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

Clinical Snapshot

Key Findings

• P Value: Not applicable for systematic review

• Effect Size: 9/12 studies achieved >80% accuracy or AUC-ROC

• Primary Outcome: Model performance for predicting immunization non-adherence

• Nnt Or Sensitivity: Sensitivity and specificity varied across studies but not pooled

• Confidence Interval: Not reported - no meta-analysis performed

Clinical Application

Requires substantial validation in Australian healthcare settings before implementation Would need integration with Australian Immunisation Register and validation against NHMRC immunization guidelines. Limited relevance given different socioeconomic and healthcare access patterns compared to study populations Limited applicability to Australian children due to predominance of LMIC data and different healthcare contexts

Abstract

UNLABELLED: Immunization is one of the most effective interventions to prevent infectious diseases. Identifying individuals at risk of non-adherence to immunization schedules could enable early interventions to improve coverage. With the increasing availability of large datasets with sociodemographic, medical, and geographic data, artificial intelligence (AI) offers new possibilities for identifying predictive patterns. We performed a systematic literature review of original studies that used AI to predict immunization non-adherence in children. Searches were conducted in PubMed, Embase, and Web of Science and no restrictions were applied regarding publication year, language, or study design. Study characteristics, dataset size, AI methods, performance metrics, and validation approaches were extracted. Risk of bias was assessed using PROBAST + AI. A total of 843 records were identified; after screening, 12 studies were included. Nine of 12 studies were conducted in low- and middle-income countries, used cross-sectional survey data, and analyzed vaccines from routine immunization schedules. Model performance was generally high; 9/12 studies reported an accuracy or an area under the receiver operating characteristic curve exceeding 80%, but no study conducted external validation. Predictors clustered into seven categories: maternal, paternal, family, child, geographic, wealth, and others. Among maternal and child-related factors, prenatal care and previous immunization history had a high predictive weight. CONCLUSION: AI demonstrates a promising potential for identifying individuals at risk of vaccine non-adherence. However, the lack of external validation raises concerns about model overfitting and limits real-world applicability. Demographic health surveys were routinely repeated across countries and time periods and could therefore represent an ideal resource for external validation. WHAT IS KNOWN: • Geographic, sociodemographic, and medical factors contribute to non-adherence to the immunization schedule.   • Identifying these individuals can guide early interventions to improve immunization coverage. WHAT IS NEW: • Artificial intelligence can reveal unknown influential factors; prenatal care and delivery setting emerged as key factors in this systematic review.

References

1. 1.Rey, R., L'Huillier, A. G., & Pittet, L. F. (2026). Leveraging artificial intelligence to predict non-adherence to pediatric immunization schedules: a systematic review. European Journal of Pediatrics. https://doi.org/10.1080/14760584.2025.2510338