Colorectal cancer is one of the most common cancers, but the current staging system is limited by the variability and paradoxical survival outcomes. Body composition is an accurate predictor of survival and can be extracted from routine computed tomography (CT) scans used in cancer diagnosis. However, despite its potential, the adoption of body composition analysis has been limited due to the challenges in generating the data. In this study, we propose a deep learning–based model that combines clinical and body composition biomarkers to predict the survival of colorectal cancer patients. Our best model, which integrates both clinical and body composition features, achieved a time-dependent concordance-index score of 0.7298 ( p < 0.001 ), demonstrating a significant improvement over models based solely on clinical or body composition biomarkers, indicating that models combining body composition and clinical markers could improve survival prediction. Additionally, we observed that increased skeletal muscle tissue area and radiodensity were associated with reduced mortality risk, while higher radiodensities of visceral and subcutaneous adipose tissues were associated with increased risk.

BACKGROUND:Cancer survivors experience excess morbidity and functional decline compared with age-matched individuals without cancer, potentially reflecting acceleration of biological aging during cancer treatment. Epigenetic clocks derived from DNA methylation (DNAm) quantify biological aging, but whether changes in these measures during chemotherapy are modifiable remains unclear. METHODS:We evaluated longitudinal changes in epigenetic age acceleration among adults with stage II-III colon cancer undergoing adjuvant chemotherapy in the FORCE randomized trial. Participants were assigned to usual care or resistance training during chemotherapy. Blood DNAm was assessed at baseline and follow-up to derive multiple clocks, including DNAmPhenoAge, GrimAge, and DunedinPACE acceleration measures. Body composition and physical function were measured longitudinally. Linear mixed-effects models estimated changes in epigenetic age acceleration and intervention effects. RESULTS:DunedinPACE acceleration showed the largest increase during treatment (0.70 SD per 6 months, 95% CI: 0.41, 0.98), followed by GrimAge acceleration (0.29 SD per 6 months, 95% CI: 0.13, 0.46; approximately 1.2 years). At baseline, higher DunedinPACE acceleration was associated with greater adiposity and poorer physical function. Resistance training attenuated increases in GrimAge acceleration: usual care participants showed significant acceleration (0.45 SD per 6 months, 95% CI: 0.24, 0.60), whereas intervention participants did not. Changes in epigenetic age acceleration were not strongly associated with body composition or physical function changes. CONCLUSIONS:Epigenetic age acceleration increased during colon cancer treatment, with resistance training attenuating GrimAge acceleration. These intervention-responsive changes suggest DNAm-based aging measures may capture treatment-period physiologic changes not fully reflected by body composition or physical function.

Obesity is linked to poor breast cancer outcomes, but evidence within molecular intrinsic subtypes is limited. We measured visceral and subcutaneous adipose tissue (VAT and SAT) on routine clinical imaging and examined survival and tumor microenvironment (TME) differences across PAM50 subtypes. We sampled 1,377 individuals diagnosed with stage II-III breast cancer at Kaiser Permanente Northern California between 2005 and 2015, enriched for HER2 + and ER- tumors. VAT and SAT areas (cm2) were quantified at L3-level from clinically acquired CT scans. Tumor RNA profiling (NanoString BC360™) provided PAM50 and TME-related signatures. Cox models estimated associations of VAT and SAT with overall and disease-specific outcomes. Differential expression and signature analyses were stratified by PAM50 subtype. Participants had a mean (± SD) age of 56 ± 13 years. Intrinsic subtype distribution was: Basal-like (31

