# Predicting risk factors of non-utilisation of postnatal care in three neighbouring East African countries: application of the decision tree

## TL;DR

This study uses decision tree models to identify risk factors for not using postnatal care in three East African countries, finding them more effective than traditional methods.

## Contribution

The study introduces decision tree models as a novel and more accurate method for predicting postnatal care non-utilisation compared to logistic regression.

## Key findings

- Decision tree models outperformed logistic regression in predicting postnatal care non-utilisation.
- Home deliveries, low-quality antenatal care, and unemployment were key risk factors across the three countries.
- High-risk groups include women with unwanted pregnancies and limited access to mass media in Kenya and Tanzania.

## Abstract

Postnatal care (PNC) is recommended for the optimal health of mothers and newborns; however, PNC uptake remains poor in sub-Saharan Africa. Traditional statistical approaches have been used to predict healthcare service utilisation more often but are limited in examining complex relationships compared with decision tree (DT) models.

This study aims to predict the main risk factors of PNC non-utilisation in three neighbouring East African countries using the DT models.

PNC non-utilisation meant that both the mother and neonate did not receive at least one postnatal check within 6 weeks after delivery. Demographic and Health Surveys data from the Democratic Republic of Congo (DRC) (2013/2014), Kenya (2014) and Tanzania (2015/2016) were used. The DT model’s predictive performance was compared with the standard logistic regression (LR) using accuracy, sensitivity, specificity and area under the receiver operating characteristic curve.

DT models exhibited higher accuracy and sensitivity than LR models. In the DRC, women with low-quality antenatal care (ANC), home deliveries and unemployment had the highest probability of PNC non-utilisation (92.0%). In Kenya, women who had home deliveries, unemployment and limited access to mass media showed the highest likelihood of PNC non-utilisation (87.0%). In Tanzania, women with home deliveries, low-quality ANC and unwanted pregnancies exhibited the highest likelihood of PNC non-utilisation (100.0%).

Women with low-quality ANC, home deliveries, unemployment, unwanted pregnancies and limited access to mass media were classified as high-risk groups of PNC non-utilisation. These findings can help prioritise interventions and enhance PNC uptake in East Africa. Additionally, DT models can be applied as valuable tools for predicting maternal and child healthcare services utilisation in other sub-Saharan African countries.