Risk prediction models for adverse drug reactions in pediatrics: a scoping review

Area of Science:

• Pharmacovigilance
• Clinical Informatics
• Pediatric Drug Safety

Background:

• Adverse drug reactions (ADRs) pose risks in clinical practice.
• Pediatric-specific ADR prediction models are understudied.
• Limited evaluation exists for their methodological quality and reporting.

Purpose of the Study:

• To map existing pediatric ADR risk prediction models.
• To describe their characteristics, methodological quality, and reporting completeness.

Main Methods:

• Systematic search across multiple databases (Embase, PubMed, etc.).
• Evaluation using CHARMS, PROBAST, and TRIPOD guidelines.
• Assessment of predictive ability (AuROC, sensitivity, specificity).

Main Results:

• 7 articles (10 models) met inclusion criteria; logistic regression was common.
• Methodological limits include missing data handling and predictor screening.
• Model discrimination varied (AuROC 0.63-0.97); TRIPOD adherence was incomplete.
• No models underwent external validation.

Conclusions:

• Existing pediatric ADR prediction models have significant methodological and reporting flaws.
• Rigorous development and external validation are crucial for generalizability.
• Improving model quality is essential for enhancing pediatric medication safety.