Novel methods for efficient surveillance and monitoring

This body of work addresses some of the challenges in surveillance and monitoring by providing means of maximizing information while minimizing time, cost, and human resources. In the first paper, we introduce a new pooling algorithm for testing for a disease, matrix pooling, which accommodates imperfect tests. Matrix pooling belongs to the class of pooling methods that have greater accuracy than individual testing, under reasonable levels of test kit sensitivity and specificity. Additionally, this increase in accuracy is achieved with fewer tests. The second paper applies matrix pooling in the context of testing for acute HIV infection, which is believed by some to be the driving force behind the epidemic. Matrix pooling is more economical than individual testing whilst improving accuracy--reducing the number of false positive and false negative test results. Although matrix pooling may require more tests compared to other pooling algorithms, the significant increase in accuracy and rapidity with which results are obtained makes this method more desirable when identifying new HIV infections, even in resource poor settings. The third paper presents Large Country-Lot Quality Assurance Sampling (LC-LQAS), a method to obtain information at the local level, with sufficient accuracy to aid program managers, while also providing central policy makers with the information they need. This is achieved by aggregating local LQAS data to provide regional or national level estimates via cluster sampling methodology. The method is exemplified with a program evaluation of the HAMSET project in Eritrea. The fourth and final paper addresses the biases inherent in estimators based on a convenience sample, which precludes the legitimate use of powerful inferential tools associated with a random sample. As an alternative to a full random sample to infer the values of an indicator for the whole population, we present an annealing methodology that combines a relatively small, and presumably less expensive, random sample with the convenience sample. Using this additional information allows us to not only take advantage of inferential tools, but also, by combining with information from the convenience sample, provides more accurate information than from just using data from the random sample alone.