Prediction of Native Liver Survival in Patients With Biliary Atresia 20 Years After the Kasai Procedure.

OBJECTIVE

To establish risk models for long-term native liver survival (NLS) in patients with biliary atresia.

METHODS

In this retrospective study, we analyzed data from 1792 patients registered in the Japanese Biliary Atresia Registry. Using multivariate logistic regression, we created predictive models for NLS at 1, 5, 10, 15, and 20 years postoperatively. Variables at the first year of age and those at 1 and 5 years postoperatively were used. The variables used in this model were selected using a forward-backward stepwise selection method. The models were further assessed using C-statistics of internal validation with 1000 bootstrapping resamples.

RESULTS

The models using only variables at the first year of age showed insufficient predictive ability. The variables used in the model for NLS at 20 years postoperatively included the age at Kasai procedure, high levels of gamma-glutamyltranspeptidase at 1 year postoperatively, and low levels of cholinesterase at 5 years postoperatively. The models with internally validated C-statistics greater than 0.8 were as follows: NLS at 5 years postoperatively based on the variables at 1 year postoperatively (internally validated C-statistics: 0.88), at 10 years postoperatively based on the variables at 1 year postoperatively (0.86), and at 20 years postoperatively based on the variables at 1 year (0.81) and 5 years (0.83) postoperatively.

CONCLUSION

Overall, we developed satisfactory risk models for NLS up to 20 years postoperatively using variables up to 5 years postoperatively.