Development and validation of a machine learning-based nomogram model integrating LASSO regression and multiple
Lili Du1,2, Li Song3, Chao Gao4
1Department of Clinical Laboratory, Huangshi Central Hospital, Affiliated of Hubei Polytechnic University, Guilin North Road, No 16, Huangshi, Hubei, 435003, People's Republic of China.
Background:
Sepsis secondary to acute abdomen is a critical driver of clinical deterioration and mortality, with outcomes closely linked to sepsis severity and the timeliness of intervention. However, existing single biomarkers or scoring systems exhibit significant limitations in early sepsis prediction. Using clinical and laboratory data from patients with acute abdomen, this study developed a nomogram prediction model via advanced machine learning to identify risk factors for secondary sepsis and comprehensively validated its performance.
Methods:
Patients admitted with acute abdomen to Huangshi Central Hospital between 2019 and 2024 were included and randomly divided into training and validation sets at a 7:3 ratio. Predictors were selected using LASSO regression with tenfold cross-validation. A nomogram was constructed based on the selected variables. Model calibration was assessed using the Hosmer-Lemeshow test; discriminative ability was evaluated via the area under the receiver operating characteristic curve (AUC-ROC) in both training and validation sets. DCA and CIC were applied to assess predictive accuracy and clinical utility.
Results:
A total of 383 eligible patients were included. The incidence of sepsis was 38.29% (103/269) in the training set and 48.25% (55/114) in the validation set. Six independent predictors were identified: AKI, shock, TCO₂, hypoalbuminemia, elevated NLR, and lymphocytopenia. The nomogram demonstrated excellent discrimination, with an AUC of 0.929 (95% CI 0.898-0.959) in the training set and 0.934 (95% CI 0.891-0.977) in the validation set. Calibration curves and the Hosmer-Lemeshow test (training set: P = 0.4197; validation set: P = 0.9922) indicated strong agreement between predictions and observations. DCA showed significant net clinical benefit across a wide range of threshold probabilities (8-98% in the training set; > 8% in the validation set). CIC further confirmed close alignment between model-identified high-risk individuals and actual sepsis cases at prediction thresholds above 0.45 (training) and 0.35 (validation).
Conclusions:
This study developed and validated a nomogram for predicting sepsis risk in 383 patients with acute abdomen. Six independent predictors were identified. The model exhibited outstanding discrimination, strong calibration, and broad clinical applicability, demonstrating considerable potential for early risk stratification and clinical decision support.
