Research In Practice Blog
A novel Computable phENotype To Identify Pulmonary Embolism in chilDrEn: The CENTIPEDE Study
Study Design and Findings
Pulmonary embolism (PE) is rare in pediatric patients, but when diagnosed, the mortality rate has been reported as high as 8%. The lack of multi-center data sources has created barriers to investigating pediatric PE, and there is a need for collaboration between health centers to conduct meaningful research and improve care.
“Creating a computable phenotype or algorithm that can identify pulmonary embolism in children across a data set can streamline the process of identifying patients with PE to study and improve treatment plans,” said Hilary B. Whitworth, MD, MSCE, lead author of the study and a Clinical Futures core faculty member.
This study utilized data from PEDSnet, a pediatric learning health system that incorporates electronic health record data from over 14 million patients from 10 major pediatric institutions including Children’s Hospital of Philadelphia. This data informed the development of a computable phenotype (CP) that can accurately identify pediatric PE in electronic health record (EHR) data. The study was supported by the Children's Foundation and included researchers from Children’s Hospital of Philadelphia and the Children's Hospital of Michigan.
“We report the first pediatric pulmonary embolism computable phenotype, which can identify at risk-patients and empower informed decision making that improves outcomes for patients," said study co-author Michael L. O'Byrne, MD, MSCE.
Methods
The initial CP was developed based on the ASPECT algorithm, which identifies adults with PE in the emergency department. Researchers looked at PEDSnet data from 1/1/2012 and 12/31/2022 and used adjudicated chart review to develop a final version of the CP. Success of the CP was measured by sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV).
“The goal of the CP was to accurately identify pediatric patients with acute PE for future observational studies in PEDSnet and for cohort identification in future prospective studies,” said Dr. Whitworth.
Researchers found that the final CP version had a sensitivity of 96 % and specificity of 88 %. PPV was 90% and NPV 95%. The CP has improved PPV and specificity compared to only using a PE diagnostic code.
Implications
This CP has the potential to systematically and efficiently identify pediatric patients with a PE. The use of the CP can provide consistent identification of affected patients, enhancing the efficiency of clinical research and treatment. Similarly, electronic health record learning systems, like PEDSnet, offer readily available large volume data that can inform optimal management and treatment strategies for pediatric PE.
“The development of a pediatric PE CP improves accurate identification of PE in EHR data, which will facilitate future, multicenter research for this rare, potentially high-risk disease,” said Dr. Whitworth.
By utilizing the CP, health care providers can leverage multi-center data for this high-risk disease where cases are limited and obstacles exist to perform randomized control trials. This will give health providers a better understanding of risk factors and outcomes associated with pediatric PE, which can inform therapies and improve care for patients.
Clinical Futures author(s): Hilary B. Whitworth, MD, MSCE and Michael L. O'Byrne, MD, MSCE