2300 Children's Plaza
Box #30 Chicago, IL 60614
08/09/2011
Multivariate data analyses have the potential to enrich the use of the complex plethora of data gathered in the care of critically ill patients, with the goal of identifying risk factors associated with increased mortality. In 2010, Dr. Chaudhury proposed a retrospective data review of the outcomes of stem cell transplant patients who were admitted to the Pediatric Intensive Care Unit (PICU). This study was designed to identify predictors for mortality in this cohort of patients. Specifically, she hoped to establish a prognostic scoring system using multiple criteria to identify patients early in the course of PICU admission who would be at risk for increased morbidity and mortality. Doing so would allow us to intervene in a timely manner to improve overall outcomes in this high risk group of patients.
After obtaining approval from Children’s Memorial’s Institutional Review Board, Dr. Chaudhury applied a hierarchical cluster analysis of a large dataset ‘XenoBase’ with heterogeneous data classes (laboratory, physiology, device settings and socioeconomic status). This data was collected as part of routine care of children who received HSCT and were admitted to the PICU for continuous monitoring. About 120 post HSCT patients who were admitted to the PICU between 2003 and 2008 were identified. Of these patients, 106 were identified in XenoBase and 85 of these patients were identified in whom complete data was captured. To facilitate this study, Dr. Chaudhury utilized XenoBase, a unique, bio-informatics system that provides greatly enhanced capacity to peer into the biologic patterns of illnesses and their treatments. It offers the means to capture and analyze specified physiologic data from the existing records of thousands of patients, giving us the means to explore clinically relevant questions related to the care these patients received and their outcome.
Dr. Chaudhury sought to provide proof of concept that the use of this approach in pediatric HSCT patients could generate new insights into complex interactions of physiologic and biochemical parameters which affect outcome after admission to the PICU with the hope of using this data to generate a prognostic score to determine probability of survival in this complex patient group. She performed a retrospective analysis on these 85 patients to determine the factors that might predict outcome by cluster analysis, comparing survivors (n=49) and non-survivors (n= 36) and identifying multiple factors that, clustered together, were different between the two groups. We obtained a cluster analysis of the first 14 days post transplant and identified an association between abnormalities in PT/PTT coagulation measures, iCa (ionized calcium), bilirubin, BUN/Cr and end tidal CO2, in addition to other unexpected associations. These abnormalities in the first 14 days clustered in the non survivors.