e13654 Background: Timely cancer case ascertainment is critical for supporting clinical trials, research, and operational workflows. Despite being the gold standard, information on people diagnosed with cancer from accredited cancer registries is often delayed well over one-year post-diagnosis. Advances in LLMs offer promising solutions but require high computational and GPU power. A novel strategy combining rule-based NLP with curated data sources and selective LLM use may offer a practical alternative. We developed a real-time CDM and evaluated two approaches: an NLP-only method and a hybrid method for identifying incident cancer cases and classifying key oncology characteristics. Methods: We identified 230,500 patients with no cancer history who had pathology reports or cancer diagnoses between 07/01/2023 and 12/31/2023 in Kaiser Permanente Northern California (KPNC). All records were processed using eMaRC, a CDC rule-based NLP model widely used by cancer registries. For the hybrid method, pathology reports indicating malignancy or suspicious findings were further analyzed using MedGemma, a generative LLM trained on medical datasets and applied using structured prompts. Both models classified malignancy, primary site, histology, and behavior. When models disagreed on, MedGemma results were prioritized. We evaluated accuracy by comparing cancer cases against those recorded in KPNC Cancer Registry, which conforms to SEER/NAACCR standards. Sensitivity, specificity, PPV, and NPV were calculated for 1631 breast, 851 colorectal, 677 lung, 687 melanoma, and 1358 prostate cancers cases, which together represent 60% of annual cases recorded in the registry. Results: Both methods – eMaRC only and hybrid approach – demonstrated high specificity ( > 99%). Adding MedGemma to eMaRC resulted sensitivities of 97%–99% across all cancer types, an average 2% increase over eMaRC alone. Sensitivity improved the most for prostate cancer (95% to 99%), followed by lung cancer (94% to 97%). Among 230,500 patients, true case prevalence ranged from 0.3% to 0.7%. Using the hybrid approach, was 91% for breast and prostate, ranged from 84%-86% for melanoma and colorectal, and lowest for lung at 71%. In practice, cases classified as breast and prostate are generally reliable, while lung cancer classification needed further review due to a higher rate of false positives. Conclusions: A hybrid method combining rule-based NLP, curated data sources, and selective LLM use achieved overall high sensitivity and specificity. Although less detailed than the cancer registry and prone to misclassification of certain cancers, the CDM’s rapid and automated process is highly scalable and efficient, making it valuable for operational use and research requiring rapid case identifications.

BACKGROUND:Frailty and sarcopenia predict postoperative morbidity, but the benefit from enhanced recovery after surgery in high-risk patients is unclear. This study assessed whether frailty is linked to early ambulation and nutrition and whether achievement of these milestones reduces the risk of adverse outcomes among frail older adults. METHODS:In a retrospective cohort study of 5,634 adults aged ≥50 years undergoing inpatient colorectal resection within an integrated health system from 2015 to 2020, frailty was measured using the Hospital Frailty Risk Score; a subset of frail patients was further stratified using computed tomography-derived muscle quantity and quality. Early ambulation and early nutrition were defined as occurring within 12 postoperative hours. The primary outcome was 30-day major morbidity; the secondary outcome was nonhome discharge. Generalized estimating equations were adjusted for demographic and clinical covariates. RESULTS:Frail patients (Hospital Frailty Risk Score >15) were less likely than nonfrail patients to achieve early ambulation (42% vs 73%) and early nutrition (31% vs 53%). Among frail patients, early ambulation was associated with a greater absolute reduction in 30-day major morbidity (risk difference, -12%; 95% confidence interval, -17% to -7%) than among nonfrail patients (risk difference, -3%; 95% confidence interval, -6% to 0%). Similar patterns were observed for nonhome discharge and for early nutrition. Sarcopenia further identified a high-risk subgroup within frail patients. CONCLUSION:Although frail patients are less likely to achieve enhanced recovery after surgery milestones, achievement of these milestones is associated with greater reductions in adverse outcomes. Combined assessment of frailty and sarcopenia may help identify high-risk older adults most likely to benefit from personalized perioperative interventions.

