JCO Clinical Cancer Informatics最新文献

{"title":"Consumer Wearable Device Measures of Gait Cadence and Activity Fragmentation as Predictors of Survival Among Patients Undergoing Chemotherapy.","authors":"Carissa A Low, Christianna Bartel, Krina Durica, Heidi S Donovan, Roby Thomas, Jennifer Fedor","doi":"10.1200/CCI-25-00111","DOIUrl":"10.1200/CCI-25-00111","url":null,"abstract":"<p><strong>Purpose: </strong>Consumer wearable devices provide new opportunities for measuring patterns of objective daily physical activity throughout cancer treatment. In addition to capturing step counts, these devices can also measure gait cadence and activity fragmentation, two metrics that may reflect functional capacity. The goal of the current study was to examine whether step count, gait cadence, and activity fragmentation predicted overall survival in patients with solid tumors.</p><p><strong>Methods: </strong>We enrolled patients (N = 213) receiving outpatient chemotherapy for any solid tumor into an observational cohort study. Patients wore a consumer wearable device to measure continuous physical activity patterns for up to 90 days and were followed for a median of 2.53 years, during which 42% of the sample died. Univariable and multivariable Cox proportional hazards regression analyses were used to evaluate associations between wearable device physical function metrics and survival.</p><p><strong>Results: </strong>In univariable analyses, higher step count (hazard ratio (HR), 0.87; <i>P</i> = .007), less activity fragmentation (HR, 1.03; <i>P</i> < .001), and faster peak gait cadence (HR, 0.81; <i>P</i> < .001) were significantly associated with lower mortality risk. Associations with activity fragmentation and gait cadence persisted after adjustment for age and cancer type and stage and after additional adjustment for clinician-rated performance status and patient-reported physical function.</p><p><strong>Conclusion: </strong>Activity fragmentation and gait cadence metrics derived from consumer wearable devices were associated with overall survival in patients receiving chemotherapy for any solid tumor. These associations remained statistically significant after adjustment for covariates, including clinician-rated performance status and patient-reported physical function. These findings suggest that wearable devices may capture important prognostic information about physical function independent of what clinicians and patients perceive.</p>","PeriodicalId":51626,"journal":{"name":"JCO Clinical Cancer Informatics","volume":"9 ","pages":"e2500111"},"PeriodicalIF":3.3,"publicationDate":"2025-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12258954/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144612408","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Benefits, Problems, and Motivations for Using the Online Patient Portal in Adolescent Oncology: Interviews With Adolescents and Parents.","authors":"Bryan A Sisk, Stephanie Chen, Christine Bereitschaft, Mark A Fiala, Lindsay J Blazin, Maya F Ilowite, Jennifer Mack, James DuBois","doi":"10.1200/CCI-25-00038","DOIUrl":"https://doi.org/10.1200/CCI-25-00038","url":null,"abstract":"<p><strong>Purpose: </strong>Communication is central to optimizing adolescent cancer care. Online patient portals are widely available tools that support communication. However, the perspectives of parents and adolescents on parental portal access has not been well studied.</p><p><strong>Methods: </strong>We performed separate semistructured interviews with adolescents with cancer and their parents, recruited from three academic pediatric cancer centers. We performed thematic analysis of benefits, problems, and motivations for parental portal use.</p><p><strong>Results: </strong>We interviewed 48 parent/adolescent dyads with cancer. Participants described the importance of allowing parents access to their child's portal, related to perceived parental needs and rights. Parental needs related to managing their child's complex medical needs. Parental rights related to their financial support for the child and their obligation to ensure their child's well-being. Although the cancer diagnosis did not change views on parental rights, it did increase parental needs for portal access. Participants described five benefits provided by portals: (1) improving parental knowledge and understanding, (2) supporting care coordination and family self-management, (3) supporting communication, (4) supporting parental roles, and (5) strengthening relationships. Participants described four problems caused by portal access: (1) complexity of portal contents and misunderstanding, (2) emotional distress, (3) loss of privacy, and (4) exacerbating family tensions. Parents described two factors influencing their portal use: (1) user experience, especially onerous enrollment processes, and (2) perceived usefulness of the portal.