Informed consent was obtained from all patients after the nature of procedures was fully explained. The Institutional Review Board in our hospital approved data collection and analysis for this study. Patients were included in this study if percutaneous RFA was performed in our center from January 2005 to July 2016. When patients had a single nodular HCC less than or equal to 5 cm or multiple nodules up to three in number with each nodule less than or equal to 3 cm and when surgical resection candidates showed clinically significant portal hypertension (i.e., the presence of varices or splenomegaly associated with thrombocytopenia), RFA was recommended as an alternative treatment modality instead of hepatic resection at our institute (12). In addition, RFA was performed based on the preference of patients against strong recommendation for surgery by clinicians. Patients who had been repeatedly treated with RFA at least 3 months from the previous ablation were regarded as new patients in this study. TACE was often performed prior to RFA when liver tumors were less conspicuous on ultrasonography or when the tumors were located in difficult locations for RFA (13).

The diagnosis of HCC was made using noninvasive criteria defined by the American Association for the Study of Liver Disease (AASLD) recommendations consisting of arterial hyper-enhancement with washout on portal- or delayed-phase images on gadoxetic acid-enhanced liver MRI and/or dynamic CT (1214). In patients who did not meet the non-invasive diagnostic criteria, HCC was diagnosed based on biopsy-proven pathological confirmation.

Percutaneous RFA was performed for inpatients under conscious sedation using a combination of intravenous fentanyl citrate (Fentanyl citrate®; Myungmoon Pharm. Co., Hwaseong, Korea) and midazolam (Midazolam; Bukwang Pharm. Co., Ansan, Korea) (15). Pre-ablation prophylactic antibiotics were not used. RFA device was selected and used in each procedure depending on the size and location of the tumor. Multitined expandable electrodes or internally cooled electrodes (single or multiple) were used as appropriate according to tumor size and location. All sonographic procedures were performed with a 3.5-MHz convex-array transducer using a free-hand technique. Percutaneous RFA was mostly performed with real-time sonographic guidance by two experienced radiologists (Y.K.C. and M.Y.K.) with eight years of experience in sonography guided ablation procedures at the start of this study. When sonographic guidance was not technically feasible, CT guidance was used. The ablation procedure was terminated when the size of ablation zone on US monitoring was large enough to achieve at least 5 mm of safety margin (2). Vital signs were monitored during the entire procedure.

TACE procedure was performed by two interventional radiologists (Y.K.C. and Y.S.A. with experience of over seven years and four years at the start of this study, respectively). Hepatic angiography was performed using 5 Fr angiographic catheters followed by superselection of tumor feeders as distal as possible using a microcatheter. An iodized oil-doxorubicin hydrochloride (Adriamycin; Dong-A Pharm., Seoul, Korea) emulsion was then administered into the feeders. The volume of iodized oil ranged from 3 to 10 mL. Once the flow became sluggish, gelatin sponge particles (Cutanplast; Mascia Brunelli, Italy) mixed with contrast material were administered into feeders until blood flow was diminished to minimum level.

CT examinations were performed with 16-, 64-, or 256-slice multidetector CT scanners. MR images were obtained from either a 1.5T or a 3.0T superconducting system using an 8-channel or a 32-channel phased-array coil, respectively. The number of tumors was determined on pretreatment CT or MR imaging. Tumor size was determined as the maximal diameter of the tumor measured on pre-ablation CT or MRI images taken within one month from the ablation procedure.

Major complications were defined as those that might threaten the patient's life, lead to substantial morbidity and disability, or result in lengthened hospital stay. All other complications were considered minor (16). Treatment mortality was defined as any death within 30 days after RFA. Information was extracted from a prospective database for RFA by a radiologist with 15 years of experience in liver imaging. Another senior radiologist with 20 years of experience in liver imaging interpreted CT or MR images with positive or equivocal findings of liver abscess. Final decision was made by consensus.

To determine prognostic factors of liver abscess formation, univariate and multivariate analyses were performed. Parameters that proved to be significant or marginally significant (p-value < 0.1) in univariate analysis were subsequently tested in the multivariate logistic regression model. To adjust for potential bias, propensity score matching was carried out between those treated without a history of multiple prior sessions of TACE in the same segment before RFA (Group 1) and those with such a history (Group 2) using chi-squared test. Propensity score matching was performed for 21 potential factors, including age greater than or equal to 65 years, sex, multinodularity, largest tumor greater than or equal to 3 cm, presence of a centrally located tumor, a dome nodule, a subcapsular tumor, adjacent gastrointestinal tract or adjacent large intrahepatic vessel greater than or equal to 3 mm, use of internally cooled single electrode, use of multiple expandable electrodes, CT guidance, Child-Pugh class, hepatitis C viral infection, alcoholics, coexistence of other malignancy, diabetes mellitus, and hypertension. A caliper of 0.03 was applied for propensity score matching. For matched patients, differences in incidence of liver abscess formation and major complications were compared between the two groups using chi-squared test. Statistically significance was considered at p < 0.05.