JCO Clinical Cancer Informatics最新文献

Purpose: With the aid of ever-increasing computing resources, many deep learning algorithms have been proposed to aid in diagnostic workup for clinicians. However, existing studies usually selected informative patches from whole-slide images for the training of the deep learning model, requiring labor-intensive labeling efforts. This work aimed to improve diagnostic accuracy through the statistic features extracted from hematoxylin and eosin-stained slides.

Methods: We designed a computational pipeline for the diagnosis of inverted urothelial papilloma (IUP) of the bladder from its cancer mimics using statistical features automatically extracted from whole-slide images. Whole-slide images from 225 cases of common and uncommon urothelial lesions (64 IUPs; 69 inverted urothelial carcinomas [UCInvs], and 92 low-grade urothelial carcinoma [UCLG]) were analyzed.

Results: We identified 68 image features in total that were significantly different between IUP and UCInv and 42 image features significantly different between IUP and UCLG. Our method integrated multiple types of image features and achieved high AUCs (the AUCs) of 0.913 and 0.920 for classifying IUP from UCInv and conventional UC, respectively. Moreover, we constructed an ensemble classifier to test the prediction accuracy of IUP from an external validation cohort, which provided a new workflow to diagnose rare cancer subtypes and test the models with limited validation samples.

Conclusion: Our data suggest that the proposed computational pipeline can robustly and accurately capture histopathologic differences between IUP and other UC subtypes. The proposed workflow and related findings have the potential to expand the clinician's armamentarium for accurate diagnosis of urothelial malignancies and other rare tumors.

Purpose: Pathologists diagnose prostate cancer (PCa) on hematoxylin and eosin (HE)-stained sections of prostate needle biopsies (PBx). Some laboratories use costly immunohistochemistry (IHC) for all cases to optimize workflow, often exceeding reimbursement for the full specimen. Despite the rise in digital pathology and artificial intelligence (AI) algorithms, clinical implementation studies are scarce. This prospective clinical trial evaluated whether an AI-assisted workflow for detecting PCa in PBx reduces IHC use while maintaining diagnostic safety standards.

Methods: Patients suspected of PCa were allocated biweekly to either a control or intervention arm. In the control arm, pathologists assessed whole-slide images (WSI) of PBx using HE and IHC stainings. In the intervention arm, pathologists used the Paige Prostate Detect AI algorithm on HE slides, requesting IHC only as needed. IHC was requested for all morphologically negative slides in the AI arm. The main outcome was the relative risk (RR) of IHC use per detected PCa case at both patient and WSI levels.

Results: Overall, 143 of 237 (60.3%) slides of 64 of 82 patients contained PCa (78.0%). AI assistance significantly reduced the risk of IHC use per detected PCa case at both the patient level (RR, 0.55; 95% CI, 0.39 to 0.72) and slide level (RR, 0.41; 95% CI, 0.29 to 0.52). Cost reductions on IHC were €1,700 for the trial, at €50 per IHC stain. AI-assisted pathologists reported higher confidence in their diagnoses (80% v 56% confident or high confidence). The median assessment time per HE slide showed no significant difference between the AI-assisted and control arms (139 seconds v 112 seconds; P = .2).

Conclusion: This study demonstrates that AI assistance for PCa detection in PBx significantly reduces IHC costs while maintaining diagnostic safety standards, supporting the business case for AI implementation in PCa detection.

Purpose: Understanding the genetic heterogeneity of Lynch syndrome (LS) cancers has led to significant scientific advancements. However, these findings are widely dispersed across various resources, making it difficult for clinicians and researchers to stay informed. Furthermore, the uneven quality of studies and the lack of effective translation of knowledge into clinical practice create challenges in delivering optimal patient care. To address these issues, we developed and launched the Precision Lynch Syndrome Knowledge Base (PLSKB), a specialized, interactive web-based platform that consolidates comprehensive information on LS.

Methods: To create the PLSKB, we conducted an extensive literature review and gathered data from reliable sources. Through an extensive literature review and survey of other reliable sources, we have extracted prominent and relevant content with a high level of accuracy, transparency, and detailed provenance. To enhance usability, we implemented an evidence-leveling framework, categorizing studies on the basis of the type of research, reliability, and applicability to clinical care. The platform is designed to be dynamic, with updates performed monthly to incorporate the latest research.

Results: The PLSKB integrates a broad spectrum of data related to LS, including biomarkers, cancer types, screening and prevention strategies, diagnostic methods, and therapeutics options. This centralized resource is intended to support clinicians and researchers in making evidence-based decisions throughout surveillance and care processes. Its interactive design and frequent updates ensure that users have access to the most current and relevant findings.

