Analytic bias specifications based on the analysis of effects on performance of medical guidelines.
Laboratory tests are key indicators for certain practice guidelines, and analytic bias can significantly alter the performance of these guidelines. Three clinical paradigms are described: serum cholesterol testing for risk assessment of cardiac disease, serum thyroid-stimulating hormone (TSH) measurement for the detection of hypothyroidism, and serum prostate-specific antigen (PSA) testing for prostate cancer risk assessment. Maximum tolerance limits for analytic bias are calculated by assessing the subgroup population fluctuations in the number of patients exceeding the guideline threshold values and limiting the analytic bias to one-half of these fluctuations. Our calculated maximum bias limits are +/-1% for cholesterol and +/-6% for TSH and PSA. Our recommended +/-1% bias limit for cholesterol allows for a -6.5% to + 5.8% change in the number of patients designated as at risk for cardiac disease, whereas the +/-3% National Cholesterol Education Program limits permit a -18.4% to +16.7% variation. Similarly, our +/-6% bias limits for TSH allow a -17.7% to +26.6% change in patients flagged for hypothyroidism, whereas the +/-10% bias values found with many commercial reagents permit a -28.2% to +49.2% variation in patient classification. Our +/-6% PSA bias limits correspond to changes from -14.2% to +11.4% in the number of men classified as at risk for prostate cancer. The +/-10% bias ranges for PSA correspond to -19.9% to +20.4% variation in patient classification. The larger tolerance limits of the CLIA-88 standards for proficiency testing would cause even wider variations in patient classifications.