Integrating Artificial Intelligence and Machine Learning Into Cancer Clinical Trials.

The practice of oncology requires analyzing and synthesizing abundant data. From the patient's workup to determine eligibility to the therapies received to the post-treatment surveillance, practitioners must constantly juggle, evaluate, and weigh decision-making based on their best understanding of information at hand. These complex, multifactorial decisions have a tremendous opportunity to benefit from data-driven machine learning (ML) methods to drive opportunities in artificial intelligence (AI). Within the past 5 years, we have seen AI move from simply a promising opportunity to being used in prospective trials. Here, we review recent efforts of AI in clinical trials that have moved the needle towards improved prediction of actionable outcomes, such as predicting acute care visits, short term mortality, and pathologic extranodal extension. We then pause and reflect on how these AI models ask a different question than traditional statistics models that readers may be more familiar with; how then should readers conceptualize and interpret AI models that they are not as familiar with. We end with what we believe are promising future opportunities for AI in oncology, with an eye towards allowing the data to inform us through unsupervised learning and generative models, rather than asking AI to perform specific functions.

Second primary breast cancer after diagnosis of breast cancer among male patients: An examination of population characteristics and overall survival.

BACKGROUND: Male patients with breast cancer (BrC) have increased risk of developing 2nd-primary BrC (2nd-BrC). Given the relative rarity of male BrC, population-based registries are needed to analyze overall survival (OS) outcomes for these patients. METHODS: Using the Surveillance, Epidemiology and End Results registry of patients diagnosed from 1975 to 2016, a cohort study of men whose only malignancy was BrC (BrC-O; n = 6,475), and men who developed 2nd-BrC after initial BrC diagnosis (BrC-2; n = 85) was performed. The standardized incidence ratio (SIR) of 2nd-BrC, Kaplan-Meier OS and multivariable Cox regression modelling were performed. FINDINGS: The SIR for 2nd-BrC was 32.95 (95%CI:[23.85-44.38],p < 0.05). The majority (88%) of 2nd-BrC for BrC-2 were contralateral from 1st-BrC; suggesting the unlikeliness of miscoding local recurrences as 2nd-BrC for most patients. There was no statistically significant difference between rates of hormone (reported in 44%) or HER-2 (reported in 33%) receptor status between BrC-O and BrC-2, albeit with limited data. The 2nd-BrC for BrC-2 was significantly more likely to be localized or distant stage (rather than regional) than BrC-O. Median OS was 103 months (95% CI: [99, 108]) for BrC-O and 62 months (95% CI [49, 128] after 2nd-BrC. When sub-grouped by BrC stage, and when analyzed by Cox regression, there was no significant difference in OS between BrC-O and BrC-2. INTERPRETATION: Patients with male BrC are at significantly increased risk of 2nd BrC, but they can expect similar post-BrC prognosis (versus those without 2nd-BrC), after adjusting for patient demographics and tumor characteristics known to affect OS. FUNDING: None.

Surgical Management and Adjuvant Therapy for Patients With Neurological Deficits From Vertebral Hemangiomas: A Meta-Analysis.

STUDY DESIGN: Meta-analysis. OBJECTIVE: To understand the benefits and limitations of surgical management and adjuvant therapies for patients presenting with neurological deficits from vertebral hemangiomas (VH). SUMMARY OF BACKGROUND DATA: VH is the most common benign spine tumor but rarely causes symptoms. Patients with back pain alone are treated with conservative management (kyphoplasty and radiation therapy), while those with neurological deficits require complex multi-modal treatment plans. METHODS: A PubMed literature search for "symptomatic vertebral hemangioma with spinal cord compression" identified 47 articles. From these articles and their references, 19 observational studies on patients who underwent surgery for VH met inclusion criteria. Meta-analyses were performed comparing outcomes of the surgical and adjuvant therapies using Stata13 software. For those with insufficient data for meta-analyses, descriptive analyses of variables were completed. RESULTS: One hundred ninety seven surgical cases of VH with neurologic deficits were identified. Surgery provided a complete remission of symptoms in 84% of patients, however 18% of patients had recurrence of hemangioma. Adjuvant interventions included radiation, embolization, and kyphoplasty. Radiation therapy (XRT) was associated with a lower recurrence rate and an increase in minor transient adverse effects. Preoperative embolization performed in 98 patients was associated with improved symptoms, reduced complications, lower recurrence rate, less blood loss, and higher incidence of pathologic vertebral fractures. Meta-analyses did not yield statistically significant results, likely due to the heterogeneity amongst the studies and small sample sizes, but the results compiled together provide insight on potential benefits of preoperative embolization for symptomatic relief and reduced risk of recurrence with XRT that deserves further study. CONCLUSION: For patients with neurologic deficits from spinal cord or nerve root compression, surgery provides improvement in symptoms. Recurrence of VH and symptoms refractory to surgery can be further reduced by adjuvant therapies such as embolization, kyphoplasty, and radiation with some unique risks to each therapy. LEVEL OF EVIDENCE: 2.

