Diabetes medication trials using deep learning techniques

Efficiently selecting participants for diabetes medication trials is crucial but challenging due to the time-consuming and error-prone manual screening of electronic health records (EHR). This study explores deep learning's potential in healthcare, particularly in patient selection, aiming to advance precision medicine and enhance diabetes treatment outcomes. The methodology involved evaluating various neural network architectures, including fully connected neural network (FCNN), recurrent neural networks (RNN), deep belief networks (DBNs), and long short-term memory (LSTM), on handling EHR data. Results demonstrate that while LSTM networks excel in modelling extended dependencies in sequential data, FCNN architecture exhibits superior performance across multiple metrics, including area under the curve (AUC), F1 score, precision, recall, and accuracy metrics. assessment using Kappa score reveals a fair level of agreement for FCNN and LSTM architectures, contrasting with poor agreement levels observed with RNN and DBNs. These findings highlight the potential of deep learning methods, particularly FCNN architectures, in revolutionising patient selection processes for diabetes drug trials and advancing precision medicine initiatives.