Multivariable analysis of predictors of unplanned hospital readmission after pancreaticoduodenectomy: development of a validated risk score.

BACKGROUND: Unplanned hospital readmission after pancreaticoduodenectomy (PD) is usually due to surgical complications and has significant clinical and economic impact. This study developed a risk score to predict 30-day readmission after PD. METHODS: Patients undergoing PD between 2009 and 2016 were reviewed from a prospective database. Predictors of readmission were identified using a multivariable logistic regression model, from which a points-based risk scoring system was derived. RESULTS: 81 of 518 patients (15.6%) were readmitted within 30 days. History of cardiac disease ([odds ratio] OR = 2.12; 95% CI: 1.12-4.56), CRP>140 mg/L on post-operative day 3 (OR = 2.34; 95% CI: 1.37-4.35) and comprehensive complication index >14 (OR = 1.74; 95% CI: 1.03-2.85) were independent predictors of readmission. The regression coefficients were used to generate a risk score with excellent calibration (p = 0.917) and good discrimination (c-index = 0.65; 95% CI: 0.58-0.71; p < 0.001). Patients were categorised as low, moderate and high risk, with readmission rates of 6.4%, 13.4% and 23.0% respectively (p < 0.001). CONCLUSION: The risk score identifies patients at high risk of readmission after pancreaticoduodenectomy. Such patients may benefit from pre-discharge imaging and/or enhanced follow-up, which may potentially reduce the impact of readmissions.