Data for Discovery
Resnick Discusses Data-Sharing Platforms for Precision Care
Vast amounts of patient data get generated over time—electronic health records across specialties, diagnostic scans, follow-up tests, care plans, doctors’ observations. Such data can reveal all sorts of health insights, even more efficiently if the information could be accessed, analyzed and shared across the health system.
Dr. Adam Resnick, director of the Center for Data-Driven Discovery in Biomedicine at Children’s Hospital of Philadelphia (CHOP), is building scalable data-sharing platforms for pediatric patients. His model connects layers of medical data into an interoperable framework toward improving cross-disciplinary research and patient care.
Resnick recently spoke at the National Institute on Deafness and Other Communication Disorders (NIDCD) Director’s Seminar series. NIDCD leaders developed the series to highlight innovative research with the potential to improve the lives of people with communication disorders.
“It’s an exciting endeavor to think about the possibility of large-scale data sets and infrastructure created for research to become proximal to care,” Resnick said.
Resnick’s team began collecting patient samples and generating molecular data sets on children with brain tumors. To advance a precision medicine framework, he said, “you have to think about repositories, data mining, biomedical informatics, hypothesis generation, genotype-phenotype connections [toward] clinical trial enrollment of subjects.”
They started small then grew rapidly with more than 5,000 participants enrolled at 34 sites, as part of the Children’s Brain Tumor Network (CBTN), in what is the largest pediatric brain tumor study of its kind. The goal then became creating a data-sharing platform accessible across disciplines, diseases and expertise.
In 2017, CHOP launched the Gabriella Miller Kids First Data Resource Center, a cloud-based repository of genomic and clinical data sponsored by NIH’s Common Fund. Its namesake Miller, a girl from Virginia, died at age 10 of brain cancer. The initial vision for CBTN was to focus on pediatric brain cancer. In the setting of Kids First, NIH dramatically expanded the scope.
NIH stipulated “not only are you going to work across all brain tumors, you’re going to work across all pediatric cancer and bring in structural birth defects and congenital abnormalities,” Resnick said.
A common framework then emerged. “It took time to build the next iteration of infrastructure that empowered cross-condition discovery,” said Resnick. “This would never have happened if it wasn’t for an NIH-based vision for data-sharing, interoperability and collaboration.”
Today, Kids First has more than 21,000 whole genomes across more than 6,000 phenotypes. It’s one of the largest whole genome and molecular data sets for pediatric cancer and structural birth defects.
“It was a learning experience for me that platforms not only support and enhance coordination around stakeholders that know each other, but that they are extremely powerful to bring together people who may not know each other,” Resnick said.
This rationale also underpins the INCLUDE project (Investigating Co-occurring conditions across the Lifespan to Understand Down syndrome) that links 21 NIH institutes and centers as research partners to study conditions that disproportionately affect individuals with Down syndrome. Resnick co-leads its Data Coordinating Center.
Photo: Nan_Got/Shutterstock
Like the Kids First platform, INCLUDE leverages the cloud, allowing others to connect and contextualize the information, explained Resnick.
Researchers can analyze across phenotypes, such as autism or hearing loss, and genomics or other molecular features. If an investigator inputs a gene of interest, that gene may be implicated in other contexts that interest other researchers. “That’s a powerful framework for prioritizing which genes matter for you or for somebody else,” he said.
When building such interoperable platforms, Resnick said he factored in that clinicians and data scientists have different ways of communicating and using information.
“They speak completely different languages from the clinical endpoints you’re trying to help and the data landscape,” he said. “They ask questions very differently. And it’s challenging to create a framework that supports what is often the rapid point-and-click requirements for someone who is not a data scientist and at the same time a user interface that meets the needs of data scientists and bioinformaticians.”
Resnick underscored that phenotypes change over time with new findings, successive diagnostics, treatments and outcomes. The key is capturing and analyzing this longitudinal data in ways that support a vision of individualized measurements.
“This is where precision medicine thrives,” Resnick said. Often, “the analysis happens in the doctor’s head based on what they know and remember and have experienced. Precision medicine requires that next layer of interactions between that patient’s data, which is suddenly more complex and multimodal. Part of it is not interpretable [now] but might be tomorrow when the next patient comes along and looks like that patient.”
An ongoing challenge has been how to integrate the notes and narratives created as part of clinical care. When a radiologist looks at a patient’s scan or an oncologist works on a care plan, they become storytellers, said Resnick. “That language embedded in the care is absent from most of our structured data.” And that information is critical.
Clinically, patients are cared for by integrated teams of specialists, but the data behind such care is often siloed without a single space for data-sharing. “We don’t want each specialist working in different places,” Resnick emphasized. “They’ll make their separate discoveries and never talk.”
Resnick’s team has spent the past year creating workflows that bring de-identified imaging data into shared project spaces in the cloud, along with genomics, so the radiologist and genomicist, for example, can each see what they want while sharing, combining and analyzing results that can advance discovery and ultimately precision diagnostics and care.
Resnick’s team is also exploring ways to facilitate real-time data exchange and the potential for artificial intelligence programs to help interpret and query data.
Photo: NIDCD
“As scientists, a promise we give to patient families is ‘if you can measure it, you can understand it—if we can understand it, we can intervene with it,’” Resnick said. “If that’s true, then our job is to accelerate that process through technology.”
During an ensuing panel discussion, leaders from NIDCD, the National Human Genome Research Institute and the Office of the Director’s All of Us Research Program weighed in on the potential of linking data science, genomics and precision medicine to advance care.
Resnick underscored the need for a feedback loop. Clinicians and researchers complain about the time-consuming task of inputting data because they usually don’t get anything back, he noted.
“Imagine if you could tell a clinician, ‘If you put your data in this way, you can immediately see your patient in the context of all other patients.’”
Data-sharing, they agreed, is a collective effort.
“We need to look not only outside of our own fields, outside of health care, but also within,” said NIDCD Clinical Director Dr. Joshua Levy. “Clinically, we are way too siloed in our own subspecialties. From a data standpoint, the same holds true. The representation on this panel speaks to the importance of breaking down these barriers. We don’t need to recreate 10 different data solutions. We can integrate and learn from each other.”