leading to prostate cancer progression and immune evasion. Importantly, targeting the sPOM121/β-catenin axis in patient-derived preclinical and syngeneic mouse models halts prostate cancer aggressiveness and enhances antitumor immunity. Taken together, these findings reveal previously unknown actionable reprogramming functions of off-pore NUPs in solid tumors. Significance: This study uncovers how oncogenic signaling programs are transcriptionally heightened by the NUP sPOM121 in metastatic prostate cancer. Localization of sPOM121 at active transcriptional nuclear condensates propels disease progression and immune evasion, offering novel anticancer therapeutic opportunities.

CDK12 primarily functions as a transcription regulatory cyclin-dependent kinase (CDK) that controls mRNA elongation, splicing, and polyadenylation. The CDK12 gene is implicated in human cancers since it is frequently mutated and/or deleted in prostate and ovarian cancer but paradoxically amplified in breast cancer. Here, we demonstrate that CDK12 promotes serine-933 phosphorylation of DNA2, a nuclease/helicase critical for replication fork stress regulation, and the phosphorylation subsequently facilitates DNA2 polyubiquitination and degradation mediated by the APC/C CDC20 E3 ubiquitin ligase. CDK12 inactivation induces but amplification suppresses genome-wide expression of interferon response and antigen processing and presentation machinery genes in ovarian and breast cancer cells, respectively. Besides causing aberrant DNA2 stabilization, replication stress, genomic instability, and cytosolic double-stranded DNA (dsDNA) accumulation, CDK12 loss also triggers cGAS–STING activation and innate immune response, which can be reversed by forced expression of replication protein A (RPA) subunits or DNA2 depletion. Our findings identify DNA2 as a phosphorylation substrate of CDK12, connecting CDK12 to cell cycle regulation. These data also reveal DNA2 protein destruction as a critical mechanism that dictates genomic instability, cGAS–STING signaling activation, and innate immune response in CDK12-deregulated cancers.

Importance Tumor-infiltrating lymphocytes (TILs) are a provocative biomarker in melanoma, influencing diagnosis, prognosis, and immunotherapy outcomes; however, traditional pathologist-read TIL assessment on hematoxylin and eosin–stained slides is prone to interobserver variability, leading to inconsistent clinical decisions. Therefore, development of newer TIL scoring approaches that produce more reliable and consistent readouts is important. Objective To evaluate the analytical and clinical validity of a machine learning algorithm for TIL quantification in melanoma compared with traditional pathologist-read methods. Design, Setting, and Participants This multioperator, global, multi-institutional prognostic study compared TIL scoring reproducibility between traditional pathologist-read methods and an artificial intelligence (AI)-driven approach. The study was conducted using retrospective cohorts of patients with melanoma between January 2022 and June 2023 across 45 institutions, with tissue evaluated by participants from academic, clinical, and research institutions. Participants were selected to ensure diverse expertise and professional backgrounds. Main Outcomes and Measures Intraclass correlation coefficient (ICC) values were calculated for the manual and AI-assisted arms using log-transformed data. Kendall W values were calculated for Clark scores (brisk = 3, nonbrisk = 2, and sparse = 1). Reliabilities of ICC and W values were classified as moderate (0.40-0.60), good (0.61-0.80), or excellent (>0.80). AI TIL measurements were dichotomized using the 16.6 and median cutoffs. Univariable and multivariable Cox regression analyses assessed the prognostic value of TIL scores adjusted for clinicopathologic variables. Results There were 111 patients with melanoma in the independent testing cohort (median [range] age at diagnosis, 61.0 [25.0-87.0] years; 56 [50.5%] male) who contributed melanoma whole tissue sections. A total of 98 participants evaluated TILs on 60 hematoxylin and eosin–stained melanoma tissue sections. All 40 participants in the manual arm were pathologists, while the AI-assisted arm included 11 pathologists and 47 nonpathologists (scientists). The AI algorithm demonstrated superior reproducibility, with ICCs higher than 0.90 for all machine learning TIL variables, significantly outperforming manual assessments (ICC, 0.61 for AI-derived stromal TILs vs Kendall W , 0.44 for manual Clark TIL scoring). AI-based TIL scores showed prognostic associations with patient outcomes (n = 111) using the median cutoff approach with a hazard ratio (HR) of 0.45 (95% CI, 0.26-0.80; P = .005), and using the cutoff of 16.6, with an HR of 0.56 (95% CI, 0.32-0.98; P = .04). Conclusions and Relevance In this prognostic study of TIL quantification in melanoma, the AI algorithm demonstrated superior reproducibility and prognostic associations compared with traditional methods. Although the retrospective nature of the cohorts limits demonstration of clinical utility, the publicly available dataset and open-source AI tool offer a foundation for future validation and integration into melanoma management.