</p><p><strong>Conclusion: </strong>Adolescents with cancer and their parents believed that parents should be permitted access to nonsensitive clinical data in the adolescent's portal. Limiting portal access could create extra burdens on parents. Electronic health record companies and hospitals must develop technologies to permit parental access to nonsensitive information through the portal, especially in oncology.</p>","PeriodicalId":51626,"journal":{"name":"JCO Clinical Cancer Informatics","volume":"9 ","pages":"e2500038"},"PeriodicalIF":3.3,"publicationDate":"2025-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144585590","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Using Early Biomarker Change and Treatment Adherence to Predict Risk of Relapse Among Patients With Chronic Myeloid Leukemia Who Are in Remission.","authors":"J Felipe Montano-Campos, Erin Hahn, Eric Haupt, Jerald Radich, Aasthaa Bansal","doi":"10.1200/CCI-25-00003","DOIUrl":"10.1200/CCI-25-00003","url":null,"abstract":"<p><strong>Purpose: </strong>There is little guidance for decision making in chronic myeloid leukemia (CML) after patients achieve molecular remission. Our study addresses this gap by developing a risk prediction model for molecular relapse using early longitudinal factors, such as BCR::ABL1 biomarker-level changes and treatment adherence.</p><p><strong>Methods: </strong>We analyzed electronic health record data of patients with CML diagnosed between 2007 and 2019 from an integrated health system. We used a time-to-event modeling framework using a Cox proportional hazards approach where we evaluated time from molecular remission to molecular relapse. The main predictors were early changes in BCR::ABL1 levels from treatment initiation to the first follow-up measurement (typically around 3 months) and treatment adherence in the first 6 months, categorized as perfect (≥0.98) or less-than-perfect (<0.98). Model performance was assessed through five-fold cross-validation combined with 100 Monte Carlo bootstrapping iterations to ensure robustness and minimize bias.</p><p><strong>Results: </strong>Patients with early improvement in BCR::ABL1 levels had a 70% lower risk relapse (hazard ratio [HR], 0.30 [95% CI, 0.15 to 0.59]) compared with those without early molecular response. Perfect adherence during this critical early phase of treatment was associated with a 56% lower relapse risk (HR, 0.44 [95% CI, 0.22 to 0.85]). Predictive accuracy was high at 6 months (AUC, 0.90; 95% CI, 0.87 to 0.95) and 1-year postremission (AUC, 0.78; 95% CI, 0.74 to 0.81). Relapse risk was significantly higher among Black, Asian, and Hispanic patients compared with non-Hispanic White patients.</p><p><strong>Conclusion: </strong>Early biomarker trends and adherence after treatment initiation are critical for accurately predicting relapse among patients who achieve molecular remission. The proposed model addresses a gap in guidance after molecular remission and has the potential to enable personalized monitoring and optimize surveillance strategies, offering transformative potential for CML care.</p>","PeriodicalId":51626,"journal":{"name":"JCO Clinical Cancer Informatics","volume":"9 ","pages":"e2500003"},"PeriodicalIF":3.3,"publicationDate":"2025-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12236434/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144585591","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Computational Morphological Assessment of Bladder Cancer Tissue Is Prognostic of Recurrence and Overall Survival Following Transurethral Resection.","authors":"Patrick Leo, Behtash G Nezami, Mahmut Akgul, Naoto Tokuyama, Xavier Farré, Robin Elliott, Vidya S Viswanathan, Holly Harper, Gregory MacLennan, Anant Madabhushi","doi":"10.1200/CCI-24-00304","DOIUrl":"10.1200/CCI-24-00304","url":null,"abstract":"<p><strong>Purpose: </strong>Current risk assessment tools for bladder cancer following transurethral resection of the bladder tumor (TURBT) depend on pathological examination of resected tissue, with the consequent intra- and inter-reviewer variability. Improved prognostic tools could enable increased monitoring and aggressive interventions for high-risk patients while reducing the frequency of invasive testing for low-risk patients.