B-160 Quality control proceduret to assure a defined level of medically relevant quality

Abstract Background A laboratory procedure comprises a measurement procedure plus control procedure that validates the measurement. Numerous quality specifications for the measurement procedure are often set in terms of total allowable error (Tea); ie, that total measurement error (TEm) be less than or equal to Tea. Recommendations are usually based on the ratio of stable random measurement error Smeas) to Tea (sigma score), and implicitely assume that Smeas is stable. A method of evaluating the efficacy of QC procedures is to calculate the undetected lost medical utility (ULMU) function for a control procedure. ULMU functions are calculated from statistical power functions as: ULMU = (lost medical utility due to measurement error) X (1 – statistical power). This is an expected value of lost medical value not detected by QC. ULMU functions always have the shape of an inverted U: ie, at some level of TEm it reaches a maximum value. This maximum value is a way of specifying the medical efficacy of the QC procedure. Methods I have developed a QC simulation programnusing Microsoft Excel and its built-in programming language (VBA). Simulations were run on a Dell XP13 laptop computer. The simulator produces Guassian-distributed random numbers that may be scaled to simulate random measurement error (RE). It introduces analytical bias (SE), with both RE and SE ((and hence TEm) specified as a proportion of Tea. It allows definition of various QC rules, also with rejection parameters defined as a proportion of Tea, generating a 2-dimensions SPF that covers a user-specified range of SE and RE. It calculates an ULMU function based on the probability that TEm exceeds Tea, ie the probability of undetected non-conformity to Tea (UNC):UNC = (Probability of non-confomity) X (1-probabilty of rejection, Prej) A number of rules and procedures may be specified, including 1 point> limit, 2 points>limit, multimean, and root mean square deviation from the target value; rule violation may reject the run or repeat 1 or 2 controls. Rule parameters can be adjusted to achieve a desired maximum UNC. Loosening cotrol limite raises the entire UNC curve, while tightening the limits lowers the curve. At each level of maximum UNC, the simulator will also calculated the expected amount of rejection (Pfr) at stable operatoing conditions of SE and RE. Results The table below indicates the QC rule rejection limit for a maximum UNC of ∼10%, as well as the stable Smeas needed to achieve a Pfr of 1% and 0.1%, based on simulations of 10,000 or 30,000 runs. Conclusion Calculation oof the maximum UNC is a method of setting QC rules and procedures to assure a defined level of medically relevant maximum error, in essence a way of capping the probability of type II error. Use of more complex QC rules can lower the Pfr without sacrificing error detection. Rule parameters can be adjusted to yield any desired undetected non-conformity to error goals, but must be weighe against the cost in false rejection of stable operation.