Prediction of Long-term Prognosis of Children with Attention-deficit/Hyperactivity Disorder in Conjunction with Deep Neural Network Regression

Abstract

Background: Although attention-deficit/hyperactivity disorder (ADHD) symptoms decrease with the factors such as age, many individuals keep suffering from the disorder in adolescence and adulthood. Objective: In this paper, a deep learning algorithm was built to forecast the prognosis of ADHD, using the patient's clinical features, the measurement of their response to treatment, and the degree of improvement seen after six years of follow-up. Participants and Settings: The clinical findings such as socio-demographic data of 433 patients followed by the child and adolescent psychiatry department for an average of 6 years with diagnoses of ADHD, and ADHD sub-type, accompanying oppositional/conduct disorders, other psychiatric and organic disorders, the effectiveness of psychotherapy and medication on attention, academic status, and behavioral problems were used to help predict prognosis. Methods: A deep neural network (DNN) learning-based regressor was used to make a prediction model. Results: The results obtained from the DNN regression model achieved accurate predictions for all outputs. The mean error for all outputs was evaluated as mean-squared error (mse) and 0.0068 mean-absolute error (mae), respectively. The R-value was evaluated as 0.91316. It was proven that the model prediction power was adequate as tested with these metrics. Conclusions: This methodology improves the prediction of ADHD prognosis and provides a robust predictive model. Studies with larger samples may support the results.