Cucchetti A et al. (2007) Artificial neural network is superior to MELD in predicting mortality of patients with end-stage liver disease. Gut 56: 253–258

The shortage of liver donors, and length of transplantation waiting lists, confer considerable importance on determining the prognosis of patients with chronic liver disease so that allocation of donor organs can be prioritized. The Model for End-stage Liver Disease (MELD) score has been adopted for this purpose, although it frequently fails to predict mortality accurately, possibly because of complex relationships between the biological variables it evaluates.

Cucchetti et al. compared the ability of an artificial neural network (ANN) with that of the MELD score to predict mortality in patients with end-stage liver disease who were on waiting lists for liver transplantation. The ANN was trained to predict 3-month mortality using clinical data from 188 patients, and then tested on data from an internal validation group of 63 patients and an external cohort of 137 patients. The mortality predictions of the ANN were compared with those of the MELD score by looking at the areas under receiver-operating characteristic curves (AUC); the ANN mortality predictions were significantly more accurate than the MELD predictions for both the internal validation group (AUCs 0.95 vs 0.85; P = 0.032) and the external cohort (AUCs 0.96 vs 0.86; P = 0.044).

The ANN was considered to be superior to the MELD score for prioritizing liver transplant candidates, and the authors suggest that it has potential as a reliable tool for reducing mortality among patients awaiting transplantation.