The translation of AI-generated brain metastases (BM) segmentation into clinical practice relies heavily on diverse, high-quality annotated medical imaging datasets. The BraTS-METS 2023 challenge has gained momentum for testing and benchmarking algorithms using rigorously annotated internationally compiled real-world datasets. This study presents the results of the segmentation challenge and characterizes the challenging cases that impacted the performance of the winning algorithms. Untreated brain metastases on standard anatomic MRI sequences (T1, T2, FLAIR, T1PG) from eight contributed international datasets were annotated in stepwise method: published UNET algorithms, student, neuroradiologist, final approver neuroradiologist. Segmentations were ranked based on lesion-wise Dice and Hausdorff distance (HD95) scores. False positives (FP) and false negatives (FN) were rigorously penalized, receiving a score of 0 for Dice and a fixed penalty of 374 for HD95. The mean scores for the teams were calculated. Eight datasets comprising 1303 studies were annotated, with 402 studies (3076 lesions) released on Synapse as publicly available datasets to challenge competitors. Additionally, 31 studies (139 lesions) were held out for validation, and 59 studies (218 lesions) were used for testing. Segmentation accuracy was measured as rank across subjects, with the winning team achieving a LesionWise mean score of 7.9. The Dice score for the winning team was 0.65 ± 0.25. Common errors among the leading teams included false negatives for small lesions and misregistration of masks in space. The Dice scores and lesion detection rates of all algorithms diminished with decreasing tumor size, particularly for tumors smaller than 100 mm3. In conclusion, algorithms for BM segmentation require further refinement to balance high sensitivity in lesion detection with the minimization of false positives and negatives. The BraTS-METS 2023 challenge successfully curated well- annotated, diverse datasets and identified common errors, facilitating the translation of BM segmentation across varied environments and providing the tools for future development of personalized volumetric reports to patients undergoing BM treatment.

PURPOSE To report the results of the multicenter, open-label IRAKLIA trial (ClinicalTrials.gov identifier: NCT05405166 ) of isatuximab subcutaneous (SC) versus intravenous (IV), plus pomalidomide and dexamethasone, in relapsed/refractory multiple myeloma (MM), to our knowledge, the first phase III MM trial using an on-body injector (OBI). METHODS Patients with ≥1 prior line of therapy were randomly assigned 1:1 to Isa OBI (1,400 mg) or IV (10 mg/kg) once weekly in cycle (C)1 and then every 2 weeks, plus pomalidomide (4 mg once daily, day [D]1-21) and dexamethasone (40 mg once weekly [age ≥75: 20 mg]) and treated until progression, unacceptable toxicity, or patient request. Coprimary end points were overall response rate (ORR; noninferiority margin, 0.839) and Isa C trough (C6D1 predose; noninferiority margin, 0.8). Noninferiority of OBI versus IV was demonstrated if both coprimary end points achieved noninferiority. RESULTS IRAKLIA randomly assigned 531 patients (OBI, n=263; IV, n=268). After 12-month median follow-up, the ORR was 71.1% (OBI) and 70.5% (IV; relative risk, 1.008 [95% CI, 0.903 to 1.126]; lower CI exceeded noninferiority margin). The mean (standard deviation) C6D1 C trough was 499 (259) μg/mL (OBI) and 340 (169) μg/mL (IV). The C trough geometric mean ratio (90% CI) was 1.532 (1.316 to 1.784); lower CI exceeded noninferiority margin. Grade ≥3 treatment-emergent adverse event incidences were 81.7% (OBI) and 76.1% (IV); infusion reaction incidences were 1.5% and 25.0%. Injection site reactions occurred in 0.4% of OBI injections (all grade 1-2); 99.9% of injections completed without interruption. CONCLUSION IRAKLIA demonstrated efficacy and pharmacokinetic noninferiority between Isa OBI and IV. No unexpected safety signal was observed, with excellent local tolerability of Isa OBI. Efficacy and safety were comparable with Isa IV in ICARIA-MM, except the lower OBI infusion reaction rate. These results support potential use of the OBI, designed to improve practice efficiency.