Development and validation of a prediction model for invasive syndrome in liver abscess patients based on LASSO regression: a multi-center retrospective cohort study in China

Background Patients with liver abscess are at high risk of developing invasive K. pneumonia liver abscess syndrome (IKPLAS), which can worsen survival and quality of life. Early identification of high-risk patients is crucial. This study aimed to identify risk factors for IKPLAS and develop a predictive model to guide early intervention. Methods We retrospectively collected data from 1,762 liver abscess patients at the First Hospital of Jilin University between 2015 and 2024. Patients were randomly divided into a training set and an internal validation set at a 7:3 ratio, and 203 patients from another hospital served as an external validation cohort. The SMOTE algorithm was applied to address data imbalance. Independent risk factors were identified using LASSO and logistic regression analyses, and the performance of different models was compared. Ultimately, a LASSO-based logistic regression model was used to construct a predictive nomogram. Model performance was comprehensively evaluated using the area under the receiver operating characteristic curve (AUC), decision curve analysis (DCA), clinical impact curve (CIC), and calibration curve. An online risk calculator was also developed for clinical use. Results Among 1,965 patients (1,304 males, 661 females; mean age 58.96 ± 13.07 years), 548 (28.9%) developed IKPLAS. Independent risk factors included CRP (OR = 1.005, 95% CI: 1.003–1.007), PLT (OR = 0.995, 95% CI: 0.994–0.997), Prior biliary disease (OR = 1.137, 95% CI: 1.025–2.571), Fever (OR = 2.196, 95% CI: 1.292–3.824), Pleural effusion (OR = 7.355, 95% CI: 4.883–14.761), Ascites (OR = 8.786, 95% CI: 5.141–9.342), Broth culture (OR = 2.264, 95% CI: 1.186–3.371), DM (OR = 2.516, 95% CI: 1.757–3.63), and TBIL (OR = 1.006, 95% CI: 1.002–1.010). The nomogram achieved AUCs of 0.960, 0.920, and 0.892 in the training, internal, and external validation sets, respectively, with good calibration and clinical utility. Conclusion We developed a nine-factor nomogram to predict individualized IKPLAS risk, demonstrating high discrimination and calibration, supporting early identification of high-risk patients and personalized management.