Dados do Trabalho

Título

Risk-adjusted length of stay to evaluate severe Community-Acquired Pneumonia (sCAP) outcomes: A machine learning analysis of 16,985 ICU admissions.

Objetivo

Severe Community-Acquired Pneumonia (sCAP) significantly impacts morbidity and mortality, leading to substantial resource allocation in ICUs. This study aims to validate the Standardized Length of Stay Ratio (SLOSR) as a measure of ICU performance in sCAP patients.

Métodos

We included 16,985 admissions with sCAP from January to December 2023 across 220 ICUs in 57 hospitals within a Brazilian private hospital network. Data were collected on demographics, source of admission, comorbidities, Simplified Acute Physiology Score (SAPS 3) variables, and ventilatory support use. Machine learning algorithms were applied to analyze the dataset, with model validations conducted using cross-validation techniques. The SLOSR, validated in 2023, was used to evaluate ICU efficiency based on individual baseline expected ICU length of stay (LOS). Verification included comparing metrics (Root Mean Squared Error [RMSE], Mean Absolute Error [MAE], Coefficient of Determination [R²]), as well as calibration plots and a funnel plot with unit-level (ICU) coefficients.

Resultados

Hospital mortality was 9.3% and ICU mortality 6.4%, with a median ICU LOS of 4 days.. The mean SAPS 3 score was 50, and 28.1% of patients required ventilatory support. The SLOSR demonstrated a grouped R² of 0.89. In the funnel plot, the median SLOSR was 1.13 (0.91-1.34), indicating that most units conformed to expected performance levels and thus suggesting effective benchmarking in ICU care for sCAP.

Conclusão

The validated SLOSR is an accurate tool for assessing ICU efficiency in sCAP. Its excellent calibration allows for reliable predictions of ICU length of stay and resource utilization, supporting value-based care strategies.