Leveraging artificial intelligence to predict non-adherence to pediatric immunization schedules: a systematic review

Area of Science:

• Public Health
• Medical Informatics
• Epidemiology

Background:

• Geographic, sociodemographic, and medical factors influence immunization schedule adherence.
• Identifying at-risk individuals enables targeted interventions to improve vaccination coverage.
• Artificial intelligence (AI) offers novel approaches to uncover complex predictive patterns.

Purpose of the Study:

• To systematically review studies utilizing AI for predicting childhood immunization non-adherence.
• To identify key predictors and assess the performance and validation of AI models.

Main Methods:

• Systematic literature review of studies using AI to predict immunization non-adherence in children.
• Searches conducted in PubMed, Embase, and Web of Science without restrictions.
• Risk of bias assessed using PROBAST+AI; study characteristics, AI methods, and performance extracted.

Main Results:

• Twelve studies were included, with most from low- and middle-income countries using routine immunization data.
• AI models generally showed high performance (accuracy/AUC >80% in 9/12 studies).
• Key predictors included maternal factors (prenatal care) and child-related factors (previous immunization history).

Conclusions:

• AI shows significant potential for identifying individuals at risk of vaccine non-adherence.
• Lack of external validation limits current real-world applicability and raises concerns about overfitting.
• Demographic health surveys are a valuable resource for external validation of AI models.