ARTIFICIAL NEURAL NETWORKS APPLIED TO OUTCOME PREDICTION FOR COLORECTAL-CANCER PATIENTS IN SEPARATE INSTITUTIONS

Background Artificial neural networks are computer programs that can b e used to discover complex relations within data sets. They permit the recognition of patterns in complex biological data sets that cannot b e detected with conventional linear statistical analysis. One such com plex problem is the prediction of outcome for individual patients trea ted for colorectal cancer. Predictions of outcome in such patients hav e traditionally been based on population statistics. However, these pr edictions have little meaning for the individual patient. We report th e training of neural networks to predict outcome for individual patien ts from one institution and their predictive performance on data from a different institution in another region. Methods 5-year follow-up da ta from 334 patients treated for colorectal cancer were used to train and validate six neural networks designed for the prediction of death within 9, 12, 15, 18, 21, and 24 months. The previously trained 12-mon th neural network was then applied to 2-year followup data from patien ts from a second institution; outcome was concealed. No further traini ng of the neural network was undertaken. The network's predictions wer e compared with those of two consultant colorectal surgeons supplied w ith the same data. Findings All six neural networks were able to achie ve overall accuracy greater than 80% for the prediction of death for i ndividual patients al institution I within 9, 12, 15, 18, 21, and 24 m onths. The mean sensitivity and specificity were 60% and 88%. When the neural network trained to predict death within 12 months was applied to data from the second institution, overall accuracy of 90% (95% CI 8 4-96) was achieved, compared with the overall accuracy of the colorect al surgeons of 79% (71-87) and 75% (66-84). Interpretation The neural networks were able to predict outcome for individual patients with col orectal cancer much more accurately than the currently available clini copathological methods. Once trained on data from one institution, the neural networks were able to predict outcome for patients from an unr elated institution.