Validação do escore prognóstico “Pediatric Risk of Mortality IV” em unidades de terapia intensiva pediátricas brasileiras

Abstract

This thesis evaluates the performance of the Pediatric Risk of Mortality IV (PRISM IV) score in a cohort of patients admitted to Brazilian pediatric intensive care units (PICUs). Three methods were used to evaluate the performance of the score: (i) the standardized mortality ratio (SMR), (ii) the discrimination of the score through the analysis of the area under the ROC curve (AUROC) and (iii) the calibration of the score using the calibration belt, being the first Brazilian study to use this technique to evaluate this score. The first article aimed to evaluate the performance of the PRISM IV score in the cohort of patients admitted to the 36 participating PICUs. The results revealed that the observed hospital mortality (2.1%) was higher than the predicted one (1.56%), with a SMR of 1.32 (95% CI 1.16 - 1.50). PRISM IV demonstrated good discrimination (AUROC 0.86 [95% CI 0.83-0.89]), distinguishing well between survivors and non-survivors, but poor calibration poor. The score underestimated mortality in a wide range of predicted risk, between 2% and 61%, although it showed good calibration at the extremes of risk (<2% and >61%). Since the start of the study coincided with the onset of the COVID-19 pandemic in Brazil, we subdivided the analyses into the pre- and post-vaccine periods, to assess how differences in the distribution of cases in the study population between the two periods would affect the performance of PRISM IV. In both periods, the analyses also indicated good discrimination (pre-vaccine: AUROC 0.84 [95% CI 0.80-0.88]; post-vaccine: AUROC 0.90 [95% CI 0.86-0.94]), but underestimation of mortality in a wide range of intermediate risk (pre-vaccine: range 2-98%; post-vaccine: range 3-46%). The second manuscript presents a performance analysis in subgroups categorized by type of management (public or private), population served (exclusively pediatric or neonatal and pediatric), and type of care (with or without affiliation to postgraduate programs). The analyses indicated good discrimination in all groups, but poor calibration in most of them. The type of PICU management was the factor that most influenced the performance of the score. Public PICUs demonstrated higher SMR and worse PRISM IV calibration, underestimating mortality in a wide risk range. In contrast, private PICUs showed good calibration in almost all predicted mortality ranges, overestimating mortality in the risk range < 9%, which corresponds to the majority of the PICU population. Variations in patient profiles (such as baseline health status and comorbidities) and structural and organizational factors of the units are possible factors related to the observed differences in PRIM IV performance and deserve in-depth investigation in future research. The study highlights the importance to evaluate the necessary adjustments to PRISM IV to adapt it to the Brazilian population, in parallel with investments in general and intensive pediatric public health, in order to reduce inequalities and achieve good score performance, translating good quality of care provided in all Brazilian PICUs.