Abstract
IMPORTANCE: This is the first Brazilian study evaluating the performance of Pediatric Risk of Mortality (PRISM) IV and the first to use the calibration belt technique. OBJECTIVES: This study aimed to evaluate the performance of PRISM IV in a large cohort of patients admitted to Brazilian PICUs. DESIGN, SETTING AND PARTICIPANTS: This is a longitudinal, prospective, multicenter study conducted in 36 Brazilian PICUs with children between 29 days and 18 years old admitted from March 2020 to March 2022. MAIN OUTCOMES AND MEASURES: PRISM IV's performance was assessed using the standardized mortality ratio (SMR), the area under the receiver operating characteristic curve (AUROC) with 95% CI, and the calibration belt with 80% and 95% CI. RESULTS: A total of 12,046 patients from 36 PICUs were included. Observed overall in-hospital mortality was higher than predicted: observed = 249 (2.1%) × predicted = 188.1 (1.56%) (SMR = 1.32 [95% CI, 1.16-1.50]); discrimination was good (AUROC = 0.86 [95% CI, 0.83-0.89]), and calibration was poor, underestimating mortality over a wide range of predicted mortality (2-61%). To explore the impact of the COVID-19 pandemic on PRISM IV's performance, we divided the study period into prevaccine and postvaccine. In the prevaccine period, the SMR was 1.38 (95% CI, 1.17-1.62), the AUROC was 0.84 (95% CI, 0.80-0.88), and the range of miscalibration was broader than in the total cohort (underestimation in the 2-98% range). In the postvaccine period, the SMR was 1.26 (95% CI, 1.03-1.51), the AUROC was 0.90 (95% CI, 0.86-0.94), and the calibration belt underestimated mortality in a narrower range of 3-46% of predicted mortality. CONCLUSIONS AND RELEVANCE: PRISM IV showed good discrimination but miscalibration across a wide range of predicted mortality and different COVID-19 pandemic periods in a large cohort. Further research with subgroup analyses are needed to develop strategies to improve the performance of PRISM IV in different and heterogeneous Brazilian healthcare contexts.