Cluster Designs to Assess the Prevalence of Acute Malnutrition by LQAS: A Validation by Computer Simulation

Background: The standard sampling design for assessing global acute malnutrition (GAM) is a 30 x 30 (30 clusters of thirty observations) cluster sample. Recently, Deitchler et al demonstrated that Lot Quality Assurance Sampling (LQAS) implemented in a cluster-sampling framework is a viable alternative to the standard methodology for assessing GAM prevalence. Yet, traditional LQAS methods require simple random sampling to guarantee valid results.

Methods: In this study, we simulate correlated binary outcomes to examine the classification error of two LQAS cluster designs, a 67 x 3 (67 clusters of three observations) and a 33 x 6 (33 clusters of six observations). Further, we explore the use of a 67 x 3 sequential sampling scheme for LQAS classification of GAM prevalence based on the Wald probability ratio test.

Results: Results indicate that, for independent clusters with moderate intracluster correlation for the GAM outcome, the three sampling designs maintain approximate validity for LQAS analysis. Moreover, sequential cluster sampling can substantially reduce the average sample size required for accurate classification. The presence of intercluster correlation, however, impacts dramatically the classification error that is associated with LQAS analysis.

Conclusion: These results suggest that LQAS cluster designs could be useful tools for assessing the prevalence of GAM, particularly in emergency situations where time is of critical importance. However, a certain degree of prior knowledge regarding the degree of homogeneity within and between clusters is a necessity when implementing such designs.