Conclusion: The PLSKB bridges the gap between research and clinical practice by offering a reliable, up-to-date repository of evidence-based information. This tool empowers clinicians and researchers to deliver precision care and advance research for LS and related conditions, ultimately improving patient outcomes.

Purpose: To determine the accuracy of large language models (LLMs) in generating appropriate treatment options for patients with early breast cancer on the basis of their medical records.

Materials and methods: Retrospective study using anonymized medical records of patients with BC presented during multidisciplinary team meetings (MDTs) between January and April 2024. Three generalist artificial intelligence models (Claude3-Opus, GPT4-Turbo, and LLaMa3-70B) were used to generate treatment suggestions, which were compared with experts' decisions. The primary outcome was the rate of appropriate suggestions from the LLMs, compared with the reference experts' decisions. The secondary outcome was the LLMs' performances (F1 score and specificity) in generating appropriate suggestions for each treatment category.

Results: The rates of appropriate suggestions were 86.6% (97/112), 85.7% (96/112), and 75.0% (84/112) for Claude3-Opus, GPT4-Turbo, and LLaMa3-70B, respectively. No significant difference was found between Claude3-Opus and GPT4-Turbo (P = .85), but both tended to perform better than LLaMa3-70B (P = .027 and P = .043, respectively). LLMs showed high accuracy for adjuvant endocrine therapy and targeted therapy indications. However, they tended to overestimate the need for adjuvant radiotherapy and had variable performances in suggesting adjuvant chemotherapy and genomic tests.

Conclusion: LLMs, particularly Claude3-Opus and GPT4-Turbo, demonstrated promising accuracy in suggesting appropriate adjuvant treatments for patients with early BC on the basis of their medical records. Although LLMs showed limitations in validating surgery and indicating genomic tests, their performance in other treatment modalities highlights their potential to automate and augment decision making during MDTs. Further studies with fine-tuned LLMs and a prospective design are needed to demonstrate their utility in clinical practice.

Purpose: The COVID-19 pandemic disrupted normal mechanisms of health care delivery and facilitated the rapid and widespread implementation of telehealth technology. As a result, the effectiveness of virtual health care visits in diverse populations represents an important consideration. We used lung cancer screening as a prototype to determine whether subsequent adherence differs between virtual and in-person encounters in an urban, safety-net health care system.

Methods: We conducted a retrospective analysis of initial low-dose computed tomography (LDCT) ordered for lung cancer screening from March 2020 through February 2023 within Parkland Health, the integrated safety-net provider for Dallas County, TX. We collected data on patient characteristics, visit type, and LDCT completion from the electronic medical record. Associations among these variables were assessed using the chi-square test. We also performed interaction analyses according to visit type.

Results: Initial LDCT orders were placed for a total of 1,887 patients, of whom 43% were female, 45% were Black, and 17% were Hispanic. Among these orders, 343 (18%) were placed during virtual health care visits. From March to August 2020, 79 of 163 (48%) LDCT orders were placed during virtual visits; after that time, 264 of 1,724 (15%) LDCT orders were placed during virtual visits. No patient characteristics were significantly associated with visit type (in-person v virtual) or LDCT completion. Rates of LDCT completion were 95% after in-person visits and 97% after virtual visits (P = .13).

Conclusion: In a safety-net lung cancer screening population, patients were as likely to complete postvisit initial LDCT when ordered in a virtual encounter as in an in-person encounter.

Purpose: Circulating tumor DNA (ctDNA) assays are promising tools for the prediction of cancer treatment response. Here, we build a framework for the design of ctDNA biomarkers of therapy response that incorporate variations in ctDNA dynamics driven by specific treatment mechanisms. These biomarkers are based on novel proposals for ctDNA sampling protocols, consisting of frequent sampling within a compact time window surrounding therapy initiation-which we hypothesize to hold valuable prognostic information on longer-term treatment response.

Methods: We develop mathematical models of ctDNA kinetics driven by tumor response to several therapy classes and use them to simulate randomized virtual patient cohorts to test candidate biomarkers.

Results: Using this approach, we propose specific biomarkers, on the basis of ctDNA longitudinal features, for targeted therapy and radiation therapy. We evaluate and demonstrate the efficacy of these biomarkers in predicting treatment response within a randomized virtual patient cohort data set.

Conclusion: This study highlights a need for tailoring ctDNA sampling protocols and interpretation methodology to specific biologic mechanisms of therapy response, and it provides a novel modeling and simulation framework for doing so. In addition, it highlights the potential of ctDNA assays for making early, rapid predictions of treatment response within the first days or weeks of treatment and generates hypotheses for further clinical testing.

