Transfer Learning for Mortality Prediction in Non-Small Cell Lung Cancer with Low-Resolution Histopathology Slide Snapshots.

High-resolution whole slide image scans of histopathology slides have been widely used in recent years for prediction in cancer. However, in some cases, clinical informatics practitioners may only have access to low-resolution snapshots of histopathology slides, not high-resolution scans. We evaluated strategies for training neural network prognostic models in non-small cell lung cancer (NSCLC) based on low-resolution snapshots, using data from the Veterans Affairs Precision Oncology Data Repository. We compared strategies without transfer learning, with transfer learning from general domain images, and with transfer learning from publicly available high-resolution histopathology scans. We found transfer learning from high-resolution scans achieved significantly better performance than other strategies. Our contribution provides a foundation for future development of prognostic models in NSCLC that incorporate data from low-resolution pathology slide snapshots alongside known clinical predictors.

Matching Patients to Accelerate Clinical Trials (MPACT): Enabling Technology for Oncology Clinical Trial Workflow.

Clinical trial enrollment is impeded by the significant time burden placed on research coordinators screening eligible patients. With 50,000 new cancer cases every year, the Veterans Health Administration (VHA) has made increased access for Veterans to high-quality clinical trials a priority. To aid in this effort, we worked with research coordinators to build the MPACT (Matching Patients to Accelerate Clinical Trials) platform with a goal of improving efficiency in the screening process. MPACT supports both a trial prescreening workflow and a screening workflow, employing Natural Language Processing and Data Science methods to produce reliable phenotypes of trial eligibility criteria. MPACT also has a functionality to track a patient's eligibility status over time. Qualitative feedback has been promising with users reporting a reduction in time spent on identifying eligible patients.

Building Research Infrastructure to Develop Greater Learning Efficiencies (BRIDGE).

In this manuscript, we outline our developed version of a Learning Health System (LHS) in oncology implemented at the Department of Veterans Affairs (VA). Transferring healthcare into an LHS framework has been one of the spearpoints of VA's Central Office and given the general lack of evidence generated through randomized control clinical trials to guide medical decisions in oncology, this domain is one of the most suitable for this change. We describe our technical solution, which includes a large real-world data repository, a data science and algorithm development framework, and the mechanism by which results are brought back to the clinic and to the patient. Additionally, we propose the need for a bridging framework that requires collaboration between informatics specialists and medical professionals to integrate knowledge generation into the clinical workflow at the point of care.

Brain tumors in United States military veterans.

BACKGROUND: Comprehensive analysis of brain tumor incidence and survival in the Veteran population has been lacking. METHODS: Veteran data were obtained from the Veterans Health Administration (VHA) Medical Centers via VHA Corporate Data Warehouse. Brain tumor statistics on the overall US population were generated from the Central Brain Tumor Registry of the US data. Cases were individuals (≥18 years) with a primary brain tumor, diagnosed between 2004 and 2018. The average annual age-adjusted incidence rates (AAIR) and 95% confidence intervals were estimated per 100 000 population and Kaplan-Meier survival curves evaluated overall survival outcomes among Veterans. RESULTS: The Veteran population was primarily white (78%), male (93%), and between 60 and 64 years old (18%). Individuals with a primary brain tumor in the general US population were mainly female (59%) and between 18 and 49 years old (28%). The overall AAIR of primary brain tumors from 2004 to 2018 within the Veterans Affairs cancer registry was 11.6. Nonmalignant tumors were more common than malignant tumors (AAIR:7.19 vs 4.42). The most diagnosed tumors in Veterans were nonmalignant pituitary tumors (AAIR:2.96), nonmalignant meningioma (AAIR:2.62), and glioblastoma (AAIR:1.96). In the Veteran population, survival outcomes became worse with age and were lowest among individuals diagnosed with glioblastoma. CONCLUSIONS: Differences between Veteran and US populations can be broadly attributed to demographic composition differences of these groups. Prior to this, there have been no reports on national-level incidence rates and survival outcomes for Veterans. These data provide vital information that can drive efforts to understand disease burden and improve outcomes for individuals with primary brain tumors.

Prostate cancer patient stratification by molecular signatures in the Veterans Precision Oncology Data Commons.