Abstract Background: Adiposity is linked to adverse breast cancer outcomes, but mechanisms connecting adipose signaling to tumor progression remain unclear. Adipokines like leptin and adiponectin, and their receptors, may influence tumor behavior via paracrine and autocrine interactions across tumor, stromal, and immune cells in the breast tumor microenvironment (TME). Using high-plex spatial molecular imaging, we mapped adipokine expression and cell type-specific signaling networks to reveal adiposity-related differences in tumor biology. Methods: Single-cell spatial transcriptomics was performed using the CosMx Spatial Molecular Imager (Human 6K Panel) on breast tumors and matched adjacent tissues. Cells were segmented and classified into tumor epithelium, stroma, and immune compartments via an AI framework. LEP, LEPR, ADIPOQ, ADIPOR1, and ADIPOR2 were quantified to derive compartment-specific positivity rates, mean/very high expression, potential 5-gene co-expression, and composite adipokine signature scores and expression was compared with non-tumor cells and cells from adjacent tissues, and correlated with select clinical variables. Results: In 259 women with stage II-III breast cancer (49% postmenopausal, 47% Black, 53% Hispanic, 62% ER-positive, 17% HER2-positive), adipokine and receptor expression was consistently higher in tumor cells than stromal or immune cells. All five adiposity genes showed elevated expression in tumor cells versus stroma and immune cells. In the tumor epithelial compartment, ADIPOR1, ADIPOR2, and LEPR exhibited markedly high-end expression versus adjacent epithelium (P <10−16), defining a tumor-intrinsic adipokine receptor program. Tumor epithelial 5-gene signature scores and co-expression of ≥2 adipokine genes were enriched in a subset of tumors and modestly inversely associated with tumor size. In stroma, ADIPOR1, ADIPOR2, and LEPR formed a tightly coordinated module; stromal ADIPOR1 positivity and higher stromal adipokine signature scores were enriched in higher-grade and ER-negative tumors, indicating a grade- and subtype-linked stromal adipokine phenotype. In the immune compartment, ADIPOR1 and ADIPOR2 were broadly detectable at lower intensity, correlated with LEPR and immune-specific adipokine scores, consistent with a distinct immune-focused axis. Across compartments, 5-gene signatures captured coherent TME-wide adipokine programs remodeled in tumor versus adjacent tissue and showing compartment-specific associations with BMI, grade, ER status, and tumor size. Conclusion: Single-cell spatial profiling reveals a compartmentalized adipokine architecture in the breast TME, including a heightened tumor epithelial receptor program, a grade-linked stromal module, and an immune-associated axis that together form integrated adipokine signatures strongly connected to clinicopathologic features. Citation Format: Samarth Singhal, Amber Rockson, Hanina Hibshoosh, Parin Shah, Fatemeh Derakhshan, Diane Chen, Alireza Salem, Benjamin Izar, Kevin L. Gardner, Coral O. Omene, Daniel Fernandez, Adrienne L. Castillo, Emma Armstrong, Ijeamaka Anyene Fumagalli, Elizabeth M. Cespedes Feliciano, Sandeep K. Singhal, Adana A. M. Llanos. Cell type-specific adipokine signaling networks in the breast tumor microenvironment uncovered by high-plex spatial imaging [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2026; Part 1 (Regular Abstracts); 2026 Apr 17-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(7 Suppl):Abstract nr 4961.

