1997 Revised Guidelines for Performing CD4+ T-Cell Determinations in Persons Infected with Human Immunodeficiency Virus (HIV)

When performing method-validation studies on the new single-platform methods for enumerating CD4+ T-cell populations, laboratorians must consider that these assays may determine the absolute CD4+ count using methodologies that are very different from multi-platform techniques. In most clinical settings, multi-platform methods do not perform at the level of a gold standard. Still, the single-platform methods must be compared with accepted methods or testing procedures. When no optimal standard exists and bias is present, the amount of error contributed by each method cannot be determined. Therefore, if results yielded from a single-platform method are significantly different from those obtained using a multi-platform method, the new method is not necessarily in error. Conducting a large-scale study correlating results from single-platform methods with clinical disease data to establish new medical decision points may be the only surrogate for comparison with a gold standard. Laboratories should not adopt methods that yield results significantly different from multi-platform methods until these studies can be performed, published, and accepted by the scientific and medical communities.

Traditional method comparison tools may be used for validation of single-platform methods that compare favorably with multi-platform methods. Single-platform methods, as the name implies, derive the absolute CD4+ T-cell counts from a single measurement and therefore have the potential to yield a less variable (although not necessarily more accurate) analysis than multi-platform methods, which utilize a combination of hematology and flow cytometry measurements. Laboratorians should utilize statistical tools that provide useful information about these new methodologies but that do not presume that either the comparative or test method is definitive. Linear least squares regression analysis must be conducted based on the assumption that no error exists in the comparative method, and regression-type scatter plots provide inadequate resolution when the errors are small in comparison to the analytical range (70, 71). The bias scatterplot may provide laboratorians with a more useful tool for determining bias (Figure 4, below). These simple, high resolution graphs plot the difference in the individual measurements of each method (X test method - X comparative method) against those by one of the methods (X comparative method) (70). Such graphs provide an easy means of determining if bias is present and distinguishing if bias is systematic, proportional, or random/non-constant. The laboratorian may visually determine the significance of these differences over the entire range of values, and when sufficient values are plotted, outliers and/or samples containing interfering substances can be identified. The laboratorian may then divide the data into ranges relevant to medical decisions and calculate the systematic error (mean of the bias), the random error (standard deviation of the bias), and total error (the greatest absolute 95% error limit of the systematic error twice the random error) to gain insight into analytical performance at the specified decision points (70, 71). Several detailed guidelines and texts can provide laboratorians with additional information regarding quality goals, method evaluation, estimation of bias, and bias scatter plots (70-76). Once a new method is accepted and implemented, the laboratory should continue to monitor the correlation between the results and the patient's clinical disease data to ensure that no problems have gone undetected by the relatively few samples typically tested during method evaluations.

(Click the image to enlarge.)