Introduction to the Study: Predicting Under-Five Mortality in Low Sociodemographic Index States of India
Under-five mortality (U5MR) is a critical global health indicator, reflecting the health status of a nation and its population. It is also a key measure of progress towards Sustainable Development Goal (SDG) 3.2, which aims to reduce U5MR to 25 deaths per 1000 live births or less in all countries by 2030. Globally, significant progress has been made, with the number of deaths among children under five decreasing from 12.8 million in 1990 to 4.9 million in 2022, and the global U5MR reducing by 61% since 1990, from 94 to 37 deaths per 1000 live births by 2023.
Despite this global decline, the burden of child mortality remains disproportionately high in regions like sub-Saharan Africa and Southern Asia, particularly in low-income and lower-middle-income nations, which account for four out of five deaths among children under five, even though only three out of five live births occur there. Southern Asia alone contributes to 26% of the world’s U5MR. In India, under-five mortality has historically been a significant concern, declining from 109 deaths per 1000 live births in 1992–1993 to 41.9 per 1000 live births in 2019–2021. However, in Low Sociodemographic Index (LSDI) states such as Uttar Pradesh, Bihar, Rajasthan, Madhya Pradesh, Chhattisgarh, and Jharkhand, the U5MR remains higher than the national average, at approximately 45 deaths per 1000 live births. These states necessitate targeted interventions to identify and address the underlying factors contributing to child mortality.
While traditional survival analysis methods like Cox regression are widely used and offer interpretability, medical data often involves complex, multidimensional, and non-linear relationships that traditional techniques may not fully capture. To address this, survival machine-learning (ML) techniques have emerged as powerful alternatives, demonstrating superior performance in handling such complex datasets. This study, titled “Survival Machine-Learning Approach for Predicting Under-Five Mortality in Low Sociodemographic Index States of India,” aims to compare various survival analysis techniques, from traditional statistical models to state-of-the-art ML algorithms, in predicting child mortality and determining its influencing factors.
Key aspects of this research include:
- The study utilized National Family Health Survey-5 (NFHS-5) data from 94,202 children in LSDI states of India.
- It was found that 4.5% (4,284) of the studied children died before their fifth birthday.
- The risk of death was significantly higher in children born to younger mothers (15–25 years), uneducated mothers, mothers with a poorer wealth index, and children with low birth weight. Other significant factors included caste, maternal BMI (overweight/obese), male gender, and children not immediately placed on the mother’s chest after birth.
- Among the tested models, the Random Survival Forest (RSF) model demonstrated superior performance in identifying these risk factors and predicting under-five mortality, exhibiting the highest C-index and time-dependent AUC.
- The study underscores the importance of empowering women through education, improving family planning, addressing poverty, and providing equitable healthcare as crucial strategies to reduce child mortality, offering insights for policy shaping and initiatives in vulnerable communities.
This study represents one of the initial attempts to predict under-five child mortality using survival ML models, providing a comprehensive understanding of the determinants of U5MR and showcasing the potential of advanced ML techniques in this critical public health area.
Reference:
Vishwakarma, M., Tyagi, G., & Radhakrishnan, R. V. (2025). Survival Machine-Learning Approach for Predicting Under-Five Mortality in Low Sociodemographic Index States of India. Journal of Research in Health Sciences, 25(3), e00653. https://doi.org/10.34172/jrhs.9033.
