JMIR Publications

ABSTRACT

Background:

Child and adolescent psychopathology diagnosis involves a multifaceted approach, which integrates clinical observations, behavioral assessments, medical history, cognitive testing, and familial context information. Digital technologies, including online platforms for administering caregiver-rated questionnaires, are increasingly utilized in this field, especially during the first screening phase. With the rise of digital platforms for data collection, advanced psychopathology classification methods such as supervised machine learning (ML) have gained prominence in both research and clinical settings. This shift, recently referred to as psycho-informatics, has been facilitated by the gradual incorporation of computational devices into clinical workflows. However, an integrated telemedicine and ML approach is yet to be developed.

Objective:

This study aims to develop ML models for the classification of Attention-Deficit/Hyperactivity Disorder (ADHD) and Autism Spectrum Disorder (ASD) diagnoses using online parent-reported socio-anamnestic data, with a primary focus on accuracy over interpretability. The objective is therefore to address the need for an innovative computational psychometrics framework that harnesses the potential of digital platforms and ML to improve the assessment of neurodevelopmental conditions.

Methods:

In this retrospective, single-center observational study, socio-anamnestic data were collected from 1688 children and adolescents referred for suspected neurodevelopmental conditions. The data included information on sociodemographic, clinical, environmental, and developmental factors, collected remotely through the first Italian online screening tool for neurodevelopmental disorders, the Medea Information and Clinical Assessment On-Line (MedicalBIT). Random Forest (RF) and Logistic Regression (LR) models were developed and evaluated using classification accuracy, sensitivity, specificity, and importance of independent variables.

Results:

RF models demonstrated robust accuracy, achieving 84% (95% CI: 82%-85%, P < .001) for ADHD and 86% (95% CI: 84%-87%, P < .001) for ASD classifications, with high sensitivities of 93% for ADHD and of 95% for ASD. In contrast, the LR models exhibited both lower accuracy (61%, 95% CI: 57%-64%, P < .001 for ADHD, and 63%, 5% CI: 60%-67%, P < .001 for ASD) and sensitivities (62% for ADHD and 68% for ASD). Intriguingly, the independent variables considered for classification differed in importance between the two models, reflecting the distinct approaches and capabilities of RF and LR.

Conclusions:

The present study provides evidence for the potential of an ML approach, especially when built on RF models, in enhancing the diagnostic process of child and adolescent psychopathology. While interpretability remains crucial, the developed approach might therefore provide valuable screening tools for clinicians, highlighting the significance of embedding computational techniques in the diagnostic process. This paves the way for the integration of AI-driven tools in clinical practice to pursue the long-term goal of reducing the workload on clinicians.