Improving LQAS for monitoring and evaluation of health programs in resource-poor settings
Originally used for industrial quality control, lot quality assurance sampling (LQAS) has become an increasingly popular tool for decentralized monitoring and evaluation (M&E) of health programs in resource-poor settings. In this thesis, we present new methods to improve upon existing practices in LQAS.
A review of LQAS is presented in the first chapter. It is intended both to establish clear operational principles for its use as an evaluation tool and to clarify a number of misconceptions which have become commonplace in the literature.
The second chapter is concerned with two innovations in LQAS motivated by specific applications in health. The first of these involves extending the traditional LQAS tool for two-way classification to three categories in order to handle the problem of intestinal schistosomiasis in schoolchildren. The second discusses the integration of cluster sampling and LQAS to assess the burden of global acute malnutrition (GAM) in children.
The following chapter is devoted to Bayes-LQAS ( B-LQAS) to allow the incorporation of prior beliefs into the classification procedure. Ultimately, we cast the B-LQAS problem in the decision theoretic framework, considering both zero-one and piecewise linear utility functions. We focus on choosing LQASdesigns, consisting of the sample size and decision rule, under a range of assumed prior distributions. Using data from Mandera District, Kenya, we show that strong prior information indicating high levels of GAM prevalence suggests that decision rules should be modified to achieve maximum expected utility. We also discuss multiple classification in the Bayesian framework. Returning to schistosomiasis, we show how to appropriately incorporate prior beliefs when performing three-way classification.
The last chapter is focused on the use of LQAS for routine M&E of health programs. An adaptive Bayesian approach to choosing decision rules over the course of a finite time horizon is presented. Data collected over the course of seven time points in Bara District, Nepal on competency to prepare oral rehydration therapy are considered, and these results indicate that optimal designs are dependent on data trends and are less sensitive to the choice of LQAS design.
0573: Public health