Table 2: Advantages and disadvantages of ANNs.


(i) Easy model building with less formal statistical knowledge required.(i) Clinical interpretation of model parameters is difficult (black boxes).
(ii) Capable of capturing interactions between predictors.(ii) Sharing an existing ANN model is difficult.
(iii) Capable of capturing nonlinearities between predictors and outcomes.(iii) Prone to overfitting due to the complexity of model structure.
(iv) Users can apply multiple different training algorithms(iv) Confidence intervals of the predicted risks are difficult to obtain.
(v) The model development is empirical. Few guidelines exist to determine the best network structures and training algorithms.