Validação do escore prognóstico Pediatric Index of Mortality (PIM 3) em uma unidade de terapia intensiva no Brasil

• Ano de defesa: 2016
• Autor(a) principal: Arantes, Karla Loyola de Oliveira
• Orientador(a): Garcia, Pedro Celiny Ramos
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Pontifícia Universidade Católica do Rio Grande do Sul
• Programa de Pós-Graduação: Programa de Pós-Graduação em Medicina/Pediatria e Saúde da Criança
• Departamento: Faculdade de Medicina
• País: Brasil
• Palavras-chave em Português: CRIANÇAS; MORTALIDADE INFANTIL; PROGNÓSTICO; UNIDADES DE TERAPIA INTENSIVA; PEDIATRIA; MEDICINA
• Área do conhecimento CNPq: CIENCIAS DA SAUDE::MEDICINA
• Link de acesso: http://tede2.pucrs.br/tede2/handle/tede/7027

Introduction: Prognostic scores are useful tools in assessing the effectiveness of treatments, mortality risk and quality of services, allowing the comparison between different Intensive Care Units as the implementation and improvement of treatment and public health policies and protocols. The PIM (Pediatric Index of Mortality) is one of the most widely used prognostic scores in pediatrics and has improved generating PIM 2 and PIM 3. The latest has not been validated in developing countries. Objectives: Validation of PIM 3 score in a tertiary pediatric hospital in southeastern Brazil, and comparison of its performance with the PIM 2, currently used. Methods: A contemporary cohort study undertaken between January 1 and December 31, 2014, at the Pediatric Intensive Care Unit of HEINSG (Hospital Estadual Infantil Nossa Senhora da Glória). The sample characterization was performed using the observed frequency, percentage, measures of central tendency and variability. The calibration of the scores was analyzed by the Hosmer-Lemeshow test setting, while the Z statistic Flora was used to evaluate the similarity between overall mortality and the one observed through the standardized mortality rate (SMR - Standardized Mortality Rate). For Flora z test, it is considered critical values for the non-null hypothesis rejected the two standard deviations (SD) (or between <1.96 and> -1.96). The area under the ROC curve (Receiver Operating Characteristic) was used to analyze the discrimination capacity of PIM2 and PIM3 models among patients who were discharged or died, and the assessment of the concordance between the measures of PIM 2 and PIM3 was performed using the Student t test for independent samples. The agreement between the measures of PIM 2 and PIM3 was evaluated by Bland & Altman plot. The significance alpha level used in the analyzes was 5% and 95% confidence interval. Data were collected in an Excel table, confirmed on medical records and later transferred to IBM SPSS software to perform all analyzes. Results: 293 patients were admitted to the PICU during the studied period, 38 of whom presented exclusion criteria. 35 (13.7%) of the 255 patients studied died. The average score PIM2 was significantly higher than the PIM3, and Flora Z statistics showed no difference between the overall mortality observed and the expected one in PIM2, but this difference was found in PIM3. The PIM2 score got an excellent discrimination (AUC = 0.830) and its sensitivity was 85.7, and the specificity was 69.1. On the other hand, the PIM 3 score had an acceptable discrimination (AUC = 0.748), while its sensitivity was 74.3, and its specificity was 67.7. The comparison between the areas under the ROC curve of PIM2 and PIM3 was significant (p = 0.015), showing that there is a difference between their areas, with better performance for PIM2 compared to PIM 3 (Z Flora 2.427). The Bland-Altman diagramme indicated that the 95% limits of concordance between the two versions of PIM ranged from -1.2 to 2.3, indicating that the measures are inconsistent. There is discordance of 10.6% above and below the limit ± 1.96 standard deviations (SD) between the mentioned values, that is about twice the tolerable 5%. Conclusion: In our study, the PIM 2 shows better results to discriminate those patients who will die. We suggest, based on these results, that data collection should be maintained using the 2 versions of the score in this unit. Than, these data could be reanalyzed with a larger sample, and these results could be compared with new studies conducted in locations where population have similar characteristics.