Fitting multilevel models using STEPS data from multiple countries for estimating health outcomes

Oregon Convention Center 

With an increased interest in global health outcomes for policy makers looking to understand their own country's health or those interested in global health, there has been increased speculation on whether the use of aggregated data is adequate for accurate predictions on the country level. An alternative to this is to use individual level survey health data from each country. The current main reason for taking this route is so health outcome predictions are made using complex survey design information which would otherwise be unusable through aggregation.

Our analysis uses a multilevel model with random intercepts for each country and models diabetes prevalence as a function of individual level predictors such as body mass index (BMI), age, gender, and highest completed education level. We also include a country's gross domestic output (GDP) class as a country level predictor. These predictions are then adjusted using complex survey design features to specify inferences of estimated diabetes prevalence to each country's population.

multilevel logit model

diabetes

chronic disease

public health

disease reporting 

Survey Research Methods Section