Development and validation of a LASSO-based predictive model for inadvertent hypothermia in ICU patients

To develop a risk predictive model for inadvertent hypothermia (IH) in intensive care unit (ICU) patients and to validate the accuracy of the model.

The data was collected at the ICU of a tertiary hospital in Zunyi from November 2022 to June 2023 for model construction and internal validation. Data collected at the ICU of another tertiary hospital in Zunyi from July 2023 to December 2023 was used for external validation. The Least Absolute Shrinkage and Selection Operator (LASSO) was used to screen for strongly correlated predictors and build a predictive model, which was presented in the form of a nomogram and perform internal and external validation.

This study included a total of 720 participants, the incidence of IH in ICU patients was 18.19%. Six predictor variables were ultimately screened to construct the model: risk of IH in ICU patients = 1/(1 + exp-(-3.631 + 0.984 × catecholamines - 3.200 × antipyretic analgesics + 1.611 × RRT + 1.291 × invasive mechanical ventilation + 1.160 × GCS + 0.096 × lactate)). The results of the prediction model evaluation showed an AUC of 0.852 (95%CI: 0.805, 0.898) and internal validation yielded a C-statistic of 0.851. The Hosmer-Lemeshow test showed that x²  = 7.438, p = 0.282 and the calibration curve showed that the actual prediction was close to the ideal prediction. The results of the DCA showed that the model is able to provide effective evidence to support clinical decision making. External validation showed an AUC of 0.846 (95%CI: 0.779, 0.913). The Hosmer-Lemeshow test showed x²  = 13.041, p = 0.071 and the calibration curve was close to the ideal prediction situation.

The IH predictive model for ICU patients constructed in this study passed both internal and external validation, and has good differentiation, calibration, clinical utility, and generalizability, which can help healthcare professionals to effectively identify high-risk groups for IH in the ICU.