</p><p><strong>Methods: </strong>We present an automated tumor risk assessment method based on quantitative features of nuclear pleomorphism and polarity extracted from digitized hematoxylin and eosin slides and compared this model with pathologist grading. Our model, incorporating six features, was trained to estimate overall survival risk on n = 189 patients and validated for recurrence prognosis on an independent validation set of n = 151 patients.</p><p><strong>Results: </strong>The model had an accuracy of 0.73 (95% CI, 0.66 to 0.81) in identifying patients who would have recurrence within 5 years of surgery. Within the validation set was a consensus set of patients (n = 94) on which three pathologists independently assigned the same grade and a nonconsensus set (n = 57) where they did not. The model had similar performance in the consensus and nonconsensus set, with accuracies of 0.70 (95% CI, 0.61 to 0.80) and 0.78 (95% CI, 0.67 to 0.89), respectively, and was able to recapitulate pathologist scoring on the consensus set (accuracy = 0.76).</p><p><strong>Conclusion: </strong>The results of this study suggest the need to incorporate both computerized analysis and pathologist grading into post-TURBT treatment planning.</p>","PeriodicalId":51626,"journal":{"name":"JCO Clinical Cancer Informatics","volume":"9 ","pages":"e2400304"},"PeriodicalIF":2.8,"publicationDate":"2025-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12303250/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144692386","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Analysis of Tumor Microenvironmental Features Between Primary and Synchronous Liver Metastases From Patients With Colorectal Cancers Using a Deep Learning Algorithm.","authors":"Oluwadunni E Emiloju, Nathan R Foster, Bahar Saberzadeh Ardestani, Rory L Smoot, Rondell P Graham, Rish K Pai, Frank A Sinicrope","doi":"10.1200/CCI-25-00004","DOIUrl":"https://doi.org/10.1200/CCI-25-00004","url":null,"abstract":"<p><strong>Purpose: </strong>The development of colorectal liver metastases (CRLMs) is associated with poor prognosis, and recent data suggest that metastasis to the liver is associated with resistance to immunotherapy. We characterized the microenvironment of primary colorectal carcinomas (CRCs) relative to their synchronous CRLMs using a validated segmentation algorithm (QuantCRC) that quantifies 15 distinct morphologic tumor features.</p><p><strong>Materials and methods: </strong>Adult patients with CRC with synchronous CRLM (N = 57) at Mayo Clinic were identified from the electronic health record. Routine tumor hematoxylin and eosin sections were digitized and reviewed for quality control. QuantCRC (Aiforia, Inc) was applied to digitized images to extract 15 GI pathologist predefined morphological features. Tumor features were compared between paired primary tumors and their synchronous CRLMs using the Wilcoxon signed-rank test overall and within subgroups. Linear regression models were used to find predictors of the paired differences in the distinct tumor morphological features.</p><p><strong>Results: </strong>The study included 57 patients (median age 59 years [IQR, 50-73], 51% female) with CRC primary and synchronous CRLM, and among primaries, 37 (65%) were left-sided. QuantCRC identified six of 15 morphological features that differed significantly between primaries and their CRLMs. Compared with the primary, CRLM showed reduced stroma, more high-grade tumor and necrosis, higher tumor-stromal ratio (TSR; all <i>P</i> < .05), and a trend toward increased tumor-infiltrating lymphocyte (TIL) density (<i>P</i> = .052). Synchronous CRLM from patients with left-sided primary tumors had significantly higher TSR, percent high-grade, percent necrosis, and TIL density compared with the primary (all <i>P</i> values ≤0.04). Among right-sided primaries, CRLM had a significantly reduced percent mature stroma (<i>P</i> = .0296) and a higher percent necrosis (<i>P</i> = .0027).</p><p><strong>Conclusion: </strong>Compared with the primary, synchronous CRLMs showed a higher TSR and increased high-grade tumor and necrosis, which are features of tumor aggressiveness and a metastatic phenotype.