Statistical Quality Control of HIV-1 ELISA Test Performance
Gentleman, R. C., Hamada, M. S., Matthews, D. E., Wilson, A. R., Journal of the American Statistical Association
A. R. WILSON*
Ensuring the quality and performance of human immunodeficiency virus, type 1 (HIV-l) enzyme-linked immunosorbent assay (ELISA) testing in routine laboratory Situations is an important practical concern. In this article we develop statistical quality control procedures designed to monitor test performance of the microplates being processed. First, the logarithm of measured optical density is shown to be a more appropriate scale on which to work. Shewhart control charts are considered but are not entirely satisfactory because of between-plate variation. Correct classification depends more on within-plate variation than on between-plate variation. Range charts are useful, but they do not directly indicate whether the results from any particular microplate are reliable. Consequently, a new statistical control chart--the separation chart--is proposed, and the assumptions on which it is based are empirically verified. Retrospective analysis of nearly 1,300 microplates, using separation and range charts, demonstrate s that these statistical process control methods may represent a potentially useful supplement to the manufacturer's recommended quality control procedures.
KEY WORDS: Control chart; Log transformation; Range and separation charts; t test.
Millions of enzyme-linked immunosorbent assays (ELISA's) are performed each year to detect antibodies to the human immunodeficiency virus, type 1 (HIV-1). These tests are carried out by agencies involved in collecting blood for transfusion purposes or by diagnostic laboratories, such as those operated by various levels of government in the interest of public health. If the recent suggestion that U.S. hospitals should test all new patients for evidence of infection with HIV-1 is implemented (see Joyce and Brown 1991), the number of HIV-1 ELISA's carried out will increase markedly.
Schwartz, Dans, and Kinosian (1988) have distinguished between test efficacy (i.e., performance under ideal conditions) and test effectiveness (i.e., performance under average conditions) and have highlighted the important practical concern that test effectiveness represents. Although the manufacturer of a serologic test ensures that the testing procedure can detect seropositive individuals with adequate precision under idealized conditions, the statistical quality control (SQC) procedures that we describe are designed to monitor the performance of the testing procedure under day-to-day conditions. This article focuses on the quality control procedures recommended by the manufacturer. Laboratories typically use additional in-house quality control procedures, which are not considered here.
We begin with a brief description of the ELISA test in Section 2, and identify the need for ongoing quality control. We discuss the fundamental statistical issues in HIV-1 testing, and summarize the implications of these considerations with respect to the monitoring of test performance in Section 3. In Section 4, we describe Shewhart control charts and outline problems that arise when these charts are used in a routine way to monitor HIV-1 ELISA test performance. In Section 5 we propose a new control chart, the separation (T) chart, and suggest that the T chart combined with the range (R) chart directly addresses some key issues in the quality control of ELISA test performance. We present concrete evidence of the value of these new statistical tools in Section 6, where we use T and R charts to retrospectively analyze data collected by the Virology Laboratory of the Ontario Ministry of Health. In the final section, we summarize the insights provided by these investigations into statistical tools for quality co ntrol of ELISA test performance.
2. DESCRIPTION OF THE HIV-1 ELISA
A discussion of methods of ensuring satisfactory assay performance on a routine basis requires an understanding of the basic steps involved in obtaining an ELISA result. …