Marcio Soares, Lunna Perdigão Borges, Gastón Burghi, Pedro Kurtz, José Raimundo A. de Azevedo, Carlos Eduardo Brandão, Aloysio Saulo Breves Beiler, Niklas Söderberg Campos, Liane Oliveira Cavalcante, Mário Diego Teles Correia, Victor de Souza Cravo, Pedro Henrique Barbosa D’Almeida, Flávio Geraldo Rezende de Freitas, Thais de Almeida Machado, Marcelo de Oliveira Maia, Edson Silva Marques, Gloria Martins, Ulisses de Oliveira Melo, Laura Herranz Prinz, Sílvia Regina Ramos, Thiago Gomes Romano, Marcos Soares Tavares, Suzana Margareth Lobo, Jorge Ibrain Figueira Salluh, Ederlon Rezende
OBJECTIVE: To compare the performance of the standard equation (SAPS 3-SE) and a customized version (SAPS 3-Custom) of the Simplified Acute Physiology Score 3 in a contemporary cohort of Brazilian and Uruguayan intensive care unit patients. METHODS: We conducted a retrospective cohort study of 262,198 adults admitted to 177 intensive care units between 2022 and 2023. Discrimination was assessed using the area under the Receiver Operating Characteristic curve (AUROC), and calibration by comparing predicted and observed mortality in calibration curves. RESULTS: Of patients 70% were medical, and 21% were scheduled for surgery; mean SAPS 3 was 46.6 ± 16.0. Median intensive care unit and hospital stays were 3 (1 - 5) and 8 (4 - 16) days, respectively. Intensive care unit mortality was 10.6% and hospital mortality was 16.4%. Predicted mortality was 19.0% for SAPS 3-SE and 16.6% for SAPS 3-Custom. Both models had excellent discrimination (AUROC = 0.841). SAPS 3-SE overestimated mortality across all risk deciles, whereas SAPS 3-Custom achieved uniform agreement between predicted and observed values. Standardized mortality rates were 0.86 (95%CI 0.85 - 0.87) for SAPS 3-SE and 0.98 (0.98 - 0.99) for SAPS 3-Custom; standardized resource use rates were 0.90 (0.90 - 0.91) and 0.98 (0.97 - 0.98), respectively. At the intensive care unit level, SAPS 3-Custom produced standardized mortality rates (0.95 [0.77 - 1.17]) and standardized resource use rates (0.97 [0.82 - 1.23]) distributions centered around 1.0, unlike SAPS 3-SE, which yielded lower values. Findings were consistent for medical and surgical subgroups. CONCLUSION: In this large, contemporary cohort, SAPS 3-Custom demonstrated superior calibration and accuracy over SAPS 3-SE, supporting its use for performance evaluation and benchmarking in intensive care units in Brazil and Uruguay.