A serum fusion-gene machine-learning model may offer early diagnostic accuracy and could help improve the 5-year survival rate in patients with hepatocellular carcinoma, according to a recent study published by Yu et al in The American Journal of Pathology.
Background
Hepatocellular carcinoma is the most common type of hepatic cancer—accounting for 90% of cases—and is one of the leading causes of cancer-related mortality. Early diagnosis of the disease may be critical to improving patient survival. Yet, the standard screening method currently used to detect the hepatocellular carcinoma biomarker serum alpha-fetoprotein is not always accurate. As a result, up to 60% of hepatic cancer cases are diagnosed at more advanced stages, resulting in a survival rate of about 20%.
“Early diagnosis of [hepatic] cancer helps save lives. However, most [hepatic] cancers occur insidiously and without many symptoms. This makes early diagnosis challenging. What we need is a cost-effective, accurate, and convenient test to screen early-stage [hepatic] cancer in human populations,” stressed senior study author Jian-Hua Luo, MD, PhD, in the Department of Pathology at the High Throughput Genome Center and the Pittsburgh Liver Research Center at the University of Pittsburgh School of Medicine. “We wanted to explore if a machine-learning approach could be used to increase the accuracy of screening for [hepatocellular carcinoma] based on the status of the fusion genes,” he explained.
Study Methods and Results
In the recent study, researchers used real-time quantitative reverse transcription polymerase chain reaction (RT-PCR) to analyze a panel of nine fusion transcripts in the serum samples of 61 patients with hepatocellular carcinoma and 75 patients with non–hepatocellular carcinoma medical conditions—with the goal of determining a more effective and efficient diagnostic tool to predict hepatic cancer cases.
The researchers found that 78% (n = 7/9) of the fusion transcripts were frequently detected in patients with hepatocellular carcinoma. Using a leave-one-out cross-validation approach, they then generated machine-learning models based on the serum fusion-gene levels in order to predict hepatocellular carcinoma.
A four fusion-gene logistic regression model produced an accuracy of 83% to 91% in predicting the occurrence of hepatocellular carcinoma. When combined with serum alpha–fetoprotein, the two–fusion gene plus alpha–fetoprotein logistic regression model produced 95% accuracy among all of the cohorts. Further, quantification of fusion-gene transcripts in the serum samples accurately assessed the impact of the treatment and was able to monitor the patients for cancer recurrence.
Conclusions
The researchers suggested that the novel screening method could help increase the 5-year survival rate of patients with hepatocellular carcinoma from 20% to 90% because of its ability to accurately diagnose early disease and monitor treatment outcomes.
“Early treatment of [hepatic] cancer has a 90% 5-year survival rate, while late treatment has only 20%. The alternative to this test is to subject every individual with some risk of [hepatic] cancer to imaging analysis every 6 months, which is very costly and ineffective. In addition, when imaging results are ambiguous, this test will help to differentiate malignant vs benign lesions,” underscored Dr. Luo. “The fusion-gene machine-learning model significantly improves the early detection rate of [hepatocellular carcinoma] over the serum [alpha–fetoprotein] alone. It may serve as an important tool in screening for [hepatocellular carcinoma] and in monitoring the impact of … treatment,” he concluded.
Disclosure: The research in this study was supported in part by grants from the National Cancer Institute; the Clinical Translational Science Institute at the University of Pittsburgh; and the National Institute of Diabetes, Digestive, and Kidney Diseases. For full disclosures of the study authors, visit ajp.amjpathol.org.
