Medical Decision Making (MDM)
Volume 38, Issue 2, February 2018
Using Cluster Analysis to Group Countries for Cost-effectiveness Analysis: An Application to Sub-Saharan Africa
Louise B. Russell, Gyan Bhanot, Sun-Young Kim, Anushua Sinha
First Published August 19, 2017; pp. 139–149
Objective. To explore the use of cluster analysis to define groups of similar countries for the purpose of evaluating the cost-effectiveness of a public health intervention—maternal immunization—within the constraints of a project budget originally meant for an overall regional analysis.
Methods. We used the most common cluster analysis algorithm, K-means, and the most common measure of distance, Euclidean distance, to group 37 low-income, sub-Saharan African countries on the basis of 24 measures of economic development, general health resources, and past success in public health programs. The groups were tested for robustness and reviewed by regional disease experts.
Results. We explored 2-, 3- and 4-group clustering. Public health performance was consistently important in determining the groups. For the 2-group clustering, for example, infant mortality in Group 1 was 81 per 1,000 live births compared with 51 per 1,000 in Group 2, and 67% of children in Group 1 received DPT immunization compared with 87% in Group 2. The experts preferred four groups to fewer, on the ground that national decision makers would more readily recognize their country among four groups.
Conclusions. Clusters defined by K-means clustering made sense to subject experts and allowed a more detailed evaluation of the cost-effectiveness of maternal immunization within the constraint of the project budget. The method may be useful for other evaluations that, without having the resources to conduct separate analyses for each unit, seek to inform decision makers in numerous countries or subdivisions within countries, such as states or counties.