Introduction

Global data sharing is essential. This is the premise of the Academic Research Organization (ARO) Council, which was initiated in Japan in 2013 and has since been expanding throughout Asia and into Europe and the United States. The volume of data is growing exponentially, providing not only challenges but also the clear opportunity to understand and treat diseases in ways not previously considered. Harnessing the knowledge within the data in a successful way can provide researchers and clinicians with new ideas for therapies while avoiding repeats of failed experiments. This knowledge transfer from research into clinical care is at the heart of a learning health system.

Methods

The ARO Council wishes to form a worldwide complementary system for the benefit of all patients and investigators, catalyzing more efficient and innovative medical research processes. Thus, they have organized Global ARO Network Workshops to bring interested parties together, focusing on the aspects necessary to make such a global effort successful. One such workshop was held in Austin, Texas, in November 2017. Representatives from Japan, Taiwan, Singapore, Europe, and the United States reported on their efforts to encourage data sharing and to use research to inform care through learning health systems.

Results

This experience report summarizes presentations and discussions at the Global ARO Network Workshop held in November 2017 in Austin, TX, with representatives from Japan, Korea, Singapore, Taiwan, Europe, and the United States. Themes and recommendations to progress their efforts are explored. Standardization and harmonization are at the heart of these discussions to enable data sharing. In addition, the transformation of clinical research processes through disruptive innovation, while ensuring integrity and ethics, will be key to achieving the ARO Council goal to overcome diseases such that people not only live longer but also are healthier and happier as they age.

Conclusions

The achievement of global learning health systems will require further exploration, consensus-building, funding aligned with incentives for data sharing, standardization, harmonization, and actions that support global interests for the benefit of patients.

Purpose

The presence of distant metastases is one of the most powerful predictors of outcome in patients with neuroblastoma. However, the pattern of metastatic spread is not incorporated into current risk stratification systems. Small case series have suggested that patients with neuroblastoma who have metastatic disease limited to distant lymph nodes (4N disease) may have improved outcomes.

Patients and methods

We analyzed retrospective data from the International Neuroblastoma Risk Group database for patients diagnosed from 1990 to 2002. 4N patients were compared with the remaining stage 4 patients (non-4N), excluding those with missing metastatic site data.

Results

In all, 2,250 International Neuroblastoma Staging System stage 4 patients with complete data were identified, of whom 146 (6.5%) had 4N disease. For 4N patients, event-free survival (EFS; 5-year, 77% ± 4%) and overall survival (OS; 5-year, 85% ± 3%) were significantly better than EFS (5-year, 35% ± 1%) and OS (5-year, 42% ± 1%) for non-4N stage 4 patients (P < .001). 4N patients were more likely to be younger (P < .001) and have tumors with favorable characteristics, including absence of MYCN amplification (89% v 69%; P < .001). In a multivariable analysis, 4N disease remained a significant predictor of outcome (hazard ratio for non-4N v 4N: 3.40 for EFS and 3.69 for OS). Within subgroups defined by age at diagnosis and tumor MYCN status, 4N disease was significantly associated with improved outcomes.

Conclusion

4N represents a subgroup with better outcome than that of other patients with metastatic disease. These findings suggest that the biology and treatment response of 4N tumors differ from other stage 4 tumors, and less intensive therapy should be considered for this cohort. Future exploration of biologic factors determining the pattern of metastatic spread is warranted.

Risk-based treatment approaches for neuroblastoma have been ongoing for decades. However, the criteria used to define risk in various institutional and cooperative groups were disparate, limiting the ability to compare clinical trial results. To mitigate this problem and enhance collaborative research, homogenous pretreatment patient cohorts have been defined by the International Neuroblastoma Risk Group classification system. During the past 30 years, increasingly intensive, multimodality approaches have been developed to treat patients who are classified as high risk, whereas patients with low- or intermediate-risk neuroblastoma have received reduced therapy. This treatment approach has resulted in improved outcome, although survival for high-risk patients remains poor, emphasizing the need for more effective treatments. Increased knowledge regarding the biology and genetic basis of neuroblastoma has led to the discovery of druggable targets and promising, new therapeutic approaches. Collaborative efforts of institutions and international cooperative groups have led to advances in our understanding of neuroblastoma biology, refinements in risk classification, and stratified treatment strategies, resulting in improved outcome. International collaboration will be even more critical when evaluating therapies designed to treat small cohorts of patients with rare actionable mutations.

Predictive modeling with electronic health record (EHR) data is anticipated to drive personalized medicine and improve healthcare quality. Constructing predictive statistical models typically requires extraction of curated predictor variables from normalized EHR data, a labor-intensive process that discards the vast majority of information in each patient's record. We propose a representation of patients' entire raw EHR records based on the Fast Healthcare Interoperability Resources (FHIR) format. We demonstrate that deep learning methods using this representation are capable of accurately predicting multiple medical events from multiple centers without site-specific data harmonization. We validated our approach using de-identified EHR data from two US academic medical centers with 216,221 adult patients hospitalized for at least 24 h. In the sequential format we propose, this volume of EHR data unrolled into a total of 46,864,534,945 data points, including clinical notes. Deep learning models achieved high accuracy for tasks such as predicting: in-hospital mortality (area under the receiver operator curve [AUROC] across sites 0.93-0.94), 30-day unplanned readmission (AUROC 0.75-0.76), prolonged length of stay (AUROC 0.85-0.86), and all of a patient's final discharge diagnoses (frequency-weighted AUROC 0.90). These models outperformed traditional, clinically-used predictive models in all cases. We believe that this approach can be used to create accurate and scalable predictions for a variety of clinical scenarios. In a case study of a particular prediction, we demonstrate that neural networks can be used to identify relevant information from the patient's chart.

The epigenome orchestrates genome accessibility, functionality, and three-dimensional structure. Because epigenetic variation can impact transcription and thus phenotypes, it may contribute to adaptation. Here, we report 1,107 high-quality single-base resolution methylomes and 1,203 transcriptomes from the 1001 Genomes collection of Arabidopsis thaliana. Although the genetic basis of methylation variation is highly complex, geographic origin is a major predictor of genome-wide DNA methylation levels and of altered gene expression caused by epialleles. Comparison to cistrome and epicistrome datasets identifies associations between transcription factor binding sites, methylation, nucleotide variation, and co-expression modules. Physical maps for nine of the most diverse genomes reveal how transposons and other structural variants shape the epigenome, with dramatic effects on immunity genes. The 1001 Epigenomes Project provides a comprehensive resource for understanding how variation in DNA methylation contributes to molecular and non-molecular phenotypes in natural populations of the most studied model plant.