Although machine learning (ML) has shown promise across disciplines, out-of-sample generalizability is concerning. This is currently addressed by sharing multi-site data, but such centralization is challenging/infeasible to scale due to various limitations. Federated ML (FL) provides an alternative paradigm for accurate and generalizable ML, by only sharing numerical model updates. Here we present the largest FL study to-date, involving data from 71 sites across 6 continents, to generate an automatic tumor boundary detector for the rare disease of glioblastoma, reporting the largest such dataset in the literature (n = 6, 314). We demonstrate a 33% delineation improvement for the surgically targetable tumor, and 23% for the complete tumor extent, over a publicly trained model. We anticipate our study to: 1) enable more healthcare studies informed by large diverse data, ensuring meaningful results for rare diseases and underrepresented populations, 2) facilitate further analyses for glioblastoma by releasing our consensus model, and 3) demonstrate the FL effectiveness at such scale and task-complexity as a paradigm shift for multi-site collaborations, alleviating the need for data-sharing.

Background Patients with chronic viral infections including human immunodeficiency virus (HIV), hepatitis B (HBV) and hepatitis C (HCV) are at increased risk of developing malignancies. The safety and efficacy of ICI therapy in patients with both cancer and chronic viral infections is not well established as most clinical trials of ICIs excluded these patient populations. Methods We performed a retrospective analysis of patients with advanced-stage cancers and HIV, HBV, or HCV infection treated with ICI therapy at 5 MedStar Health hospitals from January 2011 to April 2018. Results We identified 50 patients including 16 HIV, 29 HBV/HCV, and 5 with concurrent HIV and either HBV or HCV. In the HIV cohort (n = 21), any grade immune-related adverse events (irAEs) were 24% with grade ≥ 3 irAEs 14%. Among 5 patients with matched pre/post-treatment results, no significant changes in HIV viral load and CD4+ T-cell counts were observed. RECIST confirmed (n = 18) overall response rate (ORR) was 28% with 2 complete responses (CR) and 3 partial responses (PR). Responders included 2 patients with low baseline CD4+ T-cell counts (40 and 77 cells/ul, respectively). In the HBV/HCV cohort (n = 34), any grade irAEs were 44% with grade ≥ 3 irAEs 29%. RECIST confirmed ORR was 21% (6 PR). Among the 6 patients with known pre/post-treatment viral titers (2 HCV and 4 HBV), there was no evidence of viral reactivation. Conclusions Our retrospective series is one of the largest case series to report clinical outcomes among HIV, HBV and HCV patients treated with ICI therapy. Toxicity and efficacy rates were similar to those observed in patients without chronic viral infections. Viral reactivation was not observed. Tumor responses occurred in HIV patients with low CD4 T-cell counts. While prospective studies are needed to validate above findings, these data support not excluding such patients from ICI–based clinical trials or treatment.

Background: Screening for suicidal ideation in high-risk groups such as U.S. veterans is crucial for early detection and suicide prevention. Currently, screening is based on clinical interviews or self-report measures. Both approaches rely on subjects to disclose their suicidal thoughts. Innovative approaches are necessary to develop objective and clinically applicable assessments. Speech has been investigated as an objective marker to understand various mental states including suicidal ideation. In this work, we developed a machine learning and natural language processing classifier based on speech markers to screen for suicidal ideation in US veterans. Methodology: Veterans submitted 588 narrative audio recordings via a mobile app in a real-life setting. In addition, participants completed self-report psychiatric scales and questionnaires. Recordings were analyzed to extract voice characteristics including prosodic, phonation, and glottal. The audios were also transcribed to extract textual features for linguistic analysis. We evaluated the acoustic and linguistic features using both statistical significance and ensemble feature selection. We also examined the performance of different machine learning algorithms on multiple combinations of features to classify suicidal and non-suicidal audios. Results: A combined set of 15 acoustic and linguistic features of speech were identified by the ensemble feature selection. Random Forest classifier, using the selected set of features, correctly identified suicidal ideation in veterans with 86% sensitivity, 70% specificity, and an area under the receiver operating characteristic curve (AUC) of 80%. Conclusions: Speech analysis of audios collected from veterans in everyday life settings using smartphones offers a promising approach for suicidal ideation detection. A machine learning classifier may eventually help clinicians identify and monitor high-risk veterans.

