Machine learning algorithm for identifying the predictors of childhood immunization in rural Uganda

Abstract

Background: To understand the determinants of childhood immunization in a rural Uganda using a machine learning method of Classification and Regression Tree (CART). Methods and materials: All children aged 12–59 months in the Iganga Mayuge health demographic surveillance site (IMHDSS) database from 2010 to 2017 were considered. We extracted information on childhood immunization status, mothers’ socio-demographic characteristics, pregnancy, birth and other maternal related factors. A non-parametric method of classification and regression tree (CART) can help understand the predictors childhood immunization. Results: There were 5,412 children in this period and one out of every 10 (9.5%) children were not immunized. Place of delivery: the place where the mother gave birth from was the most important splitting variable in explaining the outcome. Only 8.1% of children born in a health facility were not immunized compared to 16.3% of those born outside the health facility. Household poverty levels: 9.6% of children from poor households that were born in the facility were not immunized compared to 4.8% from less poor households. The same pattern was true for unimmunized children born outside the health facility (poor: 17.4% vs less poor: 8.8%). Antenatal care (ANC) attendance: children whose mothers had not attended ANC, were from poor households and were born outside a health facility were more likely to be unimmunized compared to those whose mothers had attended ANC (21.1% vs 16.0%). Living with a partner: children whose mothers were not living with a partner, had attained primary level education, were from poor households and born in a health facility were more likely to be unimmunized compared to those whose mothers were living with a partner (13.4% vs 6.5%). Conclusion: Delivering from a health facility, attending antenatal care, spouse participation and the wealth index are important predictors of childhood immunization. In promoting immunization, there is need to implement interventions that ensure health equity and universal health coverage. Machine learning methods/algorithms create population segments that would ease targeting interventions to improve an outcome of interest.