Predicting hepatocellular carcinoma recurrences: A data-driven multiclass classification method incorporating latent...

Area of Science:

• Oncology
• Medical Informatics
• Biostatistics

Background:

• Hepatocellular carcinoma (HCC) surgery has high recurrence rates.
• Accurate recurrence prediction is vital for timely interventions and improved patient outcomes.
• Distinguishing early and late HCC recurrence is essential due to differing causes and prognoses.

Purpose of the Study:

• To develop a novel Bayesian network-based method for predicting distinct hepatocellular carcinoma recurrence outcomes (early, late, or no recurrence).
• To address the challenge of insufficient early-stage patient data for accurate recurrence prediction.
• To improve the accuracy and robustness of HCC recurrence prediction models.

Main Methods:

• Proposed a Bayesian network model incorporating a latent variable for 'dominant recurrence type'.
• This latent variable helps overcome information deficiencies from early-stage patient data.
• Evaluated the method using real-world HCC datasets and compared it against benchmark techniques.

Main Results:

• The proposed Bayesian network method significantly outperformed three prevalent benchmark techniques.
• Consistent improvements were observed across accuracy, precision, recall, and F-measures.
• Out-of-sample evaluation on a second dataset confirmed the method's robustness and similar performance.

Conclusions:

• The novel Bayesian network approach effectively predicts distinct hepatocellular carcinoma recurrence patterns.
• This method offers a more robust and accurate tool for predicting HCC recurrence compared to existing methods.
• The findings have significant implications for clinical practice, aiding in timely detection and treatment strategies for HCC patients.