Predicting benefit from immune checkpoint inhibitors in patients with non-small-cell lung cancer by CT-based ensemble deep learning: a retrospective study.

BACKGROUND: Only around 20-30% of patients with non-small-cell lung cancer (NCSLC) have durable benefit from immune-checkpoint inhibitors. Although tissue-based biomarkers (eg, PD-L1) are limited by suboptimal performance, tissue availability, and tumour heterogeneity, radiographic images might holistically capture the underlying cancer biology. We aimed to investigate the application of deep learning on chest CT scans to derive an imaging signature of response to immune checkpoint inhibitors and evaluate its added value in the clinical context. METHODS: In this retrospective modelling study, 976 patients with metastatic, EGFR/ALK negative NSCLC treated with immune checkpoint inhibitors at MD Anderson and Stanford were enrolled from Jan 1, 2014, to Feb 29, 2020. We built and tested an ensemble deep learning model on pretreatment CTs (Deep-CT) to predict overall survival and progression-free survival after treatment with immune checkpoint inhibitors. We also evaluated the added predictive value of the Deep-CT model in the context of existing clinicopathological and radiological metrics. FINDINGS: Our Deep-CT model demonstrated robust stratification of patient survival of the MD Anderson testing set, which was validated in the external Stanford set. The performance of the Deep-CT model remained significant on subgroup analyses stratified by PD-L1, histology, age, sex, and race. In univariate analysis, Deep-CT outperformed the conventional risk factors, including histology, smoking status, and PD-L1 expression, and remained an independent predictor after multivariate adjustment. Integrating the Deep-CT model with conventional risk factors demonstrated significantly improved prediction performance, with overall survival C-index increases from 0·70 (clinical model) to 0·75 (composite model) during testing. On the other hand, the deep learning risk scores correlated with some radiomics features, but radiomics alone could not reach the performance level of deep learning, indicating that the deep learning model effectively captured additional imaging patterns beyond known radiomics features. INTERPRETATION: This proof-of-concept study shows that automated profiling of radiographic scans through deep learning can provide orthogonal information independent of existing clinicopathological biomarkers, bringing the goal of precision immunotherapy for patients with NSCLC closer. FUNDING: National Institutes of Health, Mark Foundation Damon Runyon Foundation Physician Scientist Award, MD Anderson Strategic Initiative Development Program, MD Anderson Lung Moon Shot Program, Andrea Mugnaini, and Edward L C Smith.

Streamlining the Quantitative Metrics Workflow at a Comprehensive Cancer Center.

INTRODUCTION: The objective of the project was to describe an efficient workflow for quantifying and disseminating tumor imaging metrics essential for assessing tumor response in clinical therapeutic trials. The clinical research utility of integration of the workflow into the electronic health record for radiology reporting was measured before and after the intervention. MATERIALS AND METHODS: A search of institutional clinical trial databases was performed to identify trials with radiology department collaborators. Investigator initiated trials, or those which lacked a standardized or automated system of collaboration with the research team were selected for the study. A web based application integrated in the electronic health record platform, the Quantitative Imaging Analysis Core (QIAC) initiative was established as a divisional resource with institutional support to provide standardized and reproducible imaging metrics across the institution. The turnaround time for radiology reports before (phase 1) and after web based application workflow (phase 2) was measured. During our test period (November 2014 to June 2015), a total of 68 requests with 37 from phase 1 and 31 from phase 2 were analyzed for patients who were enrolled in prospective clinical therapeutic interventional trials. RESULTS: The mean turnaround time for generation of quantitative tumor metric results after implementation of the web based QIAC workflow (phase 2) was significantly lower than prior (phase 1) (15.9 ± 21.3 vs 31.7 ± 35.4 hours, p= 0.0005). The mean time from the scan to the preliminary assessment was 19.6 ± 25.6 hours before and significantly reduced to 8.0 ± 9.9 hours with implementation of web based QIAC workflow. CONCLUSION: Implementation of a web based QIAC workflow platform enabled significantly improved turnaround time for quantitative tumor metrics reports and enabled faster access to the reports.

Evaluation of Cancer Patients With Suspected Pulmonary Embolism: Performance of the American College of Physicians Guideline.

