Approval from the local Institutional Review Board was obtained and the requirement for patient consent was waived due to the retrospective nature of this study. Our search of patients' electronic medical records revealed 2,166 patients at risk for HCC (due to hepatitis B virus surface antigen positivity, hepatitis C antibody positivity, or liver cirrhosis of any etiology) who underwent abdominal US for HCC surveillance from January 2008 to December 2013 at a single tertiary medical center. We excluded 1,783 patients who had been previously treated for liver cancer (n=465), who received no tumor diagnosis (n=1,220), or who underwent only a single screening US examination (n=98). Of the remaining 383 treatment-naïve at-risk patients who were diagnosed with liver cancer during surveillance, we excluded 206 patients who were diagnosed with liver cancer based on the initial surveillance US (n=91) or who did not have an interval of 4 to 8 months between the final screening round and the prior screening round (failure of screening, n=115) [10]. We also excluded 22 patients who did not undergo diagnostic computed tomography (CT) or magnetic resonance imaging (MRI) within 6 months after obtaining positive surveillance US (failure of follow-up, n=6), who received a histopathologic diagnosis of a non-HCC malignancy (n=11), or who were in Child-Pugh class C (n=5). Our final study cohort consisted of the remaining 155 patients (Fig. 1). The patients were either diagnosed with liver cancer by positive US surveillance results (US group, n=82) or by CT or MRI performed as an alternative modality after obtaining negative US surveillance results (CT/MRI group, n=73).

At our institution, surveillance for HCC is performed for atrisk patients by biannual abdominal US and serum AFP testing. Surveillance US is performed by hepatology residents or fellows with 1 to 4 years of experience in abdominal US. Our abdominal US reporting form includes (1) quality of the echogenic window (acceptable, poor, or unacceptable); (2) type of liver cirrhosis (macronodular, micronodular, or no gross evidence of cirrhosis); (3) degree of fatty liver (mild, moderate, severe, or no gross evidence of fatty liver); (4) number, size, and location of any space-occupying lesion; (5) presence or absence of splenomegaly; and (6) presence or absence of vascular thrombosis. The assessment of the echogenic window is based on the estimated proportion of scanned area to the entire liver; if some portions or only small portions of the liver are visualized on US, the echogenic window quality is rated as poor or unacceptable, respectively. Cirrhosis is classified as macronodular or micronodular depending on the presence of numerous cirrhosis-associated nodules ≥3 mm or <3 mm, respectively.

If any space-occupying lesion measuring >1 cm was newly detected by US, or the serum AFP level was above 20 ng/mL, the surveillance test was considered positive, and the patient underwent dynamic contrast-enhanced CT or MRI. In some cases, CT or MRI was performed to evaluate patients’ symptoms or as an alternative tool for surveillance under clinical suspicion of liver cancer. The diagnosis of HCC was based on the radiological findings of CT or MRI or on histological evidence, according to the guidelines of the American Association for the Study of Liver Disease [2].

A board-certified radiologist with 8 years of experience in abdominal imaging (C.A.) reviewed the electronic medical records, including the prospectively written radiology reports. To test our hypothesis, the quality of the echogenic window was dichotomized into adequate (acceptable) or inadequate (poor or unacceptable) based on the US reports. In addition, a board-certified radiologist with 4 years of experience in abdominal imaging (Y.Y.K.) retrospectively reviewed the sonograms, blinded to clinical information, in order to classify the patients as with or without parenchymal macronodularity. The presence of numerous cirrhosis-associated nodules ≥3 mm that made the hepatic echotexture markedly heterogeneous was considered indicative of macronodular liver cirrhosis. For patients in the US group, the US report and images from the prior screening round were reviewed; for those in the CT/MRI group, the US report and images from the final screening round were reviewed.

The clinical information that was examined included age, sex, body mass index (BMI), etiology of liver disease, Child-Pugh class, serum AFP level, serologic laboratory data, the presence or absence of diabetes mellitus (DM), final pathological diagnosis, and survival information. Based on radiology reports and clinical data, tumor stage at the time of the initial diagnosis was determined according to the Milan criteria or the Barcelona Clinic Liver Cancer (BCLC) staging system. Surveillance failure was defined as an initial tumor diagnosis beyond the Milan criteria or at BCLC stage B or C despite regular surveillance.

