Search

Introduction: Post-Endoscopic Retrograde Cholangiopancreatography (ERCP) pancreatitis (PEP) is ERCP most frequent complication. Predicting PEP risk can be determinant in reducing its incidence and managing patients appropriately, however, studies conducted so far identified single risk factors with standard statistical approaches, with limited accuracy. Aims & Methods: The aim was to build and evaluate performances of machine learning models to predict PEP probability and identify relevant features. The “STARK project” is an international, multicenter, prospective cohort study focused on PEP-associated factors. Data were randomly split in training (80%) and test set (20%). Models used to predict PEP probability were: gradient boosting (GB) and logistic regression (LR). On the training set, a 10-split random cross-validation (CV) was applied to optimize parameters to obtain the best mean Area Under the Receiver Operating Characteristics Curve (AUC of ROC). Afterwards, the model was re-trained on the whole training set with the best parameters and then applied on the test set. Results: 1, 150 patients were included. 70 (6.1%) patients developed PEP. GB model AUC in CV was 0.7±0.076 (95% CI 0.64-0.76), LR model 0.585±0.068 (95% CI 0.53-0.63) (p-value=0.012). AUC in test for GB model was 0.671. Most relevant variables for PEP prediction were: total bilirubin, age, body mass index, procedure time, alcohol units/day, previous sphincterotomy, biliary cannulation attempts and use of Ringer’s solution. Conclusion: This is the first study applying machine learning techniques for PEP prediction, with GB significantly outperforming LR model. Relevant variables were mostly pre-procedural except for procedure time, biliary cannulation attempts and Ringer’s solution use.

Purpose: Post-Endoscopic Retrograde Cholangiopancreatography (ERCP) acute pancreatitis (PEP) is ERCP most frequent complication. Predicting PEP onset risk can be determinant in reducing its incidence. However, studies conducted so far identified single risk factors that have never been used together to predict the risk effectively. The aim of this study was to build a mathematical model to predict PEP probability through machine learning techniques. Materials and methods: “STARK project” is an international, multicenter, prospective cohort study developed within Pancreas 2000 Educational Program, carried out in 7 tertiary centers enrolling patients undergoing ERCP. Patients enrolled were followed-up to detect PEP. The data was randomly split in training set (80%) and test set (20%). Two models were used to predict PEP probability: gradient boosting (GB) and logistic regression (LR). On both models the same data preparation and the same following procedure was applied: on the training set, a 10- split random cross-validation (CV) was applied to optimize parameters in order to obtain the best mean Area Under the Receiver Operating Characteristics Curve (AUC of ROC). Afterwards, the model was re-trained on the whole training set with the best parameters and then applied on the test set. Results: 1, 150 patients were included. 70 (6.1%) patients developed PEP. GB model retrieved a AUC in CV of 0.7±0.076 with 95% CI 0.64-0.76 ; LR model retrieved a AUC in CV of 0.585±0.068 with 95% CI 0.53-0.63. The statistical comparison between the two models in CV retrieved a p value of 0.01. AUC in test for the GB model was 0.671, confirming the stability of the model. Model most relevant variables for the prediction of PEP were: total bilirubin level, age, body mass index, procedure time, units of alcohol/day, previous sphincterotomy, attempts of biliary cannulation, use of Ringer Lactate. Conclusions: This is the first study applying machine learning techniques for the prediction of PEP, with the GB model showing a significantly better performance than the LR model. The most relevant variables we observed were mostly pre- procedural variables except for the procedure time, attempts of biliary cannulation, use of Ringer Lactate.