Abstract

Background: An assessment of varices is required prior to systemic therapy in patients with HCC. However, current non-invasive criteria, including the Baveno criteria, have not been validated in patients with HCC, and performing an EGD can delay HCC treatment initiation. We aimed to develop a noninvasive algorithm for assessing varices in patients with unresectable HCC.

Methods: We performed a multicenter retrospective study from 20 centers in the US, including adult patients with BCLC stage B/C HCC from 2007-2019. We included those with Child Pugh A5-B7 cirrhosis with an EGD within 12 months of index imaging without intervening HCC treatment. We excluded patients with history of variceal bleeding or uncontrolled ascites or hepatic encephalopathy. We collected demographics, laboratory data, and CT/MRI imaging findings extracted by an abdominal radiologist including presence of abdominal varices, spleen diameter/volume, and portal vein diameter. High-risk varices per EGD were defined as large varices, those requiring banding, presence of white nipple, or presence of red wale. We used elastic net for variable selection and model building. We divided the cohort into a 70:30 training set and validation set, with the goal of maximizing negative predictive value to avoid EGD in low-risk patients.

Results: We included 707 patients, with a median age 64.6 years, 80.6% male and 59.8% White, 15.0% Black, 8.2% Asian, and 23.2% Hispanic. The most common liver disease etiologies were hepatitis C (43.6%), alcohol (39.9%), hepatitis B (6.5%), and NASH (4.7%). Patients were evenly distributed between BCLC B (54.0%) and C stage (46.0%) disease. Median time from HCC diagnosis to EGD was 47.4 (IQR: 114) days, with 24.4% of patients having high-risk varices. Our clinical model (Table) achieved an NPV of 87.0% in the validation cohort. Our model including imaging variables (Table) increased NPV to 93.0% in the validation cohort. The model would avoid conducting EGDs in 49 out of every 100 patients without significant varices. In a sensitivity analysis including other high risk bleeding diatheses (gastric varices and portal hypertensive gastropathy), the model had an NPV of 89%.

Conclusion: A model using clinical and imaging data can accurately predict absence of high-risk varices in patients with HCC and avoid EGD in many patients prior to initiation of systemic therapy, thereby expediting care for patients with unresectable HCC.