Individualized prediction models in ADHD: a systematic review and meta-regression

Area of Science:

• Neuroscience
• Psychiatry
• Biostatistics

Background:

• Increasing efforts are directed towards developing prediction models for personalized ADHD detection, prediction, and treatment.
• Current research aims to systematically review and appraise existing ADHD prediction models and assess factors influencing their performance.

Purpose of the Study:

• To systematically review and appraise available prediction models for Attention-Deficit/Hyperactivity Disorder (ADHD).
• To quantitatively assess factors impacting the performance of published ADHD prediction models.

Main Methods:

• A PRISMA/CHARMS/TRIPOD-compliant systematic review was conducted, searching studies reporting validated diagnostic, prognostic, or treatment-response prediction models in ADHD until December 20, 2023.
• Meta-regressions were used to explore factors affecting the area under the curve (AUC) of models, and the Prediction Model Risk of Bias Assessment Tool (PROBAST) assessed study risk of bias.

Main Results:

• Out of 7764 records, 100 prediction models were included, predominantly for diagnosis (88%).
• Only 8% of models had low risk of bias, and 7% were externally validated; none are clinically implemented.
• Models incorporating clinical predictors demonstrated improved performance (AUC increased).

Conclusions:

• While numerous ADHD diagnostic models exist, prediction of outcomes or treatment response is limited, with no models ready for clinical implementation.
• Clinical predictors appear to enhance prediction model performance, suggesting their importance in future model development.
• Future research must focus on high-quality, replicable, and externally validated models, followed by implementation studies to bridge the gap to clinical practice.