Forecasting Caseloads of Global Acute Malnutrition in Kenya to Support Anticipatory Action: The Role of Spatial Variation

Rebecca Luttinen , The University of Texas at San Antonio
Molly E. Brown, University of Maryland College Park
Kathryn Grace, University of California Santa Barbara

This study develops various strategies for early warning systems for child malnutrition in Kenya using patterns in climate and conflict, and facility- and community-level data. We use spatial generalized additive models with a spline fit for the spatial component of the model, as well as the climate/conflict predictors, to model Kenya ministry of health caseloads of children aged 6-49 months for global acute malnutrition (GAM) from 2012-2024.We evaluate the associations between GAM caseloads at various temporal lags up to a year prior to a month of caseloads. We also evaluate model performance at different spatial levels, building a country-wide model, and models for specific counties. We find that the country-wide model is heavily driven by the spatial component of the model, with an overall performance of a 37% explanation in the deviance of caseloads. We find that a model for Isiolo and Marsabit counties performs even better at about a 44% explanation in the deviance of caseloads, while other models for specific counties perform even worse than the country-wide model. We use this information to generate facility-level forecasts of GAM caseloads in the year of 2025.

See extended abstract

 Presented in Session P8. Demographic Trends, History, Data and Methods