Outcomes in Patients with Intact and Resected Brain Metastasis Treated with 5-Fraction Stereotactic Radiosurgery.

Purpose: Hypofractionated stereotactic radiosurgery (HF-SRS) with or without surgical resection is potentially a preferred treatment for larger or symptomatic brain metastases (BMs). Herein, we report clinical outcomes and predictive factors following HF-SRS. Methods and Materials: Patients undergoing HF-SRS for intact (iHF-SRS) or resected (rHF-SRS) BMs from 2008 to 2018 were retrospectively identified. Linear accelerator-based image-guided HF-SRS consisted of 5 fractions at 5, 5.5, or 6 Gy per fraction. Time to local progression (LP), time to distant brain progression (DBP), and overall survival (OS) were calculated. Cox models assessed effect of clinical factors on OS. Fine and Gray's cumulative incidence model for competing events examined effect of factors on LP and DBP. The occurrence of leptomeningeal disease (LMD) was determined. Logistic regression examined predictors of LMD. Results: Among 445 patients, median age was 63.5 years; 87% had Karnofsky performance status ≥70. Fifty-three % of patients underwent surgical resection, and 75% received 5 Gy per fraction. Patients with resected BMs had higher Karnofsky performance status (90-100, 41 vs 30%), less extracranial disease (absent, 25 vs 13%), and fewer BMs (multiple, 32 vs 67%). Median diameter of the dominant BM was 3.0 cm (interquartile range, 1.8-3.6 cm) for intact BMs and 4.6 cm (interquartile range, 3.9-5.5 cm) for resected BMs. Median OS was 5.1 months (95% confidence interval [CI], 4.3-6.0) following iHF-SRS and 12.8 months (95% CI, 10.8-16.2) following rHF-SRS (P < .01). Cumulative LP incidence was 14.5% at 18 months (95% CI, 11.4-18.0%), significantly associated with greater total GTV (hazard ratio, 1.12; 95% CI, 1.05-1.20) following iFR-SRS, and with recurrent versus newly diagnosed BMs across all patients (hazard ratio, 2.28; 95% CI, 1.01-5.15). Cumulative DBP incidence was significantly greater following rHF-SRS than iHF-SRS (P = .01), with respective 24-month rates of 50.0 (95% CI, 43.3-56.3) and 35.7% (95% CI, 29.2-42.2). LMD (57 events total; 33% nodular, 67% diffuse) was observed in 17.1% of rHF-SRS and 8.1% of iHF-SRS cases (odds ratio, 2.46; 95% CI, 1.34-4.53). Any radionecrosis and grade 2+ radionecrosis events were observed in 14 and 8% of cases, respectively. Conclusions: HF-SRS demonstrated favorable rates of LC and radionecrosis in postoperative and intact settings. Corresponding LMD and RN rates were comparable to those of other studies.

A Deep Learning-Based Computer Aided Detection (CAD) System for Difficult-to-Detect Brain Metastases.

PURPOSE: We sought to develop a computer-aided detection (CAD) system that optimally augments human performance, excelling especially at identifying small inconspicuous brain metastases (BMs), by training a convolutional neural network on a unique magnetic resonance imaging (MRI) data set containing subtle BMs that were not detected prospectively during routine clinical care. METHODS AND MATERIALS: Patients receiving stereotactic radiosurgery (SRS) for BMs at our institution from 2016 to 2018 without prior brain-directed therapy or small cell histology were eligible. For patients who underwent 2 consecutive courses of SRS, treatment planning MRIs from their initial course were reviewed for radiographic evidence of an emerging metastasis at the same location as metastases treated in their second SRS course. If present, these previously unidentified lesions were contoured and categorized as retrospectively identified metastases (RIMs). RIMs were further subcategorized according to whether they did (+DC) or did not (-DC) meet diagnostic imaging-based criteria to definitively classify them as metastases based upon their appearance in the initial MRI alone. Prospectively identified metastases (PIMs) from these patients, and from patients who only underwent a single course of SRS, were also included. An open-source convolutional neural network architecture was adapted and trained to detect both RIMs and PIMs on thin-slice, contrast-enhanced, spoiled gradient echo MRIs. Patients were randomized into 5 groups: 4 for training/cross-validation and 1 for testing. RESULTS: One hundred thirty-five patients with 563 metastases, including 72 RIMS, met criteria. For the test group, CAD sensitivity was 94% for PIMs, 80% for +DC RIMs, and 79% for PIMs and +DC RIMs with diameter <3 mm, with a median of 2 false positives per patient and a Dice coefficient of 0.79. CONCLUSIONS: Our CAD model, trained on a novel data set and using a single common MR sequence, demonstrated high sensitivity and specificity overall, outperforming published CAD results for small metastases and RIMs - the lesion types most in need of human performance augmentation.

Interrater Reliability in Toxicity Identification: Limitations of Current Standards.

PURPOSE: The National Cancer Institute Common Terminology Criteria for Adverse Events (CTCAE) v5.0 is the standard for oncology toxicity encoding and grading, despite limited validation. We assessed interrater reliability (IRR) in multireviewer toxicity identification. METHODS AND MATERIALS: Two reviewers independently reviewed 100 randomly selected notes for weekly on-treatment visits during radiation therapy from the electronic health record. Discrepancies were adjudicated by a third reviewer for consensus. Term harmonization was performed to account for overlapping symptoms in CTCAE. IRR was assessed based on unweighted and weighted Cohen's kappa coefficients. RESULTS: Between reviewers, the unweighted kappa was 0.68 (95% confidence interval, 0.65-0.71) and the weighted kappa was 0.59 (0.22-1.00). IRR was consistent between symptoms noted as present or absent with a kappa of 0.6 (0.66-0.71) and 0.6 (0.65-0.69), respectively. CONCLUSIONS: Significant discordance suggests toxicity identification, particularly retrospectively, is a complex and error-prone task. Strategies to minimize IRR, including training and simplification of the CTCAE criteria, should be considered in trial design and future terminologies.

Natural language processing for abstraction of cancer treatment toxicities: accuracy versus human experts.

OBJECTIVES: Expert abstraction of acute toxicities is critical in oncology research but is labor-intensive and variable. We assessed the accuracy of a natural language processing (NLP) pipeline to extract symptoms from clinical notes compared to physicians. MATERIALS AND METHODS: Two independent reviewers identified present and negated National Cancer Institute Common Terminology Criteria for Adverse Events (CTCAE) v5.0 symptoms from 100 randomly selected notes for on-treatment visits during radiation therapy with adjudication by a third reviewer. A NLP pipeline based on Apache clinical Text Analysis Knowledge Extraction System was developed and used to extract CTCAE terms. Accuracy was assessed by precision, recall, and F1. RESULTS: The NLP pipeline demonstrated high accuracy for common physician-abstracted symptoms, such as radiation dermatitis (F1 0.88), fatigue (0.85), and nausea (0.88). NLP had poor sensitivity for negated symptoms. CONCLUSION: NLP accurately detects a subset of documented present CTCAE symptoms, though is limited for negated symptoms. It may facilitate strategies to more consistently identify toxicities during cancer therapy.