Purpose: Liquid biopsy, specifically circulating cell-free DNA (cfDNA), has emerged as a powerful tool for cancer early diagnosis, prognosis, and treatment monitoring over a wide range of cancer types. Computational modeling (CM) of cfDNA data is essential to harness its full potential for real-time, noninvasive insights into tumor biology, enhancing clinical decision making.

Design: This work reviews CM-cfDNA methods applied to clinical oncology, emphasizing both machine learning (ML) techniques and mechanistic approaches. The latter integrate biological principles, enabling a deeper understanding of cfDNA dynamics and its relationship with tumor evolution.

Results: Key findings highlight the effectiveness of CM-cfDNA approaches in improving diagnostic accuracy, identifying prognostic markers, and predicting therapeutic outcomes. ML models integrating cfDNA concentration, fragmentation patterns, and mutation detection achieve high sensitivity and specificity for early cancer detection. Mechanistic models describe cfDNA kinetics, linking them to tumor growth and response to treatment, for example, immune checkpoint inhibitors. Longitudinal data and advanced statistical constructs further refine these models for quantification of interindividual and intraindividual variability.

Conclusion: CM-cfDNA represents a pivotal advancement in precision oncology. It bridges the gap between extensive cfDNA data and actionable clinical insights, supporting its integration into routine cancer care. Future efforts should focus on standardizing protocols, validating models across populations, and exploring hybrid approaches combining ML with mechanistic modeling to improve biological understanding.

Purpose: Many Natural Language Processing (NLP) methods achieve greater performance when the input text is preprocessed to remove extraneous or unnecessary text. A technique known as text segmentation can facilitate this step by isolating key sections from a document. Give that transformer models-such as Bidirectional Encoder Representations from Transformers (BERT)-have demonstrated state-of-the-art performance on many NLP tasks, it is desirable to leverage such models for segmentation. However, transformer models are typically limited to only 512 input tokens and are not well suited for lengthy documents such as cancer pathology reports. The Longformer is a modified transformer model designed to intake longer documents while retaining the positive characteristics of standard transformers. This study presents a Longformer model fine-tuned for cancer pathology report segmentation.

Methods: We fine-tuned a Longformer Question-Answer (QA) model on 504 manually annotated pathology reports to isolate sections such as diagnosis, addenda, and clinical history. We compared baseline methods including regular expressions (regex) and BERT QA. However, those methods may fail to correctly identify section boundaries. Model performance was evaluated using sequence recall, precision, and F1 score.

Results: Final test results were obtained on a hold-out test set of 304 cancer pathology reports. We report sequence F1 scores for the following sections: diagnosis (0.77), addenda (0.48), clinical history (0.89), and overall (0.68).

Conclusion: We present a fine-tuned Longformer model to isolate key sections from cancer pathology reports for downstream analyses. Our model performs segmentation with greater accuracy.

Purpose: To comprehensively analyze the association between preoperative maximum standardized uptake value (SUV_max) on 18F-fluorodeoxyglucose positron emission tomography-computed tomography and postoperative recurrence in resected non-small cell lung cancer (NSCLC) using machine learning (ML) and statistical approaches.

Patients and methods: This retrospective study included 643 patients who had undergone NSCLC resection. ML models (random forest, gradient boosting, extreme gradient boosting, and AdaBoost) and a random survival forest model were developed to predict postoperative recurrence. Model performance was evaluated using the receiver operating characteristic (ROC) AUC and concordance index (C-index). Shapley additive explanations (SHAP) and partial dependence plots (PDPs) were used to interpret model predictions and quantify feature importance. The relationship between SUV_max and recurrence risk was evaluated by using a multivariable Cox proportional hazards model.

Results: The random forest model showed the highest predictive performance (ROC AUC, 0.90; 95% CI, 0.86 to 0.97). The SHAP analysis identified SUV_max as an important predictor. The PDP analysis showed a nonlinear relationship between SUV_max and recurrence risk, with a sharp increase at SUV_max 2-5. The random survival forest model achieved a C-index of 0.82. A permutation importance analysis identified SUV_max as the most important feature. In the Cox model, increased SUV_max was associated with a higher risk of recurrence (adjusted hazard ratio, 1.03 [95% CI, 1.00 to 1.06]).

Conclusion: Preoperative SUV_max showed significant predictive value for postoperative recurrence after NSCLC resection. The nonlinear relationship between SUV_max and recurrence risk, with a sharp increase at relatively low SUV_max values, suggests its potential as a sensitive biomarker for early identification of high-risk patients. This may contribute to more precise assessments of the risk of recurrence and personalized treatment strategies for NSCLC.