Development and validation of a predictive model for assessing the risk of recurrence in patients with Klebsiella pneumoniae-induced pyogenic liver abscess

Abstract Background Pyogenic liver abscess (PLA) is a potentially life-threatening bacterial infection of the liver that can cause suppurative lesions. Klebsiella pneumoniae (KP) is the primary pathogen responsible for causing PLA in China and across the Asia-Pacific region. Although various treatment modalities, including antimicrobial therapy combined with percutaneous drainage, have significantly reduced mortality in patients with KP-induced liver abscess (KPLA), recurrence remains a persistent clinical challenge. This study identified the risk factors, developed a predictive model for assessing the risk of recurrence in patients with KPLA, and constructed a nomogram for model visualization. Methods This retrospective study analyzed clinical data from 486 patients with KPLA admitted to the Affiliated Hospital of Chengde Medical University between June 1, 2015, and June 1, 2024. The patient population was randomly stratified in a 7:3 ratio into training (n = 340) and validation (n = 146) cohorts. The risk factors for recurrence in patients with KPLA were identified through univariate and multivariate logistic regression analyses. A predictive model was constructed, and its predictive accuracy was evaluated based on the area under the receiver operating characteristic (ROC) curve (AUC) and analyses of calibration, decision, and clinical impact curves. A nomogram was additionally developed for model visualization. Results Among the 486 enrolled patients, 64 (13.2%) experienced recurrence (recurrence group), while the remaining 422 patients were included in the non-recurrence group. The predictive model integrated five independent risk factors, including a history of type 2 diabetes mellitus (T2DM), malignant neoplasm, cerebral infarction, septic shock, and the Sequential Organ Failure Assessment (SOFA) score. The model demonstrated excellent calibration (Hosmer-Lemesow χ² = 5.301, P = 0.623), achieving AUC values of 0.824 and 0.819 for the training and validation cohorts, respectively. The clinical utility of the model was further confirmed through analysis of the decision and clinical impact curves. Conclusion A history of T2DM, malignant neoplasm, cerebral infarction, septic shock, and the SOFA score serve as independent predictive factors for assessing the risk of recurrence in patients with KPLA. The predictive model, constructed based on these factors, can effectively estimate the risk of recurrence in patients with KPLA, thereby facilitating the identification of high-risk populations.