Comparative Effectiveness of Neoadjuvant Chemoradiation Versus Upfront Surgery in the Management of Recto-Sigmoid Junction Cancer.

INTRODUCTION: The optimal management of locally advanced recto-sigmoid cancer is unclear. Although some experts advocate for upfront surgery, others recommend neoadjuvant chemoradiation followed by surgery. We used the National Cancer Database to characterize patterns-of-care and overall survival (OS) associated with these treatment strategies. PATIENTS AND METHODS: Patients with clinical stage II or III recto-sigmoid cancer who underwent surgery with or without adjunctive chemotherapy and/or radiotherapy from 2006 to 2014 were identified, and dichotomized into: (1) upfront surgery, and (2) neoadjuvant chemoradiation cohorts. Patterns-of-care were assessed using multivariable logistic regression. The association between neoadjuvant chemoradiation use and OS was assessed using Cox proportional hazards analysis with propensity score-matching. RESULTS: Of 9313 identified patients, 6756 (73%) underwent upfront surgery and 2557 (27%) received neoadjuvant chemoradiation. Treatment at academic facilities and higher clinical T stage were predictors of neoadjuvant chemoradiation use. Compared with upfront surgery, neoadjuvant chemoradiation resulted in fewer positive circumferential resection margins (384 [11%] patients vs. 108 [8%] patients; P = .001), and 478 [18.7%] patients achieved a pathologic complete response at surgery. In propensity score-matched analysis, neoadjuvant chemoradiation use was associated with improved OS (hazard ratio, 0.79; 95% confidence interval, 0.69-0.90) compared with upfront surgery; 5-year estimated OS was 77.0% versus 72.0%, respectively. The improvement in OS persisted in landmark analysis of patients who survived at least 12 months. CONCLUSION: Only a small percentage of patients with locally advanced recto-sigmoid cancer receive neoadjuvant chemoradiation even though its use might result in improved OS relative to upfront surgery. Prospective research is warranted to validate and standardize therapeutic strategies in patients with recto-sigmoid cancer.

Survival implications of staging lymphadenectomy for non-endometrioid endometrial cancers.

PURPOSE: To determine, in patients with non-endometrioid endometrial carcinoma, 1) survival benefit associated with pelvic lymphadenectomy (LND), 2) survival benefit for para-aortic lymphadenectomy performed in addition to pelvic lymphadenectomy, and 3) association between number of lymph nodes removed and survival. METHODS: Patients with clinical stage I serous carcinoma, clear cell carcinoma, or carcinosarcoma who underwent hysterectomy from 2010 to 2013 were identified from the National Cancer Database. Hazard ratio (HR) for death was assessed using propensity score-weighted multivariable Cox regression models. Subgroup analyses assessed for differences in risk of death among histologic subtypes. RESULTS: 7250 patients met study criteria. 930 (13%) did not undergo LND; 2177 (30%) underwent pelvic LND alone; 4143 (57%) underwent pelvic+para-aortic LND. On propensity score-weighted analysis, pelvic LND was associated with decreased risk of death (HR=0.65, 95% CI: 0.59-0.71) compared to no LND. Pelvic+para-aortic LND was associated with decreased risk of death (HR=0.85, 95% CI: 0.79-0.91) compared to pelvic LND for patients with serous carcinoma. Removal of >15 nodes was independently associated with decreased HR for death (HR=0.86, 95% CI: 0.77-0.96); this association persisted when analysis was limited to patients with node-positive disease (HR=0.78, 95% CI: 0.63-0.95). CONCLUSIONS: LND is associated with survival benefit in patients with non-endometrioid endometrial cancers. Addition of para-aortic LND to pelvic LND may be most beneficial for patients with serous carcinoma. Systematic lymphadenectomy may be associated with survival benefit through detection and microscopic cytoreduction of occult disease.

