Type 3c diabetes (T3cD) is a distinct form of diabetes related to diseases of the exocrine pancreas. In the setting of pancreatitis, T3cD affects 4-9% of children and is associated with long-term morbidity and increased mortality. T3cD is understudied especially in pediatrics, and we cannot yet predict which children will develop T3cD. My overall goal is to better understand risk factors for this diabetes development and which combination of risk factors best predicts diabetes development. However, this work is difficult to do at a single center given the overall rarity of T3cD in pediatrics; thus, utilization of largescale databases is critical. In this mentor-guided research, I propose to use PEDSnet to address this knowledge gap. The first necessary step is to develop a computable phenotype, or an algorithm to accurately identify a cohort of patients at risk of T3cD (i.e. those with pancreatitis). My mentors and I have been refining this phenotype with our institution’s data and now seek to validate it at another institution (CHOP).
I propose to develop and validate a computable phenotype for pediatric pancreatitis using EHR data from the large PEDSnet database. My mentors and I have already been iteratively revising our algorithm and currently have a computable phenotype sensitivity of 99.6%, specificity 71.7%, positive predictive value 85.4%, and negative predictive value 99.0%. While this is an improvement over our initial definition, we are now working to improve specificity by determining common features among false positives. Our next step is to work with collaborators from a second PEDSnet institution, Children’s Hospital of Philadelphia, to validate our computable phenotype at a second site. Dr. Hatch-Stein at CHOP has agreed to collaborate with us on this work.