Abstract Bispecific antibodies are novel immunotherapies approved for the treatment of patients with relapsed or refractory multiple myeloma (MM). While these therapies induce deep and durable responses, they are associated with adverse events, such as cytokine release syndrome and immune effector cell–associated neurotoxicity syndrome, that require specific protocols for monitoring, prevention, and treatment. Due to the potential risk of cytokine release syndrome and immune effector cell–associated neurotoxicity syndrome, current US labeling recommends inpatient administration during step-up dosing of bispecific antibodies, which typically occurs in academic centers and can often restrict the number of patients that can be treated. The administration of bispecific antibodies in the outpatient and community setting can enhance the patient experience and reduce resource use, which is a barrier to the widespread uptake of these treatments. With bispecific antibodies being evaluated in earlier lines of therapy for MM, the broader use of these agents will underscore the need for more accessible and less resource-intensive options that ensure safe outpatient administration. In this Viewpoint, we aim to provide practical recommendations built from our own real-world perspectives and case examples for institutions seeking to implement remote patient monitoring (RPM) strategies to support the outpatient dosing of bispecific antibodies in patients with MM. We discuss how the use of RPM has the potential to improve adverse event detection and management, improve patient outcomes, and decrease resource use and costs. While specific approaches will vary based on the needs of each treatment center, our experiences suggest that RPM implementation can be safe and feasible for both academic and community practices.
BackgroundIdentification of metastasis status in non–small cell lung cancer (NSCLC) is a critical part of understanding disease prognosis, treatment courses, trial eligibility, and population-level cancer surveillance. However, metastasis status is inconsistently recorded in structured cancer registry fields, since manual abstraction of clinical notes is often an error-prone process. This challenge highlights an opportunity for leveraging large language models (LLMs) to conduct high-scale metastasis extraction from real-world clinical documentation. ObjectiveWe conducted a retrospective, multicohort comparative evaluation of 3 distinct LLMs for 2 independent classification tasks: overall metastasis presence at any site and brain/central nervous system (CNS) metastasis presence. We evaluated model performance on 2 independent NSCLC cohorts: one with registry labels and another with manual annotations. We further explored whether our methods could analyze clinical documentation and recover missing information in structured registry fields. MethodsPatient cohorts were derived from the Winship Cancer Institute. Cohort 1 (n=579 patients; 24,887 notes across 69 note types; 2023-2025) used registry-linked metastasis fields as the reference standard. Cohort 2 (n=22 patients; 644 radiology notes; 2010-2021) was drawn from 2 completed randomized trials and used dual-annotated labels (Cohen κ: 0.93 overall metastasis; 0.88 CNS metastasis) as the reference standard. We fine-tuned the GatorTron-base encoder model for each independent binary classification task. We evaluated MedGemma-27B-text and Llama 3.1-70B using zero-shot prompting. A separate cohort of 675 patients with missing registry labels was used for an exploratory missingness-recovery analysis, validated against annotations of a random subsample. Our best-performing model was used to conduct further site-specific metastasis classification, based on available registry data. ResultsMore than half (675/1260, 54%) of initially identified cohort 1 patients had missing or unknown registry metastasis labels. MedGemma demonstrated the best performance in overall metastasis classification (cohort 1: F1-score=0.80; cohort 2 patient-level: F1-score=1.0; cohort 2 note-level: F1-score=0.93). For brain/CNS metastasis, Llama3 performed best in both cohorts (cohort 1: F1-score=0.79; cohort 2 patient-level: F1-score=0.93; cohort 2 note-level: F1-score=0.86). The fine-tuned GatorTron model showed strong performance for classification of overall metastasis in cohort 1 (F1-score=0.72). Error analysis indicated that several model errors reflected incomplete registry labels, ambiguous clinical language, or missing documentation. In the exploratory recovery analysis, model predictions agreed with manual annotations at an accuracy of 0.90 and F1-score of 0.89. Additional site-specific metastasis classification by MedGemma demonstrated high performance for liver metastasis (F1-score=0.82) and bone metastasis (F1-score=0.79). ConclusionsAll models demonstrated relatively high performance. The zero-shot generative models were more robust to nuanced documentation and context-dependent brain/CNS metastasis extraction. The fine-tuned encoder model demonstrated strong classification performance but may have been limited by potential inaccuracies in the registry reference standards during model training. This study further demonstrated the potential of LLMs in recovering plausible structured labels from narrative text, complementing cancer registries for metastasis ascertainment.
BackgroundThyroid cancer is often labeled a “good cancer” due to high survival rates. However, many patients diagnosed with differentiated thyroid cancer (DTC) treated with radioactive iodine (RAI) experience short- and long-term symptoms as a result of treatment. These include salivary dysfunction, dry mouth, dry eyes, and nausea. Despite these challenges, patients report limited knowledge and unmet informational needs regarding RAI symptoms and symptom management. Evidence-based, patient-centered education to support symptom management during DTC survivorship is needed. Our team developed Radioactive Iodine Support (RAI Support), a prototype of a 4-week web-based intervention for RAI symptom management among survivors of DTC. ObjectiveThis study aimed to evaluate the usability of the RAI Support intervention among survivors of DTC who completed RAI treatment and assess preliminary effect sizes of change in RAI symptom burden, self-efficacy for managing chronic symptoms, informational needs, global physical and mental health, and health-related quality of life (HRQOL). MethodsWe conducted a longitudinal usability testing study with 8 participants over a 4-week period. At baseline (T1), participants completed an online survey and a 90-minute virtual usability testing session. During usability testing sessions, study staff used a think-aloud protocol to elicit feedback on the content, language, and visual appeal of the prototype across 5 tasks. Participants also completed the System Usability Scale (SUS). Participants were encouraged to continue using the intervention over the 4-week period, during which use data were recorded. A follow-up after assessment (T2) was completed at the end of the 4-week period of home use. Survey outcomes included RAI symptom burden, self-efficacy for managing chronic symptoms, informational needs, global physical and mental health, and HRQOL. ResultsAmong the 8 survivors of DTC who had undergone RAI treatment (age: mean 45, SD 8.2 years), the mean SUS score was 88.1 (SD 8.7), indicating above-average usability. All participants effectively completed each task, with ease of use ranging from 100% to 87.5% and error rates ranging from 0% to 25%. Preliminary r effect size signals of change were found for improved self-efficacy, global mental health, and HRQOL (r values ranged from 0.31 to 0.45). ConclusionsSurvivors of DTC reported above-average usability of the RAI Support prototype and completed each task successfully. Variability in error rates and ease of use suggests that refinements are needed to further optimize usability and patient engagement. RAI Support also shows preliminary promise in improving self-efficacy for symptom management, global mental health, and HRQOL among survivors of DTC. Findings support future pilot testing.
