JCO Clinical Cancer Informatics最新文献

Purpose: Patients with myelodysplastic syndromes (MDS) exhibit diverse disease trajectories necessitating different clinical approaches ranging from watch-and-wait strategies to hematopoietic stem cell transplantation. Existing risk scores like the IPSS-R or Endothelial Activation and Stress Index provide static risk stratification at diagnosis but do not capture evolving disease dynamics. We addressed this problem by introducing a dynamic, data-driven approach to repeatedly predict short-term mortality risks, across the patient's disease course.

Materials and methods: We developed a machine learning model on the basis of gradient-boosted decision trees to estimate 1-year mortality risks from both longitudinal parameters from blood values and diagnosis-based features. We trained the model on a data set of patients from the MDS Registry Düsseldorf (n = 1,024) and validated it on patients from University Hospitals Heidelberg (n = 286) and Mannheim (n = 31).

Results: Validations on independent cohorts achieved area under the receiver operating characteristic curve scores of around 0.8 and better predictive performance for 1-year mortality compared with a diagnosis-only baseline model. The model accurately predicted mortality risks as early as within the first 90 days of diagnosis. Feature importance analysis revealed clinically plausible feature-label relations, supporting interpretability. Comparison with the IPSS-R and training on 1-year AML progression revealed the advantages and generalizability of the approach.

Conclusion: This dynamic risk model enables continuous, individualized assessment of 1-year mortality risk in patients with MDS, offering a supplement to static scores used at diagnosis. Our results highlight the utility and importance of including longitudinal parameters in risk assessment analysis.

Purpose: Data sharing is necessary to advance understanding of the etiology and biology of cancer in children, adolescents, and young adults; drive therapeutic discoveries; and improve treatment outcomes. To meet this critical need, the National Cancer Institute's Childhood Cancer Data Initiative (CCDI) provides innovative, user-friendly tools and resources that enable researchers and pediatric oncologists to access and analyze the large volume of diverse childhood cancer data (over 1 million files) that has been collected and harmonized from multiple studies, including CCDI's Molecular Characterization Initiative, Pediatric MATCH, Childhood Cancer Survivor Study, etc. This article outlines how to find, request, access, download, and analyze data indexed in the CCDI Hub Explore Dashboard and Childhood Cancer Clinical Data Commons (C3DC), key components of the CCDI Data Ecosystem, to accelerate progress in pediatric cancer research.

Methods: Both CCDI resources support cohort-based analysis and use data models that include study, participant, sample, diagnosis, and treatment data. These models are updated in collaboration with field experts. Additionally, CCDI drafted a Pediatric Cancer Core common data elements list, which serves as a standard reference for researchers.

Results: The CCDI Hub is the primary access point for finding data, tools, and applications managed by CCDI. The C3DC provides harmonized, participant-level clinical data and the CCDI Hub Explore Dashboard catalogs data at the file level. These resources enable users to search for and download manifests of harmonized, de-identified participant data and build cohorts.

Conclusion: CCDI prioritizes data accessibility and interoperability and, with its resources and data, continues to aid in pediatric cancer research discovery, data-driven insights, and collaboration across the pediatric cancer community.

Purpose: Radiotherapy is a critically important cancer treatment; however, its details are often not well represented in electronic health record data sets. US Veterans' radiation courses are further distributed across a range of medical centers, both internal and external to the Veterans Health Administration (VHA), inhibiting analysis of radiotherapy treatment across this population.

Methods: We train and test a suite of supervised machine learning models for the accurate prediction of radiation course dates using billing and diagnostic codes from a combination of VHA and Centers for Medicare and Medicaid Services (CMS) databases. We use a separate heuristic algorithm to assemble course date predictions into complete radiation treatments.

Results: Our top model predicts radiation course dates with compelling accuracy (macro-average of 0.914 across classes). The retrospective application of our model and assembly algorithm to radiation procedure dates for 1,331,342 patients identified 1,526,660 predicted courses of radiotherapy.

Conclusion: The identified courses were collected into a shared resource to facilitate future VHA-based studies, and our predictive model is available for application to a wider range of non-VHA data sets, particularly those leveraging CMS data.

