A simple machine learning-derived rule to promote ERAS pathways in Liver Transplantation

Stefano Skurzak, Alessandro Bonini, Paolo Cerchiara, Cristiana Laici, Andrea De Gasperi,
Manlio Prosperi, Matilde Perego, Elena Augusta Guffanti, Giovanni Chierego,
Gaetano Azan, Roberto Balagna, Antonio Siniscalchi, Gianpaola Monti, Martina Tosi,
Ciro Esposito, Elisabetta Cerutti, Stefano Finazzi, on behalf of the GIVITI group

Journal of Liver Transplantation 12 (2023) 100179


Enhanced recovery after surgery (ERAS) is a fascinating new approach to the perioperative care of liver transplantation (LT). Being an already established pathway in other surgical fields, ERAS in LT (ERALT) is moving its first steps into a complex scenario.

Material and Methods

In this study, using an Italian multicentre database dedicated to LT (Petalo Trapianto Fegato), we compared a group of patients who had a relatively short length of hospital stay (LHoS) after LT (12 days, 569 patients) vs a group that exceeded this LHoS (1017 patients). The main aim was to find a clinical rule to select patients who could afford safely and successfully an ERAS pathway. We used several machine learning techniques to find the best model to predict a short LHoS. We used logistic regression and Boruta random forest to select the most important features to be included in a prognostic score.


According to our results, early after LT, an ERAS pathway might be confidently considered early after LT when the MELDNa is less than 10 or when the MELDNa is between 10 and 17 and the patient received ≤ 5 units of Packed Red Blood Cells intraoperatively (accuracy 72%, sensitivity 78%, specificity 66%, positive predictive value 78%).


This simple clinical rule is intended to be used as a screening tool in patient selection for centres approaching ERAS in LT focusing clinical safety and efficacy, physician confidence and patients’ satisfaction.

Torna in alto