73 Use of Probabilistic Linkage to Create a Novel Database to Study the Care of Bronchiolitis in Pediatric Intensive Care Units (PICU)

Abstract

OBJECTIVES/GOALS: Bronchiolitis is a major cause of PICU admission, yet identifying best practices is limited by the lack of existing databases containing the needed demographic and clinical variables. We used probabilistic linkage to join the Virtual Pediatric System (VPS) and Pediatric Health Information Systems (PHIS) databases to overcome this data barrier. METHODS/STUDY POPULATION: We performed a single site study joining VPS and PHIS data. These national databases contain clinical (VPS) and billing and resource use (PHIS) data. Limits on the use of patient identifiers (PI) for multi-center research, makes direct linkage techniques impossible. To demonstrate that probabilistic linkage can accurately link VPS and PHIS records, we obtained our single site VPS and PHIS records and linked them using probabilistic linkage without PI and compared this to a gold standard linkage created with PI. We also compared demographic features of linked and unlinked records to assess the ability of probabilistic linkage to create a representative sample. RESULTS/ANTICIPATED RESULTS: We obtained 920 VPS records of patients with bronchiolitis and linked 91% (839/920) to a PHIS record with 4 (0.5%) false-positive matches. Characteristics of linked and unlinked records are compared in Table 1. Comparison of probabilistically linked and unlinked records showed no difference in median age in years (0.7 [Interquartile range (IQR) 0.3-1.5] v 0.7 [IQR 0.2-1.5], p = 0.76), median number of complex chronic conditions (0 [IQR 0-1] v 0 [IQR 0-1], p = 0.16), median Pediatric Index of Mortality 3 severity of illness scores (-4.6 [IQR -4.7 – 4.4] v -4.6 [IQR -4.7 – 4.4], p = 0.44), median days of PICU stay (4 [IQR 3-6] v 4 [IQR 2-6], p = 0.36), proportion female ( 44% v 46%, p = 0.82), or proportion of patients intubated (28% v 24%, p = 0.41). DISCUSSION/SIGNIFICANCE: Probabilistic linkage creates an accurate combined VPS-PHIS database. Extending our methodology to join data from all 38 hospitals contributing to VPS and PHIS will allow creation of a large database containing the demographic, treatment, and outcome data needed to enable Comparative Effectiveness Research of bronchiolitis care.