Background/aims: In patients with hepatocellular carcinoma, prediction of survival is difficult. The aim of this prospective study was to provide a simple classification for predicting survival of patients with hepatocellular carcinoma, based on a multivariable Cox model.
Methods: Seven hundred and sixty-one patients who presented with hepatocellular carcinoma from 24 Western medical centers were enrolled over a 30-month period. Patients were randomly assigned to either a training sample (n=506, with 418 deaths) from which a classification system was established, or a test sample (n=255, with 200 deaths) for validating its prognostic significance.
Results: Five prognostic factors were selected at the 0.0001 level: Karnofsky index <80% (relative risk of death=2.2, 95% confidence interval: 1.7-2.7), serum bilirubin >50 micromol/l (relative risk=2.1, 95% confidence interval: 1.7-2.6), serum alkaline phosphatase at least twice the upper limit of normal range (relative risk=1.6, 95% confidence interval: 1.3-2.0), serum alpha-fetoprotein >35 microg/l (relative risk=1.7, 95% confidence interval: 1.4-2.1), and ultrasonographic portal obstruction (relative risk=1.3, 95% confidence interval: 1.1-1.7). Three risk groups with different 1-year survival rates (72%, 34%, 7%) were derived, and independently validated in the test sample (79%, 31%, 4%).
Conclusion: This classification could be useful in the assessment of prognosis from homogeneous groups of patients with respect to their expected outcome.