Predicting Success in Percutaneous Transhepatic Biliary Drainage

To develop a model to predict successful bilirubin decrease following percutaneous biliary drain placement.

A total of 257 patients who were identified having undergone percutaneous transhepatic biliary drain placement (PTBD) at our institution between 2002 and 2013 had their medical records and imaging reviewed. Of those, 190 of these patients met criteria and were used in the analysis. A regression model was performed on logarithm-transformed collected variables to predict post-drainage logarithmic transformed total bilirubin levels. A stepwise variable selection method based on Schwarz Bayesian Information Criterion was used to select the most closely associated variables. The model was validated with a Monte Carlo simulation. A short program was developed to calculate the point estimate using the model developed and compared to actual values.

The variables that best predicted bilirubin reduction were initial Tbl (PrTbl), INR and ALT. The selected model had a root mean squared error of 0.8. The model had a negative predictive value (PoTbl is below 2 mg/dL) of 83%.

PTBD may not achieve decreasing bilirubin in patients with a malignant obstruction. This is an initial model that can help determine which patients may not benefit from PTBD placement. With more patients, the model’s validity can be increased and provide useful clinical determinant to aide patient care.