e12566 Background: GP78 (AMFR) protein expression is associated with breast cancer (BC) progression and poor outcomes (PMC9310521), but traditional bulk protein quantification ignores spatial architecture within the tumor microenvironment (TME). We developed the GP78-Spatial Aggressiveness Index (GP78-SAI), and AI-integrated, spatially informed metric derived from multi-gene GP78 protein-based signatures. We hypothesized that the GP78-SAI would outperform genomics-markers, demonstrate race-independent prognostic utility, and be feasibly for use in resource-limited clinical settings. Methods: GP78-SAI was computed using spatial autocorrelation modeling across 651 breast tumor tissue microarray cores from patients with stage II-III primary invasive breast cancer treated with curative-intent surgery (diagnosed between 2005-2020) and validated using CosMx 6K plex data. Tumor cell neighborhoods were defined using k-nearest neighbors (k = 10) with row-standardized spatial weights. Ancestry-optimized multi-gene GP78 cytoplasmic signatures were derived for Black/African American (AA) compared to White/European American (EA) patient cohorts and applied across the same datasets. The resulting scoring system was designed to be computationally efficient and compatible with standard immunohistochemistry workflows. Results: GP78-SAI robustly stratified tumor aggressiveness, with significantly higher spatial clustering in high-grade (Grade 3) tumors versus low-grade (Grade 1/2) tumors. Using AA-optimized signatures, mean GP78-SAI was 0.196 in high-grade (HG) versus 0.136 in low-grade tumors (adjusted P = 0.0003); using EA-optimized signatures, mean GP78-SAI was 0.173 versus 0.126, respectively (adjusted P = 0.0064). In contrast, the single-gene AMFR marker demonstrated minimal spatial organization (mean ≈0.02) and no association with tumor grade (adjusted P = 0.77). Despite ancestry-specific signature derivation, GP78-SAI demonstrated consistent, race-independent discriminatory performance. Conclusions: GP78-SAI is a novel spatial aware AI biomarker that captures an aggressive tumor organization phenotype not detected by conventional molecular assays. By translating spatial protein architecture into a quantifiable metric, GP78-SAI improves tumor risk stratification while supporting equitable treatment prioritization. These findings support clinical integration of GP78-SAI into digital pathology workflows to advance precision oncology across diverse patient populations.

STUDY OBJECTIVES:Insomnia is highly prevalent among postmenopausal women and is associated with adverse health outcomes, highlighting the need to identify modifiable determinants of sleep. Diet and sleep are interrelated; however, few studies have evaluated longitudinal associations of complete dietary patterns with incident insomnia in postmenopausal women. This study evaluates prospective associations of established diet quality metrics with insomnia in the Women's Health Initiative Observational Study (WHI-OS). METHODS:The WHI-OS enrolled 93 676 postmenopausal women from across the United States. Alternate Mediterranean (aMed) diet and Dietary Approaches to Stop Hypertension (DASH) diet quality scores were quantified from a Food Frequency Questionnaire at baseline and dichotomized scores using a data-driven approach. Insomnia was assessed at baseline and Year 3 using the WHI Insomnia Rating Scale. Multivariable logistic regression models adjusted for sociodemographic, lifestyle, and health factors evaluated associations of baseline diet quality with incident insomnia and longitudinal insomnia status (stable/new onset insomnia vs. stable absence/remission of insomnia). RESULTS:Among women without insomnia at baseline (n = 50 644), good vs. poor diet quality at baseline was associated with lower risk for incident insomnia at 3-year follow-up (OR [95% CI], aMed: 0.925 [0.879-0.974], p=.003; DASH: 0.937 [0.891-0.985], p=.01]). In longitudinal analyses (n = 74 513), greater baseline adherence to aMed and DASH diets related to 6.3% (0.903-0.971) and 8.5% (0.883-0.948) lower odds, respectively, of having stable or new onset insomnia over 3 years (both p<.005). CONCLUSIONS:Better diet quality predicts lower insomnia risk in postmenopausal women. Clinical trials are needed to determine whether strategies to enhance diet quality improve insomnia symptoms. CLINICAL TRIAL INFORMATION:The Women's Health Initiative Observational Study is registered on Clinicaltrials.gov #NCT00000611: https://clinicaltrials.gov/study/NCT00000611.

