Ghana’s childhood anaemia burden still affects nearly half of preschoolers

New analysis identifies maternal anaemia, stunting, fever, age and local climate as key signals of risk, while highlighting districts where targeted action could have the greatest impact.

A new analysis of childhood anaemia in Ghana suggests the country now has a clearer picture of both who is most at risk and where the burden is heaviest. Drawing on the 2022 Ghana Demographic and Health Survey, researchers at the University of Cape Coast used a combination of multilevel regression and machine learning to examine anaemia in children aged six to 59 months. Their findings point to a problem that remains widespread, persistent and shaped by a mix of child, maternal and local environmental factors. According to the study published in BMC Public Health, the weighted prevalence of childhood anaemia was just under 49%, confirming that the condition still affects roughly half of Ghana’s preschool-aged children.

The scale of that burden matters well beyond blood counts. Childhood anaemia is linked to slower cognitive development, weaker immunity and poorer readiness for school, which is why public health researchers have continued to track it closely across sub-Saharan Africa. A separate study in The Lancet Regional Health found that anaemia among Ghanaian children fell from 76.1% in 2003 to 48.9% in 2022, but it also showed that younger children and those from poorer households remain disproportionately affected. Taken together, the newer studies suggest progress has been real, but uneven and incomplete.

What distinguishes the latest paper is its attempt to combine two analytical approaches rather than rely on one. The authors first used a two-level logistic model to account for the fact that children are clustered within districts, then compared that with three machine-learning methods: Random Forest, XGBoost and Elastic Net. To avoid inflating performance estimates, they split the data by district rather than at random, so that children from the same district did not appear in both the training and testing samples. They then used SHAP values, a method for explaining machine-learning predictions, to identify the factors most strongly associated with anaemia.

Nine variables emerged as high-confidence predictors because they were important in both the regression model and the machine-learning ranking. These included a child’s age and sex, maternal anaemia, maternal age, birth order, stunting, fever, population density and average district temperature. The researchers found that child age and maternal anaemia were among the most consistent signals, while stunting and fever pointed to the role of chronic undernutrition and illness. The appearance of temperature among the strongest predictors also raises the possibility that climate-related conditions may be feeding into malaria risk, food access or disease patterns, although the study design does not allow those pathways to be separated with certainty.

The study also found that some broader district-level variables, including nightlights, rainfall, elevation, vegetation cover, malaria prevalence and travel time, mattered in the machine-learning models even when they were not statistically significant in the multilevel regression. That difference matters because it suggests the problem may involve non-linear relationships or local interactions that conventional models do not easily capture. The authors were cautious on this point, treating the mismatch as an unresolved result rather than forcing a simple explanation. Their analysis also showed that only a modest share of the variation in anaemia was explained by district-level differences, even though geographic clustering was still present.

The geography of risk is equally important. The paper identified several districts with the highest odds of childhood anaemia, including Ho Municipal in the Volta region, Ahanta West in Western region, Nkwanta South in Oti region and Wa West in Upper West region. Those hotspots are spread across different parts of the country, reinforcing the idea that anaemia in Ghana does not follow a single regional pattern. Other recent work using geospatial methods has also found clustering in northern areas and highlighted the influence of maternal anaemia, poverty and limited maternal education, suggesting that local conditions and household disadvantage continue to shape risk.

Although the machine-learning models delivered only moderate predictive accuracy, the findings are still practical. The best-performing model in the new analysis reached an area under the receiver operating characteristic curve of 0.727, which is useful but far from perfect. That modest performance underscores a broader point: childhood anaemia is driven by a layered mix of nutrition, infection, maternal health, poverty and place, and no single dataset can capture every influence. Even so, the study offers policymakers a useful shortlist of priorities, particularly maternal anaemia treatment, action on stunting, fever management and more targeted intervention in high-burden districts.