Field testing LQAS to assess acute malnutrition prevalence

Summary of published research¹

In emergency settings, the prevalence of global acute malnutrition (GAM) needs to be assessed quickly, accurately and often repeatedly, to determine when and where to start and stop humanitarian aid. The most common approach for assessing GAM is a two-stage 30 x 30 cluster survey - which can be time-consuming and expensive. Alternative, less time-consuming and resource-intensive approaches are needed. One of the most frequently used quality control statistical methods in international health is Lots Quality Assurance Sampling (LQAS). Cumulative binomial probabilities are used in LQAS analyses to detect if a critical threshold has been reached for an indicator. To design an LQA sampling plan, the threshold of interest for an indicator, e.g. GAM prevalence, and tolerable statistical error are defined a priori.

A recently published article examines three adaptations of LQAS to assess GAM and other child-level indicators in food-insecure settings.

Computer simulations confirmed that small clusters instead of a simple random sample could be used for LQAS assessments of GAM. Three LQAS designs were developed (33 x 6, 67 x 3, sequential design)²,³, to assess GAM thresholds of 10, 15 and 20%. The designs were field-tested simultaneously with a 30 x 30 cluster-survey in Siraro, Ethiopia during June 2003. Using a nested study design, anthropometric, morbidity and vaccination data were collected on all children 6-59 months in sampled households. Hypothesis tests about GAM thresholds were conducted for each LQAS design. Point estimates were obtained for the 30 x 30 cluster-survey and the 33 x 6 and 67 x 3 LQAS designs. In order to collect data for the study, 15 interviewers were hired and trained. Teams used an odometer to record the distance travelled from the base camp to the first cluster for each work-day and used a stopwatch to record the time required to complete each survey.

Hypothesis tests showed GAM as < 10% for the 33 x 6 design and GAM as > or = 10% for the 67 x 3 and sequential designs. Point estimates for the 33 x 6 and 67 x 3 designs were similar to those of the 30 x 30 cluster-survey for GAM (6.7%, Confidence Interval (CI) = 3.2-10.2%; 8.2%, CI= 4.3-12.1%, 7.4%, CI - 4.8-9.9%, respectively) and all other indicators. The CIs for the LQAS designs were only slightly wider than the CIs for the 30 x 30 cluster-survey, yet the LQAS designs required substantially less time to administer.

The study concluded that the LQAS designs provide statistically appropriate alternatives to the more timeconsuming 30 x 30 cluster-survey. However, additional field-testing is needed using independent samples rather than a nested study design, which was the most critical limitation of the study. Ideally, data for each design should be sampled independently to allow for stricter comparison of results between designs.

Despite this limitation, the authors concluded that LQAS designs can contribute to the methodological toolkit of humanitarian agencies.

¹Deitchler M et al (2007). A field test of three LQAS designs to assess the prevalence of acute malnutrition. Int J. Epidemiology. Advance access published May 21st, 2007, pp 1-7. The article is available free from http://ije.oxfordjournals.org/cgi/content/abstract/dym092v1

²Thirty-three clusters with six children in each, sixty seven clusters with three children in each, etc.

³A Retrospective Study of Emergency Supplementary Feeding Programmes. Dr Carlos Navarro-Colarado. June 2007. ENN and SC UK. Available at http://www.ennonline.net/research/