is for cancer occurrence; andThe Transformer model, unlike the recurrent models, does not process the input as a sequence of timesteps but, rather, uses an attention mechanism to enhance the embedding vectors correlated with the outcome.
To comprehensively test different types of neural networks and the corresponding hyperparameters, we conducted a large parameter search for each of the network types . The different types of models include simple FF models and MLP) and more complex models that can take the sequential information of disease ordering into consideration . See the supplementary table with comparison metrics across different models .
The relative risk ratio is calculated as the odds of getting pancreatic cancer when classified at high risk compared to a random method that just uses the disease incidence in the population. RR is defined aswhere TP is true positives, FP is false positives, FN is false negatives and TN is true negatives. The RR score is defined at a given operational decision point along the RR curve as a function of the number of patients predicted to be at high risk (Fig.