Veterans are at an increased risk for prostate cancer, a disease with extraordinary clinical and molecular heterogeneity, compared with the general population. However, little is known about the underlying molecular heterogeneity within the veteran population and its impact on patient management and treatment. Using clinical and targeted tumor sequencing data from the National Veterans Affairs health system, we conducted a retrospective cohort study on 45 patients with advanced prostate cancer in the Veterans Precision Oncology Data Commons (VPODC), most of whom were metastatic castration-resistant. We characterized the mutational burden in this cohort and conducted unsupervised clustering analysis to stratify patients by molecular alterations. Veterans with prostate cancer exhibited a mutational landscape broadly similar to prior studies, including KMT2A and NOTCH1 mutations associated with neuroendocrine prostate cancer phenotype, previously reported to be enriched in veterans. We also identified several potential novel mutations in PTEN, MSH6, VHL, SMO, and ABL1 Hierarchical clustering analysis revealed two subgroups containing therapeutically targetable molecular features with novel mutational signatures distinct from those reported in the Catalogue of Somatic Mutations in Cancer database. The clustering approach presented in this study can potentially be used to clinically stratify patients based on their distinct mutational profiles and identify actionable somatic mutations for precision oncology.

Machine learning-based natural language processing to extract PD-L1 expression levels from clinical notes.

Introduction: PD-L1 expression is used to determine oncology patients' response to and eligibility for immunologic treatments; however, PD-L1 expression status often only exists in unstructured clinical notes, limiting ability to use it in population-level studies. Methods: We developed and evaluated a machine learning based natural language processing (NLP) tool to extract PD-L1 expression values from the nationwide Veterans Affairs electronic health record system. Results: The model demonstrated strong evaluation performance across multiple levels of label granularity. Mean precision of the overall PD-L1 positive label was 0.859 (sd, 0.039), recall 0.994 (sd, 0.013), and F1 0.921 (0.024). When a numeric PD-L1 value was identified, the mean absolute error of the value was 0.537 on a scale of 0 to 100. Conclusion: We presented an accurate NLP method for deriving PD-L1 status from clinical notes. By reducing the time and manual effort needed to review medical records, our work will enable future population-level studies in cancer immunotherapy.

Epidemiology of Cancer-Associated Venous Thromboembolism in Patients With Solid and Hematologic Neoplasms in the Veterans Affairs Health Care System.

Importance: Identifying changes in epidemiologic patterns of the incidence and risk of cancer-associated thrombosis (CAT), particularly with evolving cancer-directed therapy, is essential for risk stratification. Objective: To assess the incidence of CAT over time and to determine pertinent patient-specific, cancer-specific, and treatment-specific factors associated with its risk. Design, Setting, and Participants: This longitudinal, retrospective cohort study was conducted from 2006 to 2021. Duration of follow-up was from the date of diagnosis until first venous thromboembolism (VTE) event, death, loss of follow-up (defined as a 90-day gap without clinical encounters), or administrative censoring on April 1, 2022. The study took place within the US Department of Veterans Affairs national health care system. Patients with newly diagnosed invasive solid tumors and hematologic neoplasms were included in the study. Data were analyzed from December 2022 to February 2023. Exposure: Newly diagnosed invasive solid tumors and hematologic neoplasms. Main Outcomes: Incidence of VTE was assessed using a combination of International Classification of Diseases, Ninth Revision, Clinical Modification and International Statistical Classification of Diseases, Tenth Revision, Clinical Modification and natural language processing confirmed outcomes. Cumulative incidence competing risk functions were used to estimate incidence of CAT. Multivariable Cox regression models were built to assess the association of baseline variables with CAT. Pertinent patient variables included demographics, region, rurality, area deprivation index, National Cancer Institute comorbidity index, cancer type, staging, first-line systemic treatment within 3 months (time-varying covariate), and other factors that could be associated with the risk of VTE. Results: A total of 434 203 patients (420 244 men [96.8%]; median [IQR] age, 67 [62-74] years; 7414 Asian or Pacific Islander patients [1.7%]; 20 193 Hispanic patients [4.7%]; 89 371 non-Hispanic Black patients [20.6%]; 313 157 non-Hispanic White patients [72.1%]) met the inclusion criteria. Overall incidence of CAT at 12 months was 4.5%, with yearly trends ranging stably from 4.2% to 4.7%. The risk of VTE was associated with cancer type and stage. In addition to confirming well-known risk distribution among patients with solid tumors, a higher risk of VTE was observed among patients with aggressive lymphoid neoplasms compared with patients with indolent lymphoid or myeloid hematologic neoplasms. Compared with no treatment, patients receiving first-line chemotherapy (hazard ratio [HR], 1.44; 95% CI, 1.40-1.49) and immune checkpoint inhibitors (HR, 1.49; 95% CI, 1.22-1.82) had a higher adjusted relative risk than patients receiving targeted therapy (HR, 1.21; 95% CI, 1.13-1.30) or endocrine therapy (HR, 1.20; 95% CI, 1.12-1.28). Finally, adjusted VTE risk was significantly higher among Non-Hispanic Black patients (HR, 1.23; 95% CI, 1.19-1.27) and significantly lower in Asian or Pacific Islander patients (HR, 0.84; 95% CI, 0.76-0.93) compared with Non-Hispanic White patients. Conclusions and Relevance: In this cohort study of patients with cancer, a high incidence of VTE was observed, with yearly trends that remained stable over the 16-year study period. Both novel and known factors associated with the risk of CAT were identified, providing valuable and applicable insights in this current treatment landscape.

