Prediction Models for Infection Severity after Thermal Ablation of Liver Malignant Tumors

To identify predictors and develop models for predicting infection severity following percutaneous thermal ablation for liver malignancies.

A total of 8655 ultrasound (US)-guided percutaneous thermal ablations at a single center (January 2010-November 2023) were retrospectively reviewed. The study included 172 patients (mean age, 58.5 years [SD ± 11.4]; 145 men), with 86 infected (18 severe and 68 nonsevere) and 86 noninfected patients. Patients developing infection within 30 days were classified as severe (septic shock or infection-related death) or nonsevere. A noninfected control group was matched by the treatment date. A scoring system was developed to identify infection in patients with postablative fever. Two multivariate logistic regression models were constructed: 1 using preablative variables and another combining preablative and postablative characteristics.

Among 8655 ablations, 86 (0.99%) developed postablative infections. All infected patients developed postprocedural fever, compared with 29.1% (25/86) in controls. An infection screening score (cutoff ≥ 3) achieved a diagnostic accuracy of 0.910. For predicting infection severity, the preablation model identified metastatic tumor (odds ratio [OR], 4.98; P = .009) and subcapsular location (OR, 7.68; P = .002) as independent predictors, with area under the receiver-operating characteristic curve of 0.774, sensitivity of 0.889, and specificity of 0.500. The combined model included preprocedural monocyte ratio (monocyte count divided by white blood cell count: OR, 0.71; P < .001), subcapsular location (OR, 624.86; P < .001), and postfever procalcitonin (OR, 1.78; P < .001), outperformed the preablation model (area under the receiver-operating characteristic curve, 0.966; sensitivity, 0.889; specificity, 0.912).

Infection severity after liver tumor ablation was effectively predicted. The combined model demonstrated excellent performance and could facilitate early risk stratification for timely intervention.