We developed and evaluated neural networks as predictors of outcomes i
n alcoholic patients with severe liver disease using commonly availabl
e clinical and laboratory values, Hospital charts of 144 patients were
reviewed. Nine variables (five laboratory, four clinical) were record
ed along with in-hospital death or survival, Data were organized into
separate development and validation sets. Neural network predictions o
f survival were compared with those of the Maddrey discriminant functi
on and logistic regression models developed on the same data, Model pe
rformance was evaluated by comparing areas under receiver-operating ch
aracteristic (ROC) curves and the distributions of model scores, Survi
vors had significantly different laboratory and clinical characteristi
cs, the most important being a higher prothrombin time, lower bilirubi
n, and lower incidence of encephalopathy, Neural network performance w
as significantly better than that of the Maddrey score (ROC areas, 81.
5% vs, 73.8%; P = .04), The ROC rea for neural networks was similar to
that of logistic regression (ROC area 78.2%; P = .3), but the neural
networks were more successful in classifying patients into low- and hi
gh-risk groups (P < .001), A neural network score with laboratory data
from hospital-day 7 improved prognostic accuracy further to 84.3%, Af
ter adjusting for baseline risk, the neural network change ill illness
severity was still a significant predictor of mortality (P = .001), N
eural networks using clinical and laboratory data showed a high progno
stic accuracy for predicting mortality in alcoholic patients with seve
re Liver disease.