National probability samples in studies of low-prevalence diseases. Part I: Perspectives and lessons from the HIV cost and services utilization study.

OBJECTIVE: To examine the trade-offs inherent in selecting a sample design for a national study of care for an uncommon disease, and the adaptations, opportunities and costs associated with the choice of national probability sampling in a study of HIV/AIDS.
SETTING: A consortium of public and private funders, research organizations, community advocates, and local providers assembled to design and execute the study.
DESIGN: Data collected by providers or collected for administrative purposes are limited by selectivity and concerns about validity. In studies based on convenience sampling, generalizability is uncertain. Multistage probability sampling through households may not produce sufficient cases of diseases that are not highly prevalent. In such cases, an attractive alternative design is multistage probability sampling through sites of care, in which all persons in the reference population have some chance of random selection through their medical providers, and in which included subjects are selected with known probability. DATA COLLECTION AND PRINCIPAL
FINDINGS: Multistage national probability sampling through providers supplies uniquely valuable information, but will not represent populations not receiving medical care and may not provide sufficient cases in subpopulations of interest. Factors contributing to the substantial cost of such a design include the need to develop a sampling frame, the problems associated with recruitment of providers and subjects through medical providers, the need for buy-in from persons affected by the disease and their medical practitioners, as well as the need for a high participation rate. Broad representation from the national community of scholars with relevant expertise is desirable. Special problems are associated with organization of the research effort, with instrument development, and with data analysis and dissemination in such a consortium.
CONCLUSIONS: Multistage probability sampling through providers can provide unbiased, nationally representative data on persons receiving regular medical care for uncommon diseases and can improve our ability to accurately study care and its outcomes for diseases such as HIV/AIDS. However, substantial costs and special circumstances are associated with the implementation of such efforts.