Natural history models for primary sclerosing cholangitis (PSC) have been developed using adult case data but have never been validated in children. It is unknown how accurate such models are for PSC children. The Revised Mayo Clinic, Amsterdam-Oxford, and Boberg models were evaluated using the pediatric PSC collaboration database. Using patient data at the time of PSC diagnosis, we computed the risk strata and projected survival for each patient within each model and compared it to observed survival. In the Mayo, Boberg, and Amsterdam-Oxford models, model fit was good at one year and fair at ten years. Most children were accurately identified as low risk by the Mayo model, whereas most were wrongly labeled as high risk by the Amsterdam-Oxford model. All of the models overestimated the survival of high-risk individuals. Albumin, bilirubin, aspartate aminotransferase (AST), and platelets were the most strongly linked to outcomes. Autoimmune hepatitis was more common in higher risk categories, and AST over-weighting in these individuals explained the actual against expected survival difference.
All three models provided fair long-term discrimination of outcomes while providing good short-term discrimination of outcomes. None of the models explain the high incidence of autoimmune hepatitis characteristics in children, as well as the increased aminotransferases. It is necessary to develop a model tailored to children. AST, bilirubin, albumin, and platelets will be helpful predictors, but they must be weighted to account for the particular characteristics of PSC in children.
