Dados do Trabalho

Título

Machine Learning: Predicting the ICU Lenght of Stay Based on the SAPS3 Score at admission.

Objetivo

Estimates of mortality at admission is known to be crucial in ICU management and benchmarking, and the utilization of SAPS3 score is a widespread practice. Nonetheless, there is no previous study in literature for the assessment of a patient's length of stay (LOS) based on the severity score, even though this is also widely recognized as an elementary paradigm of efficacy at ICUs. This study has a two fold goal: firstly to investigate a connection between SAPS3 and LOS; and finally, to determine the mathematical relationship between the two, so that an ICU LOS can be real time, individually predicted, at patient admission.

Métodos

The studied population is a cohort of all 1,717 patients admitted to ICU whose SAPS3 had been determined, in single, publicly funded hospital in São Paulo, SP. This sample was taken from December 2020 to current, ongoing August 2023. Data is fully accountable, continuously collected at a MS Access (Microsoft) SQL database. All LOS data was plotted against SAPS3 index, and a function of polynomial regression was determined. Statistical analysis was made at the mentioned database, IBM SPSS Statistics (IBM Corp) and Python's Panda library.

Resultados

SAPS3 and LOS has a general positive correlation, whereas very high SAPS3 implies progressively shorter LOS. At the time of this manuscript, the function that determines the relation between the two variables is given by y= -0.0031x**2 + 0.4504x - 2.3641

Conclusão

.This is the first known study to openly describe and determine LOS estimates based on SAPS3 severity score.