This article is more than 5 years old. Information may no longer be current.
CART analysis showed superior predictability among liver failure patients vs. traditional KCC models
Click Here to Manage Email Alerts
CHICAGO — King’s College Criteria Classification and Regression Tree analysis increased prediction accuracy and specificity among acetaminophen-induced acute liver failure patients awaiting transplantation, compared with traditional King’s College Criteria models, according to data presented at Digestive Disease Week 2014.
Researchers conducted a retrospective study using a cohort of 803 patients (median age, 37 years; 76% women) with acetaminophen-induced acute liver failure (APAP-ALF) who were on the US ALFSG registry between 1998 and 2013. The Classification and Regression Tree (CART) was used by researchers to predict 21-day spontaneous survival in patients during admission and late-stage (days 3-7). Data collected from CART were used to measure prediction accuracy (AC), sensitivity and specificity. Standard logistic regression methods were used on 679 patients with complete data.
Compared with the traditional KCC model, CART criteria showed improvements in accuracy (82% vs. 70%) in late-stages and had the highest prediction of accuracy (0.86 vs. 0.82). CART used day 3-7 data and showed an increase of AC at 82%, sensitivity at 86% and specificity of 46%. At admission, CART analysis yielded an AC of 66%, sensitivity of 65% and specificity of 67%. KCC yielded an AC of 70%, sensitivity of 97% and specificity 15% for late-stage data. Overall, 588 patients survived after 21 days, while 215 died before day 21.
“The benefits of KCC-CART tree analysis is that compared to logistic regression and other discriminative analyses, the CART model has few requirements about distribution of variables, an effective way of handling observations with missing data, and allows for inclusions of variables with complex interactions,” Constantine J. Karvellas, MD, assistant professor of medicine for hepatology/critical care medicine, University of Alberta, Edmonton, Alberta, said at the conference.
For more information:
Karvellas C. #476. Presented at: Digestive Disease Week 2014; May 3-6; Chicago.
Disclosure: Karvellas and Lee report numerous financial disclosures.