Multivariate logistic regression analysis for prediction of clinically relevant pancreatic fistula in the early phase after pancreaticoduodenectomy

Abstract

Background: Postoperative pancreatic fistula (PF) remains a major complication after pancreaticoduodenectomy (PD). We aimed to investigate the predictors of clinically relevant PF after PD.

Methods: We retrospectively analyzed the predictive factors of relevant grade B/C PF using logistic regression analysis of 100 consecutive patients who underwent PD. PF was defined in accordance with the International Study Group on PF (ISGPF).

Results: White blood cell count (WBC) of 73.6 × 10(2) /μl, C-reactive protein (CRP) of 9.3 mg/dl and amylase value in drains (d-amylase) of 647 U/I on postoperative day (POD) 4 were proposed as the cut-off values for predicting grade B/C PF with high accuracy by the receiver operating characteristic (ROC) curve. Multivariate logistic regression analysis revealed that the three factors as significant predictive factors and the predicted probability of detecting grade B/C PF was calculated by the following formula; P = 1/[1 + exp{-(2.033 × WBC+3.269 × CRP+2.698 × d-amylase-4.122)}]. P > 0.5 indicates the prospective incidence of the PF. When the cut-off values of the three significant predictors were substituted into the formula, P always showed above 0.5 if more than two predictors were above their cut-off values, indicating a high probability of grade B/C PF.

Conclusions: White blood cell count, CRP and d-amylase on POD4 were predictive factors for clinically relevant PF after PD. These findings indicate that our formula is useful for management of drain after PD.

Keywords: Multivariate logistic regression analysis; Pancreaticoduodenectomy; Prediction of pancreatic fistula.