摘要 : "In almost all cases when dealing with a limit of detectionor limit of determination, the primary purpose of deter-mining that limit is to stay away from it." Thus spoke the late William Horwitz (a chemist with the FDA and AOACfor... 展开 "In almost all cases when dealing with a limit of detectionor limit of determination, the primary purpose of deter-mining that limit is to stay away from it." Thus spoke the late William Horwitz (a chemist with the FDA and AOACfor many years) during a panel discussion on detection limits.~1Indeed, any analyst would prefer to report only asurementsthat have low uncertainty associated with them. The obviousnext step, then, is deciding how far away from the detectionlimit (DL) the measurement should be in order for the numberto be "reportable." Such discussions fall into the realm of whatare often called "quantitation limits" or "reporting limits." Inthe opinion of the authors of this column, such limits: 1) areless rigorously calculated and applied than are detection limitsand 2) should be abandoned in favor of reporting every measure-ment (made without extrapolation of the applicable regressioncurve) along with its prediction interval and associated confi-dence level. (See Part 26, American Laboratory, Jun/Jul 2007,for an editorial on detection limits.) Such an approach gives theuser the statistically sound data he or she needs to make a well-informed decision on the acceptability of the noise level in themeasurements. However, other protocols for reporting data aremore widely employed. The rest of this article will discuss theseprocedures and expand on one of them (i.e., the reporting of thenumber of significant digits). 收起