BACKGROUND AND AIMS:Cancer therapy-related cardiac dysfunction (CTRCD) is a complication of contemporary oncologic treatment and a contributor to incident heart failure (HF) in cancer survivors. Although certain potentially cardiotoxic cancer therapies are known to increase risk, population-based estimates in large, diverse, and contemporary cohorts remain limited. The aim of the Kaiser Permanente Cardiovascular Health Enhancement and Monitoring for Oncology (KP CHEMO) study was to determine the incidence, timing, and treatment-specific variation in CTRCD within an integrated US health system. METHODS AND RESULTS:We conducted a retrospective cohort study of adult Kaiser Permanente Northern California (KPNC) members diagnosed with malignant tumours between 2012 and 2022 who received anthracyclines, human epidermal growth factor receptor (HER2) inhibitors, immune checkpoint inhibitors (ICIs), or tyrosine kinase inhibitors. CTRCD was defined as a >10% decline in left ventricular ejection fraction to <53% or incident HF identified by natural language processing. Cumulative incidence rates were calculated overall and by drug class. Early CTRCD was ≤12 months and late was >12 months. Among 26 646 patients (mean age 62 ± 14 years; 64% women; 57% non-Hispanic White), the cumulative incidence of CTRCD was 8.4% (95% confidence interval 7.7-9.1). Incidence was highest with HER2 inhibitors (10.7%) and lowest with ICIs (5.2%) (P < .001). Nearly half of all events occurred within the first year. CONCLUSIONS:CTRCD was common and occurred predominantly within the first year after therapy initiation, potentially reflecting both early susceptibility and more intensive early surveillance. Variation across drug classes highlights differing cardiotoxic risk profiles. These findings support risk prediction models and targeted surveillance strategies to reduce downstream HF risk.

Cancer-associated cachexia (CAC) is a chronic wasting disease typically associated with advanced cancer, resulting in progressive and debilitating loss of function and poor tolerance to anticancer therapy. Preclinical animal models have identified various potential mechanisms and mediators, which have had limited translational success in clinical trials. This review focuses on human studies and discusses the clinical phenotyping of CAC using imaging-derived body composition, quality-of-life and functional measures, existing evidence for mediators, current therapeutic options, and future directions to advance the field. Identifying mechanisms driving CAC and targeting them are expected to improve the quality of life, treatment efficacy, and survival. SIGNIFICANCE:CAC represents a significant clinical unmet need. Despite its high prevalence and associated mortality and morbidity, there are currently no globally approved effective therapies. This review provides a comprehensive overview of human studies aimed at defining CAC clinically and identifying mediators underlying it that are revealing effective health interventions. Furthermore, we highlight ongoing international efforts to advance our understanding of CAC.

Supplementary Table 6. Sex-Specific Adjusted Associations of Body Composition Measurements (Per SD Increase) with Reduced RDI and Relative Changes (%) in the Number of Moderate and Severe Adverse Events

BACKGROUND:The cancer-anorexia-cachexia syndrome (CACS) is a common and debilitating wasting disorder characterized by loss of skeletal muscle and worse morbidity and mortality. In pre-clinical studies, CACS is associated with loss of peroxisome proliferator-activated receptor alpha (PPAR-α) dependent ketone production in the liver. Fibrates are PPAR-α agonists that are commonly used to treat dyslipidemia. Treating mice with fibrates was found to prevent skeletal muscle loss. We examine whether patients with cancer treated with PPAR-α agonists experience less CACS. METHODS:We performed a retrospective cohort study of patients (N = 6922) at Memorial Sloan Kettering Cancer Center who were diagnosed with non-small cell lung cancer (NSCLC) between 2002 and 2017 and were incidentally prescribed fenofibrate or gemfibrozil at the time of diagnosis. These patients were compared to a propensity score-matched control set who were not taking either drug. The primary outcome included a composite outcome of CACS, which included significant weight loss before or after the time of diagnosis. Secondary outcomes included change in cross-sectional skeletal muscle area over time as measured in serial CT imaging studies and overall survival. Descriptive statistics, Kaplan-Meier analysis and multivariable logistic regression were performed to compare outcomes between the two groups. RESULTS:Among patients with NSCLC, 149 were taking fenofibrate or gemfibrozil at the time of diagnosis. A 2:1 propensity score-matched cohort of 298 patients was created that was well-matched with regard to baseline characteristics. Regarding the primary composite outcome, there was no significant difference in the prevalence of CACS between those taking fibrates and propensity-matched controls (49.7 vs. 46.6%). When skeletal muscle mass was measured directly using cross-sectional imaging, patients on fibrates were found to have lost significantly less muscle area over time (-3.3 vs.-4.2%, p = 0.03). There was no difference in overall survival between groups. CONCLUSION:Patients with NSCLC taking fibrates at the time of diagnosis lost less muscle area over time. In a secondary analysis, this change was not associated with a change in overall survival, though this study was likely underpowered for this analysis.