</p>","PeriodicalId":51626,"journal":{"name":"JCO Clinical Cancer Informatics","volume":"9 ","pages":"e2500004"},"PeriodicalIF":3.3,"publicationDate":"2025-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144692385","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Linear Federated Learning for Outcome Prediction With Application to Hepatocellular Carcinoma Radiotherapy.","authors":"Keyur D Shah, Harald Paganetti, Pablo Yepes, Theodore S Hong, Jennifer Y Wo, J Hannah Roberts, Eugene J Koay, Christian V Guthier, Ibrahim Chamseddine","doi":"10.1200/CCI-25-00074","DOIUrl":"10.1200/CCI-25-00074","url":null,"abstract":"<p><strong>Purpose: </strong>Federated learning (FL) enables multi-institutional predictive modeling without sharing raw patient data, preserving privacy while leveraging diverse data sets. This study evaluates the use of linear FL (LFL) as an interpretable approach to enhance sample size and generalizability in outcome prediction. As a proof of concept, we applied LFL to patients with hepatocellular carcinoma (HCC) undergoing external beam radiotherapy (EBRT), predicting hepatic toxicity and 1-year survival (SRVy1).</p><p><strong>Methods: </strong>Patient data from Massachusetts General Hospital (MGH) and Brigham and Women's Hospital (BWH) were used to train models, whereas an independent validation data set from MD Anderson Cancer Center assessed generalizability. Logistic regression was developed to predict hepatic toxicity and SRVy1 using key clinical features, including baseline albumin, bilirubin, Child-Pugh score, liver size, and mean liver dose. The LFL approach allowed each institution to train models locally without sharing raw patient data. Model performance was evaluated using the AUC and compared between the LFL model and institution-specific models.</p><p><strong>Results: </strong>For survival prediction, single-institution models were limited, with AUC = 0.55-0.63, with LFL increasing it to 0.67. For toxicity prediction, external validation showed AUC = 0.68 for the MGH model and 0.69 for the BWH model, with LFL maintaining the AUC at 0.7. The model coefficients were moderate in the LFL compared with the single-institution models, indicating an ability to mitigate bias, which was also reflected by better performance on the validation data set.</p><p><strong>Conclusion: </strong>LFL maintained or improved predictive performance over single-institution models for survival and hepatic toxicity in patients with HCC treated with EBRT while preserving model interpretability and patient privacy. These findings support LFL's role in multi-institutional collaborations.</p>","PeriodicalId":51626,"journal":{"name":"JCO Clinical Cancer Informatics","volume":"9 ","pages":"e2500074"},"PeriodicalIF":2.8,"publicationDate":"2025-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12319563/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144531078","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Deep Learning-Based Body Composition Analysis for Outcome Prediction in Relapsed/Refractory Diffuse Large B-Cell Lymphoma: Insights From the LOTIS-2 Trial.","authors":"Russ A Kuker, Juan P Alderuccio, Sunwoo Han, Mark K Polar, Tracy E Crane, Craig H Moskowitz, Fei Yang","doi":"10.1200/CCI-25-00051","DOIUrl":"https://doi.org/10.1200/CCI-25-00051","url":null,"abstract":"<p><strong>Purpose: </strong>The present study aimed to investigate the role of body composition as an independent image-derived biomarker for clinical outcome prediction in a clinical trial cohort of patients with relapsed or refractory (rel/ref) diffuse large B-cell lymphoma (DLBCL) treated with loncastuximab tesirine.</p><p><strong>Materials and methods: </strong>The imaging cohort consisted of positron emission tomography/computed tomography scans of 140 patients with rel/ref DLBCL treated with loncastuximab tesirine in the LOTIS-2 (ClinicalTrials.gov identifier: NCT03589469) trial. Body composition analysis was conducted using both manual and deep learning-based segmentation of three primary tissue compartments-skeletal muscle (SM), subcutaneous fat (SF), and visceral fat (VF)-at the L3 level from baseline CT scans. From these segmented compartments, body composition ratio indices, including SM*/VF*, SF*/VF*, and SM*/(VF*+SF*), were derived. Pearson's correlation analysis was used to examine the agreement between manual and automated segmentation. Logistic regression analyses were used to assess the association between the derived indices and treatment response. Cox regression analyses were used to determine the effect of body composition indices on time-to-event outcomes. Body composition indices were considered as continuous and binary variables defined by cut points. The Kaplan-Meier method was used to estimate progression-free survival (PFS) and overall survival (OS).