Stereotactic Body Radiotherapy for Lung Metastases from Colorectal Cancer: Prognostic Factors for Disease Control and Survival.

OBJECTIVES: To evaluate disease control and survival after stereotactic body radiotherapy (SBRT) for lung metastases from colorectal cancer and to identify prognostic factors after treatment. METHODS: Patients with metastatic colorectal cancer to the lungs treated with SBRT from 2002 to 2013 were identified from a prospectively maintained database. Patients may have received prior systemic therapy, radiotherapy to nonthoracic sites and/or resection of thoracic and/or nonthoracic metastases. Endpoints were timed from end of SBRT and included overall survival (OS), progression-free survival, distant metastases-free survival, and local failure-free survival. Univariate and multivariate analysis using Cox proportional hazard modeling was used to identify prognostic factors. RESULTS: Sixty-five patients were identified. Before SBRT, 69.2% and 33.8% of patients received systemic therapy and lung-directed local therapy, respectively, for metastatic disease. At the time of SBRT, 64.6% had lung-only involvement. Median survivals were: OS of 20.3 months (95% confidence intervals [CI], 15.9-27.0 mo), progression-free survival of 5.7 months (95% CI, 3.2-7.0 mo), distant metastases-free survival of 5.8 months (95% CI, 3.2-7.6 mo), and local failure-free survival of 15.4 months (95% CI, 8.5-21.1 mo). Nearly all (98%) patients developed distant progression. Extra lung and liver involvement at the time of initial metastases (hazard ratios [HR] 2.10) and extra lung involvement at SBRT (HR 2.67) were the only independent predictors of OS. Net gross target volume of >14.1 mL (HR 2.49) was the only independent predictor of local failure-free survival. CONCLUSIONS: Reasonable survival and local control can be achieved with SBRT. We identified several prognostic factors testable in future prospective trials that may help improve patient selection.

A population-based study of prognosis and survival in patients with second primary thyroid cancer after Hodgkin lymphoma.

Hodgkin lymphoma (HL) survivors are at increased risk of thyroid cancer (TC). We sought to determine whether increased risks of high-risk pathology or mortality are seen with thyroid cancer after HL (HL-TC) compared with first primary thyroid cancer (TC-1). From the Surveillance, Epidemiology and End Results (SEER) registry, we compared patient and tumor characteristics as well as survival outcomes between HL-TC and TC-1 and fit a multivariable Cox model to assess for a possible association between HL history and overall survival after TC. Among 139,297 TC-1 and 174 HL-TC patients, history of HL was not associated with anaplastic or sarcoma TC. Multivariable analyzes showed that history of HL was not associated with a difference in risk of death after TC (hazard ratio: 0.96, 95% confidence interval: (0.81, 1.13), p = .61). Despite a significantly increased risk of TC among HL survivors, prior HL is not associated with more aggressive pathologic subtypes or worse prognosis.

Second primary head and neck cancer after Hodgkin lymphoma: a population-based study of 44,879 survivors of Hodgkin lymphoma.

BACKGROUND: Survivors of Hodgkin lymphoma (HL) are at an increased risk of developing second malignancies. To the authors' knowledge, the risks of head and neck cancer (HNC) after HL and subsequent survival have not been thoroughly investigated. METHODS: From the US population-based Surveillance, Epidemiology, and End Results (SEER) database for 1973 through 2011, survivors of HL who developed HNC as a second cancer were analyzed. Patients with a first primary HNC were used as a comparison group. Observed-to-expected ratios and summary statistics were performed on patients with HL with squamous cell carcinoma (HL-SCC) and salivary gland cancer (HL-SGC). The impact of HL history on overall survival was assessed using a multivariate Cox proportional hazards model. RESULTS: The observed-to-expected ratio for SCC among patients with HL was 1.73 (95% confidence interval [95% CI], 1.36-2.16; P<.05), whereas it was 8.56 for SGC (95% CI, 5.82-12.15; P<.05). Using Cox proportional hazards modeling, a history of HL was found to be an adverse prognostic factor for overall survival for SCC (hazard ratio, 1.37; 95% CI, 1.08-1.73 [P = .009]) but not SGC (hazard ratio, 1.21; 95% CI, 0.82-1.79 [P = .34]). The inferior survival of the patients in the HL-SCC cohort appears to be attributable to more deaths from HL and other malignancies diagnosed after SCC. CONCLUSIONS: There is a significantly increased risk of salivary and nonsalivary HNC after HL, and worse survival for patients with HL-SCC versus those with a first primary SCC. Clinicians should be aware of the risks of HNC after HL. In the absence of evidence-based criteria, the authors recommend that survivors of HL undergo periodic head and neck examination.