Background: Excess adiposity is associated with higher rates of breast cancer recurrence and increased mortality. Insights into the mechanisms by which adiposity influences these outcomes may lead to improved disease management. A histologic marker of white adipose inflammation, crown-like structures of the breast (CLS-B), is associated with excess adiposity, adipocyte hypertrophy, and increased in-breast aromatase levels. CLS-B are comprised of dead or dying adipocytes surrounded by macrophages, and may be measured in slides obtained from breast surgical specimens. We are conducting a prospective cohort study of these inflammation-related markers and breast cancer outcomes. We describe here select correlates of CLS-B and adipocyte diameter in 630 women diagnosed with breast cancer in the Pathways Study. Methods: The Pathways Study is a prospective cohort study with 4,504 women who were diagnosed with invasive breast cancer from late 2005 through May 2013 in the Kaiser Permanente Northern California healthcare system. From women who underwent mastectomy, formalin-fixed paraffin-embedded tissue blocks from definitive surgery enriched in white adipose tissue were obtained, sectioned and stained by hematoxylin & eosin and CD68 immunohistochemistry. Slides were then examined for adipocyte diameter and CLS-B. Blood specimens collected around the time of enrollment into the cohort were also assayed for high-sensitivity C-reactive protein (hsCRP). Data related to cancer diagnosis, demographic characteristics, anthropometry, and other variables were obtained from baseline in-person visits, surveys, and electronic health records. Results: To date, 630 women have complete data on adipocyte diameter and presence of CLS-B. Most women were diagnosed with Stage I-III disease, 79% had estrogen-receptor positive-tumors, and 19% had HER2-positive tumors. Mean and median adipocyte diameters were 103.6 and 103.4 µ, respectively, with an inter-quartile range of 94.4 to 113.9 µ. CLS-B were observed in 267 (42.4%) of the 630 women, and among these women, median CLS-B density was 0.3 per cm2, with an interquartile range of 0.16 to 0.77 per cm2. In univariate associations, there was some suggestion that adipocyte diameter varies by self-reported race and ethnicity (p=0.056), with largest mean values for Blacks (n=50, adipocyte diameter 110.4 µ) and smaller for Whites, Asians, and Latinas (n=390 and 102.8 µ; n=97 and 103.8 µ; and n=76 and 103.7 µ, respectively). These differences aligned with differences in body mass index (BMI) by race and ethnicity, and were strongly associated with BMI (p<0.0001), with adipocyte diameter across increasing BMI categories (kg/m2) of 95.8 µ for 18.5≤BMI<25, 106.2 µ for 25≤BMI<30; 109.0 µ for 30≤BMI<35, and 111.7 µ for 35≤BMI. Similar associations were observed for waist circumference and waist-to-hip ratio, as well as hsCRP (all p<0.0001). Presence of CLS-B differed somewhat across race and ethnicity (p=0.097), with 58.0% of Blacks, 39.2% of Whites, 49.5% of Asians, and 38.2% of Latinas having CLS-B present. Presence of CLS-B was also strongly associated with BMI, waist circumference, waist-to-hip ratio, and hsCRP. For example, across the increasing BMI categories above, the proportions of women with CLS-B were 27.2%, 44.8%, 51.3%, and 63.6%., and across increasing hsCRP quartiles were 22.5%, 34.5%, 53.6%, and 56.9%. Summary: These observations indicate that breast adipose tissue inflammation and adipocyte size correlate with anthropometric measures of adiposity and systemic inflammation, and vary by race and ethnicity. As we continue to obtain measurements of these variables, we will determine how they relate to more precise body composition compartments and whether they are also associated with recurrence, mortality, and other outcomes after breast cancer. Citation Format: Lawrence Kushi, Lia L. D'Addario, Domenick J. Falcone, Dilip Giri, Thaer Khoury, Chi-Chen Hong, Heather Greenlee, Warren Davis, Rochelle Payne Ondracek, Janise M. Roh, Daniel F. Fernandez, Elizabeth M. Cespedes Feliciano, Bette J. Caan, Christine B. Ambrosone, Marilyn L. Kwan, Neil M. Iyengar. Correlates of Breast Adipose Inflammation after Primary Breast Cancer Diagnosis in the Pathways Study [abstract]. In: Proceedings of the San Antonio Breast Cancer Symposium 2024; 2024 Dec 10-13; San Antonio, TX. Philadelphia (PA): AACR; Clin Cancer Res 2025;31(12 Suppl):Abstract nr P1-03-18.