</p><p><strong>Results: </strong>The manual and automated SM*/VF* indices, as dichotomized, were significant predictors in univariable and multivariable logistic models for failure to achieve complete metabolic response. The manual SM*/VF* index as dichotomized was significantly associated with PFS, but not OS, in univariable and multivariable Cox models.</p><p><strong>Conclusion: </strong>The pretreatment SM*/VF* body composition index shows promise as a biomarker for patients with rel/ref DLBCL undergoing treatment with loncastuximab tesirine. The proposed deep learning-based approach for body composition analysis demonstrated comparable performance to the manual process, presenting a more cost-effective alternative to conventional methods.</p>","PeriodicalId":51626,"journal":{"name":"JCO Clinical Cancer Informatics","volume":"9 ","pages":"e2500051"},"PeriodicalIF":3.3,"publicationDate":"2025-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144651175","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Extraction of Social Determinants of Health From Electronic Health Records Using Natural Language Processing.","authors":"Zhenghua Chen, Patricia Lasserre, Angela Lin, Rasika Rajapakshe","doi":"10.1200/CCI-24-00317","DOIUrl":"10.1200/CCI-24-00317","url":null,"abstract":"<p><strong>Purpose: </strong>Social Determinants of Health (SDoH) have a significant effect on health outcomes and inequalities. SDoH can be extracted from electronic health records (EHR) to aid policy development and research to improve population health. Automated extraction using artificial intelligence (AI) can improve efficiency and cost-effectiveness. The focus of this study was to autonomously extract comprehensive SDoH details from EHR using a natural language processing (NLP)-based AI pipeline.</p><p><strong>Materials and methods: </strong>A curated set of 1,000 BC Cancer clinical documents with concentrated SDoH information served as the reference standard for training and evaluating NLP models. Two pipelines were used: an open-source pipeline trained on the annotated medical documents and an industrial pretrained solution used as a benchmark. Three experiments optimized the first pipeline's performance, assessing the effect of including subtype word positions during training. The superior open-source pipeline was then used to extract SDoH information from 13,258 oncology documents.</p><p><strong>Results: </strong>The open-source pipeline achieved an average F1 score accuracy of 0.88 on the validation data set for extracting 13 SDoH factors, surpassing the benchmark by 5%. It excelled in detailed subtype extraction, while the benchmark performed better in identifying rarely annotated SDoH information in BC Cancer data set. Overall, 60,717 SDoH factors and associated details were extracted from BC Cancer EHR oncology documents. The most frequently extracted SDoH factors included tobacco use, employment status, marital status, alcohol consumption, and living status, occurring between 8k to 12k times.</p><p><strong>Conclusion: </strong>This study demonstrates the potential of an NLP pipeline to extract SDoH factors from clinical notes, with strong performance on limited data, although data set-specific adjustments are needed for broader application across institutions.</p>","PeriodicalId":51626,"journal":{"name":"JCO Clinical Cancer Informatics","volume":"9 ","pages":"e2400317"},"PeriodicalIF":2.8,"publicationDate":"2025-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12309507/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144700323","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Using Open-Source Large Language Models to Identify Access to Germline Genetic Testing in Veterans With Breast Cancer From Unstructured Text.","authors":"Chunyang Li, Michael Stringer, Vikas Patil, Richard Mcshinsky, Deborah Morreall, Christina Yong, Kelli M Rasmussen, Zachary Burningham, Suzanne Tamang, Carolyn S Menendez, Akiko Chiba, Haley A Moss, Sarah Colonna, Kerry Rowe, Daphne Friedman, Michael J Kelley, Ahmad Halwani","doi":"10.1200/CCI-24-00263","DOIUrl":"10.1200/CCI-24-00263","url":null,"abstract":"<p><strong>Purpose: </strong>The ability of large language models (LLMs) to identify access to germline genetic testing from unstructured text remains unknown. The Department of Veterans Affairs (VA) assessed access in Veterans with breast cancer by implementing and evaluating the performance of open-source, locally deployable LLMs (Llama 3 70B, Llama 3 8B, and Llama 2 70B) in identifying access from clinical/consult notes.