BACKGROUND/AIMS: Patients with chronic hepatitis C (CHC) can achieve a sustained virologic response if they received pegylated interferon plus ribavirin therapy; however, some of them do not respond or relapse after treatment. The aim of this study was to compare the ability of two statistical models to predict treatment outcomes. METHODS: Clinical data, biochemical values, and liver histological features of 107 patients with CHC were collected and assessed using a logistic regression (LR) model and an artificial neural network (ANN) model. Both the LR and ANN models were compared by receiver-operating characteristics curves. RESULTS: Aspartate aminotransferase (p = 0.017), prothrombin time (p = 0.002), body mass index (BMI; p = 0.003), and fibrosis score of liver histology (p = 0.002) were found to be significant predictive factors by univariate analysis. The independent significant predicting factor was BMI by multivariate LR analysis (p = 0.0095). The area under receiver-operating characteristics of the ANN model was larger than that of the LR model (85 vs. 58.4%). CONCLUSIONS: It was found that BMI is an independent factor for identifying patients with favorable treatment response. A useful ANN model in predicting outcomes of standard treatment for CHC infection was developed and showed greater accuracy than the LR model. Copyright (c) 2008 S. Karger AG, Basel.
BACKGROUND/AIMS: Patients with chronic hepatitis C (CHC) can achieve a sustained virologic response if they received pegylated interferon plus ribavirin therapy; however, some of them do not respond or relapse after treatment. The aim of this study was to compare the ability of two statistical models to predict treatment outcomes. METHODS: Clinical data, biochemical values, and liver histological features of 107 patients with CHC were collected and assessed using a logistic regression (LR) model and an artificial neural network (ANN) model. Both the LR and ANN models were compared by receiver-operating characteristics curves. RESULTS: Aspartate aminotransferase (p = 0.017), prothrombin time (p = 0.002), body mass index (BMI; p = 0.003), and fibrosis score of liver histology (p = 0.002) were found to be significant predictive factors by univariate analysis. The independent significant predicting factor was BMI by multivariate LR analysis (p = 0.0095). The area under receiver-operating characteristics of the ANN model was larger than that of the LR model (85 vs. 58.4%). CONCLUSIONS: It was found that BMI is an independent factor for identifying patients with favorable treatment response. A useful ANN model in predicting outcomes of standard treatment for CHC infection was developed and showed greater accuracy than the LR model. Copyright (c) 2008 S. Karger AG, Basel.