Introduction: Pneumonitis is a clinically important toxicity associated with both osimertinib and thoracic radiotherapy (TRT) in patients with metastatic EGFR NSCLC. The objective of this study was to characterize the incidence, severity, and clinical features of pneumonitis in patients receiving TRT in close temporal association with osimertinib. Materials and Methods: A retrospective study of patients who received both osimertinib and TRT was conducted at a tertiary Canadian academic cancer centre from 2016 to 2020. TRT was defined as any course of radiotherapy with lung parenchyma in the treated field. The primary endpoint was incidence and time to symptomatic pneumonitis. Variables predictive of time to pneumonitis were determined using multivariable Cox proportional hazards regression. Results: In total, 41 patients received osimertinib and TRT to 68 lesions. Five patients developed grade ≥ 2 pneumonitis (12.2%), of which one was grade 4 (2.4%) and one was grade 5 (2.4%). Median time to any-grade pneumonitis was 56 days. Among the 41 patients, TRT preceded osimertinib in 13, osimertinib was confirmed to be held during TRT in 12, confirmed to overlap TRT in 5, and status was undocumented in 11. Two of the five symptomatic cases occurred after re-irradiation, and four occurred after treatment of centrally located lesions. Exploratory multivariable analyses were highly uncertain given the five grade ≥ 2 events; the association with confirmed osimertinib overlap had a wide confidence interval (HR 3.36, 95% CI 0.13–86.56). Concluoksion: Symptomatic pneumonitis occurred in 12.2% of this selected cohort, including rare severe and fatal events. Findings are hypothesis generating and do not establish the safety or harm of overlapping osimertinib with TRT.
Circulating tumor cells (CTCs) are established prognostic biomarkers in metastatic castration-resistant prostate cancer (mCRPC); however, conventional epithelial marker-based detection platforms may miss biologically aggressive CTC subpopulations with epithelial–mesenchymal transition (EMT) or cancer stem cell (CSC)-like features. This systematic review evaluated the clinical significance of EMT- and CSC-associated CTC phenotypes in mCRPC. Following PRISMA 2020 guidelines and prospective PROSPERO registration (CRD420261308460), PubMed, Embase, and Web of Science were searched for studies published from March 2016 to March 2026. Twelve studies involving 661 patients met inclusion criteria. Because of substantial heterogeneity in CTC enrichment methods, biomarker definitions, and outcomes, findings were synthesized qualitatively. EMT-associated phenotypes, including vimentin-positive CTCs, N-cadherin-positive CTCs, EMT-transitioning CTCs, and platelet-coated CTCs, were generally associated with greater metastatic burden, treatment resistance, and shorter progression-free or overall survival (OS). CSC-associated markers, including OCT4, CD44, and ALDH1, were similarly linked to adverse outcomes and treatment resistance. Across five studies reporting OS hazard ratios, EMT- or CSC-positive CTC phenotypes generally trended toward poorer survival, although associations varied in magnitude and statistical significance. Overall, EMT- and CSC-like CTCs are associated with aggressive mCRPC biology, supporting their potential prognostic value beyond conventional CTC enumeration.
Stereotactic radiosurgery (SRS) and hypofractionated stereotactic radiotherapy (hSRT) have become a cornerstone in the management of intracranial lesions, offering high precision and ablative dose delivery in one or a limited number of treatment fractions. As clinical indications continue to expand, patients are increasingly treated for larger numbers of brain metastases in a single course and repeat SRS/hSRT courses for new and/or recurrent lesions have become more common. These evolving treatment paradigms substantially increase planning complexity, as target delineation, cumulative dose assessment, normal tissue sparing, and treatment optimization become increasingly challenging with each additional lesion and prior course of radiation. Despite its clinical success, SRS/hSRT treatment planning and delivery remain resource-intensive, requiring expert-driven target and normal structure contouring, treatment plan optimization, and rigorous quality assurance. Recent advances in artificial intelligence (AI) present an opportunity to enhance the efficiency, consistency, and patient-specific personalization of SRS/hSRT workflows while enabling the safe and scalable implementation of increasingly complex clinical applications. This review examines the current and emerging applications of AI across the entire SRS/hSRT treatment delivery pipeline, including automated image segmentation, delineation of targets and normal tissues, treatment planning, and dose prediction. Additionally, we will explore AI-enabled workflow integration and clinical decision support systems that facilitate multidisciplinary collaboration. Challenges to widespread adoption, such as data inhomogeneity, model interpretability, regulatory considerations, and the need for clinical validation, will also be explored. By summarizing current evidence and identifying key research directions, this review outlines a path toward intelligent, streamlined SRS/hSRT workflows in which AI augments clinical expertise and ultimately improves patient outcomes.
Background: This study sought to evaluate 30- and 60-day mortality following palliative whole-brain radiotherapy (WBRT) in patients with metastatic non-small cell lung cancer (NSCLC), and to identify readily accessible clinical predictors to guide therapeutic decisions. Methods: This retrospective single-centre study analyzed 392 patients treated with WBRT. Pretreatment factors were assessed using multivariable Cox regression and translated into the six-point Bedside Early Mortality Score (BEMS). The score’s ability to predict mortality within 30 and 60 days of WBRT completion was evaluated using receiver operating characteristic (ROC) analysis and compared with Lung-specific Graded Prognostic Assessment (Lung GPA) and Recursive Partitioning Analysis (RPA). Results: Overall, 115 patients (29.3%) died within 30 days and 178 (45.4%) within 60 days. The score included four parameters readily available during bedside consultation: age ≥65 years, direct inpatient transfer, and abnormal mental status at initiation of WBRT each contributed one point, while impaired mobility contributed three points. Each additional point increased the risk of mortality by 55% (HR 1.55, 95% CI 1.46–1.64). Thirty- and sixty-day mortality according to score were, respectively, 2.2% and 8.5% (0–1 points), 10% and 36.7% (2–3 points), 57.8% and 85.7% (4–5 points), and 85.8% and 90.5% (6 points). Areas under the ROC curve for 30- and 60- day mortality were 0.855 and 0.881, compared with 0.685 and 0.740 for Lung GPA and 0.833 and 0.899 for RPA. Conclusions: The Bedside Early Mortality Score may complement clinical evaluation and support shared decision making in WBRT. Independent validation is required.
Multidisciplinary tumor boards (MDTs) guide prostate cancer decisions but are strained by rising case volumes, evolving guidelines, and growing complexity. Artificial intelligence (AI) now enters the MDT at two levels. First, it generates inputs: deep-learning magnetic resonance imaging (MRI) reading matches or exceeds radiologists’ performance in the Prostate Imaging–Cancer Artificial Intelligence (PI-CAI) study; Gleason grading reaches expert-level agreement; PSMA PET quantification is automated; and multimodal pathology biomarkers predict ADT benefit. Second, decision-support systems act on MDT output: rule-based tools reproduce standard recommendations in 86–99% of cases, mostly in populations dominated by standard cases, whereas general-purpose large language models (LLMs) agree with real urological boards in 56–93% of cases, depending largely on input quality and case mix, and perform worse in metastatic disease; agreement does not establish clinical benefit. Of 33 appraised decision-support studies, 27 were at high risk of bias, 20 of them solely because the treating board served as an unblinded reference standard; explainability, regulation, and automation bias remain unresolved. We propose that AI currently serves as a supervised pre-reader, guideline checker, and triage tool. Prospective multicenter studies with patient outcomes are needed.
