BACKGROUND: There are few model networks to predict treatment outcome in viral hepatitis. AIM: To develop an easy bioinformatics platform based on algorithm decisions (Bayesian network) for a more efficient prediction of treatment response. METHODS: Totally 385 consecutive chronic hepatitis C (CHC) treated patients were included. More than 40 variables were analysed. Data from 308 patients were used to build the variable model network using DLIFE platform based on predictive graphical models. The prediction accuracy of the bioinformatics network was compared with the true data collected in a retrospective study. The model was then validated twice with external data from CHC patients treated in other hospitals. RESULTS: The accuracy of this bioinformatics network for treatment response in our 308 patients was 83.3%, which is higher than the accuracy obtained by physicians on the basis of study of clinical data and their own experience (50-65%). The receiver operator characteristic curve areas after validation with another cohort of patients were: 0.91 for sustained virological response, one for nonresponse, and 0.81 for relapse. DLIFE offered a diagnostic accuracy of 81.3%, which is a clear improvement compared with unassisted prognosis (50-65%). CONCLUSIONS: This bioinformatics platform (DLIFE) accurately predicts the outcome of CHC combination therapy, improving treatment decisions and reducing costs. This bioinformatics platform allows integrating widespread data sources and permits predicting the clinical outcome of a particular patient using a general predictive graphical model.
BACKGROUND: There are few model networks to predict treatment outcome in viral hepatitis. AIM: To develop an easy bioinformatics platform based on algorithm decisions (Bayesian network) for a more efficient prediction of treatment response. METHODS: Totally 385 consecutive chronic hepatitis C (CHC) treated patients were included. More than 40 variables were analysed. Data from 308 patients were used to build the variable model network using DLIFE platform based on predictive graphical models. The prediction accuracy of the bioinformatics network was compared with the true data collected in a retrospective study. The model was then validated twice with external data from CHCpatients treated in other hospitals. RESULTS: The accuracy of this bioinformatics network for treatment response in our 308 patients was 83.3%, which is higher than the accuracy obtained by physicians on the basis of study of clinical data and their own experience (50-65%). The receiver operator characteristic curve areas after validation with another cohort of patients were: 0.91 for sustained virological response, one for nonresponse, and 0.81 for relapse. DLIFE offered a diagnostic accuracy of 81.3%, which is a clear improvement compared with unassisted prognosis (50-65%). CONCLUSIONS: This bioinformatics platform (DLIFE) accurately predicts the outcome of CHC combination therapy, improving treatment decisions and reducing costs. This bioinformatics platform allows integrating widespread data sources and permits predicting the clinical outcome of a particular patient using a general predictive graphical model.