The emergence of state-of-the-art large language models (LLMs), which hold the ability to generalize to diverse natural language processing tasks, has led to new opportunities in health care. Oncology is especially well-suited to leverage these resources as the journeys of patients with cancer inherently yield extensive, longitudinal data sets comprising clinical narratives, pathology and radiology reports, and genomic sequencing reports. This review begins with an overview of the fundamental concepts behind LLMs, including the definitions, architecture, training paradigm, and performance optimization through prompt engineering and retrieval-augmented generation. We also take a moment to explore the newly emerging paradigm of LLMs in a multiagentic framework. We then synthesize current research on how LLMs may benefit stakeholders within the practice of oncology, including patients, oncologists, researchers, and learners. Finally, we address the limitations and risks of LLMs, including hallucinations, inherent biases, patient privacy, and clinician deskilling. While research thus far shows significant potential for LLMs to transform cancer care, necessary future directions include studies emphasizing patient stakeholder perspectives on LLM incorporation in clinical workflows, the development of relevant clinical benchmarks for LLM evaluation, a greater focus on real-world prospective testing, and deeper exploration of LLM reasoning capabilities.

Purpose: The Commission on Cancer (CoC) seeks to expand access to high-quality care through community engagement standards targeting centers' catchment areas and efforts to accredit centers in more areas including rural hospitals. Little is known about the social, environmental, and geographic characteristics of their catchment areas. To support future investigation into the impact of CoC-accredited centers, this study compares characteristics of cancer care utilization-based catchment areas, termed Cancer Service Areas (CSAs), with and without CoC-accredited centers.

Methods: Geocoded CoC-accredited centers and cancer care patient flows extracted from Medicare claims data were used to delineate CSAs using a spatially constrained community detection method. Characteristics including environmental justice index (EJI), social vulnerability index (SVI), rurality, travel time, and localization index (LI, a ratio of cancer care received by patients within a CSA) were aggregated by CSA. A logistic regression model was created to evaluate characteristics associated with the presence of a CoC-accredited center within a CSA.

Results: Six hundred sixty-eight CSAs were defined, of which 511 CSAs had at least one CoC-accredited center. CSAs with CoC-accredited centers had lower health vulnerability (odds ratio [OR], 0.65 [95% CI, 0.427 to 0.993]) and lower racial and ethnic minority status vulnerability (OR, 0.61 [95% CI, 0.424 to 0.886]), but no differences for other components of the EJI or SVI. These CSAs also had higher LIs, meaning patients remained in their local CSA for care (OR, 9.00 [95% CI, 2.408 to 33.640] for high v low LIs).

Conclusion: Minority and comorbid populations may have more difficulty accessing cancer center care, further exacerbating observed variations in cancer outcomes. Cancer centers may address this by broadening their outreach into at-risk catchment areas.

Purpose: Partial nephrectomy has been advocated as the preferred surgical approach for small kidney tumors over total nephrectomy. However, partial nephrectomy is associated with increased perioperative risk. Estimating renal function after nephrectomy can facilitate personalized patient counseling, guide surgical approach, and identify patients who could benefit from perioperative interventions. Existing prediction models have several limitations including the lack of external validation or a user-friendly tool or application, and most have used traditional statistical methods.

Methods: We used data from two academic medical institutions and machine learning (ML) methods to develop and externally validate renal function after nephrectomy-machine learning (RFAN-ML), a model to estimate long-term renal function after partial or total nephrectomy. Boruta feature selection was used to select four routinely available clinical features, specifically age, BMI, preoperative renal function, and nephrectomy type. In the training set of 1,932 patients, we compared six ML regression models representing a set of both ensemble and nonensemble ML algorithms and optimized for root mean squared error (RMSE). This model was evaluated in a test set of 1,995 patients, and the best performing model was selected as RFAN-ML.

Results: We compared RFAN-ML with existing renal function prediction benchmarks and found that RFAN-ML outperformed or had competitive performance with benchmarks on RMSE (16.6 [95% CI, 15.6 to 17.5]), R², and mean absolute error.

