A novel Computable phENotype To Identify Pulmonary Embolism in chilDrEn: The CENTIPEDE Study

Background

Pediatric pulmonary embolism (PE) is rare and requires multicenter research to generate meaningful results and improve care. PEDSnet is a multicenter learning health system including 10 pediatric institutions and a valuable tool to study rare diseases.

Objectives

We aimed to develop a computable phenotype (CP) to accurately identify pediatric PE in PEDSnet.

Methods

We developed an algorithm for identification of pediatric PE in Children's Hospital of Philadelphia data within PEDSnet and compared it to blinded chart review. We identified a cohort of patients from 1/1/2012–12/31/2022 who were predicted to have PE based on the CP and an equal number predicted to not have a PE. Manual chart review identified true positive and negative and false positive and negative PE. Subsequent adjustments were made to improve CP performance. In patients with a confirmed PE, we identified select outcomes in PEDSnet and compared them to chart review to evaluate the ability to assess outcomes in PEDSnet.

Results

There were 152 PE patients identified by the best performing CP which included: 1) Anticoagulant prescribed, dispensed, or administered in an inpatient or emergency department encounter, 2) SNOMED-CT code for PE, 3) Chest CT or perfusion study in the encounter. CP performance characteristics were as follows: sensitivity 96 %, specificity 88 %, positive predictive value 90 %, and negative predictive value 95 %. Preliminary outcome validation highlighted areas for improvement in identifying transfusions and thrombolysis.

Conclusions

We report the first pediatric PE CP. Future research will include multicenter CP validation and improved identification of interventions and outcomes.