Predictors and nomogram of midazolam-induced somnolence: a retrospective cohort study

Abstract Background This study aimed to develop and validate a nomogram for predicting the risk of prolonged somnolence in hospitalized patients receiving intravenous midazolam. Methods We retrospectively collected clinical data from hospitalized patients at Liuzhou People’s Hospital who were treated with intravenous midazolam and and required flumazenil for sedation reversal between July 2020 and December 2021. These patients were randomly divided in a specific ratio (7:3) into a training cohort (n = 710) and an internal validation cohort (n = 303). Univariate analysis was performed in the training cohort to screen for relevant factors associated with prolonged somnolence. Subsequently, least absolute shrinkage and selection operator (LASSO) regression and multivariable logistic regression were employed to identify independent risk factors for prolonged somnolence, based on which a predictive nomogram model was constructed. The receiver operating characteristic (ROC) curve, calibration curve, and Hosmer-Lemeshow (H-L) test were employed to assess the discrimination and calibration of the model. Decision curve analysis (DCA) was performed to evaluate the clinical utility of the model. Results A total of 1,013 patients were included. Multivariate logistic regression analysis revealed that serum albumin (ALB) < 35 g/L (OR = 1.73, 95%CI 1.20–2.50, P < 0.01), male sex (OR = 1.41, 95%CI 1.04–1.91, P = 0.026), and alanine aminotransferase (ALT) > 160 U/L (OR = 3.16, 95%CI 1.11–11.37, P = 0.046) were independent risk factors for midazolam-induced somnolence in patients receiving intravenous midazolam. A nomogram was developed and validated using the three independent risk factors identified above. The model demonstrated an acceptable discrimination, with an area under the ROC curve of 0.603 in the training cohort and 0.615 in the internal validation cohort. The calibration curve and H-L test both demonstrated good consistency of the model, with the training cohort showing χ²=0.0028 and P = 1.00 and the validation cohort showing χ²=0.0054 and P = 0.997. DCA indicated that the model provided net clinical benefit when the threshold probabilities ranged from 3% to 67% for the training cohort and 3% to 85% for the validation cohort. Conclusion To facilitate the rational administration of midazolam, we developed a nomogram that can help assess the likelihood of prolonged somnolence in patients receiving intravenous midazolam.