Machine learning to predict survival in patients treated for hepatocellular carcinoma: a systematic review.

In 30 seconds

This systematic review evaluated machine learning algorithms predicting survival in patients treated for hepatocellular carcinoma (HCC). It included 12 studies with 43 algorithms, revealing that performance varied widely, with area under the receiver operating characteristic curves ranging from 0.66 to 0.92. Algorithms using gradient boosting generally performed better for both short- and long-term predictions.

Key findings

• The area under the receiver operating characteristic curves ranged from 0.66 to 0.92.
• Short-term predictions generally outperformed long-term predictions.
• Gradient boosting algorithms tended to perform better across prediction timelines.
• Performance did not significantly differ based on sample size or number of input variables.

Why it matters

Accurate survival predictions in HCC can enhance clinical decision-making and resource allocation, ultimately improving patient outcomes and reducing unnecessary interventions, such as futile liver transplants.