Linking the VPS and PHIS Databases to Create Novel Pediatric ICU (PICU) Research Datasets

Sunday, March 22, 2026

11:00 AM - 12:00 PM Central Time

Location: Connections Central - RST 18

BF

Brian F. Flaherty, MD MSCI

University of Utah, Utah

Disclosure(s): No relevant financial relationship(s) to disclose.
SO

Seth Otto, MS

Biostatistician
University of Utah, Data Coordinating Center

Disclosure information not submitted.
CO

Cody S. Olsen, MS

Senior Biostatistician
University of Utah, Data Coordinating Center

Disclosure information not submitted.
GS

Gerardo Soto-Campos, MS, PhD

Director of Analytics
Virtual Pediatric Systems, LLC

Disclosure information not submitted.
MH

Matt Hall, PhD

Principal Biostatistician
Children's Hospital Association

Disclosure information not submitted.
RW

Randall C. Wetzel, BS, MB, MBA

CEO; Professor
Childrens Hospital Los Angeles

Disclosure information not submitted.
TR

Tom B. Rice, MD

Chief Medical Officer; Professor Emeritus
Virtual Pediatric Systems, LLC; Medical College of Wisconsin, Department of Pediatrics, Division of Pediatric Critical Care

Disclosure information not submitted.
HK

Heather T. Keenan, MDCM PhD

Professor
University of Utah, Department of Pediatrics, Division of Pediatric Critical Care

Disclosure information not submitted.
LC

Lawrence J. Cook, PhD

Professor
University of Utah, Department of Pediatrics, Division of Pediatric Critical Care

Disclosure information not submitted.

Introduction: The evidence base for the care of numerous PICU conditions is underdeveloped. Observational studies using existing databases are an attractive means to rapidly improve the PICU evidence base. However, no single database contains all the variables needed for PICU specific research such as PICU diagnosis, demographics, illness severity, diagnostics, treatments, costs, and outcomes. To address this data gaps, we aimed to link the national Virtual Pediatric Systems’ (VPS) PICU clinical registry with the national Children’s Hospital Association’s Pediatric Health Information System (PHIS) administrative database.

Methods: We obtained de-identified records for patients admitted at 31 children’s hospitals participating in VPS and PHIS between 1/1/2017-12/31/2022. Data included patient identification numbers and dates of birth, admission, and discharge that were hashed/shifted consistently in both data systems and demographic variables. We used a 2-stage linkage process. First, we directly linked VPS and PHIS records that agreed exactly on patient identification number and dates of birth and admission. Unlinked records from Stage 1 were used in Stage 2, a probabilistic linkage that linked records based on their degree of matching on demographic and date variables. To assess the accuracy of Stage 2, we utilized Stage 1 linked records from 4 hospitals as a reference to estimate the percent of false negative and false positive links when these records were run through Stage 2. To assess for bias in the linkage, we also compared sex and age distributions between all linked and unlinked records.

Results: We obtained 337,756 VPS records for linkage. Our process linked 319,878/337,756 (95%) of VPS records to a PHIS record. Stage 1 generated 80% of the links. Assessment of our probabilistic Stage 2 linkage found 0.2% false negative and 0 false positive links. We found minor differences in sex (Female: Linked 45% v Unlinked 43%) and age (0 to < 2 years: 34% v 30%; 2 to < 12 years: 37% v 30%; 12 to < 18 years 24% v 23%; 18: 4% v 5%) with Cohen’s d < 0.1 for all comparisons.

Conclusions: Our linkage method linked 95% of VPS records to a PHIS record with high accuracy. Use of our linkage method will allow researchers to build large, multi-site de-identified datasets to study PICU care.