Standardization of statistical procedures and evaluation scheme in external quality-control survey

摘要

Numerous external quality assessment surveys are being conducted by variety of organizations throughout Japan, but the statistical processing and evaluation scheme are not compatible. Standardization of the procedures is essential to make comparison of results among surveys possible. A coding system is available only for names of analytes and analytical equipments. Systematic coding for analytical principles, manufacturers and standard materials is necessary. Regarding computation of peer-group statistics, the mean and SD are often biased when there are many, or wildly, outlying values. Therefore it is recommended to use an iterative method. The methodology removes a relatively large proportion of the population in the tails of the distribution and re-inflates the SD to compensate for the trimming, thus reaching an unbiased mean and SD by iteration. It is also useful to compute between-method CV and within method-CV by one-way analysis of variance. They represent overall levels of standardization and reproducibility of the analyte, respectively. The evaluation of results is usually based on the peer-group mean and SD. The scheme is unfair for those belonging to a peer-group with a narrow SD. It is recommended to use so-called "common CV evaluation scheme", which is based on a within-method CV computed from overall test results after excluding those peer-groups with large CVs. The common CV is applied to the unbiased peer-group mean to get the evaluation SD. For standardized data processing and statistical analysis, it is crucial to develop a unified, generalized soft ware. We developed its prototype named SurveyMaster I & II and herein introduce their potentials.