Complications from computed tomography-guided core needle biopsy for patients receiving stereotactic body radiation therapy for early-stage lesions of the lung.

BACKGROUND: Obtaining a tissue diagnosis has traditionally been standard practice before initiating therapy for early-stage non-small-cell lung cancer (NSCLC). In several recent studies from Europe and Asia, a substantial proportion of patients have received stereotactic body radiation therapy (SBRT) based only on the imaging characteristics of the suspicious lesion. The underlying assumption is that the risk of percutaneous needle biopsy may outweigh the benefits in a population that generally has underlying pulmonary dysfunction and other medical comorbidity. Nevertheless, there is limited information regarding biopsy-related complication rates in high-risk patients with early-stage NSCLC who are treated with SBRT. MATERIALS AND METHODS: This was a retrospective review of outcomes after biopsy in patients treated with SBRT. Complications of percutaneous core needle biopsy were analyzed in relation to patient and tumor characteristics. Each biopsy event was analyzed independently for patients with multiple biopsies. RESULTS: A total of 112 percutaneous biopsies were performed in 103 patients. Pneumothorax of any degree was observed in 40 patients (35%) (95% CI, 27%-45%), and 12 patients (10.7%) had a clinically significant pneumothorax requiring chest tube placement (95% CI, 6%-18%). The time to first fraction of SBRT was not different in patients who had a pneumothorax or placement of a chest tube. On multivariate analysis, age, performance status, smoking history, pack-years of smoking, chronic obstructive pulmonary disease history, and forced expiratory volume in the first second of expiration were not statistically significantly associated with chest tube placement. CONCLUSION: Computed tomography-guided needle biopsy in a primarily medically inoperable patient population is safe, with an acceptable degree of complications.

Cross-sectional area for the calculation of carotid artery stenosis on computed tomographic angiography.

OBJECTIVE: The use of cross-sectional area (CSA) measurements obtained from computed tomographic angiography (CTA) for the calculation of carotid artery stenosis has been suggested but not yet validated in a large population. The objective of this study was to determine whether CTA-derived CSA measurements were able to predict carotid stenosis with a level of confidence similar to CTA-derived diameter measurements, using Strandness criteria applied to carotid duplex ultrasound (CDUS) as a surrogate for true stenosis. METHODS: A retrospective review was conducted to identify patients who underwent both CDUS and CTA between 2000 and 2009. Percent stenosis was calculated using the North American Symptomatic Carotid Endarterectomy Trial (NASCET) formula with diameter measurements and again with CSA measurements. A nonparametric correlation coefficient was calculated to detect correlation between the two groups. Two-dimensional receiver-operating characteristic curves with corresponding area under the curve (AUC) statistics were generated for >50% stenosis and >80% stenosis. Three-dimensional receiver-operating characteristic plots with corresponding volume under the surface (VUS) statistics were generated to measure the comparative accuracy of diameter-based and CSA-based stenosis for <50%, 50%-79%, and >80% stenosis. RESULTS: A total of 575 vessels in 313 patients were included in the study. Spearman's correlation coefficient between diameter and CSA-derived stenosis was ρ = 0.938 (95% confidence interval [CI], 0.927-0.947; P < .0001). For diameter-derived stenosis, AUC was 0.905 (95% CI, 0.878-0.932; P < .0001) for >50% stenosis and 0.950 (95% CI, 0.928-0.972; P < .0001) for 80%-99% stenosis. For CSA-derived percent stenosis, the AUC was 0.908 (95% CI, 0.882-0.935; P < .0001) for >50% stenosis and 0.935 (95% CI, 0.908-0.961; P < .0001) for 80%-99%. The nonparametric estimate for VUS in the diameter-based stenosis group was 0.761, whereas in the CSA-based group, the VUS was 0.735. The difference between VUS was 0.026 (95% CI, -0.022 and 0.077; P = .318). CONCLUSIONS: These data support the use of CTA as an accurate method of calculating carotid artery stenosis based on agreement with Strandness criteria applied to CDUS velocities. When additional imaging beyond CDUS is necessary, we report no significant difference between diameter and CSA measurements obtained from CTA for preoperative evaluation of carotid disease.