</p><p><strong>Methods: </strong>We identified a cohort of 1,201 Veterans diagnosed with breast cancer between January 1, 2021, and December 31, 2022, who received cancer care within the nationwide VA system and had clinical and/or consult notes available. Notes from a subset of 200 randomly selected patients, reviewed by subject-matter experts to identify access to testing, were split into development and testing sets, and various hyperparameters and prompting approaches were applied. We evaluated LLM performance using accuracy, precision, recall, and F1, with expert consensus on the labeled subset serving as ground truth. We compared LLM-identified access distribution in the entire cohort with expert-identified access in the labeled subset using the chi-squared test.</p><p><strong>Results: </strong>Llama 3 70B achieved an F1 score of 0.912 (95% CI, 0.853 to 0.971), besting Llama 3 8B (F1: 0.811; 95% CI, 0.720 to 0.901) and significantly outperforming Llama 2 70B (F1: 0.644; 95% CI, 0.514 to 0.773; the test set target variable prevalence was 0.72.) We observed no significant difference between the performance of Llama 3 70B and that of the average individual expert reviewer, nor between LLM-identified access distribution across the entire cohort and expert-identified distribution in the labeled subset.</p><p><strong>Conclusion: </strong>An open-source, locally deployable LLM effectively and efficiently identified germline genetic testing access from clinical notes. LLMs may enhance care quality and efficiency, while safeguarding sensitive data.</p>","PeriodicalId":51626,"journal":{"name":"JCO Clinical Cancer Informatics","volume":"9 ","pages":"e2400263"},"PeriodicalIF":2.8,"publicationDate":"2025-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12303249/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144692387","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Machine Learning-Based Prediction of Clinical Outcomes in Patients With Cancer Receiving Systemic Treatment Using Step Count Data Measured With Smartphones.","authors":"Calvin G Brouwer, Branca M Bartelet, Joeri A J Douma, Leni van Doorn, Evelien J M Kuip, Henk M W Verheul, Laurien M Buffart","doi":"10.1200/CCI-25-00023","DOIUrl":"10.1200/CCI-25-00023","url":null,"abstract":"<p><strong>Purpose: </strong>This study aimed to investigate whether changes in step count, measured using patients' own smartphones, could predict a clinical adverse event in the upcoming week in patients undergoing systemic anticancer treatments using machine learning models.</p><p><strong>Methods: </strong>This prospective observational cohort study included patients with various cancer types receiving systemic anticancer treatment. Physical activity was monitored continuously using patients' own smartphones, measuring daily step count for 90 days during treatment. Clinical adverse events (ie, unplanned hospitalizations and treatment modifications) were extracted from medical records. Models predicting adverse events in the upcoming 7 days were created using physical activity data from the preceding 2 weeks. Machine learning models (elastic net [EN], random forest [RF], and neural network [NN]) were trained and validated on a 70:30 split cohort. Model performance was evaluated using the AUC.</p><p><strong>Results: </strong>Among the 76 patients analyzed (median age 61 [IQR, 53-69] years, 39 [51%] female), 11 (14%) were hospitalized during the study period. The median step count during the first week of systemic treatment was 4,303 [IQR, 1926-7,056]. Unplanned hospitalizations in the upcoming 7 days could be predicted with high accuracy using RF (AUC = 0.88), NN (AUC = 0.84), and EN (AUC = 0.83). The models could not predict treatment modifications (AUC = 0.28-0.51) or the occurrence of any clinically relevant adverse event (AUC = 0.32-0.50).</p><p><strong>Conclusion: </strong>A decline in daily step counts can serve as an early predictor for hospitalizations in the upcoming 7 days, facilitating proactive and preventive toxicity management strategies.</p>","PeriodicalId":51626,"journal":{"name":"JCO Clinical Cancer Informatics","volume":"9 ","pages":"e2500023"},"PeriodicalIF":3.3,"publicationDate":"2025-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12233178/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144531079","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}