Sentiment analysis of medical record notes for lung cancer patients at the Department of Veterans Affairs.

Natural language processing of medical records offers tremendous potential to improve the patient experience. Sentiment analysis of clinical notes has been performed with mixed results, often highlighting the issue that dictionary ratings are not domain specific. Here, for the first time, we re-calibrate the labMT sentiment dictionary on 3.5M clinical notes describing 10,000 patients diagnosed with lung cancer at the Department of Veterans Affairs. The sentiment score of notes was calculated for two years after date of diagnosis and evaluated against a lab test (platelet count) and a combination of data points (treatments). We found that the oncology specific labMT dictionary, after re-calibration for the clinical oncology domain, produces a promising signal in notes that can be detected based on a comparative analysis to the aforementioned parameters.

The Veterans Affairs Precision Oncology Data Repository, a Clinical, Genomic, and Imaging Research Database.

The Veterans Affairs Precision Oncology Data Repository (VA-PODR) is a large, nationwide repository of de-identified data on patients diagnosed with cancer at the Department of Veterans Affairs (VA). Data include longitudinal clinical data from the VA's nationwide electronic health record system and the VA Central Cancer Registry, targeted tumor sequencing data, and medical imaging data including computed tomography (CT) scans and pathology slides. A subset of the repository is available at the Genomic Data Commons (GDC) and The Cancer Imaging Archive (TCIA), and the full repository is available through the Veterans Precision Oncology Data Commons (VPODC). By releasing this de-identified dataset, we aim to advance Veterans' health care through enabling translational research on the Veteran population by a wide variety of researchers.

An application to support COVID-19 occupational health and patient tracking at a Veterans Affairs medical center.

OBJECTIVE: Reducing risk of coronavirus disease 2019 (COVID-19) infection among healthcare personnel requires a robust occupational health response involving multiple disciplines. We describe a flexible informatics solution to enable such coordination, and we make it available as open-source software. MATERIALS AND METHODS: We developed a stand-alone application that integrates data from several sources, including electronic health record data and data captured outside the electronic health record. RESULTS: The application facilitates workflows from different hospital departments, including Occupational Health and Infection Control, and has been used extensively. As of June 2020, 4629 employees and 7768 patients and have been added for tracking by the application, and the application has been accessed over 46 000 times. DISCUSSION: Data captured by the application provides both a historical and real-time view into the operational impact of COVID-19 within the hospital, enabling aggregate and patient-level reporting to support identification of new cases, contact tracing, outbreak investigations, and employee workforce management. CONCLUSIONS: We have developed an open-source application that facilitates communication and workflow across multiple disciplines to manage hospital employees impacted by the COVID-19 pandemic.

Machine Learning Methods to Predict Lung Cancer Survival Using the Veterans Affairs Research Precision Oncology Data Commons.

We completed a pilot study to guide the development of the VA Research Precision Oncology Data Commons infrastructure as a collaboration platform with the greater research community. Our results using a small subset of patients from the VA's Precision Oncology Program demonstrate the feasibility of our data sharing platform to build predictive models for lung cancer survival using machine learning, as well as highlight the potential of target genome sequencing data.