Submission ID 117097
| Issue/Objective | Despite significant progress in immunization, global coverage remained stagnant in 2023, failing to recover to pre-pandemic levels. Vaccine inequity remains a critical challenge, with 10 countries, , including India, comprising 59% of zero-dose children (receiving no routine vaccine). Focusing on reaching zero-dose children does not guarantee full immunization, many who begin vaccination fail to complete the schedule (referred to as drop-out), leaving them vulnerable to preventable diseases. With an annual birth cohort of 23 million, India ranks second globally in the number of zero-dose children and has a diphtheria-tetanus-pertussis-containing-vaccine(DPT) dropout rate of 7%. National analysis masks sub-national disparities, necessitating a localized approach. We use machine learning and neural-networks to model zero-dose status and vaccination dropout at national and sub-national levels. To understand the variables that put children at risk of under-vaccination in its various forms, our approach identifies common risk factors, with more granular modelling. |
|---|---|
| Methodology/Approach | Our study leverages the National Family Health Survey-5 dataset, 43,436 children in 12-23 months age (standard age to assess basic vaccination). Using socio-demographic and healthcare factors like maternal-literacy, and deprivation, we model three indicators: zero-dose status (no DTP-dose1), BCG-MCV dropout (received Bacillus-Calmette-Guérin-vaccine but not Measles-containing-vaccine), and DPT1-DPT3 dropout (received DTP-dose1 but not DTP-dose3). Class imbalance was a major challenge. Two models were employed: logistic regression and a neural network, maintaining class distribution in training and testing. Performance was evaluated using accuracy, precision, sensitivity, and F1 score. Model comparison identified the most effective approach and feature importance was analysed using model-derived weights. |
| Results | Neural networks outperformed logistic regression nationally and sub-nationally, achieving higher accuracy (0.59-0.67 vs. 0.46-0.59) and F1 scores (0.19-0.20 vs. 0.15-0.19). Neural network effectively captured interactions between vaccination factors. The national model overlooks regional disparities in vaccination trends as barriers vary across states influencing vaccination outcomes differently, emphasising the need for localised modelling. |
| Discussion/Conclusion | Our findings highlight the potential of neural networks and machine learning-driven predictive models to address vaccination inequities in India and similar low-middle-income countries. The granular level analysis highlights regional disparities, with gender and maternal illiteracy impacting vaccination differently across the country but emerging as common drivers for both zero-dose status and dropouts. |
| Presenters and affiliations | Ritika Singh Yardi School of Artificial Intelligence, Indian Institute of Technology Delhi Ritika Singh Yardi School of Artificial Intelligence, Indian Institute of Technology Delhi Mira Johri Université de Montréal Sumeet Agarwal Yardi School of Artificial Intelligence, Indian Institute of Technology Delhi |
