Developing and validating a prediction model of adolescent major depressive disorder in the offspring of depressed parents. (12th September 2022)

• Record Type: Journal Article
• Authors: Stephens, Alice; Allardyce, Judith; Weavers, Bryony; Lennon, Jessica; Jones, Rhys Bevan; Powell, Victoria; Eyre, Olga; Potter, Robert; Price, Valentina Escott; Osborn, David; Thapar, Anita; Collishaw, Stephan; Thapar, Ajay; Heron, Jon; Rice, Frances
• Is Part Of: Journal of child psychology and psychiatry and allied disciplines. Volume 64:Number 3(2023)
• Page Start: 367
• Page End: 375
• Publication Date: 2022-09-12
• Subjects: Risk prediction -- pre‐emptive -- prevention -- depressive disorder -- ALSPAC; Child psychology -- Periodicals; Child psychiatry -- Periodicals; 155.4
• DOI: 10.1111/jcpp.13704
• Languages: English
• ISSNs: 0021-9630

Abstract

Abstract : Background: Parental depression is common and is a major risk factor for depression in adolescents. Early identification of adolescents at elevated risk of developing major depressive disorder (MDD) in this group could improve early access to preventive interventions. Methods: Using longitudinal data from 337 adolescents at high familial risk of depression, we developed a risk prediction model for adolescent MDD. The model was externally validated in an independent cohort of 1, 384 adolescents at high familial risk. We assessed predictors at baseline and MDD at follow‐up (a median of 2–3 years later). We compared the risk prediction model to a simple comparison model based on screening for depressive symptoms. Decision curve analysis was used to identify which model‐predicted risk score thresholds were associated with the greatest clinical benefit. Results: The MDD risk prediction model discriminated between those adolescents who did and did not develop MDD in the development ( C ‐statistic = .783, IQR (interquartile range) = .779, .778) and the validation samples ( C ‐statistic = .722, IQR = −.694, .741). Calibration in the validation sample was good to excellent (calibration intercept = .011, C ‐slope = .851). The MDD risk prediction model was superior to the simple comparison model where discrimination was no better than chance ( C ‐statistic = .544, IQR = .536, .572). Decision curve analysis found that the highest clinical utility was at the lowest risk score thresholds (0.01–0.05). Conclusions: The developed risk prediction model successfully discriminated adolescents who developed MDD from those who did not. In practice, this model could be further developed with user involvement into a tool to target individuals for low‐intensity, selective preventive intervention.