Computational competitions are the standard for benchmarking medical image analysis algorithms, but they typically use small curated test datasets acquired at a few centers, leaving a gap to the reality of diverse multicentric patient data. To this end, the Federated Tumor Segmentation (FeTS) Challenge represents the paradigm for real-world algorithmic performance evaluation. The FeTS challenge is a competition to benchmark (i) federated learning aggregation algorithms and (ii) state-of-the-art segmentation algorithms, across multiple international sites. Weight aggregation and client selection techniques were compared using a multicentric brain tumor dataset in realistic federated learning simulations, yielding benefits for adaptive weight aggregation, and efficiency gains through client sampling. Quantitative performance evaluation of state-of-the-art segmentation algorithms on data distributed internationally across 32 institutions yielded good generalization on average, albeit the worst-case performance revealed data-specific modes of failure. Similar multi-site setups can help validate the real-world utility of healthcare AI algorithms in the future.

Background: G-DOC Plus is a data integration and bioinformatics platform that uses cloud computing and other advanced computational tools to handle a variety of biomedical BIG DATA including gene expression arrays, NGS and medical images so that they can be analyzed in the full context of other omics and clinical information. Results: G-DOC Plus currently holds data from over 10,000 patients selected from private and public resources including Gene Expression Omnibus (GEO), The Cancer Genome Atlas (TCGA) and the recently added datasets from REpository for Molecular BRAin Neoplasia DaTa (REMBRANDT), caArray studies of lung and colon cancer, ImmPort and the 1000 genomes data sets. The system allows researchers to explore clinical-omic data one sample at a time, as a cohort of samples; or at the level of population, providing the user with a comprehensive view of the data. G-DOC Plus tools have been leveraged in cancer and non-cancer studies for hypothesis generation and validation; biomarker discovery and multi-omics analysis, to explore somatic mutations and cancer MRI images; as well as for training and graduate education in bioinformatics, data and computational sciences. Several of these use cases are described in this paper to demonstrate its multifaceted usability. Conclusion: G-DOC Plus can be used to support a variety of user groups in multiple domains to enable hypothesis generation for precision medicine research. The long-term vision of G-DOC Plus is to extend this translational bioinformatics platform to stay current with emerging omics technologies and analysis methods to continue supporting novel hypothesis generation, analysis and validation for integrative biomedical research. By integrating several aspects of the disease and exposing various data elements, such as outpatient lab workup, pathology, radiology, current treatments, molecular signatures and expected outcomes over a web interface, G-DOC Plus will continue to strengthen precision medicine research. G-DOC Plus is available at: https://gdoc.georgetown.edu.

Evidence regarding the association between body mass index (BMI) and immune-related adverse events (irAEs) among cancer patients receiving immune checkpoint inhibitors (ICIs) is limited. Here, we use cross-sectional hospital-based data to explore their relationship. Pre-treatment BMI was treated as an ordinal variable (<25, 25 to ≤30, ≥30 kg/m²). The outcome of interest was irAEs after ICI initiation. A multivariable logistic regression model estimated the adjusted odds ratio (aOR) and 95% confidence interval (CI) of BMI. A total of 684 patients with stage III or IV cancer were included in the study (lung: 269, melanoma: 204, other: 211). The mean age at the first dose of ICI was 64.1 years (SD = 13.5), 394 patients (57.6%) were male, and over one-third (N = 260, 38.0%) were non-White. Overall, 52.9% of patients had BMI ≥ 25 kg/m² (25 to ≤30: 217, ≥30: 145) and 288 (42.1%) had irAEs after ICI treatment. Patients with higher BMI tended to have a higher rate of irAEs (<25: 35.7%, 25 to ≤30: 47.0%, ≥30: 49.0%). The multivariable logistic regression yielded consistent results (BMI ≥ 30 vs. BMI < 25: aOR = 1.47, 95% CI = 0.96–2.23; 25 ≤ BMI < 30 vs. BMI < 25: aOR = 1.46, 95% CI = 1.02–2.11, p-trend = 0.04). In conclusion, among patients with advanced cancer receiving ICIs, the rate of irAEs appears to be higher among those with higher BMI.