BackgroundCervical nodal metastasis remains one of the strongest determinants of treatment planning and prognosis in oral squamous cell carcinoma. Histopathology remains the reference standard, but it does not provide immediate information at the point of surgical specimen assessment. Elastic scattering spectroscopy (ESS) offers rapid, label-free optical interrogation of tissue microstructure, and may support ex vivo nodal assessment. ObjectiveThis study aimed to evaluate the diagnostic accuracy of ESS combined with machine learning classification for ex vivo detection of cervical lymph node metastasis in patients with oral squamous cell carcinoma. MethodsWe conducted a single-center observational ex vivo diagnostic modeling study in patients with oral squamous cell carcinoma undergoing primary tumor resection and neck dissection. A total of 3 ESS spectra were acquired from each bivalved lymph node before routine histopathological processing. Internal validation used repeated nested 5-fold cross-validation grouped by patient, so every spectrum and lymph node from a patient remained in the same fold. All preprocessing, scaling, dimensionality reduction, within-model-family configuration, and hyperparameter tuning, probability calibration where required, and decision-threshold selection were restricted to training data within the nested validation process; comparisons between candidate model families were exploratory and were based on pooled patient-grouped out-of-fold predictions. Spectrum-level probabilities were aggregated within each lymph node, and histopathology was the reference standard. ResultsA total of 52 patients were assessed for eligibility, 45 were included, and 280 fresh lymph nodes were assessed by ESS; 96 were pathology positive, and 184 were pathology negative. Of 840 acquired spectra, 760 passed quality control. Of the 280 nodes, 274 retained at least 2 usable spectra and comprised the analytical cohort; 6 pathology-negative nodes yielded no analyzable node-level ESS result. The hybrid model was selected because it had the highest pooled patient-grouped out-of-fold node-level area under the receiver operating characteristic curve (AUROC), with an internally validated, exploratory estimate of 0.966 (95% CI 0.929‐0.988). The same out-of-fold predictions informed model-family selection and performance reporting; this estimate does not represent independent validation of a prospectively fixed final model. At the cross-validation–derived summary threshold, secondary descriptive operating characteristics were sensitivity 93.8% (95% CI 86.4%‐97.6%), specificity 95.5% (95% CI 90.1%‐98.3%), accuracy 94.9% (95% CI 90.4%‐97.3%), positive predictive value 91.8% (95% CI 82.9%‐96.3%), and negative predictive value 96.6% (95% CI 92.4%‐98.8%). The Brier score was 0.063, with calibration-in-the-large of –0.03 and a calibration slope of 0.94. Patient-level sensitivity was 96% and specificity was 90%. Pathology-positive and pathology-negative nodes also showed differences across amplitude-related and shape-related spectral features. ConclusionsESS showed promising ex vivo discrimination of metastatic cervical lymph nodes in this single-center model-development and internal-validation study. The findings support further investigation of ESS as a potential pathology-triage adjunct rather than a replacement for histopathology.
Abstract Background Uninterrupted delivery of radiation therapy is critical to cancer treatment quality. Unplanned radiation treatment interruptions compromise tumor control and survival. Although previous studies have explored clinical and sociodemographic factors influencing treatment adherence, the role of psychosocial distress and related symptoms remains unclear. Objective This study aims to identify patient-reported psychosocial predictors of radiation treatment interruptions from patient-reported distress screening data integrated with clinical and socioeconomic correlates via traditional and machine learning–enhanced modeling. Methods This institutional cohort study included 1953 patients who underwent external beam radiation therapy between August 2020 and May 2024. Patient demographics, clinical characteristics, treatment details, and baseline distress levels were extracted from a hospital-level electronic health record data repository. Distress was assessed using the Distress Thermometer developed by the National Comprehensive Cancer Network. These data were linked with census tract–level socioeconomic indicators. Predictors of treatment interruption (measured as missed treatment days) were evaluated using negative binomial regression and an extreme gradient boosting (XGBoost) model with a Poisson objective function. Results The study cohort missed a mean of 1.19 (SD 2.18) treatment days (range: 0-15). Higher distress scores, the presence of learning or communication barriers, and genitourinary malignancy were significantly associated with increased interruption risk in the negative binomial regression model. The XGBoost model confirmed these features as top contributors to prediction performance, while yielding an incrementally lower root mean squared error of 2.19 (vs 2.21 for the regression model). Conclusions Patient-reported distress and health education barriers were modestly associated with treatment interruptions, underscoring the importance of incorporating psychosocial assessments into radiation oncology workflows. Both parametric and machine learning models provided stable and closely aligned predictions of missed radiation treatment days. Key predictors identified as statistically significant in the regression model were also among the top contributors in the XGBoost model, reinforcing the robustness of these findings and supporting integration of both approaches into future intervention strategies.
