Predictive Modeling to Identify Stunting Risk in Children

Abstract

Background

Indonesia still experiences a high stunting burden. This has both short- and long-term impacts, including higher morbidity and mortality, impaired future growth, increased chronic disease risk, and reduced productivity later in life.

Objective

This paper aims to assess the main risk factors associated with stunting in Indonesia and to develop a predictive model to identify stunting risk in children.

Methods

Data from the 2018 Indonesian Basic Health Research database were analyzed for children aged under 5 years (n = 13 106) and their mothers. Bivariate analysis was used to select variables significantly associated with stunting risk. A decision tree model was then applied to predict the risk of stunting by age group, and the data were plotted into a receiver operating characteristic (ROC) curve.

Results

The stunting rate reached 25.8%. Based on the decision tree, age, sex, birth weight, birth length, mother's highest level of education, handwashing habits, and exclusive breastfeeding were found to impact stunting risk. The prediction model demonstrated an accuracy of 73.8% for assessing the risk of stunting. The ROC curve showed an area under the curve of 63.7%, with a sensitivity of 60.1% and specificity of 59.8%.

Conclusions

This prediction model is accurate for assessing the risk of stunting. The decision tree-based prediction model performs reasonably well in differentiating between stunted and non-stunted children across different age groups, as indicated by the ROC curve.

Plain Language Summary

Why was the study done?

Stunting is a global malnutrition problem. Based on Indonesian Basic Health Research data in 2018, the prevalence of stunting was at 30.8%, but has reduced somewhat in the Indonesian Health Survey (21.5%) in 2023. It is crucial to identify particular drivers and risks for stunting in each country, and efforts to identify common predictive factors have so far been limited.

What did the researchers do?

The research team studied analyzed all children under 5 years old in Indonesia based on data from Indonesian Basic Health Research in 2018. In this study, 15 variables were analyzed that may have an association with stunting in Indonesia and were used to build predictive models as an effort to prevent stunting risk.

What did the researchers find?

The stunting rate reached 25.8%. Based on the decision tree, age, sex, birth weight, birth length, mother's latest education, hand washing habits, and exclusive breastfeeding had an impact on stunting risk. Furthermore, 73.8% of this prediction model is accurate for assessing the risk of stunting. The receiver operating characteristic (ROC) curve shows an area under the curve (AUC) of 63.7% with a sensitivity of 60.1% and specificity of 59.8%.

What do the findings mean?

This model prediction has an accuracy rate of 73.8% in determining the risk of stunting. The prediction model also performs reasonably well in differentiating between stunted and non-stunted children based on the ROC curve.