Are seroepidemiological surveys for human immunodeficiency virus infection based on tests on pools of serum specimens accurate and cost-effective?