Using Predictive Analytics to Identify Children at High Risk of Defaulting From a Routine Immunization Program:...

Area of Science:

• Public Health
• Health Informatics
• Machine Learning

Background:

• Low- and middle-income countries face challenges in achieving universal immunization coverage due to late vaccinations and dropouts.
• Lack of technology to model risk in large datasets hinders identification of at-risk children, leading to high default rates.
• Predictive analytics, using AI and data mining, can identify children likely to miss immunization visits.

Purpose of the Study:

• To test the feasibility and validate a predictive analytics algorithm for identifying children at risk of defaulting from immunization schedules.
• To assess the algorithm's accuracy in predicting non-adherence to routine vaccination visits.

Main Methods:

• Developed a predictive algorithm using 47,554 longitudinal immunization records (training and validation cohorts).
• Employed four machine learning models: random forest, recursive partitioning, support vector machines (SVMs), and C-forest.
• Evaluated models based on accuracy, precision, sensitivity, specificity, negative predictive value, and area under the curve (AUC), using variables like child's gender, language, residence, and vaccination history.

Main Results:

• The recursive partitioning algorithm achieved the highest predictive performance with an AUC of 0.791.
• All models demonstrated a C-statistic of 0.750 or above.
• The random forest model showed 94.9% sensitivity and 54.9% specificity in the validation dataset.

Conclusions:

• Predictive analytics is a feasible and accurate method for identifying children at high risk of immunization default.
• Identifying potential defaulters allows for targeted, evidence-based interventions in resource-limited settings.
• This approach can significantly contribute to achieving optimal immunization coverage and timeliness.