Predicting mental health problems in adolescence using machine learning techniques

Abstract

BackgroundPredicting which children will go on to develop mental health symptoms as adolescents is critical for early intervention and preventing future, severe negative outcomes. Although many aspects of a child's life, personality, and symptoms have been flagged as indicators, there is currently no model created to screen the general population for the risk of developing mental health problems. Additionally, the advent of machine learning techniques represents an exciting way to potentially improve upon the standard prediction modelling technique, logistic regression. Therefore, we aimed to I.) develop a model that can predict mental health problems in mid-adolescence II.) investigate if machine learning techniques (random forest, support vector machines, neural network, and XGBoost) will outperform logistic regression. MethodsIn 7,638 twins from the Child and Adolescent Twin Study in Sweden we used 474 predictors derived from parental report and register data. The outcome, mental health problems, was determined by the Strengths and Difficulties Questionnaire. Model performance was determined by the area under the receiver operating characteristic curve (AUC). ResultsAlthough model performance varied somewhat, the confidence interval overlapped for each model indicating non-significant superiority for the random forest model (AUC = 0.739, 95% CI 0.708-0.769), followed closely by support vector machines (AUC = 0.735, 95% CI 0.707-0.764). PLOS ONEPLOS ONE | https://doi.org/10.1371/journal.pone.

Can the adverse childhood experiences (ACEs) checklist be utilized to predict emergency department visits among children and adolescents?

“…Most of our predictors were categorical variables, and some reports suggest that machine learning might perform better in prediction if there are a large number of continuous variables than binary or factor variables [ 47 ]. Furthermore, it is not surprising that the small number of variables we used were not adequately predictive of the outcome [ 47 , 48 ]. We also used only one type of machine learning approach (i.e., LASSO regression) for predicting our outcome, which assumes a linear relationship between predictors and the outcome.…”

“…Future studies should assess whether the ACE items predict future ED visits by employing other machine learning approaches such as neural networks, gradient boosted trees, and support vector machines, which can take into account the nonlinearities in the relationship between the predictors and the outcome better, if any [ 47 ]. The performance of the models may also improve with a bigger sample size [ 48 ], which in our case can be achieved by analyzing a more extended period of data.…”

“…Cross-validation (in the current study, 10-fold cross-validation) is used to identify the lambda value that produced the best-fitting parsimonious model, minimizing the model deviance [ 33 , 45 , 46 ]. LASSO is believed to have a crucial advantage of consistently identifying the true underlying model and effective handling of multicollinearity between predictors enabling navigation of complex associations in health care data [ 47 , 48 ]. The dataset was split into two randomly selected sub-samples containing 75% and 25% of the total sample and named as “training” and “validation” samples.…”

“…The dataset was split into two randomly selected sub-samples containing 75% and 25% of the total sample and named as “training” and “validation” samples. Splitting allows for a more accurate determination of the model’s performance in a new dataset and reduces overfitting [ 48 ]. The training sample was used to develop and internally validate the prediction models, and the validation sample was used to validate the selected prediction models externally.…”

Machine learning prediction of dropping out of outpatients with alcohol use disorders

“…In recent psychiatric research, machine learning (ML) models have been used to predict psychiatric disorders with high accuracy, which is useful for developing clinical decision support systems and identifying influential variables [21][22][23][24]. In the United States, a study predicted success in treatment of patients with substance use disorders.…”