A new predictive model presented virtually at the European Society for Medical Oncology (ESMO) Congress 2021 was able to detect primary liver cancer (PLC) and distinguish between hepatocellular carcinoma (HCC) and intrahepatic cholangiocarcinoma (ICC).
The new study utilized plasma cell-free DNA (cfDNA) fragmentomic profiles and a machine learning model to differentiate patients with PLC from noncancer controls (overall sensitivity: 96.8% at a specificity of 98.8%). The model yielded an area under the curve (AUC) of 0.995 in the test cohort.
The model was also sensitive at detecting subtypes of PLCs, such as harder to diagnose PLCs like early-stage cancers (stage 1: 95.9%, stage 2: 97.8%) and tumors less than 3 cm (98.2%). High sensitivity was also observed for HCC (96%) and ICC (100%).
Roughly a third of the noncancer controls (104 of 335) were diagnosed with either liver cirrhosis or hepatitis B virus. The model was able to distinguish patients with PLC from these controls with high accuracy, resulting in an AUC of 0.985 (96.8% sensitivity at 96.1% specificity).
Read more about cholangiocarcinoma diagnosis
Using fragmentomic profiles, the machine learning model was also able to distinguish ICC and HCC, yielding an AUC of 0.776.
“Our predictive model has outperformed previously reported methods with solely low coverage [whole-genome sequencing] data, and therefore exhibits excellent potential in clinical practice for cost-effective PLC early detection and more accurate diagnosis of PLC subtypes,” the authors said.
A total of 381 patients with PLC (316 HCC, 52 ICC, 13 mixed) and 335 noncancer controls were recruited for the trial. Of the patients with PLC, 244 had stage 1 cancer, 93 had stage 2, and 44 had stage 3.
Whole-genome sequencing of the plasma cfDNA samples was performed for each patient, with the high-resolution fragment distribution profiles being used to construct a stacked ensemble machine learning model to predict PLCs. The model training cohort included 192 patients with PLC and 170 noncancer controls, while the test cohort included the other 189 PLC patients and 165 controls.
Zhang X, Wang Z, Wang X, et al. Ultra-sensitive and cost-effective method for early stage hepatocellular carcinoma and intrahepatic cholangiocarcinoma detection using plasma cfDNA fragmentomic profiles. Poster presented at: European Society for Medical Oncology (ESMO) Congress 2021; September 16-21, 2021; Virtual.