의약품 분석에 대한 정도관리에서의 z-스코어에 의한 평가에 대한 대안

Abstract

Proficiency test is an essential tool for ensuring analytical ability of analytical chemists and analytical institutes. Usually, the standard protocol for proficiency test is based on mean values or any other reference values and standard deviation estimated by actual analysis or Horwitz's method. In the cases of data set with a certain amount of outliers, which moves means toward them and.inflate standard deviation, the proficiency test often fails in producing satisfactory results. It is because the existence of outliers results in reduced z-score calculated from these mean and standard deviation and in turn, the process fails in detecting unsatisfactory analytical values. In addition to this, Horwitz's estimation of standard deviation accounts solely for the mass-to-mass concentration level, which is not the only factor determining standard deviation in real-world analysis of pharmaceutical formulations. To deal with this kind of problem, various methods for the calculation of z-scores were evaluated and the results were compared with those obtained from median values and median absolute deviation(MAD) introduced by Huber. Sample data were collected from quality control departments of pharmaceutical companies in Kyungln province in Korea and these data were used to evaluate conventional and new methods for proficiency test. As a result, a new method using median values and MAD gave more reliable decisions in proficiency test as applied to the data studied for this experiment.