Length of stay in pediatric intensive care unit: prediction model.

Objective: To propose a predictive model for the length of stay risk among children admitted to a pediatric intensive care unit based on demographic and clinical characteristics upon admission.

Methods: This was a retrospective cohort study conducted at a private and general hospital located in the municipality of Sao Paulo, Brazil. We used internal validation procedures and obtained an area under ROC curve for the to build of the predictive model.

Results: The mean hospital stay was 2 days. Predictive model resulted in a score that enabled the segmentation of hospital stay from 1 to 2 days, 3 to 4 days, and more than 4 days. The accuracy model from 3 to 4 days was 0.71 and model greater than 4 days was 0.69. The accuracy found for 3 to 4 days (65%) and greater than 4 days (66%) of hospital stay showed a chance of correctness, which was considering modest. Conclusion: Our results showed that low accuracy found in the predictive model did not enable the model to be exclusively adopted for decision-making or discharge planning. Predictive models of length of stay risk that consider variables of patients obtained only upon admission are limit, because they do not consider other characteristics present during hospitalization such as possible complications and adverse events, features that could impact negatively the accuracy of the proposed model.