Dados do Trabalho

Título

Validation of a new data-driven SLOSR ICU efficiency measure compared to the traditional SRU: A multicenter study of 18,267 admissions in 85 Brazilian units.

Objetivo

To validate the Standardized Length of Stay Ratio (SLOSR), a new resource use indicator based on Length of Stay (LoS) estimated from a proposed Machine Learning Model and compared to Standardized Resource Use (SRU) in a large dataset of Brazilian ICU patients.

Métodos

We included 18,267 admissions from January to December 2019, in 85 ICUs from 38 hospitals of a Brazilian hospital network. We proposed the SLOSR to evaluate ICU efficiency, which uses, as expected ICU LoS (the prediction model proposed was already published in 2021). To compare SRU and SLOSR models at the ICU level, we used R, calibration plots, and funnel plots.

Resultados

The calibration showed a higher concordance between observed and expected ICU LoS estimates per ICU for SLOSR compared to SRU (R =0.57 vs 0.90). In the funnel plot, the median SRU was 0.79 (0.69, 0.90), while the SLOSR was 1.00 (0.87, 1.10). Two ICUs (units 20 and 36) that were outside the higher 95% control limit in SRU changed to a position within the limits in SLOSR. Another three ICUs (51, 63, and 75) changed from a position within the control limits of SRU to outside the higher 95% control limit in SLOSR. Our results indicated that SRU presented an overestimation of the expected ICU LoS, whilst the SLOSR showed a higher concordance to obtain a standardized metric for ICU efficiency

Conclusão

The SLOSR is a new metric that may improve the current assessment of ICU efficiency by using
a data-driven ML technique with data from contemporaneous ICUs.