(B-238) Risk Prediction Models for Adverse Drug Reactions in Pediatric Patients: A Systematic Review

Background: Predictive models play a critical role in enhancing medication safety in clinical practice. While multiple models for adverse drug reactions (ADRs) in adults have demonstrated potential clinical utility, pediatric-specific predictive models remain understudied, with limited systematic evaluation of their methodological quality and reporting standards.

Objectives: To identify and critically evaluate risk prediction models for ADRs specifically developed or validated for pediatric patients.

Methods: Embase, PubMed, CNKI, Wanfang, VIP, and SinoMed were searched for studies on pediatric ADR prediction models. Included studies were critically appraised using the Checklist for Critical Appraisal and Data Extraction for Systematic Reviews of Prediction Modelling Studies (CHARMS), assessed for risk of bias using the Prediction Model Risk of Bias Assessment Tool (PROBAST), and evaluated for adherence to the Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis (TRIPOD) guidelines.

Results: From 12,667 screened studies, 7 articles (describing 10 models) met the inclusion criteria. The study designs included three case-control studies, one nested case-control study, two prospective cohort studies, and one cross-sectional study. Logistic regression was the primary modelling method, except for one study that employed machine learning techniques. Methodological limitations were common, including unreported methods for handling missing data and univariable predictor screening. Model discrimination, based on the area under the receiver operating characteristic curve (AUROC), ranged from 0.63-0.97, with sensitivity and specificity ranging from 52.0%-98.5% and 33.3%-98.8%, respectively. Adherence to TRIPOD guidelines varied between 62.16%-86.49%. Notably, reporting deficiencies were observed in several areas: insufficient details on blinding for outcome and predictor assessments, lack of explanation for sample size determination, and limited reporting on specific intervention details, model usage instructions, study limitations, and supplementary materials (e.g., study protocols, web calculators, and datasets). None of the models underwent external validation.

Conclusions: Current predictive models for pediatric ADRs are limited, with most tools demonstrating methodological limitations in development, validation, and adherence to reporting standards. Future research should prioritize methodological rigor in model development. Additionally, external validation of existing models is essential to explore their generalizability across diverse settings.