Adequacy of NCEP recommendations for total cholesterol, triglycerides, HDLC, and LDLC measurements.

To the Editor: We read with concern the recent paper by Fallest-Strobl, Olafsdottir, Wiebe, and Westgard (1) in which the authors state, “. . . the NCEP recommendations fail to adequately consider the quality control requirements necessary to detect medically important systematic errors.” We also note that Westgard, Petersen, and Wiebe (2) published a similar paper in 1991 questioning the adequacy of the NCEP recommendations for total cholesterol. We appreciate the importance of the issues addressed in these two papers and find no fault in the majority of the derivations. We do notice, however, that the authors did not address the conditional nature of their calculations. That is, the OPSpecs charts do not take into account the fact that a patient result will only be reported when the quality control (QC) run associated with the patient specimen is considered to be in control. Thus, the OPSpecs charts relate to the conditional probability that a measured patient result will exceed a specified limit, not to the joint probability of patient misclassification. For misclassification to occur, the measured patient result must not only exceed the specified limit, but the QC procedure must also fail to indicate an out-ofcontrol condition. Without taking into account the probability that the QC procedure will fail to detect an out-of-control condition, the authors do not accomplish their stated intention to evaluate the NCEP recommendations with regard to “quality control procedures that are necessary to detect unstable operation” (2). Incorporating QC performance into the OPSpecs charts requires inclusion of the probability that a patient result will be reported on the basis of the QC outcome of the analytic run in which the patient specimen was measured. To illustrate our point, consider the intended 0.95 probability of correct classification chosen by the authors. This conditional probability does not reflect the overall likelihood of correct classification but rather applies only to the rare condition when the QC procedure fails to alert the analyst that the laboratory instrument may have been out of control during the run in which the patient specimen was analyzed. For instance, when evaluating systematic bias, the authors chose systematic errors detectable by QC with a probability equal to 0.90. The joint probability of correct classification in this case is actually 0.995 [1.0 2 (0.05 3 0.10)]. That is, it is the complement of the product of the probabilities of two independent events: (a) The measured result will be beyond the critical decision limit (as defined by the authors, this event occurs with probability 0.05), and (b) the QC sample(s) measured during the run in which the patient specimen was analyzed will be within acceptable limits (on the basis of the example presented by the authors, this event occurs with probability 0.10). Similarly, when evaluating random error, the authors chose increases in random error detectable by QC with a probability equal to 0.80 (or 0.90). The joint probability of correct classification in this case is actually 0.99 (or 0.995). Thus, the conclusions reached by the authors about the adequacy of the NCEP recommendations are actually applicable to a more stringent correct classification requirement of at least 0.99 rather than 0.95 as they claim. Addressing the subtle difference in meaning between the conditional probability inherent in the authors’ computations and the joint probability we recommend is somewhat difficult because the authors do not actually present their results in terms of computed probabilities. Instead, they present their results in terms of the maximum allowable inherent bias and/or analytic random error corresponding to a single fixed conditional probability of correct classification equal to 0.95. The corresponding joint probability associated with their results is actually much higher than 0.95. Both the conditional and joint probabilities are applicable to an individual patient who has a specified intraperson biological variation and whose health status is being determined on the basis of a single randomly collected specimen that is analyzed in a laboratory with a specified accuracy and precision and with a specified QC procedure. The joint probability provides an overall indication of the likelihood of correct patient classification given a specified systematic bias or increase in random analytic error. The conditional probability, on the other hand, provides an indication of the likelihood of correct patient classification only under the rare circumstance that the QC procedure fails to detect the specified systematic bias or increase in random error. In addition, the authors consider only one systematic bias and one increase in random analytic error: the minimum systematic bias that is detectable with probability 0.90 by any one of several QC procedures and the minimum increase in random analytic error that is detectable with probability 0.80 (or 0.90) by any one of several QC procedures. The conditional and joint probabilities could be evaluated using operating-characteristic curves to map performance characteristics as functions of inherent and systematic biases and inherent and increased random analytic errors in the context of various QC procedures and frequencies. The authors indirectly evaluated the conditional probability for one case of systematic error and one case of increased random error via their OPSpecs charts and concluded that the NCEP recommendations for precision and accuracy are inadequate. We have developed operating-characteristic curves for LDLC, HDLC, total cholesterol, and triglycerides to evaluate the NCEP-recommended allowances for inherent bias and inherent analytic random error. These curves show that, for LDLC, HDLC, total cholesterol, and triglycerides, the current NCEP accuracy and precision recommendations are adequate to assure a high likelihood of correct patient classifications. Our analyses assume the use of standard QC procedures (e.g., Shewhart mean-andrange chart QC or Westgard multi-rule