The patient characteristics and tumor stage were compared between the US group and the CT/MRI group using the chi-square test for categorical variables and the Student t test for continuous variables. The chi-square test or the Fisher exact test was used to compare tumor stage according to the quality of the echogenic window and the presence or absence of parenchymal macronodularity in the US group; this analysis was not performed in the CT/MRI group, as tumor stage can be affected by the intervention of CT and MRI scanning and by the quality of surveillance US. Inter-observer agreement in the evaluation of echogenic window quality between the final screening round and the prior screening round was compared using kappa statistics [14]. We used logistic regression to identify clinical factors associated with adequacy of the echogenic window or the presence of parenchymal macronodularity on surveillance US. Continuous variables were dichotomized according to the Child-Pugh classification or institutional reference values (aspartate transaminase >34 IU/L, alanine transaminase >46 IU/ L, and creatinine ≥0.91 mg/dL). Factors with a P-value less than 0.2 in the univariable analysis were included in the subsequent multivariable analysis. Odds ratios (ORs) and 95% confidence intervals (CIs) were calculated. The survival outcomes of the study subjects were analyzed according to whether surveillance failure occurred. Five-year overall survival (5YOS) was examined and compared using the Kaplan-Meier method with the log-rank test. For the survival analysis, patients were censored at the time of death or the last follow-up. All analyses were performed using statistical software (SPSS Statistics version 23.0, IBM Corp., Armonk, NY, USA), and a 2-sided P-value <0.05 was considered to indicate statistical significance.

The characteristics of the patients (113 men and 42 women) are summarized in Table 1. Hepatitis B virus infection was the predominant etiology of liver disease (81.9%), followed by hepatitis C virus infection (11.0%). Most patients (93.5%) were in Child-Pugh class A. The patients in the US group and those in the CT/MRI group did not differ significantly in terms of age, sex, BMI, etiology of liver disease, and Child-Pugh classification. The proportion of patients with DM or an AFP elevation ≥20 ng/mL tended to be higher in the CT/MRI group (50.7% vs. 35.4% for DM [P=0.054] and 50.7% vs. 36.6% for AFP elevation [P=0.077]), but this trend did not reach statistical significances.

With regard to the quality of surveillance US, 16.8% (26 of 155) and 71.0% (110 of 155) of patients had an inadequate echogenic window and parenchymal macronodularity, respectively. Patients in the CT/MRI group had higher proportions of an inadequate echogenic window (Fig. 2) and parenchymal macronodularity (Fig. 3) than those in the US group (19.2% vs. 14.6% for an inadequate echogenic window [P=0.450] and 79.5% vs. 63.4% for parenchymal macronodularity [P=0.028]). The inter-observer agreement for adequacy of the echogenic window was moderate (κ=0.44).

In the entire study population, tumors were initially detected beyond the Milan criteria in 31 patients (20.0%) and at BCLC stage B or C in 52 patients (33.5%). The proportion of surveillance failure tended to be higher in the US group than in the CT/MRI group (23.2% vs. 16.4% by the Milan criteria [P=0.296] and 40.2% vs. 26.0% by the BCLC staging system [P=0.061]). The proportion of diagnoses at BCLC stage 0 tended to be lower in the US group than in the CT/MRI group (20.7% vs. 31.5%, P=0.126).

As summarized in Table 2, surveillance failure was more frequently observed in the US group in patients with an inadequate echogenic window (50.0% vs. 19.4% by the Milan criteria [P=0.046] and 60.0% vs. 37.5% by the BCLC staging system [P=0.191]). The proportion of surveillance failure did not differ between patients with parenchymal macronodularity and those without parenchymal macronodularity (23.1% vs. 23.3% by the Milan criteria [P=0.979] and 38.5% vs. 43.3% by the BCLC staging system [P=0.665]). The frequency of diagnosis at BCLC stage 0 was not significantly different according to adequacy of the echogenic window (10.0% vs. 22.2%, P=0.679) or parenchymal status (19.2% vs. 23.3%, P=0.659).

The results of the univariable and multivariable regression analyses performed to examine the relationship of various clinical factors with an inadequate echogenic window and parenchymal macronodularity are presented in Table 3. In the multivariable analyses, BMI ≥25 kg/m² (OR, 2.536; 95% CI, 0.968 to 6.643; P=0.058) showed a tendency to be associated with an inadequate echogenic window; hepatitis B virus infection (OR, 10.370; 95% CI, 3.129 to 34.270; P<0.001) and a platelet count <150×10⁹/L (OR, 3.167; 95% CI, 1.360 to 7.375; P=0.008) were independently associated with parenchymal macronodularity.

Patients with surveillance failure showed poorer survival outcomes than those without surveillance failure (5YOS, 36.4% vs. 69.9%; P<0.001 by the Milan criteria and 46.8% vs. 71.3%; P=0.001 by the BCLC staging system) (Fig. 4A, B).