Development and validation of a 90-day mortality prediction model following endobiliary drainage in patients with unresectable malignant biliary obstruction

BACKGROUND: Palliative endobiliary drainage is the mainstay treatment for unresectable malignant biliary obstruction (MBO). Despite optimal drainage, the survival benefit is arguable. This study aimed to identify factors predicting post-endoscopic drainage mortality and develop and validate a mortality prediction model. METHODS: We retrospectively analyzed data for 451 patients with unresectable pancreatobiliary cancers undergoing first endoscopic retrograde cholangiopancreatography (ERCP)-guided endobiliary stent placement between 2007 and 2017. We randomly assigned patients in a 3:1 fashion into a derivation cohort (n=339) and validation cohort (n=112). Predictors for 90-day mortality post-stenting were identified from the derivation cohort. A prediction model was subsequently developed and verified with the validation cohort. RESULTS: The overall 90-day mortality rate of the derivation cohort was 46.9%, and the mean age was 64.2 years. The 2 most common diagnoses were cholangiocarcinoma (53.4%) and pancreatic cancer (35.4%). In all, 34.2% had liver metastasis. The median total bilirubin (TB) level was 19.2 mg/dL, and the mean serum albumin was 3.2 g/dL. A metallic stent was used for 64.6% of the patients, and the median stent patency time was 63 days. A total of 70.8% had TB improvement of more than 50% within 2 weeks after stenting, and 14.5% were eligible for chemotherapy. Intrahepatic obstruction (OR=5.69; P=0.023), stage IV cancer (OR=3.01; P=0.001), pre-endoscopic serum albumin (OR=0.48; P=0.001), TB improvement within 2 weeks after stenting (OR=0.57; P=0.036), and chemotherapy after ERCP (OR=0.11; P<0.001) were associated with 90-day mortality after stenting. The prediction model was developed to identify the risk of death within 90 days post-stent placement. The AUROC was 0.76 and 0.75 in derivation and validation cohorts. Patients with a score ≥ 1.40 had a high likelihood of death, whereas those scoring < -1.50 had a low likelihood of death. Additionally, a score ≥ 0.58 provided a 75.2% probability of death, which highlights the usability of the model. CONCLUSIONS: This study proposes a useful validated prediction model to forecast the 90-day mortality of unresectable MBO patients after stenting. The model permits physicians to stratify the death risk and may be helpful to provide a proper palliative strategy.