Validation and refinement of the catheterization RISk score for pediatrics (CRISP score): An analysis from the congenital cardiac interventional study consortium

Objectives To externally validate the CRISP score, and determine if refinements might improve clinical utility. Background The CRISP score estimates risk of serious adverse events (SAEs) for pediatric catheterization. Methods Pediatric (age < 18) procedures reported to the Congenital Cardiovascular Interventional Study Consortium registry from 05/08 to 09/17 (n = 29,830, 27 centers) were divided into a development dataset of 14,784 earlier procedures, and a validation dataset of 15,046 more recent procedures. The development dataset was used to refit the original CRISP model, and to develop a revised(r) CRISP score, consisting of entirely pre-procedurally collected data. The validation dataset was then used to compare model fit and risk prediction between CRISP, rCRISP and two existing risk scores using Akaike's (AIC), Schwarz's (BIC) Bayes Information Criteria, -log Likelihood (N2LL), area under the receiver operator curve and chi-square goodness-of-fit statistic (across 5 risk categories). Results Overall 4.31% of patients experienced at least one SAE with frequency increasing from 1.08% in CRISP category 1 to 27.34% in category 5. Both CRISP and rCRISP (entirely pre-procedural) predicted risk of SAEs well, with observed to predicted ratios ranging from 0.71 to 1.18 across the 5 risk categories. Compared to the original CRISP score, rCRISP demonstrated less optimal model fit (higher AIC, BIC, and N2LL) but similar risk prediction (C-statistic = 0.71 vs. 0.70; chi-squared statistic = 6.77 vs. 6.85). Conclusion The CRISP score accurately predicts procedural risk. With minor modifications, the revised version (rCRISP) performed well with arguably greater clinical utility as an entirely preprocedural risk model.