Predicting the Risk of Postoperative Complications in Patients Undergoing Minimally Invasive Resection of Primary Liver Tumors

Abstract

Minimal-invasive techniques are increasingly applied in clinical practice and have contributed towards improving postoperative outcomes. While comparing favorably with open surgery in terms of safety, the occurrence of severe complications remains a grave concern. To date, no objective predictive system has been established to guide clinicians in estimating complication risks as the relative contribution of general patient health, liver function and surgical parameters remain unclear. Here, we perform a single-center analysis of all consecutive patients undergoing laparoscopic liver resection for primary hepatic malignancies since 2010. Among the 210 patients identified, 32 developed major complications. Several independent predictors were identified through a multivariate analysis, defining a preoperative model: diabetes, history of previous hepatectomy, surgical approach, alanine aminotransferase levels and lesion entity. The addition of operative time and whether conversion was required significantly improved predictions and were thus incorporated into the postoperative model. Both models were able to identify patients with major complications with acceptable performance (area under the receiver-operating characteristic curve (AUC) for a preoperative model = 0.77 vs. postoperative model = 0.80). Internal validation was performed and confirmed the discriminatory ability of the models. An easily accessible online tool was deployed in order to estimate probabilities of severe complication without the need for manual calculation.

Keywords: cholangiocarcinoma; hepatocellular carcinoma; laparoscopic liver surgery; risk score.

Figure 1

Performance of preoperative and postoperative models across imputed datasets Areas under the curves are shown for the preoperative model as well as the postoperative model for the primary endpoint of occurrence of major complications (A,C) and the secondary endpoint of mortality (B,D). Every curve represents one iteration of 20 imputed datasets for missing variables. The mean AUC for each model is depicted.

Figure 2

Comparison of predictive models. Comparative analysis of the AUC of the preoperative (red) and postoperative (blue) models compared to the Southampton score (green) both for the primary endpoint of major complications (A) and the secondary endpoint of mortality (B) reveals the two defined models to have a higher predictive value. AUC computation and model generation are based on the dataset, omitting cases with missing variables.