Error models and quality control performance

As an alternative to the oversimplified error schemes currently adopted in establishing quality control (QC) strategies, a complex model was assumed implying (a) the distribution of errors (critical error is regarded as a value discriminating between "effective errors" to be detected and "subcritical errors" which do not interfere with the medical decision whose detection is considered as a false-reject signal), and (b) the possibility of simultaneous losses of precision and accuracy. The control data recorded for digoxin radioimmunoassay over a one-year period were used for (1) deriving the probability density functions of random and systematic errors, through a within-run across-level normalisation procedure; (2) obtaining the functional relationships between the critical random or systematic error and the QC performance statistics (sensitivity, specificity, predictive value), weighted for the error prevalences, through integration of the probability density functions and the power functions associated with an exemplifying control rule; and (3) describing the functions which correlate the corrected performance statistics with the allowable error (whose individual values account for all possible combinations of critical random errors and critical systematic errors), by extending to the tridimensional space the above procedures. Analysis of the resulting data shows that it is necessary to revise the criteria for the choice and optimisation of QC schemes.