Cultural validation of the RCADS and use of ensemble learning for symptom profiling of anxiety and depression.

Introduction: Depression and anxiety are the most prevalent global mental health concerns, especially among children and adolescents. Numerous screening tools are available to readily detect these issues. The cultural significance of these tools in specific communities should be validated, as socio-demographic factors can influence psychopathology. Moreover, screening tools are limited to the identification of a disorder and do not highlight critical symptoms that may be more dominant in disease progression. Methods: In this study, a community sample of 237 Pakistani children and adolescents was used to validate the cultural significance of the Revised Child Anxiety and Depression Scale (RCADS) and its subscales, and develop machine learning (ML) models for profiling of the most significant symptoms of anxiety and depression. Results: Cronbach's alpha for all subscales of RCADS except Separation Anxiety Disorder (SAD) and Obsessive-Compulsive Disorder (OCD) was above 0.7. Chi-square tests between each item of RCADS and the disorders showed that only gender and grade level of patients did not have statistically significant associations with majority of the scales. Lastly, four ML algorithms were trained where Random Forests exhibited the best performance with accuracies ranging from 0.85 to 0.98. The Gini importance calculated for each item in these models highlights the most dominant symptoms contributing to each disorder. Conclusion: Overall, the study shows that all 47 individual items in RCADS are culturally significant for the screening of anxiety and depressive disorders in Pakistani populations, however, the subscales for SAD and OCD warrant some modifications due to low Cronbach's alpha values. The results of ML algorithms yield satisfactory to exceptional metrics, suggesting that these models may be adapted as efficient screening support systems in clinical settings. However, external validation of the models on unseen data is necessary before practical implementation.