Statistical analysis, the scourge of many a scientist, reaches a whole new level when evaluating multitarget clinical laboratory assays. Understanding how to assess performance of these multiplex assays is critical, noted Donna M. Wolk, Ph.D., associate professor, division chair for clinical and molecular microbiology, Arizona Health Sciences Center.
“According to CLIA regulations, verification of new laboratory methods for implementation into a diagnostics laboratory requires a rigorous evaluation of data before tests can be performed on patient samples. When implementing new technology into the laboratory, clinical microbiologists typically use what we call ‘individual analyte’ molecular tests with one or perhaps two genetic microbial target(s). The performance of these tests, compared to a reference standard method, can be done with relative ease. Clinical laboratory directors are generally familiar with the statistics used to characterize and evaluate individual analyte test methods. Now that multi-analyte testing, with detection of 3–25 targets, is a reality, statistical analysis of data presents a new challenge.”
Dr. Wolk says a case in point is multiplex testing for respiratory viruses. “There are several statistical approaches for this multi-analyte testing. One can still analyze the performance of each individual genetic target, which is tedious or one can analyze the entire dataset of results at one time by using statistical analysis like the Chi square test. For either approach, difficulties arise because of the rarity of some virus strains. Data analysis is suspect when laboratories cannot find enough clinical samples to adequately represent all virus categories.”
According to Dr. Wolk, limitations to the breadth, depth, and availability of clinical samples is a challenge. “In some cases limited availability of control material and patient samples leave the laboratorian caught in the space between what is statistically optimal and what is practical in the real world. Inequity exists between large laboratories with abundant samples and smaller laboratories. Microbial biorepositories are limited, and clinical samples are often hard to find. The use of statistics allows us to better define and characterize the limitations in our new method verification and validation.”
Will scientists have to go back to school for years of classical statistical analyses? “Not at all,” Dr. Wolk said. “We need to expand communication between biostatisticians, who have sophisticated training to perform such analyses, and also clinical microbiologists. We can establish guidelines for the statistical practices required for each particular testing application. We need to be able to select and interpret appropriate statistical analyses of clinical datasets for verification and validation of multiplex methods, as well as fine-tune our analysis of qualitative assays, quantitative methods, quality assurance projects, and evidence-based practices.”