prolonged LOS

prolonged LOS predicts the risk of prolonged length of stay (LOS) in patients with type 2 diabetes mellitus (T2DM) by constructing and validating a statistical risk model for inpatient LOS.

Key Features:

• Predictive Modeling: Employs least absolute shrinkage and selection operator (LASSO) regression for variable selection and multivariable logistic regression to construct a risk model for prolonged LOS in T2DM patients.
• Nomogram Visualization: Visualizes the predictive model as a nomogram to quantify individual predictor contributions to predicted LOS probability.
• Discrimination Assessment (ROC/AUC): Uses receiver operating characteristic (ROC) curves and area under the curve (AUC) metrics to assess model discrimination.
• Calibration Assessment: Applies calibration curves to evaluate agreement between predicted probabilities and observed outcomes.
• Clinical Utility Analysis: Uses decision curve analysis and clinical impact curves to assess net clinical benefit and practical value of the model.
• Multi-cohort Validation: Validates performance across a training set, an internal validation set, and two external validation sets with reported AUC values ranging from 0.743 to 0.803.

Scientific Applications:

• Risk Stratification: Identifies hospitalized T2DM patients at higher risk of extended LOS to inform targeted clinical interventions.
• Resource Allocation and Planning: Informs hospital bed management and resource planning by predicting patients likely to require prolonged hospitalization.

Methodology:

Variable selection was performed with LASSO (least absolute shrinkage and selection operator) regression and model building used multivariable logistic regression; the model was visualized with a nomogram and evaluated using ROC/AUC, calibration curves, decision curve analysis, and clinical impact curves across training, internal, and two external validation sets (AUC 0.743–0.803).