Ultrasound model predicts liver cancer risk

Thursday, November 30 | 2:20 p.m.-2:30 p.m. | R6-SSGI19-6 | Room E352

A model integrating clinical and surveillance ultrasound features for patients at high risk of hepatocellular carcinoma will be presented in this session.

In his talk, Yeun-Yoon Kim, MD, from Yonsei University in South Korea will present research that demonstrates how this model has superior performance compared with previously reported risk scoring systems in this area.

Predicting risk for hepatocellular carcinoma would help with triaging and strategizing best treatment options. Models integrating ultrasound features are being developed to help with such prediction.

Kim and colleagues wanted to develop and validate its model to predict a super-high-risk group of incident hepatocellular carcinoma in patients with chronic hepatitis B or C under ultrasound surveillance.

The team established its model with a developmental dataset of 7,918 patients and an internal validation dataset of 3,393 patients. It used clinical characteristics and ultrasound features to develop the model. These included the presence of cirrhosis, fatty liver, splenomegaly, ascites, and cirrhotic nodules. The group also compared the model’s performance to that of existing risk models.

The researchers found that the five-year cumulative incidence rates of hepatocellular carcinoma were 7.6%, 7.4% and 4.3% in the development, validation, and external test datasets, respectively.

They also found via regression analysis that the following clinical and ultrasound features were independently associated with cancer risk: age, sex, diabetes mellitus, serum albumin and alanine aminotransferase levels, platelet counts, cirrhotic parenchymal echotexture on imaging, and multiple cirrhotic nodules on imaging.

The team also reported that five-year cumulative cancer incidence rates in the validation and external test datasets of 24.1% and 15.5%, respectively.

“Given that the use of alternative imaging modality is limited in [cancer] surveillance, our prediction model would help focus on super-high-risk group for developing [cancer], and potentially individualize surveillance imaging modality,” it wrote.

Find out what else the model showed by attending this session.