“The anchor of all my dreams is the collective wisdom of mankind as a whole.”
Nelson Mandela
On the eve of emergency approval for children ages 5-12 years to receive the coronavirus vaccine, there is a deep appreciation for clinical research in children. There are inherent challenges in clinical pediatrics and pediatric research: the relatively small number of children with the myriad of diseases and conditions, especially with rare diseases. One of the greatest challenges for pediatric medicine has been clinical trials as well as sharing of data in children, especially in real time.
There have been notable efforts in children’s hospitals working cohesively in the overall best interest of children. The Children’s Oncology Group (COG) is a clinical trials group with over 100 clinical trials for pediatric cancer that is supported by the National Cancer Institute. This group gathers more than 10,000 experts in childhood cancer from more than 200 children’s hospitals from around the world, and this collaborative effort has led to an overall 5-year survival of about 80% for all childhood cancers. In addition, the Patient-Centered Outcomes Research Institute (PCORI) has funded 98 projects in children and adolescents, and has created a national network called Cornet to foster clinical effective research studies. Lastly, the Virtual Pediatric Intensive Care Unit (VPICU), led by Dr. Randall Wetzel of Children’s Hospital of Los Angeles, has convened a group of children’s hospitals to share ICU data (the so-called “Common Information Space”) and to leverage information science technology to gain valuable clinical insights to improve the outcome of critically ill children.
At present, most pediatric centers have data at their local storage without sharing data, so that any machine or deep learning would be limited to local learning without any external insights. Centers can also provide data and model parameters in a central location and therefore achieve central learning (as the aforementioned VPICU). In this mode of collaboration, there is full control with data security but it is vulnerable to group dynamics and leadership.
In the near future, possible solutions for collective research amongst children’s hospital that obviate the conditional requirement to share data include: 1) federated learning: the hospitals can keep the data at the edge and only share model parameters in a central location; and 2) swarm learning: the hospitals can maintain both the data and the parameters at the edge without the need for a central custodian and this is fortified by blockchain for its security. These latter collective learning methodologies will have the dual benefits of keeping the data locally while sharing the insights globally, ideal for research in children and adolescents.