Supplementary Table 1. The Comparisons of Patient Characteristics in All (N = 178), CT (N = 170), DXA (N = 162), and D3Cr (N = 118) Groups

Supplementary Table 4. The Adjusted Associations of Body Composition Measurements (Per SD Increase) with Reduced RDI Counting for Each Chemotherapy Agent

Total adiposity measured by abdominal computed tomography (CT) at the third lumbar vertebrae (L3) has been associated with breast cancer survival, but most patients undergo chest CT. If adipose tissues at the thoracic level, including those surrounding the thoracic organs, are associated with survival, they could be used to inform care for significantly more breast cancer patients. We included 2127 individuals aged 18-< 90, diagnosed with stage II-III breast cancer at Kaiser Permanente Northern California (2005–2019). Cross-sectional areas of adiposity were quantified at the fourth thoracic vertebrae (T4) and L3. Using multivariable Cox models, we estimated hazard ratios (HRs) and 95

Supplementary Table 2. Sex-Specific Standard Deviations of Body Composition Measurements (CT SMA, DXA ALM, D3Cr Muscle, CT TAT, and DXA TBF)

PURPOSECancer recurrence in clinical settings is documented in unstructured text, requiring labor-intensive manual record review to extract this outcome. A shareable natural language processing model developed at Dana-Farber Cancer Institute (DFCI)-DFCI-imaging-student-efficiently extracts cancer outcomes from radiology reports. We applied this model in a community oncology setting, aggregating report-level predictions to derive patient-level outcomes, and evaluated its performance in determining recurrence and time-to-recurrence in patients with breast cancer (BC) or colorectal cancer (CRC).METHODSWe randomly sampled 200 patients with BC and 200 patients with CRC from two cohorts at Kaiser Permanente Northern California. Patients were diagnosed with stage III disease (2005-2019) and followed until July 31, 2024, death, or disenrollment. We manually reviewed recurrence (local/regional/distant), recurrence date, and sites of recurrence using oncology, radiology, and pathology information in electronic health records. We then applied the DFCI-imaging-student model to radiology reports and compared recurrence based on the model outcomes against manual review.RESULTSA total of 7,195 radiology reports were processed. During a median follow-up of 8.4 years for BC and 6.8 years for CRC, manual review identified 78 recurrence cases in BC (39%) and 70 in CRC (35%). The DFCI-imaging-student model demonstrated high sensitivity and specificity for recurrence detection in both cancers (breast: 92.3% and 92.6%, CRC: 94.3% and 86.9%) and moderate-to-high accuracy in identifying the sites of distant metastasis. Among true positives, the median error in time-to-recurrence was 0.16 months for breast and 0.48 months for CRC.CONCLUSIONOutcomes derived from the DFCI-imaging-student model output demonstrated high accuracy, providing an efficient determination of recurrence and time-to-recurrence in large-scale research to improve recurrence surveillance and facilitate collaborative research.

Descriptive characteristics among 4,403a women diagnosed with invasive breast cancer in the Pathways Cohort, 2006–2018