Conclusion: We developed and externally validated RFAN-ML, a ML model to predict renal function after nephrectomy, and have deployed our model online. RFAN-ML has the potential to improve the care and outcomes in patients with kidney tumors by informing personalized patient counseling and guiding surgical planning.

Purpose: To examine the geospatial distribution of melanoma incidence in Pennsylvania (PA), quantify its association with agriculture practices and patterns, and consider its relevance for cancer control.

Methods: The study used an ecologic design with county-level PA data on the 2017-2021 incidence of invasive melanoma among adults 50 years and older, as well as agricultural patterns and practices, ultraviolet radiation (UVR), and demographics/socioeconomics. Spatial clustering was examined using local indicators of spatial association and Getis-Ord Gi*. Separate adjacency-weighted Conway-Maxwell-Poisson models, adjusted for UVR and social vulnerability, quantified the association between melanoma and (1) cultivated and pasture/hay acreage and (2) herbicide-, insecticide-, fungicide-, and manure-treated acreage.

Results: Melanoma incidence was 57.1% greater in a 15-county cluster (P < .05) in South Central PA; eight counties were designated as metropolitan. Compared with noncluster counties, cluster counties had significantly more cultivated land (mean 19.8% v 6.9%, P < .001) and herbicide-treated land (16.8% v 6.5%, P < .001). In adjusted models, a 10% increase in cultivated land and a 9% increase in herbicide-treated acreage each independently corresponded to a 14% increase in incidence.

Conclusion: Melanoma incidence clustered in South Central PA, an area with substantial agricultural industry. However, a majority of counties in the cluster were designated as metropolitan, challenging the concept that agriculture is primarily an industry of counties designated as nonmetropolitan (rural). Agricultural practices and patterns were associated with incidence, suggesting that cancer control adopt an integrated One Health approach to concurrently address occupational, environmental, and behavioral risks. The cluster was entirely within the 28-county catchment area of the Penn State Cancer Institute, demonstrating the utility of geospatial data and analysis for cancer control by a cancer center.

Purpose: Large language models (LLMs) show promise in assisting knowledge-intensive fields such as oncology, where up-to-date information and multidisciplinary expertise are critical. Traditional LLMs risk hallucinations and reliance on static, possibly outdated data that lack domain-specific context. Retrieval-augmented generation (RAG) has emerged as a strategy to address these issues by incorporating domain-specific information from external knowledge repositories.

Methods: We evaluated 15 LLMs, including Meta Llama-2/3, generative pretrained transformer (GPT)-3.5/4/4o variants, Claude-3, Gemini-2.0, and DeepSeek-R1. In a zero-shot workflow, each LLM answered 298 scorable questions from the 2021 American College of Radiology in-training examination. We implemented a RAG pipeline (Iridium Model) that transforms user prompts into vector embeddings, queries a specialized radiation oncology database, and merges relevant text with the original prompt to form an augmented query. We compared zero-shot versus RAG-augmented performance.

Results: Larger-parameter LLMs had higher zero-shot accuracy, with six models outscoring graduating residents (P < .01). Top scorers were reasoning models GPT-4o1, o3-mini, and DeepSeek-R1, which achieved 91.6%, 86.6%, and 91.6% without RAG, respectively. Gemini-2.0 improved 6.7% (to 79.2%), Llama-3-70b 8.4% (to 75.8%), and GPT-4o 5.7% (to 85.6%) with RAG. Top scoring reasoning models surpassed graduating resident averages by 17.7%-20% (P < .01), but had no improvement or detriment with RAG. Domain-specific gains occurred in clinical, biology, and physics. Majority voting boosted aggregate accuracy when individual model performance exceeded 50%. RAG workflows and reasoning models incurred higher computational costs.

Conclusion: Radiation-oncology-specific retrieval-augmented generation pipeline enhances nonreasoning LLM performance in radiation oncology by integrating domain-specific evidence, whereas it does not improve performance of reasoning models. These findings demonstrate that RAG can elevate clinical decision support by enabling simpler, cost-effective nonreasoning models to tackle complex tasks through retrieval capabilities-an efficient alternative to extensive model training that also yields citable, evidence-based explanations.