Time-varying predictors for clinical surveillance of small hepatocellular carcinoma

PURPOSE: Prognosis of small hepatocellular carcinoma depends on a constellation of time-varying predictors in association with liver function. We aimed to elucidate the impact of these time-dependent predictors on survival. PATIENTS AND METHODS: A total of 108 patients with hepatocellular carcinoma smaller than 5 cm in diameter were recruited. Series of laboratory data and clinical assessments were retrieved from medical records. The time-dependent scoring system for the prediction of death was developed in accordance with a time-dependent Cox regression model. RESULTS: Time trends for biologic predictors parallel cumulative survival of small hepatocellular carcinoma cases. Higher serum α-fetoprotein level was identified as the most significant time-dependent predictor. Other significant predictors included aspartate transaminase, bilirubin, alkaline phosphatase, and albumin levels and prothrombin time. Time-dependent surveillance scoring system shows the cut-off points of scores at 6 months, 1 year, 2 years, and 3 years were 42, 21, 19, and 31, respectively; the estimates of sensitivity, 100%, 100%, 100%, and 87.5%, respectively; and the estimates of specificity, 91.26%, 67.02%, 60.27%, and 78.26%, respectively. Predictive validity for this time-dependent Cox regression model, particularly within 1-year of follow-up, is good. DISCUSSION: The dynamic relationships between time-dependent predictors and risk of death were illustrated. A time-dependent predictive scoring system using these dynamic relationships was developed for real-time surveillance.