Machine learning models strong predictors of short-term survival in ACLF patients after LT
Click Here to Manage Email Alerts
Machine learning models, compared with traditional models, demonstrated good performance in the prediction of short-term prognosis following liver transplant in patients with acute-on-chronic liver failure, according to research.
“In previous studies, several scoring systems were applied to forecast the short-term outcome among [acute-on-chronic liver failure (ACLF)] patients,” Min Yang, of the Third Xiangya Hospital of Central South University in China, and colleagues wrote in BMC Gastroenterology. “The predictive value of other scores directed at ACLF, including the Chronic Liver Failure Consortium Organ Failure scores, CLIF sequential organ failure assessment scores and CLIF Consortium ACLF (CLIF-C ACLF) scores, have also been validated in ACLF patients. However, few studies revealed these scores have good predictive value for short-term outcome in ACLF patients following LT.”
In a retrospective study, researchers analyzed 132 ACLF patients who underwent LT to compare the predictive value of traditional models vs. machine learning models for short-term posttransplant survival. They used preoperative clinical variables to calculate five conventional predictive scores as well as four machine learning classifiers (support vector machine, logistic regression, multilayer perceptron and random forest).
Within 90 days after LT, 14.4% of patients had died. Analysis indicated posttransplant mortality associated with higher creatinine values ( 132.6 µmol/L) and international normalized ratio ( 2). Cox regression and multivariate analysis further identified creatinine (P = .001 and HR = 1.006; 95% CI, 1.001-1.011, respectively) and international normalized ratio (P = 0.034; HR = 1.454; 95% CI, 1.1-1.921) as independent prognostic markers of short-term outcomes.
The scores of conventional models were higher in the death group vs. the survival group, but only the model for end-stage liver disease and CLIF-C ACLF scores were significant between groups (P = .01 and P = .004, respectively). In addition, all machine learning models performed well, with random forest achieving the highest performance metric.
“This study successfully established four machine learning models for forecasting the short-term survival of ACLF patients following liver transplant. The machine learning model had better performance than the conventional models and the random forest model best predicted the short-term survival of ACLF patients following liver transplant. Machine learning algorithms could be a useful tool, facilitating better organ allocation and transplant outcomes,” Yang and colleagues concluded. “Future large-scale and multicenter are required to evaluate whether better organ allocation by machine learning algorithms could promote transplant survival.”
Collapse
Read more about
Click Here to Manage Email Alerts