BACKGROUND: Accurate risk stratification of pulmonary embolism (PE) can reduce unnecessary imaging. We investigated the extent to which the American College of Physicians (ACP) guideline for evaluation of patients with suspected PE could be applied to cancer patients in the emergency department of a comprehensive cancer center. MATERIALS AND METHODS: Data from cancer patients who underwent CT pulmonary angiography (CTPA) between August 1, 2015, and October 31, 2015, were collected. We assessed each patient's diagnostic workup for its adherence to the ACP guideline in terms of clinical risk stratification and age-adjusted d-dimer level and the degree to which these factors were associated with PE. RESULTS: Of the 380 patients identified, 213 (56%) underwent CTPA indicated per the ACP guideline, and 78 (21%) underwent CTPA not indicated per the guideline. Only one of the patients who underwent nonindicated CTPA had a PE. Fifty-seven patients underwent unnecessary d-dimer evaluation, and 71 patients with negative d-dimer test results underwent nonindicated CTPA. PEs were found in 6 of 108 (6%) low-risk patients, 22 of 219 (10%) intermediate-risk patients, and 13 of 53 (25%) high-risk patients. The ACP guideline had negative predictive value of 99% (95% confidence interval: 93%-100%) and sensitivity of 97% (95% confidence interval: 86%-100%) in predicting PE. CONCLUSION: The ACP guideline has good sensitivity for detecting PE in cancer patients and thus can be applied in this population. Compliance with the ACP guideline when evaluating cancer patients with suspected PE could reduce the use of unnecessary imaging and laboratory studies.

Preoperative Nomogram to Risk Stratify Patients for the Benefit of Trimodality Therapy in Esophageal Adenocarcinoma.

PURPOSE: To develop a nomogram that estimates 1-year recurrence-free survival (RFS) after trimodality therapy for esophageal adenocarcinoma and to assess the overall survival (OS) benefit of esophagectomy after chemoradiotherapy (CRT) on the basis of 1-year recurrence risk. METHODS: In total, 568 consecutive patients with potentially resectable esophageal adenocarcinoma who underwent CRT were included for analysis, including 373 patients who underwent esophagectomy after CRT (trimodality therapy), and 195 who did not undergo surgery (bimodality therapy). A nomogram for 1-year RFS was created using a Cox regression model. The upper tertile of the nomogram score was used to stratify patients in low-risk and high-risk groups for 1-year recurrence. The 5-year OS was compared between trimodality and bimodality therapy in low-risk and high-risk patients after propensity score matching, respectively. RESULTS: Median follow-up for the entire cohort was 62 months. The 5-year OS in the trimodality and bimodality treatment groups was 56.3% (95% confidence interval [CI] 47.9-64.7) and 36.9% (95% CI 31.4-42.4), respectively. The final nomogram for the prediction of 1-year RFS included male gender, poor histologic grade, signet ring cell adenocarcinoma, cN1, cN2-3, and baseline SUVmax, with accurate calibration and reasonable discrimination (C-statistic: 0.66). Trimodality therapy was associated with improved 5-year OS in low-risk patients (p = 0.003), whereas it showed no significant survival benefit in high-risk patients (p = 0.302). CONCLUSIONS: The proposed nomogram estimates early recurrence risk. The addition of surgery to CRT provides a clear OS benefit in low-risk patients. The OS benefit of surgery in high-risk patients is less pronounced.

Quality and Value of Subspecialty Reinterpretation of Thoracic CT Scans of Patients Referred to a Tertiary Cancer Center.

INTRODUCTION: Patients referred to tertiary cancer centers often present with imaging studies performed and interpreted at other health care institutions. Although reinterpretation of imaging performed at another health care institution can reduce repeat imaging, unnecessary radiation dose, and cost, the benefit is uncertain. The purpose of this study is to evaluate the quality of initial imaging studies of patients seeking a second opinion at a tertiary cancer center, to compare the accuracy of initial interpretations to reinterpretations performed by subspecialty trained radiologists at a tertiary oncologic center, and to determine the potential impact on patient management. METHODS: An institutional review board-approved retrospective, single-institution database review was performed in 120 new patients presenting to the thoracic surgery clinics at our institution from 2010 through 2013, with initial chest CTs performed at another institution. Two thoracic radiologists blinded to the interpretation independently assessed the quality and performed a reinterpretation of 52 CTs. Fisher's exact tests were used to compare the frequency with which clinically important staging parameters appeared in the reinterpretations and initial reports. Discrepancies between the reinterpretations and initial interpretations were adjudicated independently by two thoracic radiologists at different tertiary cancer institutions to determine which interpretations were more accurate. The impact of discrepancies on clinical management was evaluated based on National Comprehensive Cancer Network guidelines. RESULTS: Of the 52 CTs, 32 (62%) were of inadequate image quality for staging. In 17 of 52 (33%), discrepancies were identified between reinterpretations and initial interpretations. For discrepancies, the reinterpretation was judged to be more accurate for staging than the initial interpretation. In nine of these patients, staging parameters were omitted in the initial interpretations that precluded adequate staging. In the remaining eight patients, six were upstaged, one was downstaged, and one was unchanged by the reinterpretation. CONCLUSIONS: Imaging studies from outside institutions are of variable image quality and often not adequate for appropriate staging of thoracic malignancies. Reinterpretation can decrease repeat imaging and associated technical costs. Additionally, the accuracy of staging is improved by reinterpretation of CTs by subspecialty trained radiologists and can significantly impact clinical management.