To develop prognostic models for identifying children with
hepatitis B who are likely to respond to
interferon-alpha (IFN-alpha) or to spontaneously seroconvert, we evaluated results of a multinational controlled trial comprising 70 children with
chronic hepatitis B who received IFN-alpha and 74 children who did not receive
therapy. Prognostic models were developed using SMILES (similarity of least squares), which is a data analysis network that incorporates multidimensional relationships in the clinical data of complex diseases. Commonly collected clinical data included age, gender, serum
aminotransferase (aspartate aminotransferase [AST] and
alanine aminotransferase [ALT]) and hepatitis B virus (HBV)
DNA levels, and IFN-alpha dose. Additional data included pretreatment directional information (e.g. increases or decreases in serum
aminotransferase and HBV
DNA levels), liver biopsy results, race and transmission mode. Using data available prior to initiation of treatment, the SMILES models achieved prospective predictions of 89% for responders, 96% for non-responders, 100% for seroconverters and 93% for non-seroconverters. Although not predictive by themselves, the variables that had the greatest impact on predictions for IFN-alpha response were HBV
DNA pretreatment direction, baseline HBV
DNA, IFN-alpha dose and gender. The variables that had the greatest impact on predictions for spontaneous seroconversion were ALT pretreatment direction, baseline HBV
DNA level, age and AST pretreatment direction. Therefore, these models may be useful in determining, in children with
hepatitis B, the likelihood of response to IFN-alpha and spontaneous seroconversion.