Linking the Virtual Pediatric System and Pediatric Hospital Information System Databases to Create Novel PICU Research Datasets

BACKGROUND: Lack of a robust, national PICU database hinders epidemiologic and quasi-experimental research of PICU care. Combining the virtual pediatric system (VPS) and pediatric health information system (PHIS) databases would overcome this data gap. However, limits on the use of variables containing protected health information (PHI) have prevented prior large-scale VPS-PHIS linkage. We sought to link VPS and PHIS records from 31 hospitals without using PHI. METHODS: We performed a database linkage study utilizing a sequential direct + probabilistic linkage approach with records from 31 hospitals contributing to VPS and PHIS for patients discharged between January 1, 2017 and December 31, 2022. Using directly linked records as a reference, we assessed the accuracy of probabilistically linked pairs with data from four hospitals by describing the percent of false-negative and false-positive links. We also compared demographic and clinical variables between linked and unlinked records using Cohen's Effect size (CES). RESULTS: We obtained 337,836 VPS and 3,515,595 PHIS records without PHI. In total, we linked 319,955 of 337,836 (94.7%) VPS records. Direct linkage contributed 274,345 of 319,955 (85.7%) of all record linkages. Probabilistic linkage contributed 45,610 of 319,955 (14.3%) of all record linkages. Assessment of the probabilistic linkage found 0.21% false-negative links and no false-positive links. We found negligible imbalance by CES for gender (% female in linked vs. unlinked: 44.8% vs. 43.6%, CES 0.024), age in years (median [interquartile range (IQR)]: 5.1 [1.21-13.22] vs. 5.53 [1.53-13.38], CES 0.054), Pediatric Risk of Mortality 3 score (median [IQR]: 0.44 [0.3-0.9] vs. 0.37 [0.29-1.02], CES 0.096), or days in PICU (median [IQR]: 1.55 [0.8-3.47] vs. 1.33 [0.79-3.45], CES 0.031). CONCLUSIONS: Using a combined direct and probabilistic linkage approach, we created a highly accurate linkage VPS and PHIS database records from 31 hospitals without using PHI. This combined database can be used to create novel datasets for PICU research.