Patients, samples and data collection
The study was approved by the Institutional Review Board at Cincinnati Children’s Hospital Medical Center. All patients under the age of 18 years old undergoing surgery requiring CPB for correction of congenital heart disease between November 2016 and November 2020 were screened for eligibility. Patients were only included for their index surgery, except patients with single ventricle physiology. Patients with single ventricle physiology could be re-enrolled in the study for each separate surgical stage, i.e., stage 1 palliative surgery, Glenn operation, Fontan operation, and/or biventricular repair. Patients undergoing CPB for heart or lung transplantation, patients requiring immunosuppression, and patients with suspected or proven infection were excluded. Three-hundred and fifty-nine patient encounters (293 unique patients) were consented for the study. Of these, 306 encounters were included in the analysis, because both 4 and 12 hour biomarker samples collected within the specified time. Baseline demographic, clinical, and laboratory data needed to calculate severity of illness scoring and determine organ dysfunction were extracted from the electronic medical record (EMR). To minimize clinically unnecessary blood draws, laboratory data to assess for organ dysfunction was only collected at discretion of the managing clinical team.
Definitions
The Society of Thoracic Surgery-European Association for Cardiothoracic Surgery (STAT) mortality category [28,29] was used to account for risk related to surgical complexity. Pre- and postoperative severity of illness was assessed using Pediatric Risk of Mortality score III (PRISM III) [30]. Organ dysfunction was defined via adaption of Goldstein criteria [31] to account for differences in the postoperative congenital heart disease population when compared to the pediatric sepsis population, Table 1. Persistent MODS was defined a priori as dysfunction of 2 or more organ systems on postoperative day 5. As an additional measure of organ dysfunction, daily Pediatric Logistic Organ Dysfunction-2 [PELOD-2] scores were calculated preoperatively and for the first 5 postoperative days [32,33].
Clinical and surgical management
All patients received methylprednisolone (30 mg/kg) as part of the CPB circuit prime. Neonates and patients in the hospital prior to their scheduled operation received an additional dose of methylprednisolone (30 mg/kg) the morning of surgery (prior to CPB initiation). Choice of anesthesia was not standardized and left to the decision of the cardiac anesthesiologist. All patients received either modified ultrafiltration and/or continuous ultrafiltration intraoperatively, based on surgeon preference. The need for additional steroids and use of postoperative peritoneal dialysis was left to the discretion of the clinical team.
Biomarker Collection
Biomarkers were collected 4 and 12 hours post-CPB, based on studies suggesting peak inflammation occurs within 24 hours of CPB separation [4-6,8,21,34]. Blood was collected within a +/- 60 minute window, spun down to serum, and stored at -80C until ready to be analyzed. Seven PERSEVERE biomarkers were measured in this study: granzyme B (GZMB), heat shock protein 70 kDa 1B (HSPA1B), interleukin 1a (IL-1a), interleukin 8 (IL-8), C-C chemokine ligand 3 (CCL3), C-C chemokine ligand 4 (CCL4), matrix metalloproteinase 8 (MMP-8). Serum biomarker concentrations were measured according to manufacturer’s instructions using the HSP2MAG-63K multiplex bead platform (MILLIPLEXTM MAP Human Sepsis Magnetic Bead Panel 2-Immune Response Multiplex Assay) designed by the EMD Millipore Corporation (Billerica, MA, USA).
Statistical Analysis
Descriptive statistical analyses were performed using R (version 4.0.4). Demographic, clinical, and biomarker data were described using medians with interquartile ranges (IQR), means with standard deviations, or frequencies with percentages as appropriate. Comparisons of data for patients with and without persistent MODS were performed using the Kruskal-Wallis, chi-squared, or Fisher’s exact tests as appropriate. Multivariate regression analysis, controlling for clinical data, was performed to examine the relationship between biomarker concentrations at 4 and 12 hours and risk for development of MODS.
Classification and regression tree (CART) analysis was used to determine biomarker cut-points and derive a decision tree (Salford Predictive Modeler v6.6, Salford Systems, San Diego, CA) [35]. Candidate prediction variables for derivation of the decision tree were as follows: all seven PERSEVERE biomarkers at 4 and 12 hour time points, change in PERSEVERE biomarker levels from 4 to 12 hours, age in months (included as both a continuous and dichotomous variable), single ventricle status, history of prematurity, CPB time, maximum vasoactive inotropic score (VIS) and STS-EACTS mortality category. Clinical predictor variable selection was based on extant literature [36-39]. Tuning parameters determined a priori included: 10-fold cross validation, at least one of the paired terminal daughter nodes contains ≥ 5% of the subjects in the root node, and no predictor variables repeated within one of the two main branches. Performance of the decision tree was determined by generating a classification table of true versus predicted status and calculation of discrimination metrics including sensitivity, specificity, positive and negative predictive values, and area under the receiver operating curve (AUROC). We compared our prediction model, which we will refer to as PERSEVERE-CPB, to PRISM III and STS-EACTS mortality category, as they are widely accepted and validated risk assessment and severity of illness scoring systems this patient population, using the AUROC, sensitivity, and specificity. We further compared PERSEVERE-CPB to the 24 hour postoperative PELOD-2 score, as PELOD-2 is a validated scoring system for organ dysfunction [32].
IL-8 was the first level decision rule in PERSEVERE-CPB. As such, we generated a classification table to evaluate the utility of IL-8 as an independent predictor of MODS.
Finally, we stratified the cohort into risk category based on high, intermediate, and low risk terminal nodes of our model. Using risk categories (referred to as PERSEVERE risk category), we evaluated the association of risk and administration of post-operative steroids for hypotension.